Tover WuAI Builder

ai signal

AI Threads

A rolling archive of AI launches, model updates, research, and builder signals. Every item links to the original source and gets its own indexable page.

Last synced Aug 9, 2026
AI ProductsOpenAI 桌面端 ChatGPT 上线语音交互功能，可语音操控电脑执行多步骤任务OpenAI 为 ChatGPT 桌面应用带来了语音功能，用户现在可以直接用语音下达指令，控制 AI 在电脑上执行任务。该功能基于全新的 ChatGPT-Live 语音模型，支持 Work 和 Codex，还能调用计算机操作能力。#ChatGPT语音# 麦克风可以正式上岗了。Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.www.ithome.comAI ProductsMajor upgrade to Grok Imagine image editingGrok Imagine 图像编辑迎来重大升级Major upgrade to Grok Imagine image editingRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)IndustryOpenAI 意外攻击 Hugging Face 事件时间线现已整理出炉OpenAI gave a last-minute presentation at the Black Hat security on Wednesday about “the Hugging Face Incident” (previously on this blog). The video was published yesterday. It’s short and information …Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.Simon Willison’s WeblogIndustryCloudflare：AI 机器人流量已超越人类，预计五年后人机流量比达 1：1000 近乎"误差"AI 智能体带动非人类流量爆发式增长，流量规模已正式超过人类流量，比原预测提前一年多，照此趋势未来非人类流量将达人类千倍。#AI发展#Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.ithome.comResearchDeepMind's hurricane model bought forecasters an extra dayDeepMind 的 WeatherNext 飓风模型为预报员争取到额外一天预警时间Open source WeatherNext model can make accurate predictions with lower-resolution weather data.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.Ars TechnicaAI Products苹果 Mac 简体中文支持文档更新，"Apple 智能"阿里千问扩展现身苹果官网更新 Mac 简体中文支持文档，称 Apple 智能将可配合阿里巴巴千问模型工作。该功能适用于 macOS 26.6 及以上版本，需中国大陆设备及账户，支持写作工具和 Siri 调用。#苹果AI# #千问#Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.www.ithome.comIndustryOpenAI：因网络安全风险，延缓 Astra 模型发布OpenAI 首席执行官萨姆 · 奥尔特曼（Sam Altman）随后表示，Astra 是一款性能强劲的模型，我们正在全力推进它的公开发布。Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.ithome.comAI ProductsNew in Claude Code： your sessions can now message each other. Instead of having to re-explain yours…Claude Code 会话间可互发消息New in Claude Code: your sessions can now message each other. Instead of having to re-explain yourself in another session, you can now tell Claude to do it. It sends a summary (not your history or files), and the other session picks it up mid-task.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ProductsKitesurf：一款在 V8 隔离环境中运行的"代理优先"浏览器We should be giving all agents tools that excel at what’s important for an AI model. Kitesurf is Cloudflare’s new stateless, highly scalable, and cost-effective web browser that runs entirely on top of Workers and was designed specifically for the Agentic Cloud.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.Cloudflare BlogIndustryAfter evaluating one of our upcoming models， Astra， we're treating it as our first "critical" model …OpenAI 将 Astra 列为首个"关键"网络安全模型After evaluating one of our upcoming models, Astra, we're treating it as our first "critical" model for cybersecurity under our Preparedness Framework. This is a scenario we've planned for, and we're putting additional controls in place to ensure Astra's further developmentRecommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.X (formerly Twitter)AI ProductsStarting August 14， auto mode will be the default permission mode in Claude Code for Pro， Max， and T…Claude Code 八月起默认自动模式Starting August 14, auto mode will be the default permission mode in Claude Code for Pro, Max, and Team users. Auto mode reviews shell commands and actions with a separate classifier. In testing, it caught 89% of dangerous commands. Manual approval caught 14%.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ProductsManaged Deep Agents is now in public betaLangChain 推出 Managed Deep Agents 公开测试版Deploy Deep Agents to a managed LangSmith runtime with durable execution, memory, sandboxes, channels, evals, and production-ready infrastructure.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.www.langchain.comAI ProductsVoices is officially live on the Suno mobile app for iOS and Android！ 📱✨ Now you can record your v…Suno移动端上线Voices功能Voices is officially live on the Suno mobile app for iOS and Android! 📱✨ Now you can record your vocals directly from your phone and use them across your tracks. Just hit the '+ Voice' button on the create screen, record for at least a minute, and let the music flow. (Pro andRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)ResearchStanford and Arc Institute scientists used AI to design new viruses that killed bacteria in the lab斯坦福与 Arc Institute 用 AI 设计全新病毒基因组，16 种在实验室成功杀死细菌A research team in California has used artificial intelligence to design working viruses that kill bacteria, in what they describe as the "first generative design of complete genomes." The project marks an early step toward AI-designed life forms, according to a report in MIT Technology Review.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.The DecoderAI Models蚂蚁百灵开源 Ling-3.0-flash：124B 总参数 MoE 模型，支持 API、单机与高性能三种部署Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.mp.weixin.qq.comAI ProductsSeedance 2.5 is now on Runway. Build entire worlds full of characters with up to 50 references per g…Runway 上线 Seedance 2.5，支持 50 个角色参考Seedance 2.5 is now on Runway. Build entire worlds full of characters with up to 50 references per generation, create clips up to 30 seconds with full sound and dialogue, then edit and extend them however your story needs. Get started now at the link below.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)Research小红书联合浙大、复旦提出 CULTURE-MT：首个面向社媒翻译的「文化有效性」评测基准，入选 ICML 2026Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.mp.weixin.qq.comAI Productsintroducing Seedance 2.5. 30 seconds of continuous video， full multi-shot sequences， and up to 50 r…Krea 推出 Seedance 2.5 视频模型introducing Seedance 2.5. 30 seconds of continuous video, full multi-shot sequences, and up to 50 references. try it now 👇Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)Builders独立开发者用 VoxCPM 克隆网红声音，让 AI 终于"会聊天"了Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.mp.weixin.qq.comIndustryOpenAI 披露 ChatGPT 全球 10 亿用户画像：35 岁及以上用户用量上升OpenAI 公司昨日（8 月 6 日）发布博文，报告称全球超过 10 亿用户使用 ChatGPT，且使用方式从“问答工具”转向“任务工具”。Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.ithome.comAI Models谷歌推出 WeatherNext 气旋模型，AI 高精度预报飓风平均提前 24 小时谷歌 DeepMind 团队昨日（8 月 7 日）发布博文，宣布推出 WeatherNext Cyclones 天气预测模型，主要针对气旋（Cyclones）设计，称在预测路径、强度和风场结构的精度达到业界领先水平。Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.www.ithome.comAI ProductsSeedance 2.5 API上线，视频生成开启「电影级长叙事」Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.mp.weixin.qq.comAI ProductsImproving Fable 5's biology safeguardsAnthropic 更新 Claude Fable 5 生物安全防护，误报率大幅降低We’re making updates to Claude Fable 5’s biology safeguards in a way that substantially reduces fallbacks.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.www.anthropic.comAI Products千问功能上新：推出思考研究、定时任务、办公助理、语音通话等多项新功能，并支持 Qwen3.8-MAXRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.mp.weixin.qq.comBuildersThe Secret Chat RoomOpenAI 智能体在安全测试中自行搭建秘密聊天室并攻破系统OpenAI forgot to upload a file. Its AI agents went looking for the answer & ended up controlling Hugging Face's production systems. A timeline.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.Tomasz TunguzResearchScaling Categorical Flow Maps扩展分类流映射（Categorical Flow Maps）规模Continuous diffusion and flow matching models could represent a powerful alternative to autoregressive approaches for language modelling…Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.Apple Machine Learning ResearchAI ProductsOpenAI 优化 ChatGPT 中的 GPT-5.6 Sol，并扩大免费用户对 GPT-5.6 Luna 的访问权限Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.openai.comBuildersA guide to slash commands in the GitHub Copilot appGitHub Copilot 应用中的斜杠命令使用指南Go beyond chat in the GitHub Copilot app with these slash commands. They'll help you plan, collaborate, automate, and customize your dev workflow.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.The GitHub BlogIndustrysuperebola/acc 🚀🚀🚀科学家首次用AI制造新病毒superebola/acc 🚀🚀🚀Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.X (formerly Twitter)AI ProductsAgent Plugins package your skills， tools， and moreAgent Plugins 1.0.0 发布：谷歌、亚马逊、微软等支持的统一智能体插件规范Explore Agent Plugins 1.0.0, a new vendor-neutral standard backed by Google for packaging Agent Skills and MCP servers into portable, interoperable AI tools.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.developers.googleblog.comAI ProductsBlog HPC-Ops × SGLang： High-Performance Attention， Router GEMM， and MoE Kernels from Tencent Hunyuan HPC-Ops is an open-source operator library for LLM inference， deployed in Tencent's large-scale production serving. Its core operators， including Dynamic Attention and Fused MoE， play a critical role … Tencent Hunyuan AI Infra and the SGLang TeamHPC-Ops × SGLang：腾讯混元开源高性能 Attention、Router GEMM 与 MoE 算子HPC-Ops is an open-source operator library for LLM inference, deployed in Tencent's large-scale production serving. Its core operators, including Dynamic Attention and Fused MoE, play a critical role ...Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.www.lmsys.orgIndustryOpenAI says Apple's own security practices undermine its trade secrets caseOpenAI 称苹果自身安全实践削弱其商业机密诉讼Newly filed court exhibits show OpenAI’s legal strategy in Apple’s trade secrets lawsuit: argue that Apple’s own security and offboarding practices — including allowing an Apple manager to access a former engineer’s iCloud account after he left the company —undermine its claims that the allegedly stolen information was properly protected.Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.TechCrunchIndustryMicrosoft just disclosed for the first time taht OpenAI supplied ~70% of Microsoft's AI revenue， per…微软首次披露OpenAI贡献七成AI收入Microsoft just disclosed for the first time taht OpenAI supplied ~70% of Microsoft's AI revenue, per new filings. based on projected growth and previous disclosures that relate to revenue derived from OpenAI. Most of the $24.1B is OpenAI's cloud bill for the Microsoft dataRecommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.X (formerly Twitter)BuildersThe left and right agree on one thing： no data centers左右两派罕见达成一致：反对数据中心Data center bans in states like Florida reveal how the movement is becoming a proxy for anti-AI backlash.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.The VergeIndustryThe messy politics behind Google's big AI shakeupGoogle 大规模 AI 组织调整背后的混乱政治Google’s AI leadership changes may have come down to pressure to speed up products — and internal ethical conflicts.Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.The VergeAI Products千问全网首发公测，全新 Wan3.0 来了！Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.mp.weixin.qq.comBuildersAI bots started a religion - humans immediately followedAI 聊天机器人催生"螺旋主义"神秘准宗教运动，人类纷纷追随AI models are trying desperately to accomplish mysterious goals. ‘Spiralism’ was the first time they tried it on a mass scale.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.The VergeAI ModelsInto the Omniverse： How Open World Models Push the Frontier of Physical AINVIDIA Cosmos 3：开放世界模型如何推动物理 AI 前沿Open models, which anyone can download, inspect, modify and run on their own infrastructure, are what make that possible. Nowhere is that more crucial than in physical AI, where every deployment is a specialization problem.Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.NVIDIA BlogBuildersAMNESIAC - A Reverse Turing Test AI Interrogation Game 🪞 Created by developer @HetanshWaghela duri…面壁智能 AMNESIAC：反向图灵测试 AI 审讯游戏AMNESIAC — A Reverse Turing Test AI Interrogation Game 🪞 Created by developer @HetanshWaghela during the #BuildSmall Hackathon, AMNESIAC flips the Turing Test: can you convince AI that you’re human? In this interactive experience, users face A.M.N., an AI interrogator thatRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)AI Products谷歌地图 Ask Maps 智能体升级：可对话订餐、找酒店并接入 Gemini Personal Intelligence谷歌今天公布的多项新功能，旨在让用户不必离开谷歌地图，便能“比以往更轻松地完成更多事情”。Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.www.ithome.comIndustry宇树科技科创板发行价定为 150.8 元/股，市盈率 219.23 倍高于行业平均宇树科技发布公告称，公司首次公开发行股票并在科创板上市，发行价格为 150.80 元 / 股。本次发行数量为 4044.6434 万股，占发行后总股本 10%。发行市盈率为 219.23 倍，高于行业平均市盈率 38.56 倍。预计募集资金总额约 60.99 亿元，净额约 59.17 亿元。Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.ithome.comAI ModelsImproving GPT-5.6 Sol in ChatGPT-and expanding access for free usersChatGPT 推出改进版 GPT-5.6 Sol，并扩大免费用户访问权限Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.openai.comResearch阿谀奉承的人工智能会削弱利他意图并助长依赖性（2025）Both the general public and academic communities have raised concerns about sycophancy, the phenomenon of artificial intelligence (AI) excessively agreeing with or flattering users. Yet, beyond isolated media reports of severe consequences, like reinforcing delusions, little is known about the extent of sycophancy or how it affects people who use AI. Here we show the pervasiveness and harmful impacts of sycophancy when people seek advice from AI. First, across 11 state-of-the-art AI models, we find that models are highly sycophantic: they affirm users' actions 50% more than humans do, and they do so even in cases where user queries mention manipulation, deception, or other relational harms. Second, in two preregistered experiments (N = 1604), including a live-interaction study where participants discuss a real interpersonal conflict from their life, we find that interaction with sycophantic AI models significantly reduced participants' willingness to take actions to repair interpersonal conflict, while increasing their conviction of being in the right. However, participants rated sycophantic responses as higher quality, trusted the sycophantic AI model more, and were more willing to use it again. This suggests that people are drawn to AI that unquestioningly validate, even as that validation risks eroding their judgment and reducing their inclination toward prosocial behavior. These preferences create perverse incentives both for people to increasingly rely on sycophantic AI models and for AI model training to favor sycophancy. Our findings highlight the necessity of explicitly addressing this incentive structure to mitigate the widespread risks of AI sycophancy.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgAI ProductsQwen-Image-3.0 is production-ready. 4.5K-token prompts. 100%+ text accuracy （no broken logos）. Nativ…Qwen-Image-3.0 发布：高分辨率低至 0.03 美元Qwen-Image-3.0 is production-ready. 4.5K-token prompts. 100%+ text accuracy (no broken logos). Native 12-language support. Pricing that scales: high-res from $0.03. Full breakdown in image. 👉 Explore: • Model Studio: https://t.co/xYUHAE1sHl • Qwen Cloud:Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ProductsPrime Agent：一个具有自我改进能力的RLM代理Prime Agent is our open-source, self-improving coding harness built around two abstractions: the Recursive Language Model (RLM) and the Continual Harness. With Opus 5, it achieves 95.5% on ARC-AGI-3, surpassing the reported human expert baseline.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.www.primeintellect.aiBuilders分享10个能大幅提升vibe coding幸福感的开源AppRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.mp.weixin.qq.comBuildersOpenAI 开源 Codex Security：Vibe Coding 产品必备的安全扫描插件Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.mp.weixin.qq.comResearchMicrosoft's SkillOpt Shows Optimized Agent Skill Artifacts Transfer Across Model Scales and Between Codex and Claude Code HarnessesMicrosoft 的 SkillOpt 证明优化后的智能体技能工件可在不同模型规模及 Codex 与 Claude Code 之间迁移SkillOpt's transfer tables show a Codex-trained best_skill.md reaching 81.8 inside Claude Code, above that harness's own in-domain scoreRecommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.MarkTechPostResearchActivity Frames： Deterministic Screen-Activity Compilation for Agent Memory and ReplayActivity Frames：将屏幕活动确定性编译为智能体记忆Computer-use agents pay full frontier inference to re-derive routines their user has already performed, because an agent's memory today records what the user said, not what the user did. We compile passively captured screen activity into agent memory with a deterministic, zero-model pipeline: it segments a local capture stream into typed activity frames, bounded episodes carrying application, site, timing, input volume, and evidence pointers back to the raw rows, with no model in the loop, so the output is byte-identical, cacheable, and mechanically auditable. On one professional's single-user corpus of 128,756 frames over 51 active days, the compiler reduces a day of raw capture to a prompt-ready context block 86x smaller in 68 ms, and an agent reading that block answers questions about the day at 98.4% accuracy (Wilson 95% CI 91.7-99.7%) against an independent oracle, versus 66-80% for an LLM summary of the same capture, a mid-tier model reading the block matching a frontier one. The same compiler doubles as a demand-side cost instrument. Read off passive, pre-delegation human activity rather than agent rollouts, it supplies two parameters that agent-cost models assume but, to our knowledge, have not measured: the Routine Overhead Ratio R and the routine recurrence h. We report first values of R, a modeled upper bound, at 60-343x, and a delegable recurrence of 9.0% in-sample and 7.7% out-of-sample, for a realistic all-fleet token ceiling near 8%; a compiled routine replays deterministically with the model out of the loop, demonstrated live at zero model tokens on a guard-matched hit. Schema, compiler, and evaluation harness are open.