The last two points are disputed/sound significantly more reasonable in [0]. So from what I gather, Buckmaster realizes sometime during the call that the biggest result of his career is going to get steamrolled (the blowup of Navier Stokes is a much bigger deal than the blowup of 3D Euler), and on the other hand the openAi guys realize that they are basically talking about internal results with Anthropic and probably have to call corporate right after this call. Between these two stressor the conversation appears to have gone somewhat poorly.
Aww, script kiddies first technical document. ... (checks calender) Actually it's 2026, probably someone just typed "Stop me eating ice cream" into claude code instead of the chat interface.
Checking Amazon UK[0], it seems there are 18g, 39.4 g and 51 g Mars bars (and 57 g Mars Protein bars). So it's not clear weather a 60 g Mars bar is that much of a good example of shrinkflation, but 40 g seems to be the wrong comparison.
The paper is from the 30th of April this year, openAi announced the counter example to the unit distance problem on the 20th of May. That is to say this paper seems to have aged not much but quite poorly.
It impressive how prescient the technology feels. I mean in 2020 I would have thought the fully automated kitchen and the voice that asks which poem one wants to hear as fanciful, now it's kinda tech demo level.
Related, a writing advice I stole from Neal Stephenson is to write the first draft by hand. The thing is, there are a lot of small corrections where you kinda should change the text but nah, and if you already committed to copy the entire thing than you are already working at that sentence anyhow.
The author is a mathematician and I think to a certain extend the tweet reflects the current panic among (some) mathematicians. So ~~second sentence~~ third paragraph the claim that somehow ai could right now write a phd in physics or sociology is something we don't observe (at the moment). What we observe is, that ai can find counter examples to well established conjectures in mathematics quite well, but the thing is the other fields don't have the kind of well established riddles that currently produce the flashy results in mathematics.
To show my ignorance in mathematics a bit: do you feel that having such neatly defined riddles gives the AI an advantage in solving them?
A lot of Innovations or insights are obvious in hindsight, but no one thought to consider the problem, and put the pieces of the solution together. In this sense a well defined problem is a large portion of the solution as well.
I mention this because I feel AI software agents have a huge advantage due the body of prior work available to them and how provable solutions can be. This I feel gives the impression that the agents are more generally intelligent than they actually are.
That's just lack of observation. Which PhD candidate is going to say "Chat wrote it for me?" We know a lot of academic articles are AI written. And a very, very large part of the student essays. Unless intercepted, they'll end up in the thesis. And in sociology, the texts are so vague, that it becomes even harder to pick out slop.
There are good reasons to assume PhD students see an advantage to using AI, so they will.
> The approach, detailed in a paper presented at the International Conference on Machine Learning, achieved 100% accuracy in identifying models specialized for CSAM generation.
I know some reasons for 100% accuracy in machine learning, first of all the test set leaking into training data. Or you just accept a silly high false positive rate.
When I was an admin I liked to joke that if you guarantee more than 5 nines, then you are an insurance company and you are planning to pay the penalty instead of actually fulfilling your promise, here the principle is probably the same.
[0] https://x.com/SebastienBubeck/status/2097379411691516310
reply