+- MKLab (https://mklab.gr)
+-- Forum: [INDEX] (https://mklab.gr/forumdisplay.php?fid=1)
+--- Forum: ARTFICIAL INTELLIGENCE (AI) (https://mklab.gr/forumdisplay.php?fid=5)
+---- Forum: ARTICLES (https://mklab.gr/forumdisplay.php?fid=33)
+---- Thread: Maths experts discover what artists already know (/showthread.php?tid=2021)
Maths experts discover what artists already know - mklabgr - 09-16-2026
Quote:Thid an article from The Daily Telegraph @ ARTICLE
Maths experts discover what artists already know – AI eats you up
ANDREW ORLOWSKI
OpenAI and Anthropic are ruthless commercial entities hurtling towards a big executive payday
WELCOME to the party, maths. This week, top mathematicians are discovering what artists, creative industries and businesses have all learnt the hard way: you don’t consume AI – AI consumes you. Or at least, some do.
“Is it ethical to use a customer’s data to try to scoop their customer?” a maths professor asked last week. Of course it isn’t, but it’s important to know which AI company you can trust.
Posing the question was Tristan Buckmaster, an Australian mathematician at New York University, who, along with Levent Alpoge, a Turkish mathematician who works at Anthropic, had been tackling a gnarly unsolved problem called the “Navier-Stokes existence and smoothness problem”.
It is one of seven Millennium Prize challenges, where successful proofs command prestige and a $1m (£740,000) prize. In private, and at their own expense, they had hit upon a novel approach and had used a variety of AI models to help them.
By mid-August, Buckmaster and Alpoge found they had made significant progress on Navier-Stokes, but it was nowhere near ready to present to the world.
A couple of weeks later, however, Buckmaster began hearing troubling rumours. Apparently, details of their work had been obtained by OpenAI, whose models the pair had also been using. He politely asked OpenAI whether this was true.
The company stalled, but last week OpenAI declared it had cracked the problem too and had produced a foolproof method using a similar approach to the one Buckmaster and Alpoge had explored, but with much more computing power.
This was too much of a coincidence for the Aussie. It was the same “route... Levent and I had quietly chosen to attack. Almost nobody else I know of was working on it. It is not the direction one arrives at in a few days by giving a model the problem statement”, Buckmaster claimed.
Buckmaster suggested that OpenAI’s systems may have harvested his private conversations and calculations with the chatbot about the maths problem, helping it solve a conundrum in hours that the professor had been working on for months.
Things then grew darker. “Why would you ruin your career?” an OpenAI scientist had asked him. OpenAI denies making the threat.
Yes, Sam Altman, the OpenAI chief executive, had heard about Buckmaster and Alpoge’s work. Altman admitted: “We were curious if ours could do it too.”
It was then claimed that OpenAI might have trained its systems on the work of Andreas Thom, another maths professor at the University of Milano-Bicocca, who had been sharing his calculations in conversations with the chatbot.
“We are no longer looking at isolated incidents.
“We may be looking at one of the greatest intellectual scandals in the history of science,” wrote Valerio Capraro, Thom’s friend and professor.
“AI is not discovering new mathematics. AI is stealing human discovery.”
OpenAI says no specific user data were purloined in creating its Navier-Stokes solution, but it “cannot rule out that de-identified data” have leaked into its models.
Scientists now question OpenAI’s motives for giving 10,000 (and next year, 100,000) researchers free access. Is it a Trojan Horse? Once they have all that academic data, what will they do with it?
“Welcome to our world,” replied the copyright holders. After scraping everything they could from the public web to train their models, they moved on to business.
When you use Anthropic and OpenAI’s technology, they are extracting the underlying patterns of how companies work. This includes the “workflows, domain knowledge and accumulated judgment”, Satya Nadella, the Microsoft chief executive, warned in June.
This knowledge is then commoditised by being offered to everyone else. Unable to make a profit, the original company finds itself replaced by Anthropic or OpenAI.
Palantir’s Alex Karp is even more scathing about what he calls a “parasitic” relationship between “Dario [Amodei, the Anthropic chief executive] and the EA [effective altruism] crew” and their customers.
“They are trying to drug us into a future that only they control... where we own nothing and our businesses aren’t profitable,” he says.
Earlier this year, I called this the “Unicompany” proposition: profit-making will become impossible for many sectors that deal with information, from retail to healthcare, once AI commoditises them.
Nadella and Karp are speaking out because they can host safe AI models for a customer and firewall them against any knowledge extraction. Both also promise to honour the customer’s privacy.
“We stole all your knowledge and art, and now we’re gonna put a meter on it and sell it back to you. You’re welcome,” is how Ananyo Bhattacharya, a science writer of the London Institute, sums up the maths controversies.
A close reading of the story confirms that it is not AI that the mathematicians are objecting to, per se.
Many I spoke to last week were cautiously positive. Large language models are becoming useful for generating examples, as well as conducting the kind of brute-force attack that OpenAI applied to Navier-Stokes, they say.
However, several, along with Terence Tao, the prize-winning mathematician, warn of a deskilling effect, as AI cannot build the practical, enduring theories future mathematicians, physicists and engineers need.
So it is less an AI problem than an ethical one about the behaviour of two companies: OpenAI and Anthropic.
Both pretentiously style themselves as “labs”, when in reality they are ruthless commercial operations accelerating towards the biggest executive payday in history. Trust in business is hard to recover.