This past week, OpenAI dumped a staggering trove of answers to elusive math problems solved by its internal artificial intelligence models – the mathematics world isn’t happy about it. “What are they trying to achieve? I don’t know. They’re just destroying early-career mathematicians with a PR stunt,” says NYU math professor Tristan Buckmaster. Buckmaster and his collaborator Levent Alpöge had made significant progress on solving an historic math equation before OpenAI’s model beat them to it. They’ve “destroyed the whole incentive system for science and mathematics.”
The most interesting bit in this interview for me is when Tristan Buckmaster explained how OpenAI stole his PhD students’ work: the students didn’t even have to use a LLM. All they had to do was apply for a grant for their work, and for whoever reviewed the grant to write the summary using ChatGPT, and hey-presto, OpenAI got the outline of their work in progress to feed to their model and front run them.
I’m going to make my company’s CTO watch this. He thinks our software is safe because the company uses paid models and the contract says the material submitted to them won’t be used to train other models - and he believes it…
deleted by creator
I really wonder if they don’t understand that they change from a distributed dependence for labour to a single dependency for labour and what that means for their position in their eco system.
Dear boss, 50 people won’t strike or leave the company all at once for no reason at Monday 9:38am. Your ai overlords might decide change the terms of the contract randomly on a Monday morning and there is nothing you can do. You had effectively a crown and gave it away.
My company is not like that. They care about the employees. It’s actually written in black and white in the company’s charter: the purpose of the company is to do good by its employees and the community it operates in.
I’ve brought up my concerns about AI before and they’ve reacted: they invested in machines capable of running models locally for the engineers (I have one on my desk) and they formed a committee on AI safety and building local infrastructure to untie ourselves from the big cloud AI providers.
My company is one of a rare few such companies. Most other employers are like yours. Yet as good as mine is, AI is like cocaine to those who use it and to the bean counters: they know it’s bad but they just can’t stop.
deleted by creator
No idea why they would ignore such direct existential danger for their companies.
Risk Management (that’s what they named the LLM after they fired the people responsible for that) is probably telling them that there will be a huge pool of willing applicants to work for less money than the people they let go, that talent is fungible, etc.
The real fun part is when people start asking for hiring bonuses to come in and unfuck the company slop. They won’t start trying to hire people until they start losing customers which means the quality of the product is lower than it was when they took on the customer. If it takes them longer to fix their codebase than it would for a competitor to start from scratch and outdo them on marketing or advertising, then they’ll go under.
This is so fucked up.
…the contract says the material submitted to them won’t be used to train other models - and he believes it…
This is what I just don’t get, that people trust major companies when they claim to not analyze or use your data in any way they can once you have uploaded it to them.
It is a simple equation, how much does the actual breach of contract cost, how likely is it that we get sued and loose, and how much better will our models get by analyzing and training on everything?
Unless you can honestly say that a huge corporation like Microsoft/Google/Amazon would pay more to you and every other customer for missusing your data than they will earn by improving their models on all their customer’s data, you can’t use proprietary data in an LLM/AI.
I miss the old network shares, now it is all OneDrive and SharePoint, it sucks.
The techbros don’t care, if anything they will welcome to minimum wage workers into their factories.
It’s no longer safe to have open research since slop merchants are going to steal it from under you and pretend they made it.
I hope "Open"AI gets distilled into bankruptcy.
Jensen said that people don’t need to be able to count, they just need to be able to shovel coal, and be turned into soylent green.
At least Soylent Green was predicated on the overpopulation scare of the past half century, now with humanity facing population collapse by the next half century the hoarding of wealth from the dwindling population is even more egregious.
Maybe we just need to make wealth redundant?
It feels like there’s a lot of doublespeak going on in this clip. One sentence says “OpenAI didn’t solve shit and doesn’t deserve any credit for anything” and the next says “LLMs are revolutionizing how we do math.” I’m not sure we can really conclude whether LLMs have any actual application in this kind of advanced mathematics… or anywhere else for that matter.
When I machine a part, the lathe does most of the heavy lifting, but it’s a tool and it doesn’t deserve any credit for anything. Yet without it, it would take me hours to file a round part imperfectly that takes me 10 seconds to turn to perfection on the lathe. The lathe is absolutely essential to my workshop.
Two things can be true at the same time.
Maybe. But it’s not as if all the LLM companies don’t have a history of making things up out of whole cloth to make their products seem more useful for real-world problems than a puddle of vomit.
“We solved a millennium problem” sounds impressive until you find out they just stole it and then pushed the one who deserves the credit to cut out the other one who deserves credit. So what are we going to find out about the other 722 that makes it clear the whole story was a scam?
I don’t think that’s what doublespeak is. That’s just contradicting statements. Doublespeak is when you’re using language that obscures the meaning of the words used.
Feels so fake, a math problem, like which one… and solved? In what way, what are we even talking about, it seems to me just a bunch of buzzwords put together.
So yeah I was completely right…
They didn’t solve anything, they just stole the data from a team and elaborated the data in a very very efficient way, which is way faster than humans could.
First of all: The main data was done by humans, OpenAI just acted as a calculator.
Second: The repository of OpenAI makes no sense since some results are, again, just stolen data with no actual added argument.
And third: AI here is not replacing, it’s just causing damage to people’s career, cause scammers use AI to steal data from actual scientists.
If and when AI will be an available tool like a calculator to every team, everyone can speed up their game at the same rate; what we’re watching here is just humans being humans in a capitalistic society: Scammers see a new advantage, scammers use the advantage early before it gets regulated.
I haven’t watched the video, but I just have to get the word out
https://arxiv.org/pdf/2610.08144
From the abstract:
(…) we provide several examples of AI mistranslations of NL statements and proofs into Lean in practice, resulting in mismatches between NL proofs and their Lean ‘verifications’. These include OpenAI’s announced Navier-Stokes proof.
Wow, yet another way AI will kill progress and make us all stupid
That last line. That’s exactly the point.
Link to article?
