… wait, that’s it? Let me show you the whole thing.
A 23-Year-Old With Zero Math Training Just Cracked a 60-Year-Old Problem — In ONE ChatGPT Prompt
A 23-Year-Old With Zero Math Training Just Cracked a 60-Year-Old Problem — In ONE ChatGPT PromptHe didn’t even know it was famous. He was bored on a Monday. Now Terence Tao (the math GOAT) is talking about it.
The receipts: 1 amateur (Liam Price, 23, no advanced math). 1 prompt. 80 minutes. A problem that smart people had been stuck on since the 1960s. Solved with a trick nobody tried in 90 years.
Look, this isn’t another “AI is scary” story. This is a regular dude feeding a hard question to a chatbot and accidentally making math history. The whole thing is written up in Scientific American. And it cracks open a door that’s been locked your whole life.
🧩 Dumb Mode Dictionary (read this first, no shame)
| You’ll hear… | What it actually means |
|---|---|
| Erdős problem | A list of brain-buster math puzzles left behind by a legendary mathematician. People have been chipping at them for decades. There’s even a whole website tracking which are still open. |
| “Primitive set” | The puzzle’s topic — a bunch of numbers where none of them divides cleanly into another. That’s it. Sounds simple. Wasn’t. |
| GPT-5.4 Pro | The beefy paid version of ChatGPT. The one that actually thinks for a while before answering. |
| Terence Tao | Basically the Michael Jordan of living mathematicians. When he says “huh, neat” — the whole field listens. |
| “Vibe maths” | Slang for what this guy did: throwing problems at AI to see what sticks, no formal training. |
📖 How a bored Monday turned into history
Real talk: Liam Price wasn’t trying to be a hero. He was just feeding random problems into GPT-5.4 Pro to see what it’d spit out.
- He typed in Erdős Problem 1196 — didn’t know world-class brains had been stuck on it for 60 years.
- The AI cooked for about 80 minutes and handed back a fresh solution.
- He posted it to erdosproblems.com and pinged a buddy — a 2nd-year Cambridge undergrad — who looked at it and went “yo… this is real.”
- Experts confirmed it. Game over.
🧠 Why this one's actually wild (not hype)
Here’s the thing. Every human who ever touched this problem started the SAME way — same opening moves, like everyone runs the same chess opening.
The AI? It just… ignored all that. Pulled a formula from a totally different corner of math that nobody had thought to point at this problem in 90 years.
Quote from Tao himself in the write-up: “There was kind of a standard sequence of moves that everyone who worked on the problem previously started by doing. The LLM took an entirely different route.”
Translation: the wall was never that high. Everyone was just climbing the same spot.
📊 The numbers that matter
| Thing | Number |
|---|---|
| Age of the problem | ~60 years |
| Years that formula sat unused on it | ~90 |
| Prompts it took | 1 |
| Time to crack | ~80 min |
| Liam’s math degree | none |
| People who can now try this | literally you |
🗣️ What the timeline's saying
- Math nerds split into two camps: “this is the future” vs “the AI got lucky on an easy-but-overlooked one.”
- Tao’s take (the diplomatic one): the problem was “maybe easier than expected” — like there was a mental block the whole time.
- The spicy part nobody wants to say out loud: if ONE bored amateur did this, there are thousands of half-forgotten open problems sitting in public databases right now. Free. Waiting.
Cool. A Random Guy Beat 60 Years of Experts From His Couch… Now What the Hell Do We Do? (•̀ᴗ•́)و
🕳️ The Open-Problem Bounty Bridge
Look, public databases are STUFFED with unsolved problems that come with credit, prizes, or pure clout if you crack them. Most people never look because they assume “I’m not smart enough.” That assumption is now outdated.
The play: grab open questions from erdosproblems.com, the OEIS (a giant database of number patterns), or open MathOverflow threads. Feed them to a frontier AI. When something sticks, post it with your name on it. Reputation = your bag.