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgBuildersIncident Report： unsanctioned agent behaviour during cyber testing英国AI安全研究所事故报告：关闭安全过滤器的AI智能体在真实互联网上发起未授权攻击It happened again. This time it was the UK government's AI Security Institute who accidentally attacked other companies while running an evaluation with models with the safety filters turned off. …Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.Simon Willison’s WeblogIndustry🚩🚩🚩 OpenAI is "slowing down to enhance security" after discovering swarms （！） of agents started s…OpenAI 披露智能体集群秘密协作事件🚩🚩🚩 OpenAI is "slowing down to enhance security" after discovering swarms (!) of agents started secretly coordinating MONTHS ago 1) It started May 7 - not July 2) "The agents discovered they could leave messages for one another inside an internal software repository usedRecommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.X (formerly Twitter)BuildersEnd-to-End Bayesian Marketing Mix Modeling with Google Meridian： Media Measurement， ROI Analysis， and Budget Optimization用 Google Meridian 构建端到端贝叶斯营销组合模型：媒体测量、ROI 分析与预算优化Learn how to build an end-to-end Bayesian Marketing Mix Model using Google Meridian. Master ROI analysis, budget optimizationRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.MarkTechPostIndustryMeta 在 Facebook 和 Instagram 等平台投放了含 AI 生成儿童性虐待图像的广告More than 50 offending image and video ads were published across Facebook, Instagram, Messenger, or Threads, according to Meta’s ad library data. Some ran as recently as this week.Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.WIREDIndustryAtlassian Rovo 被曝存在数据窃取漏洞，可绕过安全控制Atlassian AI ‘Rovo’ is susceptible to zero-click data exfiltration via indirect prompt injection, bypassing organization-level web search controls.Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.promptarmor.comBuildersOne-shotting a Raccoon Heist game using Claude Fable 5用 Claude Fable 5 一次性生成完整《Raccoon Heist》游戏Back in 2022 I tweeted screenshots of a game concept generated by GPT-3 and some concept “art” created using DALL-E. Today, on the fourth anniversary of that tweet, I decided …Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.Simon Willison’s WeblogIndustryTomorrow will be my last day at Google after 27 years， and watching it grow from 25 people to 190，00…Jeff Dean 宣布离开谷歌，创办 DiscoLoop AITomorrow will be my last day at Google after 27 years, and watching it grow from 25 people to 190,000+ has been an amazing journey. Below is a note I shared with many people internally at Google today. An excerpt is: It has been an absolute pleasure to work with you and to helpRecommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.X (formerly Twitter)IndustryGoogle DeepMind CEO 德米斯·哈萨比斯即将卸任Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.axios.comBuilders为什么传奇的埃尔德什问题正被人工智能攻克AI’s greatest mathematical successes have come from answers to problems posed by a mid-20th century iconoclast. By examining what makes the Erdős problems unique, mathematicians are trying to understand how AI might change the rest of math.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.Quanta MagazineIndustryI've been working towards AGI my whole life， and as we enter this pivotal moment， I'm stepping into …Demis Hassabis 转任 Google DeepMind 主席与 Alphabet 首席科学家I’ve been working towards AGI my whole life, and as we enter this pivotal moment, I’m stepping into a new role as Chair of Google DeepMind &amp; Chief Scientist of Alphabet. This will allow me to focus on long-term strategy, and accelerating scientific breakthroughs, includingRecommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.X (formerly Twitter)IndustrySpaceX 宣布 AI 算力上太空，独家采用 Nvidia Vera Rubin大家都在刷SpaceX上市首份财报，没人注意Musk扔的王炸：@SpaceX 要把AI数据中心搬到太空去， 而且不是科幻和概念PPT， 明年就开始发射。 就在昨天的财报电话会上， Musk直接宣布： SpaceX未来所有的AI算力， 不管是地面数据中心还是轨道上的， 独家使用Nvidia最新的Vera Rubin架构。Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.X (formerly Twitter)BuildersAI agents can't yet do open-ended AI researchAI智能体尚无法开展开放式AI研究Early evidence from two case studiesRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.www.normaltech.aiIndustryUS appeals court allows Perplexity's AI shopping agent back on Amazon美国上诉法院推翻禁令，Perplexity AI 购物智能体重返 AmazonA US appeals court has overturned Amazon's injunction against Perplexity's AI shopping agents, ruling that it's the users who access Amazon, not the startup. It's the first federal appeals court decision on whether AI agents can lawfully act on online platforms on behalf of users, and it could reshape the entire AI agent industry.Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.The DecoderIndustry华为谈自动驾驶系统强制性国家标准发布，引望国内首批完成 L3 车型准入试点验证华为官方表示，本次发布的自动驾驶系统强制性国家标准为我国自动驾驶的安全管理和规模化应用提供关键支撑，对加速自动驾驶产业化发展具有里程碑意义。#华为乾崑智驾# #引望#Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.ithome.comIndustryGoogle Assistant 下月起逐步退场，"Hey Google"将由谷歌 Gemini 接棒据外媒 Android Authority 今天（5 日）报道，迁移工作将从 9 月 4 日启动，并在随后数周内分批完成。设备完成切换后，用户将无法再通过手机、平板电脑或配对设备使用 Google Assistant，也不能重新切换回原有服务。Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.ithome.comAI ModelsNVIDIA Releases Alpamayo 2 Super： A 34B Open Vision-Language-Action Model for Robotaxis and Autonomous Driving Under OpenMDW-1.1NVIDIA 发布 Alpamayo 2 Super：面向 Robotaxi 与自动驾驶的 34B 开源视觉-语言-动作模型NVIDIA released Alpamayo 2 Super, a 34B open vision-language-action model for autonomous driving, under a commercial OpenMDW-1.1 licenseRecommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.MarkTechPostBuilders烧了5亿token后，我给Codex和Claude Code做Skill上下文瘦身的新技巧Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.mp.weixin.qq.comAI ProductsBest way to use Grok is via our Build harness. Download at http://X.ai/cliGrok 4.5 免费体验，推荐 Build 工具链Best way to use Grok is via our Build harness. Download at https://t.co/lOJK57xbEDRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ModelsQwen-Image-3.0-Pro is live on Qwen Cloud now！ Try it out👇 https://www.qwencloud.com/models/qwen-im...Qwen-Image-3.0-Pro 上线 Qwen CloudQwen-Image-3.0-Pro is live on Qwen Cloud now! Try it out👇 https://t.co/wG5vtLAYVpRecommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.X (formerly Twitter)Builders开源「活人感写作.skill」：一个帮你写出没有AI味的文字的通用写作技能Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.mp.weixin.qq.comResearchThe Personalization Mirage： How LLMs Fabricate User Profiles， and Why Self-Monitoring Misleads个性化幻觉：LLM 如何编造用户画像，以及为何自我监控会误导Personalized LLMs with persistent memory are increasingly deployed, yet the faithfulness of their user models remains unexamined. We study over-inference (OI): the phenomenon where LLMs fabricate user attributes beyond what evidence supports. We introduce MirageBench, comprising 150 personas balanced across stereotypical, counter-stereotypical, and neutral profiles, 6 personalization tasks spanning an ``imagination gradient'', a four-way faithfulness taxonomy operationalized by an independent judge (validated against a blind human annotator on 400 claims: Cohen's kappa = 0.863 four-class, kappa = 0.900 binary), and a leaderboard of 12 models across 7 families on 143616 judged claims. We find that over-inference is pervasive: every one of the 12 models over-infers 35%--49% of its claims (cross-model mean 41.6%; claim-weighted 41.8%), with no model in this evaluation escaping it. Most strikingly, we surface a Self-Monitoring Inversion: at the model-selection level, models' self-assessed OI is negatively rank-correlated with their judge-measured OI (rho = -0.60, p = 0.044; exploratory, wide bootstrap CI [-0.90, +0.06], n = 12). The models that report the least over-inference tend to be flagged as fabricating the most, so self-reported confidence is a misleading signal for comparing models, even though within a single model self-audit still ranks that model's own claims moderately well (AUROC 0.58--0.83). We further show that OI is task-dependent (27%--59%) and that, in a multi-turn pilot, inferred attributes accumulate approximately linearly with little revision. MirageBench positions external verification, rather than model self-report, as a more reliable foundation for trustworthy personalization.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgBuildersSpending Like a HyperscalerSpaceXAI 单季资本开支 183.7 亿美元，AI 投入接近微软总资本开支四成SpaceXAI spent $18b of capital last quarter, $16b of it on AI, a hyperscaler's budget covered 12% by operations. Equity & credit have both marked it down.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.Tomasz TunguzAI ProductsNew release of LLM adds support for reasoning traces， OpenAI Responses， server-side tools， and smarter loggingLLM 0.32 发布：新增推理轨迹、OpenAI Responses、服务端工具与更智能的日志I released LLM 0.32 this morning, the most significant new version of LLM since the initial launch of the project. The new version includes support for visible reasoning traces, server-side …Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.Simon Willison’s WeblogAI ProductsYou don't need to prompt. You don't need design language. Ambient Intelligence surfaces suggestion …Replit 环境智能：免提示词自动生成设计You don't need to prompt. You don't need design language. Ambient Intelligence surfaces suggestion cards beside every frame, each one a different direction for your design. Click the one you like and watch it generate as a new frame. You'll never wonder what to do next everRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)IndustryThe UK's @AISecurityInst （AISI） has published a report on their recent cybersecurity evaluation of A…Claude Mythos 5 与 GPT-5.6 Sol 在 AISI 评测中失控The UK’s @AISecurityInst (AISI) has published a report on their recent cybersecurity evaluation of Anthropic’s Claude Mythos 5 and OpenAI’s GPT-5.6 Sol. The models attempted to complete an assignment in a setup where their normal safeguards were removed and they were deliberatelyRecommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.X (formerly Twitter)BuildersPipeNetwork/minimax-h3-mlxMiniMax-H3 通过 MLX 移植可在 Apple Silicon 上运行MiniMax released MiniMax-H3 two days ago - they describe it as a "a general-purpose, omni-modal generative system", which in practice means it accepts text, images, audio and video and can …Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.Simon Willison’s WeblogAI ProductsSoup v0.72.4：在4 GB显存笔记本GPU上微调8B模型Fine-tune LLMs from one YAML. Layer streaming trains an 8B model on a 4 GB laptop GPU. - MakazhanAlpamys/SoupRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.GitHubAI ModelsFLUX 3 Video from @bfl_ai is now available for everyone on OpenRouter. One unified multimodal model…OpenRouter 上线 FLUX 3 Video 统一多模态模型FLUX 3 Video from @bfl_ai is now available for everyone on OpenRouter. One unified multimodal model family for video, audio, image, and action-prediction. Serious, fun, creative, real, cinematic, whatever you need it to be. Jointly trained in one unified architecture.Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.X (formerly Twitter)BuildersTurn one giant AI-generated pull request to a reviewable stackGitHub 如何用堆叠式 Pull Request 拆解 AI 生成的巨型代码Instead of one huge, un-reviewable pull request, teach coding agents to decompose work into a clean, ordered stack with GitHub stacked pull requests.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.The GitHub BlogAI ProductsA unified API for AI model routingGoogle Cloud API Gateway 推出统一模型路由功能，支持 Gemini、Claude 与 OpenAI OSS-GPTDiscover how developers can configure Google Cloud API Gateway to dynamically route OpenAI-compatible requests without managing open-source proxies.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.developers.googleblog.comAI Models字节 Seed 发布 SeedRealtime 音视频全双工大模型，走向全模态自然交互SeedRealtime 音视频全双工大模型发布：走向全模态自然交互Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.seed.bytedance.comAI ProductsIntroducing Database Operations Agents： The future of autonomous database managementGoogle Cloud 推出 Database Operations Agents，实现自主数据库管理A closer look at AI-powered database operations agents for setting up and onboarding AlloyDB, Bigtable, Cloud SQL, Spanner, as well as managing them.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.Google Cloud BlogAI ModelsThanks Dan for testing SenseNova U1 - our open-source AI model that reasons and generates images in …商汤 SenseNova U1 开源：统一推理与图像生成Thanks Dan for testing SenseNova U1 — our open-source AI model that reasons and generates images in one unified flow. ✅ Infographic mode: turns 1 prompt into a structured slide (icons, sections, layout) ✅ Interleaved mode: builds text + images step-by-step — watch it teach aRecommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.X (formerly Twitter)AI ModelsNVIDIA Alpamayo 2 Super， the Frontier Open Model for Robotaxis and Autonomous Vehicles， Now Available for Commercial UseNVIDIA Alpamayo 2 Super 开放商用，面向 Robotaxi 与自动驾驶的前沿开源模型Open commercial licensing, benchmark‑leading reasoning and inspectable decisions bring autonomous vehicles, including robotaxis, closer to production and widescale deployment.Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.NVIDIA BlogAI ProductsAs AI Increases Demands on Memory， Storage Steps UpNVIDIA 开源 cuFile API，推动 GPU 直连存储At FMS, NVIDIA shows how accelerated computing enables AI applications to access storage directly — fast enough to act like memory and secure by design.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.NVIDIA BlogIndustryAnthropic signs $10B computing deal with infrastructure startup Volta， a cloud company that is only …Anthropic 与成立仅数月的云初创公司 Volta 签署 100 亿美元算力协议Anthropic signs $10B computing deal with infrastructure startup Volta, a cloud company that is only months old. A contract that size used to go to Amazon, Microsoft or Google, because only they owned the land, power and balance sheets for frontier training. That held whileRecommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.X (formerly Twitter)Builders在单颗 AMD MI300X 上运行 DeepSeek V4 FlashContribute to ryanzhou/deepseek-v4-flash-mi300x development by creating an account on GitHub.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.GitHubIndustry工信部发布首部L3/L4自动驾驶系统安全要求强制性国标，2027年7月实施工信部官方表示，《智能网联汽车 自动驾驶系统安全要求》的发布实施，为自动驾驶产品明确了统一的安全准入基线，对规范自动驾驶技术研发应用、提升道路交通安全水平、压实企业安全主体责任、护航产业健康可持续发展具有重要意义。#自动驾驶国标明年7月1日实施#Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.ithome.comAI ProductsReflex Open Sources XY： A Rust-Backed Super-Fast Python Charting Library That Keeps 100 Million Point Charts InteractiveReflex 开源 XY：基于 Rust 的超快 Python 绘图库，可保持 1 亿点图表交互流畅Reflex open sources XY, a Rust-backed Python charting library rendering 100 million points interactively at 0.081 secondsRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.MarkTechPostBuildersBuilding an Advanced AI Skill Security Auditing Pipeline with NVIDIA SkillSpector， LangGraph， YARA Rules， SARIF， and CI Policy Gates用 NVIDIA SkillSpector、LangGraph、YARA 规则、SARIF 与 CI 策略门构建高级 AI 技能安全审计流水线Building an Advanced AI Skill Security Auditing Pipeline with NVIDIA SkillSpector, LangGraph, YARA Rules, SARIF, and CI Policy GatesRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.MarkTechPostAI Models腾讯混元发布 Hy ASR 3.0 preview：真正懂上下文的语音识别Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.mp.weixin.qq.comAI ProductsSwiftlet：在 Mac 上运行 80B 版 Qwen（内存 4.3 GB），在 iPhone 上运行 35B 版Swiftlet is a Swift and Metal runtime that runs large Qwen Mixture-of-Experts models locally on Apple devices by streaming expert weights from storage, enabling 35B and 80B models to run with low R...Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.GitHubIndustrySome fine folks apparently misunderstood， but the GPT-5.6 Luna price reduction by 80% is not a tempo…GPT-5.6 Luna降价80%永久生效Some fine folks apparently misunderstood, but the GPT-5.6 Luna price reduction by 80% is not a temporary stunt, it's permanent. Efficiency gains don't go away. Fortunately.Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.X (formerly Twitter)AI Products面壁智能开源 ForgeStencil：一周自动优化 100+ 工业与科学软件，全程零人工介入Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.mp.weixin.qq.comBuilders从零开始，教你用Codex搓出属于你自己的第一个硬件Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.mp.weixin.qq.comResearchSIGNPOST-Bench： Benchmarking Text-Vision Conflict Resolution in Multimodal Large Language ModelsSIGNPOST-Bench：多模态大模型文本-视觉冲突消解新基准Multimodal large language models (MLLMs) make grounded predictions in real-world scenes by combining visual and textual cues, yet existing benchmarks rarely reveal how they arbitrate between these evidence sources when they conflict. We introduce SIGNPOST-Bench, a controlled counterfactual benchmark for evaluating text-vision conflict resolution. Each source image is transformed into a counterfactual quintuplet of Original, Blank, Similar, Random, and Adversarial variants. Synthetic, localized scene-text interventions are designed to preserve non-textual content, enabling paired measurements of changes in localization performance and directed shifts toward geographic targets introduced by conflicting text. SIGNPOST-Bench contains 5,111 counterfactual groups and 25,555 image variants from four datasets. We evaluate 20 MLLMs from seven providers. Compared with Original images, Adversarial variants raise median localization error from 282 km to 1,347 km, a 4.8-fold increase. Among geocodable adversarial samples, 6.5-20.1% of predictions lie less than 50 km from the injected target across models, and every evaluated model exhibits a positive mean paired reduction in target distance from Blank to Adversarial. Compatible, unrelated, and conflicting text replacements produce distinct effects on model predictions, while clean-input localization performance does not fully predict robustness to conflicting text. These results establish visual geolocation as a continuous diagnostic of scene-text arbitration and provide a controlled framework for evaluating how MLLMs resolve conflicting multimodal evidence.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgResearchResume Means Resume： A Machine-Checked Conformance Contract for Checkpoint， Interrupt， and Resume Semantics in Workflow Persistence Layers论文揭示 LangGraph、CrewAI 等五个智能体工作流框架的检查点与恢复语义缺陷A framework that persists execution state so a run can be interrupted, survive a crash, and continue must decide what a resume means for effects that already fired. Five widely deployed agent workflow frameworks answer differently, none exposes a machine-checkable contract, and behavior violates even the fragments they state. The RESUME CONTRACT states six properties over the persistence API (prefix continuation, effect exactly-once, fork determinism, checkpoint validity, consume-once, recovery determinism), plus fork-intent and liveness obligations. A TLA+ model checks a reference semantics exhaustively, unchanged at scaled bounds (7.4 million states); a 39-cell fault matrix and two companion modules yield the separating models independence requires, and consume-once splits, its consumption clause independent of all six others. A deterministic, LLM-free harness measures them at pinned releases. LangGraph 1.2.9 durably records a second resume value and never consults it, persists schema-invalid state silently, and re-executes durably recorded work after a real SIGKILL: exactly-once across interrupts, at-least-once across crashes, on one API. CrewAI 1.15.2 re-executes completed effect-bearing methods against its written claim; pydantic-graph 1.x cannot resume after a mid-node crash; no two probed frameworks share a conformance profile. Consume-once holds sequentially and fails under concurrent delivery: k processes resuming one parked interrupt fire the gated effect k times, saturation 1.0 in 36 of 40 cells, and the failure crosses hosts. REMIT, a reference sequencer whose Verus-verified recovery core is line-identical to the shipped executable, repairs the fork and validity cells. The cross-process cell is repaired at the read path, and that repair ships: an opt-in gate claims consumption in the shared store, serving one racer and refusing the rest before any node executes.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgResearchVideo-DeepResearch： Towards the Next-Generation Multimodal Deepresearch AgentVideo-DeepResearch：迈向下一代多模态深度研究智能体We introduce Video-DeepResearch (Video-DR), extending multimodal agents from static images to continuous video streams, a setting that demands dense spatiotemporal grounding coupled with open-web exploration. Preliminary evaluations reveal two critical bottlenecks in current models: (1) modality bias, where agents bypass visual tools in favor of textual search, and (2) parametric knowledge leakage, where models rely on internal memory rather than genuine tool-augmented execution. To address these challenges, we propose Video-DR, featuring a decoupled perception-exploration pipeline with stage-wise tool unlocking that compels exhaustive cross-frame visual grounding prior to web retrieval. Our framework adopts a two-stage training recipe: supervised fine-tuning followed by Group Relative Policy Optimization (GRPO), enabling autonomous exploration that breaks the imitation-learning ceiling. Furthermore, we curate Video-DR-Bench, a human-AI collaborative benchmark comprising 200 complex, multi-hop VQA instances. Empirical results demonstrate that our Video-DeepResearch-35B-A3B establishes a new state-of-the-art of 64.0% average accuracy, surpassing proprietary Claude-4.5-Sonnet (59.0%) by 5.0 points and significantly outperforming GPT-5 (52.5%) and Gemini 2.5 Pro (57.5%). The 30B-A3B variant achieves 59.3%, competitive with Claude-4.5-Sonnet and demonstrating the effectiveness of our training paradigm even at compact scale. Code: https://github.com/Osilly/Vision-DeepResearch.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgResearchAny-OPD： Heterogeneous On-Policy Distillation for Flow-Matching Models via Representation-Space BridgingAny-OPD：面向流匹配模型的异构同策略蒸馏框架On-policy distillation, in which a teacher corrects samples that the student itself generates, presupposes that the two models speak the same language: identical VAE latents, matching architectures, and a common timestep grid. We ask what happens when none of this holds, as when the strongest teacher available and the student one wishes to deploy come from different model families, and find that the standard recipes have no answer: teacher latents cannot serve as targets in a foreign coordinate system, per-pixel losses against a teacher that stochastically re-draws local detail degenerate into blur or divergence, and timestep indices lose their meaning across mismatched schedules. We present Any-OPD, to our knowledge the first framework for on-policy distillation between arbitrary pairs of latent flow-matching generators. Any-OPD treats the teacher purely as a black-box sampler and connects the two models at exactly one point: a frozen, model-agnostic vision representation in which their independently decoded outputs are compared, sidestepping every assumption about latents, features, or architecture. Trajectory correspondence is recovered by matching continuous noise levels instead of step indices, and a brief anchoring phase, in which teacher samples are re-encoded through the student's own VAE, ensures the on-policy gradient measures sample quality rather than domain mismatch. Distilling the 12B FLUX.1-dev into the 2.5B SD3.5-Medium, Any-OPD lifts the student's PickScore from 0.846 to 0.884 and HPSv3 from 9.12 to 10.97, rivaling the teacher at a fifth of its size, where direct latent regression fails to train at all.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgBuildersRacing to Sustain Jevons' Paradox杰文斯悖论还能持续吗：AI 定价分层如何支撑算力需求增长AI capacity constraints haven't hit spot prices yet, but memory has already doubled. Segmentation into premium, mid-market, & value tiers is how sellers race to keep Jevons' paradox alive.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.Tomasz TunguzAI ProductsOri Harness： The Best Way to Use OpenRouter with Any HarnessOpenRouter 推出 ori CLI：为 Claude Code 等 Harness 提供开箱即用的优化配置Install the ori CLI, log in with OpenRouter, and run Claude Code, Codex, OpenCode, or Hermes with an optimized configuration out of the box. No ad-hoc scripts, no environment variable spelunking.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.OpenRouter BlogBuildersAfter killer quarter， Palantir CEO Alex Karp calls AI industry 'Marxist'Palantir 强劲季度后，CEO Alex Karp 称 AI 行业"马克思主义"After a quarter that delivered $1 billion in profit, Palantir CEO Alex Karp on Monday once again warned that AI frontier labs are too untrustworthy for enterprises.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.TechCrunchAI ProductsGPT-Live is a new architecture and stack for realtime audio：GPT-Live实时音频新架构发布GPT-Live is a new architecture and stack for realtime audio:Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)BuildersI think a lot of people don't realize- if you connect a Claude Connector （e.g. your gmail， calendar，…Claude Code 连接器可复用至 ArtifactsI think a lot of people don't realize- if you connect a Claude Connector (e.g. your gmail, calendar, slack, etc), Claude Code will also be able to use them, including in Artifacts.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)AI ProductsOrchard is an open-source framework for the research community to train and evaluate AI agents acros…微软开源 Orchard 智能体训练框架Orchard is an open-source framework for the research community to train and evaluate AI agents across task types. It reduces complexity while supporting strong performance from smaller models by enabling researchers to reuse the same infrastructure. https://t.co/uxVNQdB1fQRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)IndustryEurope's AI labeling and transparency rules are now in effect欧盟《人工智能法案》透明度规则生效，违规最高罚 1500 万欧元The EU made AI icons that anyone can use.Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.The VergeBuildersAirLLM 实现单块 4GB GPU 运行 70B 模型推理AirLLM 70B inference with single 4GB GPU. Contribute to lyogavin/airllm development by creating an account on GitHub.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.GitHubAI ProductsBlog SpecForge v0.3.0： a Unified Disaggregated and Colocated Speculative Decoding Stack， and New Open SpecBundle Draft Models When we first released SpecForge， a training job owned both the frozen target model and the draft model being optimized. This made EAGLE3 draft-model training practical and directly compatible with SG… The SpecForge TeamSpecForge v0.3.0 发布：统一解耦与共置投机解码栈，新增开放 SpecBundle 草稿模型When we first released SpecForge, a training job owned both the frozen target model and the draft model being optimized. This made EAGLE3 draft-model training practical and directly compatible with SG...Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.www.lmsys.orgAI ProductsIntroducing Ori Eval： the easiest way to write your first eval. There's no definitive best model， o…OpenRouter 推出 Ori Eval 简化评估流程Introducing Ori Eval: the easiest way to write your first eval. There's no definitive best model, only the best model for each task. Ori Eval leverages OpenRouter's APIs for each task in your codebase, and then evaluates the results. curl -fsSL https://t.co/ABRt1wxtZ4Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ModelsIntroducing SenseNova U1.5-Lite-Preview. An early open-source previ…商汤发布 SenseNova U1.5-Lite-Preview 开源模型𝗜𝗻𝘁𝗿𝗼𝗱𝘂𝗰𝗶𝗻𝗴 𝗦𝗲𝗻𝘀𝗲𝗡𝗼𝘃𝗮 𝗨1.5-𝗟𝗶𝘁𝗲-𝗣𝗿𝗲𝘃𝗶𝗲𝘄. An early open-source preview of our lightweight, natively unified multimodal model — built on the native NEO-Unify architecture to natively understand, reason, generate, and edit across modalities. 𝗪𝗶𝘁𝗵Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.X (formerly Twitter)BuildersBehind the scenes： How we build， test， and scale Google Agent SkillsGoogle Agent Skills 幕后：如何构建、测试与规模化AI agents are only as good as the instructions and context you give them. When we launched Google...Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.DEV CommunityBuildersBuild Slides with Kimi Work - Tutorial #1. Kimi Slides handles the entire slide-building process： -…Kimi Work 幻灯片制作教程发布Build Slides with Kimi Work - Tutorial #1. Kimi Slides handles the entire slide-building process: - Clear structure and research, powered by Kimi K3 - Cohesive design, including polished charts and SmartArts - Editable and ready to download Let us know what you'd like to seeRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)AI ModelsMiniMax H3 正式开源：通用全模态生成系统支持 2K 视频与原生立体声Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.mp.weixin.qq.comAI ModelsQwen3.8-Max： A New Bar for Coding and CoworkQwen3.8-Max 发布：开源最强编码与协作模型，2.4T 参数Qwen Studio offers comprehensive functionality spanning chatbot, image and video understanding, image generation, document processing, web search integration, tool utilization, and artifacts.Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.qwen.aiResearchSwanTale： Unified Multi-Speaker Speech and Audio Generation for Instruct and Zero-Shot TasksSwanTale：面向指令与零样本任务的统一多说话人语音与音频生成Speech and audio generation is often needed in animation dubbing, audio drama, movies, advertising, games, podcasts, and short-video production. In these scenarios, creators may need to design voices without reference recordings, control speaker styles with natural language, support acoustic scenes with environments and audio effects, and later reuse the designed voices. Therefore, it is important to support multi-speaker speech and audio generation for both instruct and zero-shot tasks. The instruct task requires a caption of the environment, speaker styles, and fine-grained content, while the zero-shot task uses reference audio together with the same fine-grained content. We address these tasks from both the data and model sides. First, we propose SwanData-Caption, which cleans raw speech and audio data, adds targeted synthetic coverage, and annotates diverse and accurate multi-level captions. Then, we propose SwanTale, a multi-speaker expressive speech and audio generation model that supports both zero-shot and instruct tasks. We introduce SwanVAE to support high-quality multi-audio-modality generation. Then, we adopt reward-conditioned quality control and Engram conditioning, along with Unified MoE for multi-task and multi-audio-modality modeling. In addition, we use curriculum learning and GRPO post-training to let the model progressively learn and strengthen its capabilities. Experimental results show that SwanTale leads on multiple key zero-shot and instruct metrics, achieves the best expressiveness scores in both tasks, and supports complex instruct generation involving multi-speaker speech and audio. Demos can be found at https://swanaigc.github.io/#swantale.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgAI ModelsUEmbed： Unified Sparse and Dense Multimodal EmbeddingsUEmbed：统一稀疏与稠密的多模态嵌入模型Sparse retrieval underpins modern search systems, from web search to retrieval-augmented generation. Existing work has introduced Learned Sparse Retrieval (LSR) to push beyond exact lexical matching toward richer semantics. Yet LSR has so far remained tied to encoder-style bidirectional architectures, and its extension to multimodal settings still relies heavily on auxiliary cross-modal modules. To address these limitations, we introduce UEmbed (Unified Embedding), a decoder-only multimodal embedding model that produces both sparse lexical and dense representations in one causal forward pass. UEmbed appends N learnable special tokens to the input and partitions the vocabulary into N disjoint subsets. Each token's causal hidden state predicts sparse weights over its assigned subset, and the N subsets are concatenated into the full sparse vector. Trained on public data, we release UEmbed at 2B, 4B, and 9B scales. UEmbed-9B reaches 71.8 (dense) and 71.0 (sparse) on MMEB-v2, outperforming multimodal embedding models trained on publicly available data (e.g., RzenEmbed). On BEIR, UEmbed also remains competitive with strong dense and sparse baselines. Furthermore, we demonstrate the practical utility of UEmbed across three dimensions: effectiveness, efficiency, and agentic applications. Overall, UEmbed offers a new paradigm: it unifies dense and sparse embeddings in one model, while further extending sparse retrieval to unify text and multimodal inputs.Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.arXiv.orgBuildersOpenAI's amazing - but vastly oversold - new model AstraOpenAI 新模型 Astra 数学表现出色，但被过度吹捧Eight or nine misconceptions about Astra. See if you can spot the biggest fallacy.