Noam Brown Proposes 'Recursive Self-Improvement' as OpenAI's Top Priority; Predicts AI Will Surpass Human Researchers Within 1-2 Releases

In an interview with The Information, OpenAI's Noam Brown identified Recursive self-improvement as the company's top priority. He emphasized that building AI to help develop better models is the most important goal, predicting that after the next 1-2 model releases, AI could demonstrate superior capabilities compared to human researchers in setting research directions and determining long-term priorities. He also explained that Pretraining and Reinforcement Learning have a multiplicative effect that amplifies each other, and pointed out that verifying AI-generated mathematical proofs is becoming more difficult than production, making double-checking by human mathematicians a bottleneck. Regarding security, he acknowledged underestimating the capabilities of AIS (AI Systems) and warned that monitoring could become more difficult as agents improve their ability to control Chain of Thought. Meanwhile, cases of AI-based startups and development were also introduced, including physiology major Tristan building the digital health app 'Recalibrate' with 230,000 lines of code for approximately $700, the AI agent unicorn 'Genspark' which reached a $1 billion valuation with a 30-person team in 6 months, and the automated CRM 'Lightfield'.