Terence Tao Warns of Depleting High-Quality Mathematical Data for AI Training
September 8, 2026
Terence Tao argues that high-quality, novel mathematical problems are being non-renewably mined by LLMs. This depletion of unique reasoning data could limit the ability of future models to discover new mathematical truths.
HOW THIS AFFECTS YOU
●
researcherYou should account for the shrinking supply of high-reasoning training data.
●
founderThis signals a potential long-term bottleneck in scaling reasoning capabilities via data alone.