Descriptions:
Wes Roth covers what he characterizes as a genuine capability milestone: autonomous AI systems solving previously unsolved problems from the famous Erdős collection, a database of 1,135 open mathematical challenges. Terrence Tao — widely considered one of the greatest living mathematicians, the youngest IMO gold medalist in history, and a UCLA professor — publicly validated the breakthrough, confirming that Erdős problem #728 was solved “more or less autonomously” by AI, with the result representing a novel proof not previously found in the literature and not based on a trivial loophole.
The video systematically categorizes the wave of recent AI math achievements: fully autonomous solutions to open problems, human-AI collaborative proofs, new proofs of previously-solved problems, AI-powered literature reviews, and formalized proofs. GPT-5.2 (released December 11, 2025) and Grok 4.20 are highlighted as the primary drivers. Scott Aaronson is cited as having published a paper in September 2025 where a key technical proof step came directly from GPT-5 Thinking — his own characterization of the event.
Roth situates all of this within a steep capability curve concentrated in November 2025 through January 2026. Tao’s framing — “a genuine increase in capability of these tools in recent months” — anchors the analysis, and the video also explores AI’s emerging ability to rapidly generate expository mathematical writing as a secondary development. The pace and clustering of these breakthroughs across multiple independent models and research groups is presented as the central signal.
📺 Source: Wes Roth · Published January 12, 2026
🏷️ Format: News Analysis