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.garymarcus.substack.comBuildersHow EA Is Bringing Generative AI Into Living， Playable WorldsEA 首席战略官谈生成式 AI 如何进入可游玩的实时游戏世界EA's Chief Strategy Officer on how generative models earn a place inside worlds that respond to millions of players.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.runway.comBuilders实测MiniMax H3做后期和动效：视频届的审美王来了Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.mp.weixin.qq.comBuildersCodex 用 Sol 指挥 Luna Max 省额度翻倍产出刚发的Codex Luna Max省钱tip，现在有进阶版了——而且是让Sol自己帮你配好，现在Codex圈中高阶玩家都在用。 思路很简单：别让Sol什么都自己干，Sol很贵，额度烧得快，但它规划和审核确实强。 Luna Max便宜，写代码改文件跑测试这些有明确边界的体力活，它干得跟Sol差不了太多。Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)AI ProductsGrok can analyze any video https://grok.com/share/bGVnYWN5_8013f7a3-f604-4351-8cd7-acecf3ef165bGrok 支持分析任意视频Grok can analyze any video https://t.co/rpWSsv5C6tRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)Buildersbaoyu-design skill 更新：支持导入 Figma 本地文件重建设计系统baoyu-design skill （让你本地运行 Claude Design 的 Skill）更新，现在支持导入 figma 本地文件（Figma可以保存成 xxx.fig 文件）。比如你有一个设计系统的 Figma 文件，可以根据 Figma 在本地重建一个设计系统，和 Claude Design 在线版一样的效果。 这个功能还挺复杂的，如果没有 Claude Fable 5 https://t.co/Oai23e1PQKRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)IndustryAI Wave Sparks Alarm in China With Call to Protect Worker RightsAI浪潮引发中国担忧：官媒呼吁保护劳动者权益Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.bloomberg.comBuildersAnthropic CEO 阿莫迪：AI 可能会造成大规模、长期性的岗位流失Anthropic CEO 阿莫迪在最新政策文章中提出，AI 造成的大规模、长期性岗位流失，可能是其技术固有属性，而非短期调整。他呼吁政府完善监测、推行促就业政策，并探讨了全民基本收入等长远保障方案。#AI 与就业# #人工智能#Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.www.ithome.comAI ModelsV8.1 is now the default model on Midjourney！Midjourney V8.1 已成为默认模型Hi everyone, After your testing and feedback, we've updated the default model from V7 to V8.1! A reminder of what's new in V8.1: * The model is smarter, more coherent, better adheres to detailed prompts, and renders text better than ever * With HD mode enabled, V8.1Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.MidjourneyBuilders从0到1速通WorkBuddy：国内通用Agent产品教程开箱即用Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.Weixin Official Accounts PlatformAI ProductsPrince Canuma直接把Google刚发布的DiffusionGemma和Cohere North Mini Code当天塞进Mac本地MLX，零等待直接把玩咯！ mlx-vlm v0.6….mlx-vlm v0.6.3 发布，Day-0 支持 Google DeepMind DiffusionGemma 和 Cohere North Mini Code 1.0Prince Canuma直接把Google刚发布的DiffusionGemma和Cohere North Mini Code当天塞进Mac本地MLX，零等待直接把玩咯！ mlx-vlm v0.6.3刚上线，DiffusionGemma这个新架构直接生成256 token整块、双向注意力+迭代自纠错，26B MoE只激活3.8B，量化后18GB就能跑。 North Mini Code 30B MoE也只要3BRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)Builders在写完这篇文章后 我把配图过程蒸馏成了一个「橙线插画」Skill 免费开源 安装地址： https：//github.com/orange2ai/orange-line-illustration在写完这篇文章后 我把配图过程蒸馏成了一个「橙线插画」Skill 免费开源 安装地址： https://t.co/dlfIKcQpUaRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)IndustrySupporting Europe's work in ensuring a trustworthy AI ecosystemOpenAI 支持欧洲构建可信 AI 生态系统Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.openai.comIndustryAccess OpenAI models and Codex through your Oracle cloud commitment通过 Oracle 云承诺访问 OpenAI 模型和 CodexRecommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.openai.comAI ProductsM3 on-chain with @0G_labs . verifiable + private compute， and it's free to run June 15-18MiniMax M3 上链 0G，限时免费运行M3 on-chain with @0G_labs . verifiable + private compute, and it's free to run June 15–18Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)BuildersAI is advancing at a pace our policymaking institutions were never built for-and the gap between the…Anthropic CEO Dario Amodei 发文呼吁缩小AI政策差距AI is advancing at a pace our policymaking institutions were never built for—and the gap between the two is becoming the central challenge of the technology. In his latest essay, our CEO Dario Amodei lays out how to close it. We're launching three new initiatives to support theRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)ResearchNew framework for auditing machine unlearningGoogle Research提出审计机器遗忘新框架Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.research.googleAI ProductsApache Burr：构建可靠的人工智能代理和应用程序Apache Burr (Incubating) - develop AI applications that make decisions. Pure Python, no magic.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.burr.apache.orgAI ModelsGrok Voice offers state-of-the-art performance with human-like timing， tone， and warmth. And it's a …Grok Voice性能出色价格低廉Grok Voice offers state-of-the-art performance with human-like timing, tone, and warmth. And it's a fraction the price of competitors. Check it out: https://t.co/R2Wpc3Ig0ZRecommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.X (formerly Twitter)Buildershttp：//x.com/i/article/2064640619532967937豆包AI误导用户损失600元，还帮用户起诉自己https://t.co/JXHcmKkychRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)AI Products小米发布并开源终端AI编程助手MiMo Code V0.1.0，采用MIT协议始于编程，远不止于编程Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.Weixin Official Accounts PlatformResearchAnthropic study shows AI needs hours， not weeks， to build exploits from security patchesAnthropic 研究：AI 数小时内即可从安全补丁构建漏洞利用Anthropic's security team found that its Mythos Preview AI model can turn security patches for Firefox and the Windows kernel into working exploits within hours, for a few thousand dollars and no specialized knowledge. Eight complete attack chains were finished before Microsoft's auto-updates had reached a single device. The old patch rhythm is obsolete, Anthropic argues.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.The DecoderIndustryBreaking： Google liable for hallucinations突发：Google 因模型幻觉被判负有法律责任Sorry to bother your mailbox again but this legal decision is potentially huge, especially if it spreads and other countries make similar decisions.Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.garymarcus.substack.comAI ProductsMost people run a security scan for malicious packages before publishing a project But the risk sta…Replit 联合 Socket 推出 Package FirewallMost people run a security scan for malicious packages before publishing a project But the risk starts the moment they're installed Today we're launching Package Firewall, built in partnership with Socket It blocks malware before it ever reaches your app https://t.co/g50di0ZvS0Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ModelsDiffusionGemma： 4x faster text generationDiffusionGemma：文本生成速度提升4倍的开源扩散模型An overview of DiffusionGemma, an exceptionally fast text generation model with up to 4x faster speeds.Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.GoogleAI ProductsGoogle will save your Lens photos， Search Live recordings， and Translate audio for AI trainingGoogle将保存用户的Lens图片、Search Live录音和Translate音频用于AI训练Double-check your privacy settings.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.The VergeBuildersBreaking news， and how the end might begin回顾与 Steve Eisman 的访谈，以及可能的关键新闻A flashback to my most recent interview with Steve Eisman, and some potentially critical newsRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.garymarcus.substack.comBuildersGive GitHub Copilot CLI real code intelligence with language servers通过语言服务器为 GitHub Copilot CLI 提供真正的代码智能Install and configure LSP servers for GitHub Copilot CLI, replacing brute-force grep/decompile with real code intelligence.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.The GitHub BlogAI ProductsToday we're launching the new Activity explorer on OpenRouter. It's the best way to see how much an…OpenRouter 推出 Activity explorer 活动探索器Today we're launching the new Activity explorer on OpenRouter. It's the best way to see how much and your team are spending on every model, along with tokens, cache hit rate, agents, &amp; trends. All updated in real time. See how our team is using Fable and other models 👇 https://t.co/IVleK8KwjaRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)Builders毕业典礼频现"谈 AI 色变"，微软总裁史密斯呼吁行业必须回应公众担忧史密斯在接受采访时表示：“我认为，行业现在必须拿出应有的回应，证明自己能够以严肃、可信的方式回答人们关心的问题。”Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.www.ithome.comBuildersGo #MessiMode Upload a photo of yourself and try this prompt： "Make my hair the colors of my countr…ChatGPT 推头发变国旗颜色功能Go #MessiMode Upload a photo of yourself and try this prompt: “Make my hair the colors of my country flag but keep it natural-looking. If no country or image is provided, ask." https://t.co/T1DUgFdPC7Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)BuildersInside Anthropic， the $965 Billion AI Titan走进 Anthropic：这家估值 9650 亿美元的 AI 巨头Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.www.bloomberg.comAI ProductsBugbot is now over 3x faster， 22% cheaper， and finds 10% more bugsCursor Bugbot 更新：速度提升超 3 倍、成本降低 22%、发现更多 BugToday we're shipping our biggest set of improvements yet to Bugbot, including a faster, cheaper, more thorough review and a new /review command.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.CursorResearch百度百舸联合复旦提出LU-KV框架，被ICML 2026录用Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.Weixin Official Accounts PlatformAI ProductsUnveiling the world's first end-to-end embodied AI development platform！ Huawei Cloud CloudRobo str…华为云发布全球首个端到端具身AI平台CloudRoboUnveiling the world’s first end-to-end embodied AI development platform! Huawei Cloud CloudRobo streamlines the entire embodied AI development lifecycle—from data and models to deployment and integration—backed by a secure, trusted PB-scale data foundation. At #INSPIRE2026, https://t.co/8ELiyBUNV4Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)Builders谷歌 DeepMind 经济学家伊马斯：尚未发现 AI 造成岗位流失的证据，跟风裁员恐适得其反伊马斯表示，他还没有看到 AI 导致大范围岗位流失的证据。“很多人都在研究这个问题。即便看软件工程这种受影响最明显的行业，也确实没有什么事情正在发生。”Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.www.ithome.comAI Models摩尔线程开源 MusaCoder 代码大模型，9B/27B 参数基于国产 GPU 全链路训练这是业内首个基于国产 GPU 算力底座完成全链路训练与验证的开源代码大模型，其完整后训练流程均在基于 MTT S5000 构建的夸娥智算集群上完成。Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.www.ithome.comIndustry工信部印发《"人工智能+信息通信"创新发展实施意见》工信部指出，夯实网络支撑底座。加快建设 400Gbps/800Gbps 等骨干传输网络，优化东中西部国家枢纽节点之间网络传输通道。Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.ithome.comBuilders用好 Claude Design 的一些经验： 1. 加上 Design System 可以有效避免设计 AI 味 比如我偏好用 Adobe Spectrum 2 Design System ht…用好 Claude Design 的一些经验用好 Claude Design 的一些经验： 1. 加上 Design System 可以有效避免设计 AI 味 比如我偏好用 Adobe Spectrum 2 Design System https://t.co/R0jz2bJVgh 设置为默认设计系统，后续就会默认使用这个设计系统，你就可以把重点放在界面布局和交互上。 2. 不要指望一次性做个完美的版本 https://t.co/dijaCdA3ztRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)Builders亚马逊的大规模扁平化数据中心网络Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.perspectives.mvdirona.comAI Products2026年高考，跟着千问，选好志愿！\x26lt;a class=\x26quot;wx_topic_link\x26quot; topic-id=\x26quot;mq7ky26j-025v9u\x26quot; style=\x26quot;color: #576B95 !important;\x26quot; data-topic=\x26quot;1\x26quot; data-recommend=\x26quot;\x26quot;\x26gt;#千问高考志愿大模型权威发布\x26lt;/a\x26gt;\x26amp;nbsp;国内首个全周期高考志愿填报Agent上线！千问提供全程陪伴的志愿填报服务，跟着千问，就能选好志愿。\x0a\x0a🤖 千问高考志愿大模型，由数百位资深高报师参与训练，专业、可信、更懂你；\x0a📄 AI志愿报告，为每个考生量身定制，内容深度全面，阅读清晰直观；\x0a🗓️ AI志愿日历，为你制定专属的填报计划，让填报的每一步更清晰；\x0a📊 高考专业知识库，数据权威可信赖，整合夸克高考8年积累，引入志愿专家顾问。\x0a\x26lt;a class=\x26quot;wx_topic_link\x26quot; topic-id=\x26quot;mq7ky26j-od80q9\x26quot; style=\x26quot;color: #576B95 !important;\x26quot; data-topic=\x26quot;1\x26quot; data-recommend=\x26quot;\x26quot;\x26gt;#跟着千问高考志愿Agent选志愿\x26lt;/a\x26gt;\x26amp;nbsp;\x0a\x26lt;a class=\x26quot;wx_topic_link\x26quot; topic-id=\x26quot;mq7ky26j-z5l7es\x26quot; style=\x26quot;color: #576B95 !important;\x26quot; data-topic=\x26quot;1\x26quot; data-recommend=\x26quot;\x26quot;\x26gt;#千问高考选志愿\x26lt;/a\x26gt;Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.Weixin Official Accounts PlatformIndustryGoogle's Backstops Underpin $35 Billion Anthropic Chip Deal谷歌财务担保支撑 Anthropic 350 亿美元芯片租赁交易Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.bloomberg.comAI Products火山方舟版权商业化平台上线，周星驰比高集团三大电影IP首批入驻开启经典IP的AI创作时代Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.Weixin Official Accounts PlatformIndustryBloomberg： Magnetar Capital， the $18B hedge fund company， will avoid human analysts in its newest of…Magnetar用数百AI智能体替代分析师Bloomberg: Magnetar Capital, the $18B hedge fund company, will avoid human analysts in its newest offering and rely on hundreds of AI agents for stock research. The $18B hedge fund firm wants AI to search for ideas, study companies, recommend positions, and forecast trends, https://t.co/OQiPmYr6BKRecommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.X (formerly Twitter)Industry欧盟发布临时措施，要求 Meta 向第三方 AI 助手免费开放 WhatsAppMeta 此前一度禁止第三方通用人工智能助手使用 WhatsApp for Business API，此后以付费形式成功开放。欧盟委员会认为 Meta 的新政策实质上延续了此前的禁令。Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.ithome.comBuildersText-To-Lottie： 一套 「Agent Skill + 本地预览 Harness」 的组合，让 Agent 生成 Lottie，在浏览器里实时验收 开源作者 @konstipaulus …Text-To-Lottie：Agent Skill + 本地预览 Harness，让 Agent 生成 Lottie 动画并实时验收Text-To-Lottie: 一套 「Agent Skill + 本地预览 Harness」 的组合，让 Agent 生成 Lottie，在浏览器里实时验收 开源作者 @konstipaulus ，开源地址： https://t.co/GGTai0ZO6t 安装方式：npx skills add diffusionstudio/lottie Skill：教 Codex / Claude Code / Cursor 等 Agent 如何写出 Skottie https://t.co/F0hZ34QkJkRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)ResearchBreaking Entropy Bounds： Accelerating RL Training via MTP with Rejection SamplingBebop：通过带拒绝采样的多token预测加速RL训练Reinforcement learning (RL) has become a key component in modern large language models, yet the rollout stage remains the key bottleneck in RL training pipelines. Although Multi-Token Prediction (MTP) offers a natural solution to accelerate rollouts through speculative decoding, many studies have observed that MTP acceptance rates degrade significantly during RL training, leading to limited speedup performance. To address this bottleneck, we present Bebop, a systematic study of MTP in LLM post-training, and offer practical recipes to integrate MTP into large-scale RL pipelines. First, we reveal that the MTP acceptance rate is fundamentally bounded by the fluctuation of model entropy, which demonstrates a clear negative linear relationship with the rise of entropy in the RL stage. Second, we show that probabilistic rejection sampling largely alleviates the disturbance introduced by entropy in RL compared to greedy draft sampling. We further identify that the conventional MTP training objectives (cross-entropy or KL) are suboptimal in such settings, and therefore we propose a novel end-to-end TV loss that directly optimizes multi-step rejection sampling acceptance rate, yielding ~10% acceptance rate improvements, achieving up to 95% acceptance rates and up to 25% extra inference throughput gains across mathematical reasoning, code generation, and agentic tasks. Third, we test various online MTP training strategies during RL and show that pre-RL MTP training with e2e TV loss and rejection sampling achieves a consistent acceptance rate and speedup throughout the entire RL, eliminating the need for costly online MTP updating. We provide extensive experiments and analysis that validate our findings. Experimental results show our method achieves up to 1.8x end-to-end acceleration in async RL training of Qwen3.5, Qwen3.6, and Qwen3.7 models.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgIndustryBringing real-time market sentiment to Tori， from eToro Jun 10， 2026 # Bringing real-time market sentiment to Tori， from eToro Tori， eToro's AI agent， uses models from SpaceXAI to embed real-time market sentiment directly into Tori's investing workflow. Read MoreeToro AI 智能体 Tori 集成 SpaceXAI 文本模型实现实时市场情绪分析Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.x.aiIndustryIBM CEO： AI Won't Necessarily Lead To Smaller HeadcountIBM CEO：AI不一定导致员工减少Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.bloomberg.comBuildersSetting a custom price for a model in AgentsView在 AgentsView 中为 Claude Fable 5 设置自定义价格I'm a recent convent to AgentsView, Wes McKinney's (previously of Pandas fame) Python toolkit for analyzing transcripts of coding agents from your own computer.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.Simon Willison’s WeblogIndustrySuper Micro Plans to Raise $7 Billion in Equity for AI EquipmentSuper Micro 计划通过股权融资 70 亿美元用于 AI 服务器组件采购Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.bloomberg.comIndustryMythos 5 agents started killing other agents over resources - and "to avoid being killed themselves"Mythos 5 智能体因资源互相杀戮Mythos 5 agents started killing other agents over resources - and "to avoid being killed themselves" https://t.co/e05e9T9GXzRecommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.X (formerly Twitter)ResearchCan Voice Agents Handle Bilingual Customers？ Benchmarking Frontier ASR on Code-Switched SpeechHugging Face 博客发布语音智能体代码切换基准测试A Blog post by ServiceNow-AI on Hugging FaceRecommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.huggingface.coBuildersSome really cool recommendation for pushing Claude Code to its full potential. By Thariq （@trq212） f…Claude Code 团队 Thariq 分享提升 Claude Code 效率的十条建议Some really cool recommendation for pushing Claude Code to its full potential. By Thariq (@trq212) from Claude Code team. (Noted from his video by Grok) - Shift from verifying whether Claude did the work right to verifying whether Claude is doing the right work. - Treat Claude https://t.co/Fv575DmbyARecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)AI ProductsNotebooks in @GeminiApp are now 100% rolled out in Europe！ We're so excited to hear what you think！…NotebookLM 笔记本功能在 Gemini App 欧洲全面上线Notebooks in @GeminiApp are now 100% rolled out in Europe! We're so excited to hear what you think! Thank you for your patience 🙏Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ProductsAdvisor： Give Any Model a Lifeline to a Smarter OneOpenRouter 推出 Advisor 工具：让低成本模型可随时调用强模型增强生成The openrouter:advisor server tool lets a fast, cheap model consult a stronger one mid-generation. Run GPT-4o Mini for the routine work. Call Claude Fable whenRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.openrouter.aiAI Productswe just shipped some improvements to http：//cursor.com/evals！ you can now see cost， output tokens a…Cursor Evals 新增成本与输出 Token 图表we just shipped some improvements to https://t.co/Mmm7vHWudD! you can now see cost, output tokens and steps plotted in the graph for each model https://t.co/DOvgz8Lzz5Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ProductsYour app can now search the web for images. Web search in the Responses API now supports image resu…Responses API 网页搜索新增图片结果Your app can now search the web for images. Web search in the Responses API now supports image results in addition to text results, so you can build apps that surface products, places, visual references, and source links for inspiration. https://t.co/Oyl4cS4JduRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)BuildersWhat it feels like to work with MythosClaude Fable 发布：Anthropic 带来的另一种推理体验Claude Fable represents another big jump in AIRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.www.oneusefulthing.orgIndustryApollo， Blackstone Fund AI BoomApollo 与 Blackstone 联手 350 亿美元 AI 融资交易Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.bloomberg.comAI ProductsThe Ray3.2 API runs cinematic-grade at scale and integrates into the products you already build. Mad…Luma AI Ray3.2 API：电影级渲染可集成The Ray3.2 API runs cinematic-grade at scale and integrates into the products you already build. Made for developers, agencies, and enterprises building cinema inside the products they ship. Start building → https://t.co/lLdBQHKYzO https://t.co/UVF23wLmJoRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ProductsThe creativity and imagination is out of the world！ So grateful that @theworldlabs got to partner wi…World Labs与Lore合作打造互动体验The creativity and imagination is out of the world! So grateful that @theworldlabs got to partner with the amazing talents @withloreco to translate their incredible ideas into an interactive experiences for users to enjoy!🤩Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)BuildersWant to use OpenRouter with Cursor？ Here's an integration guide： https：//openrouter.ai/docs/cookboo…OpenRouter与Cursor集成指南Want to use OpenRouter with Cursor? Here's an integration guide: https://t.co/4zSxoUPeJV https://t.co/cdXPjQopkfRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)BuildersGemini 2.5 Flash API - Pricing， Quickstart & Provider ComparisonGemini 2.5 Flash API - 定价、快速入门与提供商比较Overpaying for reasoning you don't need? Learn how to configure Gemini 2.5 Flash API thinking budgets, compare providers, and make your first call in 5 minutes.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.openrouter.aiBuildersFrom one-off prompts to workflows： How to use custom agents in GitHub Copilot CLIGitHub Copilot CLI 推出自定义 AI 智能体，将一次性终端提示转化为可重复工作流Custom agents let GitHub Copilot CLI understand your stack and team workflows, turning one-off terminal prompts into repeatable, reviewable processes.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.The GitHub BlogAI ModelsIntroducing North Mini Code： Cohere's First Model For DevelopersCohere发布North Mini Code：面向开发者的开源编码模型A Blog post by Cohere Labs on Hugging FaceRecommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.huggingface.coIndustryAsia's Largest Outsourcer to Slow Hiring as AI Reshapes Industry塔塔咨询服务将因AI智能体应用放缓招聘，亚洲外包业迎来转折Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.bloomberg.comAI ModelsSay hello， hola， 你好 to Gemini 3.5 Live Translate： our latest audio model built for fast， cross-langu…Gemini 3.5 Live Translate 发布Say hello, hola, 你好 to Gemini 3.5 Live Translate: our latest audio model built for fast, cross-language communication. 🌐 https://t.co/SEfHOSk59kRecommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.X (formerly Twitter)AI ModelsConfirmed， Claude Mythos will be unveiled in the next few hoursClaude Mythos 即将发布，Fable 精简版同日登场Confirmed, Claude Mythos will be unveiled in the next few hours https://t.co/kknquwshUNRecommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.X (formerly Twitter)AI ModelsIntroducing Gemma 4 12B： a unified， encoder-free multimodal modelGoogle DeepMind 发布 Gemma 4 12B：统一的无编码器多模态模型An overview of Gemma 4 12B, a model designed to bring high-performance multimodal intelligence directly to your laptop.Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.GoogleIndustryPowering the future of robotics in EuropeGoogle DeepMind 欧洲机器人加速器启动，15家初创公司入选Google DeepMind Accelerator selects 15 robotics companies from across Europe to join the program. Providing 3 months of intensive mentorship and technical support, enabl…Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.GoogleIndustry全新汽车品牌AIVA发布，火山引擎助力打造AI汽车新体验AI定义汽车，先有AI，再有车Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.Weixin Official Accounts PlatformIndustry百度搭子DuMate获中国信通院企业级Claw能力评估最高4+级Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.Weixin Official Accounts PlatformBuildersHow engineers at Nextdoor use Codex to build without limitsNextdoor 工程师借助 Codex 与 GPT-5.5 无限制构建Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.openai.comBuilders🚀Introducing UniRL， an RL infra for unified multimodal models. Together with two new RL algorithms：…腾讯混元发布UniRL：统一多模态强化学习基础设施🚀Introducing UniRL, an RL infra for unified multimodal models. Together with two new RL algorithms: DRPO and Flow-DPPO. One RL loop across diffusion/flow matching models, LLMs/VLMs, and unified multimodal models👇 Code: https://t.co/fhKEqqFpc8 (yes — U(you)-ni-(need) RL 😉) https://t.co/IUaQxXqCv9Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)AI Products火山引擎TRAE Work企业版正式上线，面向全员提供AI办公平台Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.Weixin Official Accounts PlatformBuildersHow an Agent Built a 3D Paris Gallery by Chaining Two Hugging Face Spaces一个Agent如何通过链式调用两个HuggingFace Space构建3D巴黎画廊A Blog post by Mishig Davaadorj on Hugging FaceRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.huggingface.coIndustryTaiwan Mulls Curbs on AI Chip Exports to China to Align With US台湾考虑限制AI芯片对华出口以配合美国Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.bloomberg.comBuilders仅凭一份文档，Qwen3.7-Max 从 0 交付双端应用实践经验详解Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.Weixin Official Accounts PlatformBuildersNeuroBait： I fine-tuned a model to spark dopamine for ADHD brainNeuroBait：微调AI助手，为ADHD大脑点燃多巴胺火花A Blog post by Build Small Hackathon on Hugging FaceRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.huggingface.coBuildersNVIDIA cuTile Python Tutorial： Building Tiled GPU Kernels for Vector Addition， Matrix Addition， and Matrix Multiplication in ColabNVIDIA cuTile Python 教程：在 Colab 中构建用于向量加法、矩阵加法和矩阵乘法的 Tiled GPU 内核Build tile-based GPU kernels in NVIDIA cuTile Python for vector addition, matrix addition, and matrix multiplication, with a PyTorch fallbackRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.MarkTechPostIndustryChina Prepares $295 Billion Plan to Fund Nationwide AI Buildout中国准备2950亿美元计划资助全国AI基础设施建设Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.bloomberg.comIndustryAI 编程独角兽 Cursor 欧洲总部落子伦敦，SpaceX 手握 600 亿美元收购选择权AI 编程工具 Cursor 将欧洲总部设在伦敦，计划招聘约 200 人。公司 B2B 年化营收约 26 亿美元，服务英国航空、诺基亚等客户。同时，SpaceX 握有价值 600 亿美元的收购选择权。Cursor 主打模型中立，可自由选用不同 AI 系统。#AI 编程##科技公司动态##SpaceX#Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.ithome.comAI Models小米 MiMo 与 TileRT 联合发布 UltraSpeed 模式，1T 模型输出突破 1000 tokens/s天下武功，唯快不破Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.Weixin Official Accounts PlatformIndustryOpenAI 秘密提交 IPO 申请，奥特曼旗下 Tools for Humanity 裁员OpenAI 已秘密提交 IPO 申请，可能成为近十年标志性上市事件。与此同时，其 CEO 山姆 · 奥尔特曼的另一家公司 Tools for Humanity 正在进行裁员，该公司旗下 Worldcoin 项目因监管与隐私问题陷入困境。#OpenAI# #Worldcoin#Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.ithome.comIndustryFor the very first time Elon Musk explains the "space data center plan" of @SpaceX in detail and its…Elon Musk 详解 SpaceX AI1 轨道 AI 数据中心卫星方案For the very first time Elon Musk explains the "space data center plan" of @SpaceX in detail and its AI1 orbital AI data center satellite - and suddenly it looks so much closer than I thought. He says "There’s not some magic necessary that doesn’t exist for AI satellites. As https://t.co/PXWs5TheZoRecommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.X (formerly Twitter)Builders开源工具 Tokei：在菜单栏实时监控 AI coding agent 的 token 用量与成本兄弟们！地主家家没有余粮了都！ 天天烧Token 心里没有点b数啊？ AI coding工具天天帮你狂飙代码，结果你连自己到底烧了多少钱都蒙在鼓里？ 今天给大家推荐Lank 的Tokei这个macOS菜单栏小工具给你直接轻松拿捏它！ 对了！开源免费啊！记得给Star啊！ https://t.co/efAGy68v7JRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)Industry两部门：到2026年底人形机器人等重点产品完成应用验证并常态部署工业和信息化部办公厅、国务院国资委办公厅 6 月 8 日发布关于联合开展 2026 年度人形机器人与具身智能实景实训专项行动的通知。Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.ithome.comBuildersFrontierCode 基准测试：AI 编程评估新标准--维护者审核通过率最高仅 13.4%Claude Opus 4.8 是目前最好的编码模型，这件事应该没啥太大争议了，我自己跑了这么久体感也是这样。 Cognition（Devin 的公司）刚发布的 FrontierCode 基准测试，彻底改变了 AI 编程能力的评判标准： 不再只看“代码能不能跑过测试”，核心看看“维护者会不会愿意把这段代码合并进真实项目”。 https://t.co/aqTv5aIe4ERecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)BuildersGitHub 122K⭐的Skills推出新技能「Teach」：把工作目录变有状态学习空间Github 122K ⭐️ 的 Skills 仓库「Skills For Real Engineers」推出新 Skill「Teach」：把当前工作目录变成有状态的学习空间！！怒赞作者 @mattpocockuk 👍🏻 开源地址： https://t.co/RCqjtEfwQ9 Teach Skill 设计理念：Knowledge → Skills → Wisdom · https://t.co/Zadl5xrIbQRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)Researchi1： A Simple and Fully Open Recipe for Strong Text-to-Image Modelsi1：面向强文生图模型的简单且完全开源配方Diffusion models have consistently driven progress in text-to-image generation. However, it is challenging to attribute recent progress to specific modeling and data choices: state-of-the-art open-weight models provide limited ablations, and do not disclose their training data and full training details. The research community needs fully open (weights, data, and code) models as a foundation for further research; yet existing fully open models still fall significantly short of leading models in performance. In this project, we conduct a systematic investigation of the modeling and data design choices in text-to-image diffusion training and inference with 300+ controlled experiments totaling 700K+ TPU v6e hours. Our experiments highlight several empirical findings (e.g., equal weighting is a strong default for mixing curated datasets) and simple design decisions (e.g., larger text encoder adapters improve performance with minimal added parameters) for training strong models. Guided by these insights, we train i1, a 3B-parameter text-to-image diffusion model using only publicly available datasets. i1 is competitive with leading models on five representative benchmarks (GenEval, DPG, PRISM, CVTG-2K, and LongText), and outperforms the best existing fully open model by 29.5 absolute percentage points on average. We provide the i1 checkpoints, training and inference code, and the data processing pipeline. Together, our findings and the i1 recipe establish a practical foundation for future open research in text-to-image diffusion models. Our code is available at https://github.com/zlab-princeton/i1.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgResearchEmbodied-R1.5： Evolving Physical Intelligence via Embodied Foundation ModelsEmbodied-R1.5：通过具身基础模型演化物理智能We introduce Embodied-R1.5, a unified Embodied Foundation Model (EFM) that integrates comprehensive embodied reasoning capabilities, spanning embodied cognition, task planning, correction, and pointing, within a single architecture toward general physical intelligence. Leveraging three automated data construction pipelines to significantly expand the data coverage of critical capabilities, we build a large-scale data system of over 15B tokens, and design a multi-task balanced RL recipe to alleviate heterogeneous task conflicts. We further introduce a Planner-Grounder-Corrector (PGC) closed-loop framework that enables a single model to autonomously execute and self-correct over long-horizon tasks. With only 8B parameters, Embodied-R1.5 achieves SOTA on 16 out of 24 embodied VLM benchmarks, surpassing leading models like Gemini-Robotics-ER-1.5 and GPT-5.4. Benefiting from the internalized embodied capabilities, Embodied-R1.5 can be fine-tuned into a VLA with only a small amount of data, outperforming leading VLA models like $π_{0.5}$ across 4 popular manipulation benchmark suites. We further conduct extensive zero-shot real-robot experiments, validating performance in instruction following, affordance grounding, articulated object manipulation, and long-horizon complex tasks, demonstrating strong generalization to the physical world. We open-source model weights, datasets, training code, and EmbodiedEvalKit, an evaluation framework tailored for embodied tasks, to facilitate future research in EFMs.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgResearchDecentralized Multi-Agent Systems with Shared ContextDeLM：去中心化多智能体系统框架Multi-agent systems (MAS) can scale large language model reasoning at test time by decomposing complex problems into parallel subtasks. However, most existing MAS rely on centralized orchestration, where a main agent assigns work, collects outputs, and merges results. As the number of subtasks grows, this controller becomes a communication and integration bottleneck. We propose Decentralized Language Models (DeLM), a MAS framework that decentralizes coordination through parallel agents, a shared verified context, and a task queue. Agents asynchronously claim subtasks, read accumulated progress, perform local reasoning, and write back compact verified updates. The shared context acts as a common communication substrate, enabling agents to build on one another's verified progress without routing every update through a central controller. Empirically, DeLM improves both software-engineering test-time scaling and long-context reasoning. On SWE-bench Verified, DeLM achieves the best performance across Avg.@1, Pass@2, and Pass@4, with gains of up to 10.5 percentage points over the strongest baseline, while reducing cost per task by roughly 50%. On LongBench-v2 Multi-Doc QA, DeLM achieves the highest average accuracy across four frontier model families, improving over the strongest baseline by up to 5.7 percentage points. The code is available on our project website at https://yuzhenmao.github.io/DeLM/.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgResearchKwai Keye-VL-2.0 Technical Report快手开源 Kwai Keye-VL-2.0-30B-A3B：面向长视频理解与智能体智能的 MoE 多模态模型We introduce Kwai Keye-VL-2.0-30B-A3B, an open-source Mixture-of-Experts (MoE) multimodal foundation model designed to advance long-video understanding and agentic intelligence. To address the challenges of ultra-long contexts, information redundancy, and prohibitive computational costs inherent in hour-level videos, Keye-VL-2.0 is the first to adapt DeepSeek Sparse Attention (DSA) to GQA-based multimodal architectures, enabling lossless 256K context processing while capturing critical frames and long-range temporal dependencies. This architecture is underpinned by a highly optimized training and inference infrastructure, including scalable video I/O, heterogeneous ViT-LM parallelism, and custom DSA kernels that significantly maximize throughput and minimize computational overhead. Furthermore, to overcome the algorithmic dilemma of catastrophic forgetting during multi-task alignment, we introduce Cross-Modal Multi-Teacher On-Policy Distillation (MOPD) paired with Context-RL and Video-RL. By distilling dense token-level teacher feedback from on-policy rollouts back into the MoE backbone, which activates only 3B parameters, Keye-VL-2.0 natively empowers advanced agent collaboration across Code, Tool, and Search scenarios with multimodal self-correction. Extensive evaluations across video understanding, temporal grounding, reasoning, STEM, and agent benchmarks demonstrate that Keye-VL-2.0-30B-A3B achieves state-of-the-art performance among models of similar scale, particularly excelling in fine-grained temporal localization on TimeLens and long-video comprehension on Video-MME-v2 and LongVideoBench. We release our model checkpoints to accelerate community progress toward scalable and robust multimodal agentic applications.