The Mathematicians Are the Canary

OpenAI has dumped around 719 maths manuscripts onto GitHub, claiming answers to problems people have spent years, sometimes decades, trying to solve. Some of the mathematicians closest to those problems are, understandably, devastated.

If your first reaction is "brilliant, more maths, that's progress", imagine opening your laptop and finding the work you hoped would define your career already sitting on GitHub. You might be delighted by the discovery. You might also be wondering what you're doing next Monday morning…

I think the mathematicians are the canary for the rest of us.

Writers and artists have already felt it and programmers are living through it right now. Next comes scientific research, then the professions: legal, medical, financial, accountancy. If you sell your expertise in any way, shape or form, watch how this plays out.

What OpenAI actually released

A couple of weeks ago OpenAI announced an independent advisory group of mathematicians to help it "responsibly share advances in AI and mathematics", including how to assess and communicate new results. Sounds great. In the very same moment it put 719 manuscripts, organised into 372 families, on GitHub, with Lean libraries so the proofs can be checked.

So: we want to learn how to communicate better with you. Oh, and by the way, we solved all this maths you’ve been struggling with. Suckers!

Back in August I covered Astra's ten maths advances, and the argument then was whether ten results held up. This is 719. They won't all be unique, they won't all be correct and they won't all be useful. But the machine is now producing proposed answers faster than the people in the field can absorb them, and every one of those papers is work the profession has to go through.

Highly recommend reading (from the horse’s mouth) the reactions of various mathematicians here https://proofsandprompts.com/2026/10/08/100-reactions-to-100-solutions/

Matt Zaremsky has spent four years on the Boone-Higman conjecture in group theory, with a breakthrough roughly once a year. Then according to him OpenAI spiked the ball. The missing piece looked a lot like his team's own 2025 breakthrough, and he reckons they could have got there in another year or two. To quote Matt:

The skeleton has arrived. The process of uncovering it has gone.

Tristan Humbert woke up to news of a proof of Katok's entropy conjecture, the three-decade-old problem his PhD was about. He's applying for postdocs, so he spent the morning rewriting a research plan that was suddenly obsolete.

Plenty of people online are mocking them: you wanted it solved, it's solved, why are you whinging?

Andrew Curran put it better: "Please resist the urge to mock the mathematicians struggling through this moment. Whatever your field, your expertise, your passion, you too shall one day experience what they are going through."

Your version of this

Suppose you're a consultant. You've spent years building a model: workshops, examples, a way of helping clients. You charge thousands because it's rare. Then a cheap tool turns up producing a decent version that gets you 80% of the way there.

Your work can still help people. But the deal you thought you had with the market has changed. What you charged was set by how hard it was to get that expertise anywhere else, and now it's available for £20 a month. The market doesn't care how many years, degrees or clients you have behind you. If something almost as good is cheaper, you're done. So you have to work out who is going to pay you now.

And if you're reading this thinking "yes, I can see this happening to loads of people, but never to me, I'm a special snowflake"... you're not paying attention. Sorry.

"We'll just move up the chain"

Terence Tao has a great interview on Numberphile about maths moving from scarce proofs to abundant proofs, with human understanding struggling to keep up.

Something gets lost too: the years spent on a problem are where people learn how the subject works and find the next questions. In his view the work left is checking, explaining, connecting results with other ideas and deciding what to investigate next.

You can see the business version. Ask AI for 20 client proposals and you'll have 20 polished ones in minutes. You still have to decide which fits the customer, which makes promises you can actually deliver, and what happens after they say yes. So the value moves up the workflow to judgement. That's the line a lot of AI experts take: the machines can't do discernment, they haven't lived in our world, so we'll hand them the grunt work and move up a level.

I don't think this is going to work. It’s cope.

Again and again we've drawn a line and said this bit is for humans. Then we redraw it. The machines will get better at deciding which of those 20 proposals is best. In maths they'll get better at checking, explaining, connecting and choosing what to investigate next, probably within a year or so. I rate Tao enormously and recommend anything he does, but on this one I think he's in too deep - he can’t see what’s happening around him. Or he does and can’t accept it.

So I'd be very wary of building your future on "AI can do this bit, but it will never manage the bit I'm paid for". We humans keep getting that wrong. Repeatedly!

Sell what people do afterwards

So what do WE do?

Information is free. It has been for a while with the internet, and with generative AI it's free and instant. You can't monetise it for long . Stop trying - that’s a dumb game to play now.

What's still in short supply is action, and there's even less of it now, because people faced with this much information and this many options tend to freeze.

So the questions to ask about your own work are about what changes for the customer. If you teach, what can your students do after working with you? If you advise, what gets decided or improved? If you build software, what can the customer now accomplish? Habits, behaviour change and community are person to person, and still hard. I went through how to package your process around the result in What Will Clients Pay You For?, from a simple skill up to a full product.

The other half is paying attention. If you see a trend coming that will wipe your business out, get out of the damn way. I did this with my digital marketing agency. When GPT-3.5 came out I thought this could be a problem. By the time GPT-4 arrived, which was genuinely good, I'd shut the business down. I knew we weren't going to survive it.

You need to be keeping your head on a swivel too. And not (like Dr. Tao) ignoring the evidence in front of you.

Do this today

Greg Isenberg tweeted about a dinner where someone asked the table: if a new model drops next month, does your company get better or worse? 

Half the table went quiet.

SUCH a good question. Shit…it is THE question.

Ask it of yourself, whether you're employed, freelance or running a business.

Paste this into ChatGPT or Claude and answer its questions properly:

Open it in ChatGPT or in Claude

If the answer is "less valuable", take the three changes seriously and pick one to start on this week.

If you only watch one bit

• 06:23 The archaeological dig, and the trillion dollar company with the TNT

• 13:28 Why I don't buy "we'll just move up the chain"

• 17:16 The one question to ask about your business

The full video goes through more of the mathematicians' responses (collected by Proofs and Prompts) and the Terence Tao interview.

Don't compete with the models. You won’t win.

Instead build something that gets better every time a new one drops.

To the Task,

Kyle