🧮 Terence Tao: AI in math is real, but not magic

Several researchers recently claimed generative AI helped solve previously open math problems, including some from the famous Erdős list. Some of these AI-written proofs have checked out.

Terence Tao, widely considered the world’s leading mathematician, says the progress is impressive but overstated. In his view, AI has mostly been picking off easier problems from a long list rather than cracking the hardest ones.

AI is good at systematically scanning hundreds of obscure problems and applying known techniques. Many of the recent wins were “cheap wins,” problems that a human expert could likely solve in half a day.

Where AI shines today:

✔️ Grinding through tedious computations humans avoid
✔️ Applying standard techniques at scale
✔️ Assisting with structured reasoning
✔️ Acting like a junior co-author on technical work

Tao compares AI to a helicopter that drops you at the solution. You get the result, but you miss the journey, the insights and trail markers that human mathematicians build along the way.

Still, he sees real value. AI can help mathematicians work at larger scales, moving from handcrafted case studies to broader “population-level” explorations of problems.

He predicts that AI will soon function as a trusted co-author, especially for routine and heavy analytical work. But he also warns about limitations:

❌ Models lack creativity
❌ They make subtle mistakes
❌ They overstate confidence
❌ They are not designed for deep interactive collaboration

Tao argues that math does not need fully autonomous “push a button and solve” systems. It needs better human-AI conversations.

AI is not solving the deepest problems yet.

But it is changing how mathematics gets done.

✅ Follow @math