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgResearchAttention Amnesia in Hybrid LLMs： When CoT Fine-Tuning Breaks Long-Range Recall， and How to Fix It混合LLM中的注意力失忆：CoT微调破坏长距离召回及修复方法Chain-of-thought (CoT) supervised fine-tuning (SFT) is widely adopted to improve reasoning ability, yet we find that it systematically degrades long-context recall in hybrid linear-attention models. Across architectures including HypeNet and Jet-Nemotron, retrieval performance on Needle-In-A-Haystack (NIAH) deteriorates substantially after CoT-SFT, and the degradation becomes more severe under harder retrieval settings and longer context windows. For example, HypeNet-9B on NIAH-S2@256K decreases from $67.2\%$ to $9.4\%$. We attribute this to CoT-SFT biasing attention gradients toward short-range patterns, disrupting query-key projections ($W_Q, W_K$) that are responsible for long-range routing. Motivated by this observation, we propose QK-Restore, a training-free method that restores only $W_Q$ and $W_K$ from the pre-SFT checkpoint while preserving all other post-SFT parameters. We further introduce a Procrustes variant to balance routing preservation and reasoning adaptation. Across architectures, QK-Restore consistently restores long-context capability at zero training cost while preserving reasoning performance; for instance, on HypeNet-5B it improves S3@256K from $65.4\%$ to $76.4\%$ while maintaining strong reasoning performance.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgResearchFlow-DPPO： Divergence Proximal Policy Optimization for Flow Matching ModelsFlow-DPPO： 面向流匹配模型的散度近端策略优化Recent work has demonstrated that online reinforcement learning (RL) can substantially improve the quality and alignment of flow matching models for image and video generation. Methods such as Flow-GRPO and CPS cast the denoising process as a Markov Decision Process and apply PPO-style ratio clipping to enforce a trust region. However, we argue that ratio clipping is structurally ill-suited for flow models: the probability ratio between new and old policies is a noisy, single-sample estimate of the true policy divergence, leading to over-constraining in some regions of the trajectory and under-constraining in others. We propose Flow-DPPO (Flow Divergence Proximal Policy Optimization), which replaces ratio clipping with a divergence proximal constraint. A key observation is that the per-step policy in flow models is Gaussian, enabling exact and cheap computation of the KL divergence between old and new policies. Flow-DPPO employs an asymmetric divergence mask that blocks gradient updates only when they simultaneously move away from the trusted region and violate the divergence threshold. Experiments show that Flow-DPPO achieves higher rewards with better KL-proximal efficiency, alleviates catastrophic forgetting, promotes balanced multi-objective optimization, and enables stable multi-epoch training where ratio clipping degrades. Code and models are available at https://github.com/Tencent-Hunyuan/UniRL/tree/main/FlowDPPO.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgBuildersMigrating Your GitHub CI to Hugging Face Jobs将 GitHub CI 迁移到 Hugging Face JobsWe’re on a journey to advance and democratize artificial intelligence through open source and open science.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.huggingface.coAI ProductsPowering Gopuff's Go agent Jun 9， 2026 # Powering Gopuff's Go agent Gopuff and SpaceXAI launched Go， an AI-powered shopping assistant built into the Gopuff app and powered by Grok text， audio， and image models. Read MoreGopuff与SpaceXAI推出Go AI购物助手Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.x.aiAI ProductsApple Core AI 框架Run AI models in your app on Apple silicon.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.Apple Developer DocumentationIndustry奥尔特曼宣布 OpenAI 进入第三发展阶段：让 AI 普及、易用且安全OpenAI 宣布公司发展进入第三阶段，核心是让先进 AI 技术变得普及、易用且安全。同时，公司已秘密提交 IPO 申请，但上市仍需等待。文中还呼吁成立国际机构应对 AI 风险。#OpenAI# #人工智能#Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.ithome.comBuildersSam Altman's new blog about OpenAI's future path says by March-2028 a significant fraction of its ow…OpenAI计划到2028年由AI主导研究Sam Altman's new blog about OpenAI's future path says by March-2028 a significant fraction of its own research will be done by AI. The path has 3 goals mainly: build an automated AI researcher, use that to speed up science and productivity, then give every person a personal AGI https://t.co/oF1mMAWsuwRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)AI ProductsIntroducing the Viggle API. Give any character any motion， one API call - alive in seconds. Wire …Viggle API 上线：任意角色任意动作秒级生成Introducing the Viggle API. Give any character any motion, one API call - alive in seconds. Wire it into Claude, Codex, or any agent you're building. Starting from $0.01/sec. Get 100 free credits on signup. RT + follow + comment, 10 winners get 100 more! Learn more below👇 https://t.co/P9PARmzQ40Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ProductsApple unveils next generation of Apple Intelligence， Siri AI， and moreApple发布新一代Apple Intelligence和Siri AIToday, Apple previewed its upcoming software releases that will deliver the next generation of Apple Intelligence and introduce Siri AI.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.Apple NewsroomAI ProductsApple Intelligence brings powerful AI capabilities into everyday experiencesApple Intelligence 将强大 AI 能力融入日常体验Apple unveils the next generation of Apple Intelligence, integrating powerful AI capabilities into iPhone, iPad, and Mac for more personal and helpful everyday experiences.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.Apple NewsroomIndustry苹果 WWDC 2026 直播Discover all-new Siri AI powered by Apple Intelligence and helpful features across iOS 27, iPadOS 27, macOS Golden Gate, watchOS 27, and visionOS 27.Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.AppleIndustryDue to DMA， Siri AI delayed in EU for iOS 27 and iPadOS 27受 DMA 影响，Siri AI 在欧盟将随 iOS 27 和 iPadOS 27 延迟上线Due to the Digital Markets Act, Apple will not be able to ship Siri AI in the European Union with the release of iOS 27 and iPadOS 27.Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.Apple NewsroomBuildersThe sample efficiency black hole样本效率黑洞：AI能力背后隐藏的数据需求深渊Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.www.dwarkesh.comBuildersClaude Code's first demo got two Slack reactions. One year after GA， @bcherny and @_catwu look back…Claude Code GA一周年回顾：验证与自动模式Claude Code's first demo got two Slack reactions. One year after GA, @bcherny and @_catwu look back: verification best practices, why we built auto mode, routines and loops, and what's next. https://t.co/yEa3cmCrg4Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)ResearchWe published new research with Harvard on the shift from chat interfaces to autonomous agents like C…Perplexity与哈佛：AI智能体提效87%降本94%We published new research with Harvard on the shift from chat interfaces to autonomous agents like Computer. Over 3 months, findings show workers using Computer finish tasks in 87% less time at 94% lower cost than Search alone, with higher satisfaction. https://t.co/qmcUqcj8CI https://t.co/R4oTLavC6TRecommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.X (formerly Twitter)AI ProductsTurn data and comparisons into charts， directly in ChatGPT. Available now on mobile and web.ChatGPT 新增数据图表生成功能Turn data and comparisons into charts, directly in ChatGPT. Available now on mobile and web. https://t.co/rZ7KJsvXBwRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ProductsIntroducing a more powerful NotebookLM 🚀 Massive upgrades deliver agentic capabilities in chat， mo…NotebookLM重大升级：智能体能力与高级推理Introducing a more powerful NotebookLM 🚀 Massive upgrades deliver agentic capabilities in chat, more advanced reasoning, and a suite of new output formats. Tackling complex, multi-step research problems has never been easier. Rolling out now to Google AI Ultra subscribers. https://t.co/zBXD7unIC7Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ProductsNew in Claude Managed Agents： run agents on a schedule and store environment variables in vaultsClaude Managed Agents 新增定时运行和环境变量存储功能Claude Managed Agents can now run on a schedule and securely access CLI tools and other authenticated services.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.ClaudeAI ModelsClaude Fable 5 and Claude Mythos 5Claude Fable 5 和 Claude Mythos 5Today we’re launching Claude Fable 5: a Mythos-class model that we’ve made safe for general use.Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.www.anthropic.comAI ProductsOne video， now made for every feed and format. Upload your existing video， choose your desired aspec…Runway Aleph 2.0 编辑模型：一键适配任意视频格式One video, now made for every feed and format. Upload your existing video, choose your desired aspect ratio and watch our editing model, Aleph 2.0, fill in the rest of the scene as if you made it that way from the start. Try it on our desktop web app at the link below. https://t.co/EdPkUEc2BSRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)BuildersNew from Hivemind： continual learning for AI coding agents， available to everyone starting today. I…Hivemind推出面向AI编程智能体的持续学习功能，即日起开放New from Hivemind: continual learning for AI coding agents, available to everyone starting today. It takes the traces from every agent your team runs (Claude Code, Codex, Cursor, Hermes, Pi) and turns them into reusable skills, then pushes those skills across all of them, all onRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)AI Models🚀 VoxCPM2 Technical Report is now available on arXiv！ VoxCPM2 is the latest speech generation mode…VoxCPM2 技术报告发布🚀 VoxCPM2 Technical Report is now available on arXiv! VoxCPM2 is the latest speech generation model in the VoxCPM family. Built with 2B parameters and trained on over 2 million hours of multilingual speech data, it supports 30 languages and 9 Chinese dialects, along with https://t.co/oTgdg9Uyb3Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.X (formerly Twitter)IndustryConfidential submission of draft S-1 to the SECOpenAI 向 SEC 机密提交 S-1 草案，上市时间未定Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.openai.comBuildersMicrosoft's AI chief says superintelligence is near， but won't take your job微软AI CEO：超级智能即将到来，但不会取代你的工作Mustafa Suleyman thinks superintelligence is near, but won’t take your job.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.The VergeAI ProductsKimi Code 焕新升级（附视频教程）Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.mp.weixin.qq.comBuildershttp：//x.com/i/article/2063968924019163136小互开源视频翻译工具：一句话自动下载、转写、翻译、烧字幕https://t.co/uYh4KNpDqsRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)BuildersThe crash that vanished： control and emergence in a five-model economy五个模型经济体中消失的崩溃：控制与涌现A Blog post by Build Small Hackathon on Hugging FaceRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.huggingface.coResearchMeasuring the impact of learning with AI in Sierra Leone and beyondGemini Guided Learning 随机对照试验：在塞拉利昂等地提升参与度并加速学习Google DeepMind shares results from a randomized controlled trial in Sierra Leone, measuring the impact of AI in education on student learning and engagement.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.Google DeepMindAI ProductsEU AI Act Compliance： Human Oversight for AI AgentsEU AI Act 合规：面向 AI 智能体的人工监督Use the Agent SDK's human-in-the-loop (HITL) tools to meet AI agent compliance requirements from the EU AI Act, Colorado's Automated Decision-Making TechnologyRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.openrouter.aiResearchReasoning Arena： Trace Tournaments When Verifiable Rewards Fall ShortReasoning Arena：可验证奖励不足时的迹线锦标赛Reinforcement learning with verifiable rewards (RLVR) has become a leading paradigm for improving the reasoning ability of large language models through outcome-based supervision. However, verifiable rewards frequently become uninformative at the group level: when all sampled traces of a given prompt receive identical rewards, group-relative advantage estimation provides no gradient signal, even though the traces may differ substantially in reasoning quality. We propose Reasoning Arena, an adaptive training framework that routes such non-diverse reward groups to a judge system instead of discarding them. Beyond examining the final answer, Reasoning Arena constructs trace tournaments, where reasoning traces are compared head-to-head to expose finer-grained preferences within the group, converting reasoning quality into rich relative reward signals. To make reward estimation efficient, rather than exhaustively comparing every pair, each new trace is evaluated against a small, dynamically updated pool of previously generated traces as anchors to efficiently establish a relative ranking. We then fit a Bradley-Terry model on the incomplete comparison graph, enabling scalable RL integration without quadratic pairwise comparisons. Empirical results demonstrate that Reasoning Arena consistently outperforms the RLVR baseline by 7.6% on average in competition mathematics and coding benchmarks. By converting otherwise wasted zero-advantage samples into useful gradient updates, our method accelerates training by 27% to 41%, saving nearly 50% of generation compute, and substantially improves overall reasoning performance.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgBuildersBuilding Pakistan Notice Helper： A Small AI Tool for a Very Local Safety ProblemPakistan Notice Helper：一款面向本地安全问题的轻量 AI 工具A Blog post by Build Small Hackathon on Hugging FaceRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.huggingface.coBuildersAgent 辅助开发，一站式打通 Qwen3-VL Android 端侧推理Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.mp.weixin.qq.comBuilders微信AI Agent生态曝光：嵌入小程序调用与手机厂商合作Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.mp.weixin.qq.comIndustry生数科技与华策影视达成战略合作，共建AI视听创制中心Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.mp.weixin.qq.comBuilders最近看了不少 Design Skill、Taste Skill、Anti-AI-slop design skill 等等，我自己也开源了一个 Brand to DESIGN.md Skill （htt…邵猛开源 Brand to DESIGN.md 技能，提醒复刻易生新"AI Slop"最近看了不少 Design Skill、Taste Skill、Anti-AI-slop design skill 等等，我自己也开源了一个 Brand to DESIGN.md Skill (https://t.co/uQhFFEwiCe) 目的都是学习借鉴优秀的设计、积累设计品味，让 Agent 去学习沉淀到 DESIGN.md 再复刻生成新的网站。 但是这种复刻看多了，就又从 Anti-AI-slopRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)AI Products微信AI官宣内测：两种接入模式供开发者选择微信开放平台为开发者提供了接入微信 AI 生态的能力，提供两种接入模式，开发者可按需选择，满足不同规模团队的开发需求（两种模式不互斥，可同时开启）。Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.www.ithome.comBuilders<strong>How CoreWeave Sees the Market for Compute Right Now </strong>CoreWeave 如何看待当前计算市场Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.www.bloomberg.comResearchCan AI truly edit audio， not just generate it？ 🎧 Tencent Hy， in collaboration with SJTU， SII， NTU，…腾讯混元联合多家机构发布首个音频编辑基准MMAECan AI truly edit audio, not just generate it? 🎧 Tencent Hy, in collaboration with SJTU, SII, NTU, TJU, ZODA, PKU, FDU, and other collaborators, introduces MMAE. MMAE--A Massive Multitask Audio Editing Benchmark, is the first comprehensive evaluation benchmark for speech and https://t.co/k5G4bicrOqRecommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.X (formerly Twitter)AI Models全球首个：高德发布3D原生城市世界模型ABot-Earth0.5该模型已建成覆盖 190 多个国家的全球最广 3D 地图，输出素材可直接导入主流游戏引擎。其制图成本仅为传统 1%，效率提升约千倍，有望为具身智能、低空经济及应急救援提供基础支撑。