Example: A 21-year-old comp-sci student in Manila pulls 5 “open” tagged problems off MathOverflow, runs them through GPT-5.4 Pro over a weekend, gets one partial result that a professor confirms is new. First real citation with his name on it before he’s even graduated.
Timeline: First credible attempt in days. The easy/overlooked ones get hoovered up fast now — real window is maybe 6-12 months before everyone’s doing it.
🔍 The Proof Checker Middleman
Here’s the gap nobody’s filling: tons of amateurs are about to generate AI “solutions” they can’t actually verify. A wrong proof that LOOKS right is worthless — and embarrassing.
So become the bridge. Connect the AI-generators (excited, clueless) with the people who CAN check (grad students, math PhDs broke as hell). You take a finder’s fee for matching them up. You don’t need to understand the math — you understand the market.
Example: A 24-year-old in Lagos sets up a simple Discord + a Stripe link. Charges $40 to pair an amateur’s AI proof with a vetted math grad for a sanity check. 30 checks in month one. The grads are happy for side cash, the amateurs avoid looking dumb. Everybody eats.
🪟 The Patch-Window Sprint
Real talk: right now, math and science competitions haven’t fully figured out their “AI allowed?” rules yet. There’s a messy in-between window where AI-assisted submissions are either welcomed or just… not banned yet.
Hit open competitions — the AI Mathematical Olympiad, Kaggle science challenges, Polymath collab projects — while the door’s open. First movers stack prizes and recognition before the rulebooks tighten.
Example: A 22-year-old in Kraków enters a Kaggle math-reasoning comp using a frontier model + clean prompting. Doesn’t win first, lands top-5%, walks away with prize money and a portfolio piece that gets him a remote AI-tutoring gig.
📖 The Vibe-Math Cheat Sheet (be the dictionary)
When something brand new blows up, the FIRST person to write the clear “how to actually do this” guide owns the search results. Right now there’s no clean playbook for “how to throw open problems at AI and not embarrass yourself.”
Build it. A tight, free guide + a paid pack of battle-tested prompts for poking at unsolved problems. Host the free version, sell the prompt pack on Gumroad. The free guide pulls traffic; the pack pays rent.
Example: A 25-year-old in Jakarta writes a 2,000-word “Vibe Math 101” post, gives away the basics, sells a $12 prompt pack of 30 tested templates. Ranks on Google for “solve math problems with AI” within weeks. ~150 sales month two.
📡 The AI-Shaped Target List
Here’s the picks-and-shovels move. Everyone’s hyped, but nobody knows WHICH problems are actually “AI-friendly” — the overlooked, low-hanging ones like 1196 was. That curation is the real gold.
Don’t just do math. Comb open datasets in chemistry, biology, materials science (Zenodo, Kaggle Datasets, uni open-research lists) for problems that smell crackable. Sell the curated hit-list to students and indie researchers who’d kill to skip the hunting.
Example: A 23-year-old in Nairobi compiles 50 “probably AI-solvable” open questions across 4 fields into a clean Notion doc. Sells access for $9 on Gumroad. Researchers and grad students buy it to save weeks of digging. Slow drip, but pure profit — it’s just a list.
🛠️ Follow-Up Actions
| Want to… | Do this |
|---|---|
| See the open problems | Browse erdosproblems.com & OEIS |
| Try the actual tool | Get GPT-5.4 Pro / ChatGPT (the thinking one) |
| Find checkers | Lurk MathOverflow & math Discords |
| Enter a comp | Check AI Math Olympiad + Kaggle |
| Sell your guide/list | Set up Gumroad — free to start |
Quick Hits
| You want… | Then do this |
|---|---|
| To understand the news |
Read the Scientific American piece |
 An open problem to attack |
Pick one off erdosproblems.com |
 The right AI for it |
Use the Pro / reasoning model, not the free quick one |
 To monetize the wave |
Be the checker, the guide, or the list — not just another solver |
| Proof you don’t need a degree |
Liam Price didn’t have one. That’s the whole point. |
The wall was never that tall. Everyone was just standing in the same spot. Go stand somewhere else.
!