Daily Brief: OpenAI Internal Models Succeed on Six of Ten 'First Proof' Research-Level Math Problems

OpenAI Internal Models Succeed on Six of Ten 'First Proof' Research-Level Math Problems

OpenAI's internal models tackled ten unpublished research-level math problems known as the "First Proof" challenges, designed to test AI's ability to generate novel proofs. In a one-week sprint with minimal human oversight, the models delivered promising solutions to at least six, which experts deem likely correct. This marks a leap from basic math capabilities to frontier research, as highlighted by Sam Altman.[1][2][3]

Altman called it a key evaluation for next-gen AI, while OpenAI's Chief Scientist noted the 6/10 success rate. Sebastien Bubeck shared the challenge publicly to benchmark systems, and reactions on X exploded—Greg Brockman echoed the excitement, with some users hailing it as an "AGI achieved internally" moment.[1][2][3]

OpenAI's GPT-5.2 Derives Novel Gluon Interaction Result in Theoretical Physics Preprint

On February 14, GPT-5.2 co-authored a preprint with researchers from IAS, Vanderbilt, Cambridge, and Harvard, proposing a closed-form formula for "single-minus" gluon tree amplitudes—interactions long assumed zero in textbooks. An internal OpenAI model proved it, with humans verifying up to n=6 cases. The result reveals non-zero amplitudes under specific conditions like all-plus helicity for other gluons, potentially simplifying quantum field theory computations.[4][5][6]

OpenAI shared the news on X, emphasizing how GPT-5.2 challenged assumptions: a gluon interaction "many physicists expected would not occur can arise under specific conditions." Kevin Weil noted decades of assumptions overturned, and Bo Wang (@BoWang87) quipped that GPT-5.2 essentially said, "What if they can—under these conditions?"[5][6]

US Department of Labor Releases First National AI Literacy Framework

The US Department of Labor dropped the nation's first AI Literacy Framework on February 13, outlining core content areas and delivery principles for AI education. It ties into the White House's America's AI Action Plan from July 2025, pushing Workforce Innovation funding toward AI skills programs and aiming to lead global AI deployment distinct from EU regulations.[7][8][9]

While X reactions were muted among big names, discussions highlight its ripple effects on K-12 and beyond, positioning the US to shape AI literacy norms against the EU AI Act's stricter approach.[7][8][9]