Research
Pretraining data filtering in open-weight biological models
英国 AISI 实测 ESM-2 预训练数据过滤可削弱目标能力但约 80 美元算力即可恢复
Pretraining data filtering in open-weight biological models | AISI Work
AI Security InstituteWe assess the strength of pretraining data filtering as a safeguard for open-weight protein models, and find it to be effective and selective, but not necessarily durable against fine-tuning.
Open sourceRecommended because
This is worth tracking because it is a concrete research signal, not just a passing headline. The source preview points to a research result, method, evaluation, dataset, or safety finding. For builders and operators, "Pretraining data filtering in open-weight biological models" can be used as a checkpoint for technical due diligence, roadmap bets, agent design, and evaluation strategy. I keep this thread indexed so future searches around AI research papers, technical methods, and applied AI systems can land on a source-linked page instead of disappearing into a fast-moving feed from AI Security Institute.
What to take from this signal
Context
"Pretraining data filtering in open-weight biological models" is archived here as a source-linked AI signal from AI Security Institute. The useful part is the connection between Pretraining, data, filtering, open-weight, biological and technical due diligence, roadmap bets, agent design, and evaluation strategy, which makes the item more actionable than a normal feed headline. The source context says: We assess the strength of pretraining data filtering as a safeguard for open-weight protein models, and find it to be effective and selective, but not necessarily durable against fine-tuning.
Builder takeaway
For an AI builder, the main takeaway is to watch how this signal changes practical decisions around technical feasibility, evaluation design, safety limits, and product primitives. It can inform what to test next, which product surface to compare, and whether the underlying workflow is ready for real users.
Source context
AI Security Institute remains the authoritative source for the original claim. This page adds a stable archive URL, a short builder interpretation, and related search language so the item can be found later when the original feed has moved on.
Search angles
- Pretraining data filtering in open-weight biological models Research context
- AI Security Institute AI research
- Pretraining, data, filtering, open-weight, biological builder takeaway
- AI research papers, technical methods, and applied AI systems
This page keeps a source preview and a stable archive URL for search discovery. The original source remains authoritative.