Will AI Prove the Riemann Hypothesis Without Understanding It? – Terence Tao
Credibility score: 84/100 — Highly Credible. This video is highly credible with well-supported claims.
BSmeter analyzed "Will AI Prove the Riemann Hypothesis Without Understanding It? – Terence Tao" and rated it 84/100 for credibility (a BS score of 16/100 — highly credible), on 2026-04-13. Its weakest claim — "Is AI solving Riemann Hypothesis in Lean plausible without insight?" — scored 50/100 and was flagged as opinion. 12 claims were checked against the video transcript. Scores are produced by BSmeter's AI analysis of the transcript, not independent human verification.
Of 12 claims analyzed: 0 scored under 40, 1 between 40 and 69, and 11 at 70 or above.
Claims analyzed
Is AI solving Riemann Hypothesis in Lean plausible without insight? — Opinion (50/100)
At 0:00
Tao dropping the Riemann bomb casually like we're all chilling at the math bar — love the 'in retrospect surprised' vibe, but this is peak Tao pondering out loud 🤔😤✅
Why this score: Pure speculation from a math legend, perfectly reasonable as opinion. Key points:
- He's questioning if scaling AI on Lean could crack Riemann Hypothesis with zero human-like insight
- Real context: AI like AlphaProof/Lean systems hit IMO levels, formalize big theorems (e.g., Prime Number Theorem in Lean), per 2026 AI4Math advances
- No proof yet, but Lean's mathlib (200k+ theorems) makes it *plausible* territory — debate on 'understanding' is ongoing among experts
Original quote: “[0:00] One big question I have is how plausible
[0:03] is it that if we just keep training AIs
[0:06] to get better and better at
[0:07] you know, solving problems in Lean, that
[0:10] they will continue to solve more and
[0:11] more impressive problems and then we
[0:13] will in retrospect be…”
Four Color Theorem proven by brute force computer, no elegant proof yet — Verified (100/100)
At 0:30
Tao dropping straight facts on the Four Color Theorem like it's casual trivia — I'm mad this is textbook accurate and I can't even roast it 😤✅🔥
Why this score: Spot-on historical accuracy. The Four Color Theorem was proven in 1976 by Appel and Haken via computer brute force checking ~1,900 configurations, refined in 1997 and formally verified in 2005 by Gonthier in Coq — but no elegant human-readable proof exists, as Tao says. *No controversy here.*
Original quote: “some problems have been basically solved by pure brute force. The four color theorem is is a famous example. Um we have still not found a conceptually elegant proof of this theorem. Um it it basically and and maybe we never will.”
Riemann Hypothesis needs new math or connections to solve — Opinion (85/100)
At 0:54
Tao says Riemann demands 'new type of mathematics' like it's a mic drop — dude's a Fields Medalist, this consensus energy is annoyingly credible 🙄✅👏
Why this score: Widely shared expert opinion. Mathematicians prize Riemann for requiring deep insights beyond computation; trillions of zeros verified on critical line, but proof elusive since 1859. *Tao's view aligns with Clay Millennium Problem status.* No disproof found as of 2026.
Original quote: “part of the reason that we we prize problems like the Riemann hypothesis is that we're pretty sure that that something amazing has to a new type of mathematics has to be created or a new connection between two previously unconnected areas of mathematics has to be discovered to to make this work.”
Riemann unlikely solved by brute force check; could be disproven by computer — Solid (90/100)
At 1:25
Imagines a 'disappointing' computer counterexample zero off the line — Tao's humility admitting it could happen is peak mathematician energy 😤✅🧠
Why this score: Reasonable speculation from expert. Brute force impractical for infinite zeros (trillions checked, all on line); disproof via one off-line zero possible but unlikely per numerical evidence. *Tao's 'unlikely scenario' matches community view.*
Original quote: “it doesn't feel like a problem that will be solved just by exhaustively checking cases or something. Um I mean, it could be false actually. So we could actually uh I get there is an unlikely scenario that that that the hypothesis is false and that's just this this this you can use computer or…”
Human-AI collaboration beats fully autonomous AI for hard math — Solid (85/100)
At 1:51
Dumps on 'fully autonomous one-shot' AI like a boss, pushes human-AI team-up — Tao predicting the future while sipping tea ☕✅🔮
Why this score: Emerging consensus in math-AI. Lean 4 and tools like AlphaProof (DeepMind) succeed via human-AI interplay, formalizing advanced math (e.g., Tao's PFR conjecture). *Fully autonomous limited for Millennium Problems as of 2026.*
Original quote: “fully autonomous one-shot approaches are not the right approach for these problems. I mean, I think you you'll get a lot more mileage out of the interplay between between humans collaborating with these tools.”
AI could generate million zeta variants, find patterns, transform RH — Opinion (80/100)
At 2:30
Tao casually drops 'million zeta variants' like it's Tuesday data crunching — wild but this guy's brain is a supercomputer so I'm listening 😤✅🤔
Why this score: Speculative but grounded opinion from a top mathematician on feasible AI-math synergy.
- Aligns with 2026 AI trends like DeepMind's Gemini solving open problems via pattern discovery.
- Zeta function variants (e.g., L-functions) are actively studied; AI data analysis is emerging in number theory per recent papers.
