Background
On September 24, UCLA cryptographer Amit Sahai published a guest post on Terence Tao's blog *What's new* titled "We're gonna need a lot more mathematicians". Tao added a note at the top explaining it was a guest post, with an initial draft converted via AI from another file format. The post accumulated over 300 points on Hacker News.
Sahai opens with a memory: every math department has students who can understand new concepts but need twice as long as the top students — and most of them eventually leave research mathematics. He has been thinking about them while watching AI systems, which he collaborates with in his own work, produce genuinely beautiful new ideas — far beyond fast computation or reproducing arguments strong human researchers could already understand.
Key points
- A humbling era. The mathematical community is entering an era requiring humility: every mathematician will know what it feels like to fall behind. Sahai warns that if the community collectively abandons the work of understanding, that is a profound betrayal of human responsibility — while each individual retains the right to choose a different life.
- A terawatt-scale fusion plant. Sahai's thought experiment: imagine a future AI system proposing a novel gigawatt-scale fusion plant design, operating on principles no human has ever conceived, with robots ready to build it. The more novel the design, the less confidence can be inherited from decades of operating experience. Before approval, he wants a human collective to understand why the design works and what its safety assurances rest on — not by redoing AI's calculations, but by understanding the model, the experimental evidence supporting it, and the model's uncertainties.
- Deployable intellectual reserve. The article names its proposal: groups of mathematically sophisticated humans who can be summoned to understand major AI breakthroughs, even outside their normal research interests. Sahai rejects the objection that AI makes each person efficient enough that fewer humans suffice: individuals are biological and limited; deep understanding takes time and sustainable working rhythms, and serious collaboration goes further than any individual can.
- The conclusion is the title: we will need a lot more mathematicians.
A month-long debate
The post is the latest entry in an ongoing discussion on Tao's blog. In September 2026 alone the blog carried 35 posts, including pieces on serious AI misalignment in mathematics, why human mathematicians remain necessary, an ICIAM statement on mathematics and AI, recommendations from a summit on PhD mathematics education in the AI era, and the announcement of an advisory group on mathematics and AI. In mid-September, a joint letter from more than twenty Fields medalists joined the debate.
A notable detail: the article itself was drafted with assistance from GPT 6 Astra, with acknowledgments to Tao as well as Sahai's family members Anant Sahai and Gireeja Ranade. A piece arguing that humans must preserve understanding was itself AI-assisted — a footnote to its own thesis that using AI to help understand AI is exactly the collective understanding method he proposes.
The Hacker News debate
The main battleground in the comments was not philosophy but employment. Young researchers asked what this vision means for the job market five years out and whether the tenure pipeline would narrow. Others suggested Sahai was really calling for engineers, not mathematicians. Sahai appeared under his own name, did not dodge the concerns, and said he is co-organizing a Simons Foundation-supported project whose report will soon offer concrete action items. In the same thread he wrote a line that could serve as the article's subtitle: humanity stands at a fork — either gradually cede the effort of understanding to machines, or treat machines' thinking as a great gift for deepening our own understanding.
The question this article pins to the wall: when machines produce ideas faster than all humans can digest them, is understanding left to the machines — or to us?