OpenAI Announces Recursive Self-Improvement for AI Models

OpenAI CFO Sarah Friar announced at a Goldman Sachs tech conference in San Francisco that the company's largest models are now capable of training smaller models. This process, known as recursive self-improvement, is viewed by investors as a way to significantly reduce training costs.

The announcement triggered immediate alarm among some AI researchers. Evan Hubinger, the alignment science lead at Anthropic, stated on X that there is a greater than 10% chance AI could kill all humans within the next decade. Samuel Marks, a lead at Anthropic, attributed the pursuit of such risky technology to a combination of commercial incentives and a competitive race against less responsible developers.

Not all industry figures share these concerns. Brad Gerstner, an investor at Altimeter Capital, dismissed the claims of existential risk as nonsense.