The last generation of mathematicians?
Very candid interview. Despite achieving the highest honor in mathematics, he is very quick to admit when he doesn’t know or understand things outside of his narrow areas of expertise. I guess I didn’t realize how broad and esoteric math has gotten.
He readily admits that he can’t easily understand the 10 proofs recently generated by OpenAI without having to dedicate some time to studying the problems. We are quickly going to run into a situation where there will not be enough human verifiers capable of understanding the proofs generated by AI.
He stopped taking on graduate students before leaving academia because he wasn’t sure it was ethical to potentially mislead young students into a career that may no longer exist in its current form. He also thinks it will be increasingly difficult to teach the problem solving skills honed before this age of immediate answers.
He also discusses his skepticism (which I have long shared) that humans will be able to pivot to X job directing AI faster than AI will also pivot to doing X much better. This is a top 0.01% monkey brain- if he is cooked we’re all cooked.
Overall a very fascinating interview although it has left me quite depressed about the future of humans. 🐵. I worry this will only accelerate the push towards socialism. How do you redefine merit when that which best distinguished us from lower primates is so quickly commoditized….
That's ok! Just approve it with LGTM. There is no reason to stop doing that now all of the sudden. It was good enough for the past 50 years, so it is good enough now. The only difference between the times is that now no one actually has to trust your word that something is good anymore, because we will just trust the bot's word.
Thinking about it, AI is amazing. No one to be angry at. Zero liability. FANFO: simply route all the complaints to the AI too.
"Looks good to me"... had to look that one up.
Yeah, that's a fair point. It's not like human referee reports are known to be always right or correct, either. With proper prompting by the editor, he may even remove several of the biases we have been struggling with for years (with attempts at resolving them through double-blind refereering, etc)--while introducing new ones, for sure[1].
I guess a new framework will emerge where papers will be triaged by AI before sending them out for human review... with unintended dirty side effects (#1551663).
Coz AI is so woke, except for Grok. \s ↩
I'm doing this for code. It's very time consuming though, because you have to do multiple rounds. I was complaining to k00b last night that 2 bots flagged up a design decision in my code after literally 2 weeks of work, with on average 2-4 bot reviews per day, but I woke up this morning realizing that what it really means is that I'm starting to have code that is free of quirks now, so that they have to make up really nasty shit to deliver on the "find issues" part of the instruction.
I wish they did that in the beginning though. lol
if you ask ai for feedback, does it ever stop nitpicking? that's a big issue i find with using ai for review, it doesn't know when to stop nitting. If it doesn't have a reasonable stopping rule, you can't really trust the feedback
It used to but not anymore. There's been something added to training data, I guess to Claude 4.7, that then propagates through traces of Claude executions to all the other models. The solution is usually to scope the review out by section demarcation and then asking it to review each section in subtasks, bring it all together, and do a final scan of the entirety.
This way, you can get a lot of "no findings" on individual sections, which is more reliable. I never instruct literally for a no findings case though, that goes at the cost of completeness because you're giving the bot an out - the "cheating" is trained in too.
I disagree with that. In code, you can address a finding with a clarifying comment too.
Really fascinating.
But perhaps it says more about high level esoteric math than it does about AI vs human. It sounds like the math has become overly focused on proving esoteric problems and less focused on communication, intuition, and understanding.
Look at 3blue1brown. This may be blasphemous, but he's probably a more influential mathematician for most people than Terrence Tao. But not because he's more brilliant, or solved more problems, but because he's able to communicate and teach so beautifully.
Expanding peoples' minds is more important than formally proving something new without any understanding.
Also tagging @south_korea_ln since he might be interested in this topic
Thanks for the tag. I'll try to (speed)watch it at some point today.
First, @gmd, strongly recommend reading this related article by Tao: #1551655. It mentions several of the fears that Tsimerman is alluding to. But Tao is less pessimistic; he's been embracing AI for a long time in his research, and pushes one to think about why one does math, and that by properly (re)defining the goals for said endeavour, math may come out of it better and stronger.
And he also clearly agrees with @SimpleStacker that without properly explaining them, a solved problem is worthless. This is best illustrated with this screenshot from his article.
I believe we'll get to a point where each problem will implicitly be assumed to have been solved (by AI), but unless it has been digested by humans, incorporated into the curriculum, it may very well be considered unsolved. Until it makes it into a technical audience presentation by a human, who is able to explain it to its peers, its status will remain that of an incomplete problem without a solution. Now more than ever, humans will have a role of communicating the science. There will just be an abundance of AI-solved problems to chose from on which you can spend the necessary time to further the steps from Fig. 4 above. Not saying this will attract the same kind of people that got attracted to it in the first place, but I believe that a new market of mathematicians will emerge that will see value in doing so for the right problems.
In my last paper submission, I provided theory support to an experimental group. It is my first paper where I would not have been able to provide the level of support I did if I had not had access to AI. They were deeply satisfied, unlike several previous interactions with other groups where the tacit understanding is that we provide _minimal_ theory support, just enough to get them to publish their experimental findings in Nature or Science journals. The actual theory is expected to come later in separate theory-only papers where one has time to do everything properly. Now, in this latest paper, the quality of theory was at the same level as the quality of experiments. I'm still trying to come to terms with this new role I am taking on, using AI as a partner, but in the end, it's all beneficial to the reader of the paper, who will have been given a fuller understanding of the experimental findings.
💯 he is more influential because he discusses topics that normal intelligent people can comprehend. Most humans cannot grasp topics on the frontier where mathematicians and AI now play
But are the topics by nature more comprehensible, or are the comprehensible because he does the work to make them comprehensible?
Fatalistic thinking probably but having spent some time trying to tutor very basic math and physics in the past I'm afraid I believe many things are simply incomprehensible to people below certain abilities and IQ. And if you don't interact with the "average" human regularly you don't realize how functionally illiterate many of the population is (I'm amazed more people don't off themselves with the insulin I prescribe).
In the interview he talks about many topics outside of his expertise he says it would take him weeks or months to understand, as someone from an elite competitive math background. Versus I could spend my entire life and not get closer to answering IMO questions.
It's hard to tell if they're really just that unable to do math or the interest is just literally zero to invest any effort in it. Like, I'm often surprised at how sophisticated a person can be in discussing football (and able to do math in that context) while being completely blanked out about it in other contexts.
I do think the number of people who actually find pleasure in thinking about something and understanding something is pretty rare.
I dunno. I really loved math and physics until I hit a wall where I could clearly see I wasn't in the top % of my cohort in university. I didn't stop loving the basics but the upper level stuff got too hard for me.
🔗 Privacy-friendly: https://inv.nadeko.net/watch?v=6uIJdXmB4vE