METR Embedded to Assess New 'Ant' Model for Fooming, Card Reveals
dfrsrchtwts · x · 2026-09-23
The author highlights his favorite part of new AI model cards: how METR assesses the models. The card for a model referred to as 'ant' includes fun evaluation tasks, and notably a METR team was brought in to assess whether the model is 'fooming' (rapid self-improvement risk). They also ran a questionnaire and interviewed a researcher — a sign frontier labs are disclosing third-party autonomy-risk evaluation details in model cards.
Related event: Anthropic Model Card Shows METR Assessing Whether Model Is 'Fooming'(2 posts)→
More from Models
- Claude Opus 5.5 Debuts: 40% Cheaper, ~30% Faster Than Opus 5, Stronger Agentic Coding — nicolechirps · 2026-09-23
- Fireworks Launches Specialized Intelligence Index; DFS Model Hits 62.2% Vuln Detection Recall — nicolechirps · 2026-09-23
- Sol 6 matches Sol 5.6 quality with half the reasoning time, better Plus limits — Simple-Diver-2192 · 2026-09-23
- Founder: OpenAI should ship its stronger internal model and cut Astra prices 50% — bindureddy · 2026-09-23
- No Single Champion: Model Rankings Shift by Domain, from Legal to Finance to Healthcare — sophiamyang · 2026-09-23
- Quintin Pope asks if Claude shows in-context emergent misalignment — QuintinPope5 · 2026-09-23