A team of approximately 10,000 OpenAI AI agents solved the Navier-Stokes Millennium Prize Problem in under four days using 130 billion tokens. According to researcher Noam Brown, multi-agent coordination contributed less than 10% to this breakthrough. Current AI agents function as automated research interns, producing 3.1 times more output than human researchers on multi-day tasks, but over 50% still require human oversight. The missing ingredient, according to Brown, is research taste, the intuitive ability to judge whether a research direction is genuinely significant and original. OpenAI has set an internal goal to achieve a fully autonomous AI researcher by March 2028 through recursive self-improvement.
Source: Read the original article
