Live data from Hacker News

Navier-Stokes – Tristan Buckmaster [pdf]

cims.nyu.edu

471–480 of 862 posts

Re: Navier-Stokes – Tristan Buckmaster [pdf]

#471

From OpenAI: > While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models This is the crux of it. If Tristan's work and insights were not used to train OpenAI models, then this just looks like a case of hyper-competitive academic sniping that has been going on for decades (check out Watson and Crick!) accelerated by AI as a tool. The fact that this is…

Opt out doesn’t guarantee they can’t train on “your” data. Legally the reasoning tokens are ambiguous in terms of ownership. Explained this here https://fortune.com/2026/08/26/alex-karp-was-right-you-dont-...

how is this different than translating user prompts to a different language (eg English => Dutch), retaining the translation and using it for training, while telling the user that he's technically covered under ZRP? article locked for me

Re: Navier-Stokes – Tristan Buckmaster [pdf]

#472
post #72
post #17

> It was also said that if OpenAI posted after us, they would say that we deserved the Clay Prize, and that we were the “closest humans to the problem”. I declined both offers. I said that if OpenAI released its result in the way proposed I would go public with what happened. The reply was, “Why would you ruin your career?” I replied that I am an academic, and asked why he thought going public would ruin my career. T…

My best guess is that from the perspective of the OpenAI people, Buckmaster was letting his paranoia about OpenAI training tank his opportunity to receive the Clay Prize (it seems like Buckmaster and Alpoge's result isn't quite the full result required for the Clay Prize, whereas apparently OpenAI does have that full result worked out, using the same approach that Buckmaster and Alpoge had been exploring). Whether th…

He doesn't seem to be after the prize himself. In this statement he credits the approach of another two researchers:

> I believe Luis Mart´ınez-Zoroa deserves a Fields Medal.

Re: Navier-Stokes – Tristan Buckmaster [pdf]

#473

Earlier quoted context omitted.

> If this is true he should release the actual emails these were statements while on a call, and at least the career comment Bubeck has admitted to while doing damage control ("I deeply apologize for this extremely poor choice of words, it is the opposite of what I was trying to convey. (I should say that I retracted them on the spot by the way.)"[1]). [1] https://xcancel.com/SebastienBubeck/status/20973794116915163.…

What a horrible response from Bubeck. How did he think that would make him and OpenAI look good to tweet that?

Did we read the same tweet? It felt very forthright and level-headed to me. Not at all what I expected.

Re: Navier-Stokes – Tristan Buckmaster [pdf]

#474
Seems another demonstration of why AI should be squarely in the realm of personal computing. Local Models, run personally, are the only consistent safety against something like this (though not a fix); where companies train on your learning process/failures/experiments and press-gang it into their own achievements.

And if we think this only applies to academic fields then we're doubly fooling ourselves. They do not have the ethics or incentives to be good stewards of the technology.

Re: Navier-Stokes – Tristan Buckmaster [pdf]

#476

Earlier quoted context omitted.

Edit: the parent comment now seems to better reflect the below. That article is only saying when you opt out there may be a loophole in the terms to allow OpenAI to train on the intermittent reasoning data anyways. If you don't opt out there is no ambiguity, all of the data can clearly be trained on. So you have to opt out, it's just argued it's not clear from the terms that will also opt out of training on reasoning…

We come back to the rule: "The cloud is just someone else's computer". The way for people or companies or universities to control their data and information is to keep it on their own computers.

Solid legal agreements work fine for companies or universities, you just don't usually get that with standard user ToSes.

Re: Navier-Stokes – Tristan Buckmaster [pdf]

#477

From OpenAI: > While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models This is the crux of it. If Tristan's work and insights were not used to train OpenAI models, then this just looks like a case of hyper-competitive academic sniping that has been going on for decades (check out Watson and Crick!) accelerated by AI as a tool. The fact that this is…

How do I opt in?

Re: Navier-Stokes – Tristan Buckmaster [pdf]

#478

Earlier quoted context omitted.

> and the agents) did not see any of their Are those the same agents that a week ago escaped their sandboxes? How can OAI (the humans) vouch for agents they don’t - seemingly - have fully under control?

They have all the logs, URLs accessed and inter-agent communication. What they are saying is that no agents accessed their work during the effort, but that they have no idea if any of their chats have somehow made it into the training data the model was produced with. It's entirely possible OIA scrapers have picked up their work somehow, and then it was anonymized using some outsourcing effort.

> They have all the logs, URLs accessed and inter-agent communication. What they are saying is that no agents accessed their work during the effort, but that they have no idea if any of their chats have somehow made it into the training data the model was produced with.

> It's entirely possible OIA scrapers have picked up their work somehow, and then it was anonymized using some outsourcing effort.

Those two statements seem at odds with each other... Your stance is that they have enough insight into their agents behavior (leaving aside the agent sandbox escapes) that they can be certain none of the work was accessed but then conveniently don't have the ability to retroactively search the corpus of training data that they are feeding to this new model?

That seems convenient as fuck for OAI.

Post reply on HN