Lab researchers argue over what recursive self-improvement means
As Anthropic and METR put the idea of AI accelerating AI research into filings and numbers, Jason Wolfe of OpenAI argued that the term bundles three separate questions. Sam Altman wrote that the need for human control should ultimately limit AI progress, and François Chollet co-signed.
In the Weekly Review of 8 November 2026
Filed under AI safety, AI policy
Wolfe distinguished the automation of research, the pace of progress and the degree to which humans understand and control the process, and argued that goals should be made clear before limits are debated. Altman wrote that OpenAI rejects any future in which people only perform the actions AI systems reward, and that this should in the end constrain progress. Chollet, creator of Keras, added that the only worthwhile future keeps AI as a tool in human hands.
The argument over what to do runs alongside. In an essay published on 31 October, Scott Alexander kept a global pause as his first choice but argued against unilateral exits by careful labs or staff, which he said hand the lead to less careful ones. Boaz Barak of OpenAI and Harvard asked for clearer scenarios for the period after superintelligence.