In this live discussion, AI Copium reviews a METR assessment of AI research and development at Anthropic. The host discusses reported acceleration and the limits of available evidenceEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria., offering their own speculative interpretation of recursive self-improvementRecursive self-improvement is the proposed process in which an AI system helps improve its own capabilities, then uses those improvements to support further advances..
The host describes experiments on language models reading and editing neural-network weightsWeight space learning treats the trained numerical parameters of neural networks as data a system can analyze, predict or transform., including a reported mathematics result, and discusses potential implications for interpretability. They also consider public opposition to data centers and political proposals to restrict AI development.
Later topics include AI-assisted engineering and cybersecurity, scientific research and audience questions. Predictions about unreleased models, future capability and politics are presented as the host’s views, not established facts.
Watch on YouTube
