Press Enter to search · ESC to close

AI × Crypto

Altman Says OpenAI Can’t Safely Pursue Recursive Self-Improvement Yet

OpenAI CEO Sam Altman said on Nov. 19 at the Semafor summit that OpenAI has used its own technology to speed up next-generation model development, but has not reached the point of letting models self-evolve unsupervised. He said he could have pushed recursive self-improvement (RSI) more aggressively but cannot do so safely now. Same day, OpenAI safety committee member Paul Christiano, speaking personally on the Dwarkesh Patel podcast, put the probability of superintelligence capable of overthrowing human rule within 12 months at about one-third, and permanent loss of human control near 10%, citing hundreds of mathematical research results from OpenAI and Anthropic models in the past month.

Share

Related News

Cointime share card
Long-press the image to save or share
Pitch us Community Feedback