In this video titled "We need to talk about this... Matthew Berman
The Key Announcement
- The Claim: OpenAI announced that an internal, unreleased AI model tackled the Navier-Stokes equations and generated a 165-page proof demonstrating a finite-time singularity.
- Scale & Cost: The proof was produced by running approximately 10,000 concurrent AI agents for 88 hours (generating over 130 billion tokens) at an estimated compute cost of roughly $10 million.
- Verification Workflow: OpenAI used its newly launched flagship model, GPT-6 Astra, to formally verify the 165-page paper in an automated 17-hour AI-to-AI process.
Controversies & Industry Reactions
- Prior Work & Credit: Mathematician Tristan Buckmaster (NYU) and other researchers noted that OpenAI's work leaned heavily on established directions and ongoing human research, leading to debate over whether OpenAI rushed the announcement to claim credit for a marathon finished in the final mile.
- Practical Impact vs. Theoretical Math: Viewers and researchers point out that solving the Millennium Prize problem primarily answers the theoretical existence of smooth solutions or singularities.
It does not instantly replace applied numerical fluid dynamics or current engineering simulations used in physics, aerospace, or formula racing. - Open Source vs. Proprietary AI: The video highlights the massive discrepancy in compute resources—raising concerns about closed-loop AI development controlled by major labs versus open-source research democratization.
- Impact on Human Capability: Berman addresses broader philosophical questions around whether relying on AI agents for high-level reasoning and scientific discovery will make humans less capable, drawing parallels to historical tech tools like taming horses or mechanical automation.
Apparently OpenAI models are learning from its users (and RSI, Recursive Self Improvement),
and in this case may have "stolen" ideas and taken credit for solution.
No comments:
Post a Comment