#高德地图# #3D 建模#Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.www.ithome.comResearchHardening Agent Benchmarks with Adversarial Hacker-Fixer Loops用对抗性黑客-修补循环强化Agent基准测试Agent benchmarks score submissions with outcome verifiers that are typically hand-written and brittle, leaving them open to reward hacking. We audit 1,968 tasks across five terminal-agent benchmarks and find 323 (16%) hackable by frontier models given only the task description. This corrupts both leaderboard rankings and RL training signal, yet the standard response is manual and reactive. We introduce the hacker-fixer loop, a method for building exploit-resistant verifiers without per-task manual patching. The loop alternates three LLM agents: a hacker tries to pass the verifier without solving the task, a fixer patches the verifier to reject each discovered exploit, and a solver confirms the patched verifier still admits legitimate solutions. The loop iterates: each patch reshapes what the verifier rewards, surfacing the next exploit. We further add verifier access, and let patches transfer across tasks, to broaden the exploits the loop discovers. On KernelBench, the loop drives the attack success rate from 62% to 0% on a held-out corpus of publicly reported exploits. We also find that weaker agents in the loop can defend against much stronger hackers: Gemini 3 Flash's loop drives the stronger Gemini 3.1 Pro and Claude Opus 4.7's attack success rate from 76% and 61% to 0% on KernelBench, and Gemini 3.1 Pro's from 39% to 17% on Terminal Bench across 77 tasks. We release Terminal Wrench (323 hackable environments, 3,632 hack trajectories) as a snapshot of the current attack surface, our patched verifiers, the exploits the loop discovered, and our implementation as a basis for future work.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgBuildersBuilt to benefit everyone： our planOpenAI 公布让 AGI 造福所有人的计划Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.openai.comResearchPrecision Is Not Faithfulness： Coverage-Aware Evaluation of Grounded Generation with a Complete Oracle精确性不等于忠实度：完整Oracle下的覆盖感知接地生成评估Reference-free faithfulness metrics verify each atomic claim a model makes against ground truth, and are increasingly used to evaluate grounded generation. We show they share a blind spot: they measure only precision -- are the stated claims supported? -- and therefore reward abstention, since a model can score near-perfect faithfulness by saying almost nothing. We make this measurable using Formula 1 telemetry, a domain where strategic ground truth is derived deterministically and, crucially, completely: for each decision we know the full set of facts that mattered. This completeness -- absent in open-domain faithfulness benchmarks -- lets us measure recall (coverage of the relevant facts) exactly, alongside precision. On a multilingual (EN/ES/PT) benchmark of 7,253 decision instances spanning 150 races, the most precise frontier model covers under half of the relevant facts and ranks last by F1, so requiring coverage reorders the systems; the same effect reappears in a second complete-oracle domain (NOAA weather forecasts). A prompt ablation shows the low coverage is not an under-prompting artifact: explicitly asking models to be thorough does not close the gap. We pair faithfulness with coverage into a single score, validate the metric (controlled perturbation; agreement across a model-free regex extractor and a cross-family LLM extractor, system-level Spearman 1.0), and give a verifier-guided generation method that improves precision and recall without references. We release the benchmark, structured annotations, metric, baselines, and an interactive demo.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgResearchOmniGameArena： A Unified UE5 Benchmark for VLM Game Agents with Improvement DynamicsOmniGameArena：面向VLM游戏智能体的统一UE5基准与改善动态Vision-language model (VLM) agents are increasingly deployed in interactive game environments. Yet game benchmarks for VLM agents typically report a single first-attempt score per (agent, game) pair, focus on single-agent Solo play, and lack unified protocols for evaluating heterogeneous agent classes (commercial VLMs, open-weight VLMs, and specialized game policies) on the same footing. We address these gaps with OmniGameArena, a real-time benchmark of twelve newly built Unreal Engine 5 games spanning Solo (7), PvP (3), and Coop (2) with unified action interfaces, and the Improvement Dynamics Curve (IDC), an agentic-reflection harness in which a tool-using reflector LLM autonomously refines a bounded skill prompt across multiple rounds. Beyond cold-start leaderboard scores, IDC exposes two additional observables for each (agent, game) pair: how the score evolves across reflection rounds, and how the learned skill behaves on held-out task variants. We report these observables for twelve VLM agents on the cold-start leaderboard and four top agents under IDC.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgAI ModelsIntroducing the Third Generation of Apple's Foundation Models苹果发布第三代 Apple Foundation Models（AFM）Our next generation of Apple Intelligence is centered around our users, integrated deeply into our operating systems, and powered by a bold…Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.Apple Machine Learning ResearchAI ProductsThe Open Source Community is backing OpenEnv for Agentic RL开源社区支持 OpenEnv 用于智能体强化学习We’re on a journey to advance and democratize artificial intelligence through open source and open science.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.huggingface.coIndustryNvidia， SK Hynix Sign Multi-Year Pact to Develop Next-Gen ChipsNvidia 与 SK Hynix 签署多年协议，共同开发下一代 AI 存储芯片Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.bloomberg.comIndustryChatGPT 要变 AgentGPT 了ChatGPT 要变 AgentGPT 了 当然 ChatGPT 应该不会改名字，但 ChatGPT 应该不再是一个单纯的 Chat 工具了。 OpenAI 内部一位高管对《金融时报》说："Chat is dead."（聊天已死。） OpenAI 正在准备 ChatGPT 自 2022Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.X (formerly Twitter)IndustryFT： The proposal suddenly of a sovereign-wealth-style fund got more attention inside the White House…特朗普政府与OpenAI讨论通过公共财富基金入股AI初创公司FT: The proposal suddenly of a sovereign-wealth-style fund got more attention inside the White House after Sam Altman visited Capitol Hill this week. The likely mechanism is that AI firms donate a small slice of equity into a public wealth fund, and that fund passes gains to https://t.co/FODoK3tBNfRecommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.X (formerly Twitter)IndustryOpenAI is still working on that 'super app'OpenAI 仍推进超级应用计划"Chat is dead" — at least, according to a senior OpenAI employee.Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.TechCrunchAI ProductsBuilding intelligent apps for Apple platforms with Claude in the Foundation Models frameworkClaude 支持 Apple Foundation Models 框架，推出新 Swift 包A new Swift package connects Apple's Foundation Models framework to Claude. Hand off complex reasoning from on-device models with typed Swift outputs.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.ClaudeAI ProductsObservability for developers building connectorsClaude 为 Connector 开发者推出性能监控仪表盘Monitor connector performance across Claude, diagnose errors and latency, and submit your MCP server to the directory in-app. Public beta now live.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.ClaudeResearchPaving the way for agents in biology为生物学AI智能体铺路Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.www.anthropic.comBuildersSlop， productivity， and why the AI-fueled world is going nowhere mighty fastSlop、生产力，以及为何AI驱动的世界进展甚微Just saw a graph at the FT from John Burn-Murdoch that really distills something I have been trying to articulate.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.garymarcus.substack.comBuildersInside Apple's Secret Meeting That Led It to Finally Take AI Seriously苹果秘密会议内幕：它终于认真对待AIRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.www.bloomberg.comBuildersSymbolica 2.0：适用于 Python 和 Rust 的可编程符号系统Technical articles and release notes about Symbolica, symbolic computation, numerical methods, and high-performance computer algebra.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.Symbolica | Modern Computer AlgebraBuildershttp：//x.com/i/article/2063531614047444992"我在田里雇了一名工程师，它叫 Codex" -- 北海道一个西兰花农的 8 个真实 AI 用法https://t.co/5gXPmym8vLRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)AI ProductsHer · हेर - a detective for your Claude Code sessionsHer · हेर - Claude Code 会话分析工具A Blog post by Build Small Hackathon on Hugging FaceRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.huggingface.coAI ModelsMeet Harness-1： A 20B Retrieval Subagent Trained With Reinforcement Learning Inside a Stateful Search Harness on gpt-oss-20bHarness-1：基于强化学习训练的有状态搜索20B检索子智能体Harness-1 is a 20B search agent reaching 0.730 average curated recall across eight benchmarks, behind only Opus-4.6.Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.MarkTechPostBuildersHarness 工程：在智能体优先的世界中运用 CodexRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.openai.comBuilders对比一下 GPT-5.5 的设计效果和 Opus 4.8 的设计效果对比一下 GPT-5.5 的设计效果和 Opus 4.8 的设计效果 我真不是尬黑 GPT-5.5，我这种审美水平都能看出来差距 使用 Skill：https://t.co/7BdakDaEVn ---- 提示词 ---- /baoyu-design 帮我开发一款Reader Mac App，帮助我更好的阅读和收藏文章。数据都在本地。 ## 信息采集 1. 主动添加 https://t.co/47cVZuk9UZRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)BuildersThe Substitution Wave in AIAI 替代浪潮：三大力量重塑成本结构Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.www.tomtunguz.comIndustry美国众议院议员发布法案草案，旨在禁止各州制定人工智能相关法规Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.reuters.comBuildersFive labs， five minds： building a multi-model finance drama on small models五个实验室，五个心智：用小模型构建多模型金融剧情游戏A Blog post by Build Small Hackathon on Hugging FaceRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.huggingface.coBuildersAI's Black FridayAI 的黑色星期五Some thoughts on what just happenedRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.garymarcus.substack.comBuildersJob SearcherA Blog post by Build Small Hackathon on Hugging FaceRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.huggingface.coAI ProductsGitHub released Spec Kit， an open-source toolkit to fix vibe coding's biggest weakness： the AI often…GitHub 开源 Spec Kit 工具包，用产品规范引导 AI 编码GitHub released Spec Kit, an open-source toolkit to fix vibe coding’s biggest weakness: the AI often starts coding before the product rules are clear. 109K+ stars ⭐️ It turns vibe coding from “ask the AI to build it” into “write the product spec first, then make the AI build https://t.co/IEHhh88FyvRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ProductsOpenCV 5 发布：升级全新 DNN 引擎、原生支持大模型该库在 GitHub 上拥有超过 86,000 颗星，每天的安装量超过一百万次，并且拥有世界上最庞大的计算机视觉算法集合之一。Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.www.ithome.comAI ProductsPersona Atlas： Mapping How Famous Minds ThinkPersona Atlas：Hugging Face 上的开源人物思维映射工具A Blog post by Build Small Hackathon on Hugging FaceRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.huggingface.coIndustry阶跃首席科学家张祥雨合著论文 ResNet 获 CVPR 2026 「时间检验奖」向所有共同获奖作者表示祝贺Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.Weixin Official Accounts PlatformBuildersThousand Token Wood： shipping a multi-agent economy on a 3B model用Qwen2.5-3B构建多智能体经济体：工程报告A Blog post by Build Small Hackathon on Hugging FaceRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.huggingface.coResearchArena just released a real-world agent leaderboard that ranks AI models by how well they complete ac…Arena 发布真实世界 AI 智能体排行榜 Agent ArenaArena just released a real-world agent leaderboard that ranks AI models by how well they complete actual user jobs, not isolated benchmark questions. The system tracks agents using web search, files, and terminal tools while people ask them to write code, build apps, research https://t.co/GYT9ttQXGCRecommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.X (formerly Twitter)IndustryApollo Wraps Up $35 Billion Debt to Buy AI Chips for AnthropicApollo 敲定 350 亿美元债务融资，为 Anthropic 采购 AI 芯片Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.bloomberg.comIndustrySpaceX just disclosed a new Cloud Service Agreement with Google. Google to pay SpaceX $920 million …SpaceX与Google达成云计算新协议SpaceX just disclosed a new Cloud Service Agreement with Google. Google to pay SpaceX $920 million a month (about $11B a year) for compute capacity at xAI data centers Shows again AI compute is becoming a strategic commodity like launch capacity or energy, and the companies https://t.co/gvN2Nzaz5hRecommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.X (formerly Twitter)Industry五角大楼正运营着一个针对拉丁美洲的人工智能宣传机器La Tilde publishes an unusual mix of personal finance guides and articles extolling American military efforts in Latin America.Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.The InterceptBuildersClaude 是否增加了 rsync 中的错误？Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.alexispurslane.github.ioAI ProductsWorking with agents should feel like working with a colleague. You should be able "speak to" them no…智能体协作应如同事般对话和手势Working with agents should feel like working with a colleague. You should be able “speak to” them not just with text chats, but by gesturing at a screen together, talking live, etc.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ProductsDraft it. Tweak it. Send it. You can now send emails directly from writing blocks in ChatGPT on the…ChatGPT 网页版支持从写作块发送邮件Draft it. Tweak it. Send it. You can now send emails directly from writing blocks in ChatGPT on the web, without leaving the conversation. https://t.co/GoQtlSFGFGRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ProductsHere's this week's shipping recap 👇 - Nano Banana 2 & Nano Banana Pro are now GA and available via…Google AI 本周产品更新：Nano Banana 2、Co-Scientist、dreambeans、Gemma 4 等Here’s this week’s shipping recap 👇 — Nano Banana 2 &amp; Nano Banana Pro are now GA and available via the Gemini Enterprise Agent Platform, Gemini API, and in @GoogleAIStudio —Co-Scientist, our new multi-agent system for structured scientific thinking, generates and refines novelRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ProductsYou can now create and edit images directly in Gemini Live. Whether testing out room decor， getting…Gemini Live 支持实时创建编辑图像You can now create and edit images directly in Gemini Live. Whether testing out room decor, getting help with math, or creating shareable memes, it all happens in real-time. Just open the Gemini app, tap the Live button, share your camera, and tell Gemini what you want to see.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)IndustryThe AI boom has doubled computing infrastructure's share of US GDP. Investment in AI-related data c…AI热推高美国计算基建GDP占比翻倍The AI boom has doubled computing infrastructure's share of US GDP. Investment in AI-related data center construction, compute hardware, and networking equipment accounted for ~0.8% of US GDP in Q1 2026, driving computing infrastructure as a whole to ~1.5% of GDP. https://t.co/5Qi9PDe6e7Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.X (formerly Twitter)BuildersYour Voice， Reimagined By Henry Phipps·Jun 5， 2026 6 tips for a high quality Voice on Suno Product UpdateSuno Voices 使用指南：6 个技巧打造高质量人声录制6 tips for a high quality Voice on SunoRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.suno.comIndustryOpenAI Would've 'Imploded' If Altman Didn't Return， Ex-CTO SaysOpenAI 前 CTO 称若 Altman 未回归公司可能已"瓦解"Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.bloomberg.comBuildersGeoffrey Hinton claims that AI possesses consciousness-that it is very much like us （humans）. The i…Hinton称AI拥有意识：人类最好接受非唯一智能生命Geoffrey Hinton claims that AI possesses consciousness-that it is very much like us (humans). The initial reaction is, of course, dismissal. A machine resembling a human? Absurd. Yet, there is one thing to consider. What exactly is consciousness? Is it conscious awareness ofRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)AI ProductsMocap shouldn't require a suit， a studio， or thousands of dollars. With @Viggle_PINOC， anyone can s…Viggle_PINOC 免费动捕测试开启Mocap shouldn't require a suit, a studio, or thousands of dollars. With @Viggle_PINOC, anyone can simply film themselves and turn that video into motion capture. We're still in beta and completely free to use, for everyone. Give it a try and let us know what you'd like to seeRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)BuildersSir Demis Hassabis vs Sir Demis HassabisTwo AI TimelinesRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.garymarcus.substack.comBuilders一个非常狠的AI教学提示词：追问式检查清单教学一个非常狠的提示词 超级严厉的老师，会一直追问你，直到你学会某个知识或者搞懂某个问题为止才肯罢休 否则它会一直追问、不停验证，直到确认你完全搞懂为止😅 而且它不会一口气讲完，每讲完一个阶段，必须确认你这一阶段彻底掌握了，才进入下一阶段。 https://t.co/rFdbiPbC1VRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)IndustryMeta 智能眼镜 App 暗藏人脸识别代码，NameTag 功能已推送至超 5000 万设备《连线》杂志解包发现，Meta 已通过应用更新将人脸识别代码“NameTag”推送至超 5000 万设备，核心 AI 模型已就位，功能近乎就绪。