Tao's Second Math Age: Proof Plenty Meets Its Critics

Last night I watched a nearly hour-long slide-by-slide reading by mathematician and thinker Samuel Alexander. He walks through the Caltech talk Math 2.0 by Terence Tao, one slide at a time. Tao ranks among the most famous living mathematicians; Alexander presents himself as a theory builder with modest problem-solving glory. Both agree on one thing: AI models shake the mathematics world to its roots. In this piece I lay out both views in plain words; the goal is to grasp the order behind the quarrel, not the quarrel itself.

Tao's first claim: classic mathematics, or Math 1.0 in his words, rested on proof scarcity. Cracking an open problem was hard and rare; posts, journals, and prizes followed that scarcity. Picture taxi drivers when ride-hailing arrived; Tao says mathematicians live through the same shock now. The field barely moved for almost a century; even the web left its order intact. Tao shared a short outline on his WordPress blog, with the same verdict: scarcity ended, the old order aged out.

Why do AI models excel in mathematics above other fields? Tao's answer shrinks to two words: objective checking. Whether a proof holds is not a matter of taste; liking it and rating it right are separate acts. A physics claim demands dear trials; a math claim asks only for pen, paper, and shared stores. Tools like Lean let a machine test each formal step. The slide deck sits open on Tao's GitHub archive; the curious can study it line by line.

The second cause is digitization. Mathematics lives in writing; it gathers no tubes, wires, or samples. Access once meant entering a university library; outsiders stayed at the door. Today millions of pages stand scanned, ready feed for models to read. Tao notes that leading AI labs ranked math problems first for this reason: checkable, digital, rich in data. Where the three meet, learning systems find their ideal ground.

Result: an age of proof plenty. With enough compute, many problems once deemed hard fall within short spans; the asker needs no deep craft to win answers. Alexander likens it to Christmas morning; I agree — for eager learners every day feels festive. For its part, OpenAI shared fresh mathematics results from an internal model on October 6, 2026; such notes fan excitement but hide the full frame. Plenty is real, yet the display window and the back room differ.

Tao adds a warning: AI talent cuts like a serrated blade. Razor sharp one way, blunt the next. The flood of solved problems on social feeds misleads for this reason; winners fill the window, misses stay unseen. Honest measurement calls for outside tests. The method paper lives on arXiv; field testing runs through 1stProof. That crew feeds models hidden research items and hunts genuine outcomes, not window dressing.

Tao borrows an old war story here; the slide shows a bomber dotted with bullet marks. Instinct says plate the marked spots, yet armor belongs where no marks show; planes struck there never came back. Statistician Wald framed this logic, now a textbook case of survivor error. A Wald-based outline stands on Wikipedia; reading it sharpens Tao's point: press triumphs are returning planes; nobody counts the fallen.

To Tao, problem solving works like a lighthouse: a guide, not a harbor. True worth gathers along the path — fresh insight, wider views. Yet when a measure turns into a target, the order bends; all boats anchor at the light and none explore the coast. Tao holds that blind problem hunting now harms the future of mathematics. Alexander dissents; in his view the victim is not mathematics but the old pay-and-title order. I grant each a half: knowledge grows, but a guild locked on one measure narrows.

The most debated slide runs a cancer thought trial. Picture this: researchers ask a strong AI model for a drug that clears stage-three clinical tests. The model names a blend of unknown compounds; the blend passes. Nobody grasps why it works. Tao asks: would you take it gladly, or would you seek a doctor who grasps the method? Alexander snaps back: medicine already runs this way. Nobody fully grasps common painkillers or sleep drugs; the dying try unknown cures today under special access rules. Snubbing a tested cure counts, to him, as a rich man's worry. I lean toward Alexander: mystery is old news in medicine, yet Tao's trust query keeps worth; soon models will draft prescriptions, and every home will debate it.

Tao cheers loudest for large-scale survey mathematics: datasets once beyond reach now stand built. Exhibit one is the Equational Theories project: over 22 million links among 4,694 equations, settled true or false by volunteers hand in hand with AI. Maps and Lean checks run through GitHub; visitors can roam equation by equation. Similar stores grow around Mathlib; the story of that two-million-line collection lives on the Lean Lang site. Such work marks a new field survey style; what once outlasted a lifetime now runs overnight on token-fed systems.

Math 2.0 samples start with exposition items: finding vivid accounts for results whose proofs stay opaque. Next come living webs: tidy, updating charts of linked findings. Tao also cites MathOn, a November 13–15 student gala where open AI tools hunt fresh, finer proofs of stale results. Alexander adds a sour note: exposition once paid no rent in hiring; fine accounts earned small journals at best. In my eyes Math 2.0 can untie this knot; an order that also applauds vivid teachers opens mathematics to all.

Alexander saves his anger for the hiring and journal order. For a century, applause and posts went to problem solvers alone; theory builders waited aside. He counts himself among the builders and greets the old order's fall with laughter. Tao tells the same story softly; Alexander names names: whoever owes a wage to scarcity fears plenty. I read this brawl as larger than mathematics. Every trade sets gatekeepers against newcomers; when AI models lift the latch from keeper hands, noise erupts.

Tao closes where Alexander agrees: the tight bottleneck is imagination, not tools. Both love the camera analogy: painters kept work, and cinema — an art nobody foresaw — was born on top. If today's AI is the camera, our cinema has yet to arrive; its shape stays beyond our gaze. My lesson: embrace proof plenty, keep measures honest, save hours for teaching. For those who do, the next decade turns festive; for the rest, a hard settling. Each guild will choose its side.