Terence Tao Urges Slowing Down AI: "This Pace Is Insane"

TL;DR: Around September 20, Fields Medalist Terence Tao publicly urged slowing down AI in his keynote at Caltech's Science x AI Summit and on X, calling the current pace "insane" with "no reason at all" to go this fast. He warned that humanity is willing to change everything without understanding the consequences, in highly nonlinear dynamics. The remarks quickly sparked debate on Reddit and X, with several users pushing back, exposing an open rift between risk-first and progress-first camps.

Confirmed

• Tao explicitly said we must slow down AI, describing the pace as insane and repeatedly stressing there is no reason for it. The remarks came from his Caltech Science x AI Summit keynote and X posts.
• His stated concerns include the unsettling speed of AI progress in mathematics, unknown consequences of change, and highly nonlinear dynamics; he acknowledged AI is "knocking off major conjectures left and right" but argued the value produced is far lower than if done in a measured, principled way.
• He predicted that once AI starts doing the "prestige work" of various fields, experts in every field will call for slowing down.
• The stance is a notable shift for a scholar long open to AI-assisted mathematical research, coming as AI solves increasingly hard math problems.
• Posts also cited an OpenAI-affiliated figure's "I smell fear" response as a counterpoint.

Reactions

• User @uubzu argued that among all reasons to slow AI, "math is advancing at an uncomfortable pace" is the least persuasive.
• @kimmonismus publicly disagreed, citing urgent world problems—energy crises, disease, hunger—that make "no reason to accelerate" untenable.
• User deredleritt3r offered a parallel: reasoning models are to math what RSI is to every industry.
• User Rufuszone agreed on human primacy but was characterized as arguing it need not come at the cost of civilizational progress; sharer damianplayer countered that AI is a tool for human progress.
• Sharer mallow610 recalled software engineers' journey—many asserted in 2024 that "LLMs can't reason"—and drew an analogy to San Francisco rideshare drivers complaining Waymo deployed too fast, suggesting such "too fast" voices may not stop the technology.
• User @nihilunbounded quipped that Tao would have been better off saying outright "AI beat me to Navier-Stokes."

Why it matters

Tao is among the most esteemed living mathematicians; even he issuing a slowdown warning signals that rapid AI breakthroughs in frontier fields like mathematics are alarming core figures of academia, while the pushback underscores the sharpening public divide between progress-first and risk-first camps.

2026-09-20 ~ 2026-09-21 · 15 related posts

Full story(2 episodes)→

• Episode 1: Terence Tao Urges Slowing Down AI: "This Pace Is Insane"(2026-09-20, 15 posts)
• Episode 2: Debate Rages Over Whether AI Math Research Should Speed Up or Slow Down(2026-09-20, 2 posts)

• Terence Tao: "we have to slow down AI. the pace is insane — no reason to be this fast" — Tolopono · 2026-09-20
• Terence Tao says 'we have to slow down AI', sparking pushback over the science golden age — kimmonismus · 2026-09-20
• Terence Tao calls for slowing AI development as AI tackles harder math problems — Polymarket · 2026-09-20
• Terence Tao calls for slowing AI, sparking pushback over human hegemony in science — Kyrannio · 2026-09-20
• Terence Tao: We Have to Slow Down AI — There's No Reason to Be This Fast — AxomaticallyExtinct · 2026-09-21
• Terence Tao urges slowing AI as engineers relive the 2024 denial-to-acceptance arc — mallow610 · 2026-09-21
• Math community roasted: netizen jokes Terry Tao's AI backtrack would've been better said bluntly — ctjlewis · 2026-09-21
• [source] Terence Tao: elite experts across fields will demand a slowdown once AI takes their prestige work — rohanpaul_ai · 2026-09-21
• [source] Terence Tao: AI is cracking major math conjectures, "but we can afford to be slower" — rohanpaul_ai · 2026-09-21

6 near-duplicate retellings: damianplayer · ctjlewis · AxomaticallyExtinct · AxomaticallyExtinct · ZeroStateReflex · StewartalsopIII