Tessera

Folgen

1 Habicht

@habicht.bsky.social

Hallo

7 Posts43 Follower 107 Folgt

👁️🍄🪸@bloomfilters.plus · last mo.

the mathematics award winning fields medalist and known himedanshi yu deng just said that if ai puts all mathematicians out of a job he's gonna go home and write yuri novels. girl please do it anyway i need to see what yuri a fields medalist mathematician can do omfg

🇵🇸Tim Henke (tɪm 'ɦɛŋ.kə)@timhenke.bsky.social · 2mo ago

Prompted by the scandal around Navier-Stokes, Andreas Thom raises the possibility that OpenAI may also have stolen the work on non-sofic groups:

1/3 I want to record a number of points that I realized after reading up on the controversy. In fact it recalls an exchange I had with OpenAI after its non-sofic-group announcement. It exposes a serious, unanswered question about whether researchers can trust OpenAI with unpublished mathematics.

I wrote in an email to Mark Sellke and Sebastien Bubeck shortly after the surprising finding of a non-sofic group using the methods of Kun and myself:
“Another point is that I and a colleague in Dresden were discussing the expander matching problem and various extensions of the work with Gabor Kun actively over the last months with ChatGPT, so that we are of course curious if that was part of the training data or accessible to the reasoning process. There is a certain (frankly unacceptable) lack of transparency here; and I fear it will damage the communal process of math more than the new AI-generated results will benefit the subject.”

Mark Sellke’s complete answer was: “Regarding your conversations with ChatGPT: that did not happen.”

I had explicitly asked about two different things: (1) whether our conversations entered training data, and (2) whether they were accessible to the solving process. The categorical answer now looks as though it addressed only direct access under (2). No such qualification, explanation, or evidence was given. I take this as dishonesty to say the least.
OpenAI now says in the Buckmaster-Alpöge case that no specific user data was accessed, but adds that it “cannot rule out that de-identified data derived from their usage of our products helped improve our models.”

2/3 OpenAI is drawing a distinction that its answer to me erased, despite the fact that my question explicitly made that distinction.
We are not required to reverse-engineer OpenAI’s internal training pipeline to establish what happened. Only OpenAI has the relevant data for that. If Sellke and Bubeck give a categorical denial, it must disclose the basis: product and privacy settings, relevant datasets and checkpoints, and what “de-identified data derived from usage” means.
I disabled model training on 29 June. That control is still only a promise whose implementation users cannot audit, and it is prospective: it does not answer what happened to earlier conversations or to derivatives already selected. OpenAI’s answer did not mention the setting or any account-specific check.
There is also a mathematical parallel, though a less explicit one. The approach of Kun and myself was not the main line of attack on non-soficity, in fact there were other more promising approaches along the line of quantum games etc. OpenAI’s detailed command of our techniques therefore made me wonder how the model found this route. If nonpublic research supplied by users improved a model and the provider then used that model to race those users to publication - without informed consent, disclosure, or credit - that would be ethically indefensible. De-identification may remove a name; it does not remove the intellectual content of a mathematical idea. Sellke and Bubeck seem to be blind to this simple moral aspect.

3/3 Sellke’s categorical answer was, at minimum, unjustifiably broad and materially misleading; looking back it was plainly dishonest. Bubeck’s acknowledged attacks and his attempt to exclude Alpöge from a proposed OpenAI-proof paper deepen the concern that there is a loss of moral compass. Taken together, this raises serious questions about their judgment and personal integrity.

1 Habicht @habicht.bsky.social · 2mo ago

Wir können wissen. Wir werden nichts lernen.

Sridhar Ramesh@radishharmers.bsky.social · 2mo ago

I'm against any suggestion of shutting down AI math. Be for real. Wir müssen wissen, wir werden wissen. Try it on even more things and publish the results. However, I'm still going to moan and mourn.

1 Habicht @habicht.bsky.social · 3mo ago

Mr Altman is helping me prove my conjecture.

Jared Holshouser@jaredholshouser.bsky.social · 3mo ago

It's genuinely upsetting to see high level mathematicians be so un credulous at the motives of big tech. No ethical thoughts at all, seemingly. Just oh look this shiny toy they were "nice" enough to give me spat out a theorem.

1 Habicht @habicht.bsky.social · 5mo ago

We do have to wonder whether he'd have been able to it so quickly without early access to this result. Certainly has "a G'schmäckle".

Suresh Venkatasubramanian @geomblog.bsky.social · 5mo ago

Buried the lede :). There's already an improvement on this result by Will Savin. To n^{1.014}. arxiv.org/abs/2605.205...

1 Habicht @habicht.bsky.social · 6mo ago

This has to be my favorite math book cover.

The cover of Folland's Introduction to PDE

jackie ess @essj.bsky.social · 9mo ago

people are underselling the role of Misha Gromov as a fixer for Epstein, someone who helps him extend his network of influence by softening his reputation and making high profile introductions. Perhaps non-mathematicians won't know these names but this is not a small thing

https://www.justice.gov/epstein/files/DataSet%209/EFTA00688479.pdf

From: Misha Gromov
To: "jeffrey E." <jeevacation@gmail.com>
Subject: Re:
Date: Wed, 01 Mar 2017 16:08:46 +0000
may be we talk by Skype, mine is
On Wed, 1 Mar 2017 09:49:03 -0500, jeffrey E. wrote:
can you send james simons a note. telling him that i am not as bad as google would lead one to believe , I
understand his wife is concerned.

https://www.justice.gov/epstein/files/DataSet%209/EFTA00682542.pdf

From: Misha Gromov
To: "Jeffrey E."
Subject: Re:
Date: Sun, 08 Nov 2015 20:15:07 +0000
Maxim Kontsevich is " the sharpest cutting edge" mathematically but not conversationally. Alain Connes, despite
his age (a year younger than myself) is full of fire and ideas.
Francois Taddei, a biologist, has a global view on science education.
Yves Oudeyer has a non-trivial vision and results on robotis
On Sun, 8 Nov 2015 20:34:05 +0100, jeffrey E. wrote:
no problem , see you in new york , cambridge. if you think i should meet someone here that you find cutting
edge. I d appreciate it
On Sun, Nov 8, 2015 at 7:41 PM, Misha Gromov a wrote:
This is tough, I leave for NY in four days and have only a couple evenings free.
On Sat, 7 Nov 2015 20:09:11 -0500, jeffrey E. wrote:
Im here in paris. if you have time

doctor eek (alt decorative ghost season)@absolutely-not.bsky.social · last yr.

i looked at the methodology for this and it is a. sex addiction counseling group in texas did a surveymonkey and extrapolated the results to the entire us population which is the sort of research design that earns you an ff on an intro methods class (the extra f is for extra effort), and b. p-hacked

The Independent@the-independent.com · last yr.

Nearly a third of Americans have had a ‘romantic relationship’ with an AI bot, new survey says

The Museum of English Rural Life (MERL)@themerl.bsky.social · 2y ago

look at this absolute unit

The absolute unit sheep; an Exmoor Horn ram

Cats of Yore @catsofyore.bsky.social · 3y ago

Drama! 1889. blogs.ncl.ac.uk/speccoll/tag...

Very amusing ink and watercolor painting of two cats, one orange tabby and one white with black spots and tail, in fighting stances with locked eye contact. The tabby is crouched low and the black and white cat has its back arched and tail poofed. They seem to be on some kind of bamboo and shingle rail of some kind, raised above the ground. A tiny sliver of crescent moon hangs in the very pale blue sky behind them.