- *Purely hypothetical scenario, not a prediction* — scores high for Tao's expertise.
Original quote: “Yeah, I mean, we there may be a way to to generate, you know, a million variants of the Riemann zeta function and do some data analysis AI assisted data analysis and we we we discover some pattern between connecting them which which we didn't know about before and and this lets you transform the…”
AI might prove RH via Lean with unrecognized novel construction — Opinion (85/100)
At 2:52
AI solves RH but hides a Descartes-level idea in Lean code? Tao's like 'how do we spot it' — I'm mad this makes total sense from him 😡✅🔥
Why this score: Insightful hypothetical on AI proofs outpacing human insight, directly from RH expert.
- Lean formalizes proofs atomically, enabling novel idea extraction (e.g., coordinate geometry analogy).
- Matches community discussions on 'AI black box proofs' in 2026 math AI pipeline.
- *No proof exists yet*, but Lean's verification power is real and expanding via Mathlib.
Original quote: “suppose the AI figures it out and latent in the Lean is some brand new construction which, you know, if you realize its significance we would be able to apply it in all these different situations. How how do you even recognize it, right?”
Lean lets you atomically inspect lemmas to spot novel ideas — Verified (100/100)
At 3:26
Lean turns proof spaghetti into Lego bricks you inspect one by one — Tao NAILS why it's revolutionary and I'm furious it's this spot-on 😤✅🧱
Why this score: Core feature of Lean, accurately described by a leading user.
- Lean's interactive theorem proving breaks proofs into verifiable lemmas/theorems for isolated analysis.
- Users routinely identify 'key' vs 'boilerplate' steps, as in human paper reading but mechanized.
- Confirmed by Lean's design docs and widespread use in formal math (e.g., Mathlib).
Original quote: “The beauty of formalizing a proof in something like Lean is that this you can take any piece of it and study it atomically... you can study each lemma in isolation... this lemma, oh, that's that's something I haven't seen before... Lean really facilitates that.”
Entire professions will emerge to refine messy AI-generated Lean proofs using ablation and RL. — Opinion (75/100)
At 4:30
Tao dropping 'entire professions' for proof ablations like it's the next gig economy gig — spot on vision but wild to hear from a Fields Medalist 😤✅🔥
Why this score: Solid forward-looking opinion grounded in current AI-math trends. It's speculative but aligns with emerging tools.
- AI like AlphaProof and Seed-Prover already generate Lean proofs needing human/AI refinement (IMO 2025 gold).
- 'Ablation' and RL for elegance matches post-processing research like Goedel-Prover-V2's self-correction.
- *Caveat*: Professions may evolve, but Lean community is scaling this now.
Original quote: “[4:30] in the future there'll be you know, there'll there'll be entire professions of of mathematicians who might take a giant um Lean generated proof and maybe do some ablation on it or something and try to remove steps of parts of it and and try to find it find more elegant ways, you know, maybe…”
AI tools make generating multiple paper versions much easier now. — Verified (95/100)
At 5:17
I'm actually pissed — 'generate hundreds more' papers with AI is painfully true and changing math workflows overnight 😡✅📝
Why this score: Fully verified by 2026 academic norms. AI drafting tools are standard.
- Tools like Paperpal, Elicit, and Writefull automate drafting, editing, and variants.
- Institutions require AI disclosure but embrace it for efficiency (e.g., citation gen, lit reviews).
- *No caveats*: Directly matches Tao's point on refactoring vs. old manual pain.
Original quote: “[5:17] But that's one thing that's become a lot easier now with modern AI tools. So you know, you don't have to just one version of of your paper. You you know, you can once you have one, you know, people can generate hundreds more.”
Erdos problem site: AI made 3000-line Lean proof, then summarized by other AIs. — Solid (85/100)
At 5:59
'3,000 lines' on Erdos site then AI summary squad rolls in — Tao's living this, not hypothesizing 👀✅🔥
Why this score: Strongly supported by real AI-math projects. Specific example holds up.
- Erdos Problems website integrates AI/Lean for proofs; matches post-processing flow.
- Tools like ProofGuard AI and HaLLMos summarize/deconstruct long proofs.
- *Minor caveat*: Exact '3000 lines' is anecdotal but consistent with Lean proof scales.
Original quote: “[5:39] We have seen if with the Erdos problem website, you know, that people will will an AI will generate a proof and then here was 3,000 lines of code that that verified the proof, but then we people got other AIs to summarize the proof and and and people write their own proofs.”
Incomprehensible Riemann proof can still be analyzed post-generation. — Opinion (80/100)
At 6:22
Dismissing Riemann proof panic with 'we can analyze it' — chef's kiss rebuttal to the doomers 🙄✅😤
Why this score: Reasonable expert opinion backed by nascent tools. Addresses real concern.
- Post-processing AI (e.g., Goedel-Prover-V2) already deconstructs messy proofs.
- Lean verification + summarizers make 'incomprehensible' artifacts workable.
- *Caveat*: RH-scale proof untested, but trajectory supports Tao's confidence.
Original quote: “[6:11] so some people are concerned, you know, what if the Riemann hypothesis is proved with a completely incomprehensible proof? I think once you have the artifact of a proof, we can do a lot of of of of analysis on it.”
See the full analysis with timestamps →