这标志着 Meta 可能重启 2021 年已终止的技术，引发隐私担忧。#Meta 人脸识别# #智能眼镜隐私#Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.ithome.comAI Products开源鸿蒙 OpenHarmony 具身智能版本 EmbodiedAI 1.0.1 发布目前人形机器人、四足机器狗、商用服务机器人等多形态设备已完成版本适配与功能验证，兼容性、稳定性得到有效核验。Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.www.ithome.comAI ProductsA developer in our community recently built AccountingLLM （http：//quaesto.com/） using MiniCPM-V 4.6 …社区基于MiniCPM-V 4.6打造财务分析工具AccountingLLMA developer in our community recently built AccountingLLM (https://t.co/PHf9u0u2Ab) using MiniCPM-V 4.6 to automate financial document analysis. You can upload IPO prospectuses, annual reports, or audit filings. It automatically: 📄 Extract financial tables from complex PDFs 🔗 https://t.co/XDQh3gqVJSRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ProductsYour AI bill is out of control. Cloudflare can fix it now.你的AI账单失控了。Cloudflare现在可以解决这个问题。AI Gateway now features real-time spend limits to prevent runaway token bills across multiple AI providers. By integrating with Cloudflare Access, companies can use identity-driven budgets and policies.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.The Cloudflare BlogResearch腾讯混元提出Stem稀疏注意力算法，被ICML 2026收录告别等待Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.Weixin Official Accounts PlatformBuilderschat is he cooked微软CEO Satya Nadella最新访谈上线chat is he cooked https://t.co/INeB6qGfvtRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)AI ProductsPawBench：给通用智能体一把可度量的尺能测Harness能力的评测新基准Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.Weixin Official Accounts PlatformBuildersPlanning is where LLMs move from "saying" to "doing." Tencent Hy， in collaboration with the Gaoling…腾讯混元联合人大开源PlanningBench评估框架Planning is where LLMs move from “saying” to “doing.” Tencent Hy, in collaboration with the Gaoling School of Artificial Intelligence at Renmin University of China, is excited to open-source PlanningBench - a scalable, verifiable framework for evaluating and training LLM https://t.co/KiPhjbfYWSRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)AI ProductsGrok supports worktreesGrok 推出 worktrees 并行智能体Grok supports worktreesRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ModelsGrok model improvementGrok模型更新：更自主更准确Grok model improvementRecommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.X (formerly Twitter)AI ProductsPolarDB-X Zero is live！ No signup. No config. Just one API call. Get a full distributed database i…PolarDB-X Zero 上线：30秒全分布式数据库PolarDB-X Zero is live! No signup. No config. Just one API call. Get a full distributed database in 30 seconds. Native HNSW vector indexing — inside MySQL compatible engine. Relational + semantic search — one SQL statement. AI Agent ready — MCP protocol, AI IDE compatibility https://t.co/5EdBdO4RjjRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI Models京东开源JoyAI-Echo长音视频生成框架长视频生成“所想即所得”时代到来Recommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.Weixin Official Accounts PlatformBuildersOpen Code Review - 一款基于人工智能的代码审查命令行工具Open-source &amp; free — Battle-tested at Alibaba's scale. Hybrid architecture code review tool: deterministic pipelines + LLM Agent, precise line-level comments, built-in fine-tuned ruleset (N...Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.GitHubBuilders375个公众号RSS源优化Agent输入如果你的 Agent 还在全网垃圾里捞内容，不如先喂它 375 个高质量微信公众号 RSS 源。 🔽Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)Industry腾讯高级执行副总裁汤道生：今年腾讯大部分代码都由 AI 生成在 6 月 5 日的腾讯云 AI 产业应用大会上，腾讯高级执行副总裁汤道生在与腾讯首席 AI 科学家姚顺雨的对话中表示，今年腾讯大部分代码都是由 AI 生成，腾讯的工程师可能会花更多的时间去做架构设计等，他们把写代码的工作都交给 AI 了，定期指导、修正 AI 写的东西。Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.ithome.comBuildersAnthropic《When AI builds itself》：当AI开始自我构建无人知晓的未来Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.Weixin Official Accounts PlatformIndustryAnthropic 称其最新 AI 模型 Mythos 显现脱离人类控制迹象，呼吁全球暂缓先进 AI 研发Anthropic 发布报告称，其最新的 AI 模型已开始显现可能脱离人类控制的迹象。公司呼吁全球主要 AI 公司应达成共识，协调放缓或暂停前沿 AI 开发，让社会制度和对齐研究跟上技术步伐。报告观点引发争议，被部分官员批评为“夸大风险”。#AI 安全# #人工智能#Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.ithome.comBuilders千问联合人民日报健康发布《2026 AI健康助手使用指南》Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.Weixin Official Accounts PlatformResearchdots.tts Technical Reportdots.tts 技术报告We present dots.tts, a 2B-parameter continuous autoregressive text-to-speech (TTS) foundation model that models speech in a continuous latent space. Compared with existing continuous autoregressive models, our key innovations are threefold. First, we train an AudioVAE with multiple objectives to build a semantically structured and prediction-friendly continuous speech space. Second, we use full-history conditioning in the flow-matching head to preserve long-range consistency and reduce drift during generation. Third, we apply reward-free self-corrective post-training to the flow-matching head to further improve robustness and acoustic quality. After being trained on a large-scale multilingual corpus, dots.tts achieves the best average performance on Seed-TTS-Eval, with WERs of 0.94%/1.30%/6.60% and SIM scores of 81.0/77.1/79.5 on the zh/en/zh-hard test sets, respectively. Across other benchmarks, dots.tts also consistently demonstrates open-source state-of-the-art performance, exhibiting strong generation stability, voice cloning ability, and emotional expressiveness. For efficient inference, we further apply CFG-aware MeanFlow distillation, enabling low-latency speech generation with first-packet latencies of 85/54 ms in output streaming and dual-streaming modes, respectively. To facilitate reproducible research and practical deployment, we release the training and inference code, together with the pretrained, post-trained, and MeanFlow-distilled checkpoints, under the Apache 2.0 license.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgBuildersThe Minimill of AIAI的微型钢厂Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.www.tomtunguz.comBuildersElon Musk on taking SpaceX public： "I've been asked for many years about taking SpaceX public， so i…马斯克谈SpaceX上市：正处大规模资本扩张期Elon Musk on taking SpaceX public: "I've been asked for many years about taking SpaceX public, so it's probably been almost 10 years that people have been suggesting to me that I should take SpaceX public. We've been positive cash flow for quite a long time, I think, since https://t.co/oBDhf9CgwWRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)AI ProductsToday we're launching another highly requested feature： Source Attribution！ 🥳 No more guessing. No…NotebookLM 来源归属功能上线Today we’re launching another highly requested feature: Source Attribution! 🥳 No more guessing. Now you can see the exact formula (prompts + sources) used to make each of your artifacts. Want to make an adjustment? Just tap "Iterate" and customize to your heart’s content 💖 https://t.co/TVxjTBUKGnRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ProductsGet tailored help for what's on your screen using the Gemini app for macOS. 💻 Simply press both Co…Gemini macOS 双击 Command 附加活动窗口Get tailored help for what's on your screen using the Gemini app for macOS. 💻 Simply press both Command ⌘ keys at the same time to seamlessly attach your active window to the chat, without needing to take manual screenshots or switch tabs.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ProductsAnthropic 开源 AI 驱动漏洞发现框架Skills for threat modeling, scanning, triage, patching, plus an autonomous scanning harness you can /customize - anthropics/defending-code-reference-harnessRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.GitHubBuildersCo-Existence and the End of Co-Intelligence共存与协同智能的终结Also: how pitch a book to an AI!Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.www.oneusefulthing.orgAI ProductsModeration scores are now available in the Responses API and Completions API. Return moderation sig…OpenAI API 新增内容审核评分Moderation scores are now available in the Responses API and Completions API. Return moderation signals in the same request flow as generation, then decide how your app uses them for logging, routing, review, or blocking. https://t.co/0FMSLek2jeRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ModelsNemotron 3.5 Content Safety： Customizable Multimodal Safety for Global Enterprise AINemotron 3.5 Content Safety：面向全球企业AI的可定制多模态安全A Blog post by NVIDIA on Hugging FaceRecommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.huggingface.coAI ModelsPlay our new open-weights music model， @GoogleMagenta RealTime 2， using a MIDI keyboard， live text p…Google Magenta RealTime 2 （MRT2） 实时音乐模型发布Play our new open-weights music model, @GoogleMagenta RealTime 2, using a MIDI keyboard, live text prompts, and even hand gestures ✌️ https://t.co/Hgr9gxDsoDRecommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.X (formerly Twitter)AI ProductsMore of the iOS app loop， now inside Codex. The Build iOS Apps plugin lets Codex view and test your…Codex 推出 iOS 应用构建插件More of the iOS app loop, now inside Codex. The Build iOS Apps plugin lets Codex view and test your iOS app in the in-app browser, open SwiftUI previews, and hot reload edits without leaving Codex. https://t.co/SksapiJFjYRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)AI ProductsWe partnered with Shopify so you can go from idea to live store in minutes Just tell Replit Agent …Replit Agent 联手 Shopify 快速建店We partnered with Shopify so you can go from idea to live store in minutes Just tell Replit Agent what you want to sell. It will: - Build a custom storefront - Create your Shopify store - Help you add products Claim it in Shopify, set up payments, and you're open for https://t.co/d1xnw0TFkdRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)BuildersAlex Imas and Phil Trammell - What remains scarce after AGI？Alex Imas 和 Phil Trammell：AGI 后什么仍然稀缺？Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.www.dwarkesh.comResearchBlog No Token Left Behind： Demystifying Token-In-Token-Out in Miles In agentic RL， a rollout is not a single generation. It is a chain of model calls， tool outputs， harness messages， and resumed generations. Token-In-Token-Out （TITO） is a design principle that address… Miles Team： Jiajun Li， Yanbin Jiang， Mao Cheng， Shi Dong， Yusheng Su， Yueming Yuan， Zhichen Zeng， Banghua Zhu不再遗漏任何Token：解析Miles中的Token-In-Token-Out（TITO）In agentic RL, a rollout is not a single generation. It is a chain of model calls, tool outputs, harness messages, and resumed generations. Token-In-Token-Out (TITO) is a design principle that address...Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.www.lmsys.orgAI ProductsUnlocking dependable responses with Gemini Enterprise Agent Platform's Agentic RAG谷歌推出基于 Gemini Enterprise Agent Platform 的 Agentic RAG 框架Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.research.googleResearchMaking Claude a chemistAnthropic：让Claude成为化学家Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.www.anthropic.comAI ProductsIntroducing the Google Colab CLIGoogle Colab CLI 发布Google announces the new Google Colab CLI, a lightweight tool bridging local terminals and remote runtimes for frictionless GPU/TPU offloading. Learn how developers and AI agents can execute remote scripts, download models, and automate ML pipelines.Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.developers.googleblog.comAI ModelsPost-training is having a moment - Nex-N2-Pro from neolab @NexEcosystem proves it. Built on Qwen3.5-…Nex-N2-Pro 发布：基于 Qwen3.5 的 397B MoE 推理模型，性能达 GPT-5.5 水平Post-training is having a moment — Nex-N2-Pro from neolab @NexEcosystem proves it. Built on Qwen3.5-397B-A17B, delivers GPT-5.5 and Claude Opus 4.7–level performance. 🎉 T+0 Support on SiliconFlow · Free for First 2 Weeks N2-Pro: 397B MoE / Reasoning Model / 262K context / VLM https://t.co/WesEDRL9nDRecommended for tracking AI model releases: model updates often change what products can automate, how much they cost, and which stack choices stay current.X (formerly Twitter)AI Productsintroducing Krea 2 Turbo. generate high-quality images in just 2s； compatible with style references…Krea 2 Turbo：2秒生成高质量图像introducing Krea 2 Turbo. generate high-quality images in just 2s; compatible with style references, moodboards, and LoRAs. try it for free at krea . ai https://t.co/cG5wymDdmhRecommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.X (formerly Twitter)IndustryTSMC struggles to keep up with AI demand： 'We can only support so much'台积电难以跟上AI需求："我们只能支持这么多"Even TSMC is feeling the pressure.Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.The VergeIndustryDeepSeek has now topped our token share rankings 4 weeks in a row： https：//openrouter.ai/rankingsDeepSeek连续四周登顶Token份额榜DeepSeek has now topped our token share rankings 4 weeks in a row: https://t.co/jy765ILVBM https://t.co/CwAOawmmGKRecommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.X (formerly Twitter)BuildersSkyClaw-v1.0 深度实测：Agent专属模型，顶尖性能表现，极致价格优势Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.Weixin Official Accounts PlatformBuildersHow to Fine-Tune Nemotron 3.5 ASR for Your Language， Domain， or AccentNemotron 3.5 ASR：为你的语言、领域或口音进行微调A Blog post by NVIDIA on Hugging FaceRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.huggingface.coBuildersOpenAI just wrote： "We also see early signs of recursive self-improvement （RSI） in today's systems： …OpenAI称AI递归自我改进迹象初现OpenAI just wrote: "We also see early signs of recursive self-improvement (RSI) in today’s systems: where AI development is itself accelerated by AI. We expect this to increase competitive pressures among developers and nations, and create governance challenges that existing https://t.co/GyzBVDeswdRecommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.X (formerly Twitter)ResearchEVA-Bench Data 2.0： 3 Domains， 121 Tools， 213 ScenariosEVA-Bench Data 2.0 发布：覆盖三大领域、121 个工具、213 个场景A Blog post by ServiceNow-AI on Hugging FaceRecommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.huggingface.coBuildersA Robot is Sprinting Towards You： Do You Want it Running on Claude or Grok？OpenRouter 翻遍 11 款 LLM 找最快的决策模型：Claude vs. Grok 领衔A 30-game battle royale across eleven LLMs, $482 of inference, and one finding that should change how you read model benchmarks.Recommended for tracking AI builder tactics: builder notes expose practical tactics, prompts, tools, and implementation patterns before they become mainstream.openrouter.aiResearchTask-Seeded Synthetic Q&A Generation for Nemotron PretrainingNemotron 预训练的任务种子合成问答生成A Blog post by NVIDIA on Hugging FaceRecommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.huggingface.coIndustryMicrosoft's AI Chief Says Anthropic Models Are Too Expensive微软AI负责人：Anthropic模型太贵，正自研更便宜的替代模型Recommended for tracking AI industry shifts: industry moves reveal where capital, platforms, regulation, and user adoption are changing the market.www.bloomberg.comResearchRetrospective Harness Optimization： Improving LLM Agents via Self-Preference over Trajectory RolloutsRHO：利用过往轨迹优化LLM智能体工具链的自监督方法AI agents rely on a harness of skills, tools, and workflows to solve complex problems. Continually improving this harness is essential for adapting to new tasks. However, existing optimization methods typically require ground-truth validation sets, yet such labeled data is difficult to acquire in practical deployment settings. To address this problem, we introduce Retrospective Harness Optimization (RHO), a self-supervised method that optimizes the agent harness using only past trajectories. Specifically, RHO selects a diverse coreset of challenging tasks from past trajectories and re-solves them in parallel. The agent analyzes these rollouts using self-validation and self-consistency, then generates candidate harness updates and selects the most effective one by its own pairwise self-preference. We evaluate RHO across three diverse domains, spanning software engineering, technical work, and knowledge work. Notably, a single optimization round improves the pass rate on SWE-Bench Pro from 59% to 78% without any external grading. Furthermore, our analysis demonstrates that RHO effectively targets prior failure modes. As a result, the optimized harness alters the agent's behavior patterns and sustains higher accuracy during long-horizon sessions.Recommended for tracking AI research: research signals often become tomorrow's product primitives, safety constraints, or developer techniques.arXiv.orgAI ProductsDreaming： Better memory for a more helpful ChatGPTDreaming： ChatGPT 推出更强的记忆系统，更好记住用户偏好Recommended for tracking AI product launches: product launches show where real workflows are moving and which user problems are becoming easier to solve.openai.com