Marcin J. Suskiewicz

A new group photo on the banks of Loiret ☀️

@davidbessis I was in the audience; it was very thought-provoking and inspiring.

If you want to clarify your thoughts on a complex topic, nothing beats teaching and public speaking. This conversation in February forced to me articulate my ideas on math and AI—most of the substance for my "theorem economy" piece crystallized as I was mentally preparing for it.

@davidbessis This is another very good essay. I found that, for me, it (and the previous one about attention) resonate in some way with this essay by Simone Weil, André Weil's sister and a religious thinker (perhaps you won't agree): themathesontrust.org/papers/christi…

Thank you, everyone, for the incredible feedback on "the fall of the theorem economy"! The subject is of course bigger than just AI and math—it's about the future of human cognition. A few remarks that didn't make it to the published version:⤵️

@johnmilbank3 @BijanOmrani Bought the book, looking forward to reading it!

@BijanOmrani speaking at Southwell Minster this Sunday!

@Eirikvi It's great to see these photos, with a glimpse of a wooden Norwegian house through a window in the background. You gotta love @johnmilbank3

Philosophy evening about enchantment

@johnmilbank3 This seems to align with some of the wiser insights of the entrepreneurial culture - perceiving risk as opportunity etc. (however limited these insights might be overall).

We often best solve the problems in our lives by seeing that the problems are in fact advantages. In a way we need less ethical anxiety and more trusting in providence.

@Ananyo @cg_geometry @spectator The tech industry's constant appropriation of credit that is at least largely due to the source human data/artwork etc. - and not, or at least not just (definitely not mostly), to the extraction method needs to be counteracted.

@Ananyo @cg_geometry @spectator That's true, and a very good point. But perhaps it's still worth always stressing that the apparent 'successes' of AI are mostly about the potential of accumulated human knowledge to yield further insights (that can be extracted with computational tools).

I wrote about what the recent remarkable advances in AI for mathematics means for the subject--and for every creative human endeavour. "If mathematics can be hollowed out in this way, then every creative field is vulnerable." @spectator

@Lptomov82 @martinmbauer For what we know, it generally shares the prejudices of its training sets.

@martinmbauer You are missing one great advantage of AI, that we see from chess and go games - it can make moves, or connect ideas we couldn't because it has no prejudices, it can explore the solution space better than us. We limit our thinking by clinging on prejudices.

Yes, this is a significant result and a solid research paper. And it would’ve been much harder to achieve without GPT. While I understand the instinct, I think it is more interesting to evaluate what type of contribution the AI has made as opposed to focussing on how relevant the result is. ChatGPT generalised a previously derived result for an amplitude that was assumed to vanish for all physical kinematics people care about. These amplitudes are very complicated, lengthy expressions with certain structures and symmetries that are sometimes hidden and difficult to see. This kind of problem is exactly where AI shines! AI is better at detecting breast cancer than a clinician because it has seen millions of scans and detects structures where humans -which are limited by their lifetime exposure- can't. AI that systematically surveys many large amplitudes has a similar advantage Similar to the Erdős problems, it was mostly really an attention bottleneck that left this problem unsolved. This calculation was considered another elaborate way of arriving at zero. So few people were interested in the result and even fewer were working on it. Most if not all physicists would therefore consider the insight of the human physicists that there is in fact a kinematic region where these amplitudes are not zero the most meaningful piece of progress here Both these points aren't to diminish the result. It is seriously impressive and deserves the publicity. It shows where the strengths of modern models can significantly accelerate science and I'm convinced there will be even more relevant discoveries in the future.

This thread highlights what I fear will be the degradation of human knowledge due to AI. If I were being charitable, I would say that here Gemini is playing a hybrid role of textbook and tutor, where your tutor has a photographic memory and world class speed reading abilities, but might not be smart enough to get that PhD. So sure, access to that type of tutor is infinitely better than reading baby Rudin in a vacuum. But there is no way AI here is replacing Rudin. Unfortunately though, to a lot of people, it gives the illusion of being better while actually being worse. And thus people will think they are learning more when they are actually learning less.

@nasqret @sama I think they mainly got involved to weaponise the outcome in service of their business (which, by the way, relies on numerous questionable practices, as enthusiasts keep forgetting).

I am very glad you got engaged deep into this experiment. Mathematical community needs strong signal from the AI labs that science is a serious engagement for you. Mathematics in its full proof-driven form is a pinnacle of human ingenuity and knowing how well the models can grasp this challenge is fundamental for the progress.

We went from AI systems that struggled to do grade school math to AI systems that can solve research-level math problems in just a few years. I agree with Jakub this is perhaps the most important eval now. I am also pretty sure the main reaction will be "it's not that hard" :)

@SebastienBubeck How to square this enthusiasm with the ethical and legal practices of this particular company and the whole industry, e.g. in terms of (not) paying for the data used for training?

Well I guess we better deliver now 😅 But srly as I have said before, OAI is just the PERFECT research environment. Ambitious (in particular, non-incremental) research is in the DNA of this place. If that sounds more fun than your current frontier lab, consider joining us!

@davidbessis It was a very nice talk (discussion). Thanks a lot!

C'est demain !

@IvanAhelLab I'd also like to highlight a recent publication by Bernheim and Porier groups et al showing FAM118B, which they call "SIRal", is essential for innate immunity in cellular models; they also have great phylogenetic analysis and insights into biochemistry. science.org/doi/10.1126/sc…

@IvanAhelLab FAM118s have long interested Ivan (my postdoc PI), who, like in many other cases, had a good intuition about their importance. They have been a difficult biochemical target, though. We brought in the observation that they form filaments, an insight explored collaboratively here.

Very happy to share our collaborative project with Ahel and Huet labs et al on FAM118 proteins - noncanonical sirtuins that form filaments and process NAD in human and other vertebrate cells. nature.com/articles/s4159…

@Eirikvi Have a good sojourn!

Things will only get stranger