Well, it's over. Time to panic, the writing is on the wall for math and humanity. Scariest shit I have seen.
>>16965934What bothers me more than this is that mathematics used to be very egalitarian in the sense that if you were smart and worked hard enough you could make something of yourself in it with a much lower barrier to entry than in biology for example, which requires access to expensive labs. It was easier said than done, but it was the fairest system we had in any subject.Now there’s a real risk that you’ll be locked out of a hope of doing mathematics unless you’re born in the wealthiest parts of the wealthiest countries. If only those who are born in California have reliable access to institutions shelling out subscriptions to the latest models then it may be completely infeasible for even people from slightly poorer countries to ever hope to get into this field again. I’m not even talking just about countries like Somalia, but imagine a poorer Eastern European country where historically you were poor but you had a strong mathematical culture. But there’s no hope if LLM access is now a force multiplier. A rich kid from San Francisco beats you every time because he had the right parents.
>>16965938But a talented somalian mathematician can still write an email to Brown University or whatever, asking to pretty please be granted access to a ChatGPT shell, it's not like they have to travel to the US for that.
>>16965934We already have chatgpt 5.5
>>16965934ChatGPT has been solving Erdös problems for months already.The important insight is that almost nobody cares about Erdös problems to the point that at least two "solutions" by ChatGPT were>It's in a paper from the 70s
>>16965947Well, this can be said about any problem in mathematics.The solution is always compiled of many previously proven things (up to some transformations).The thing is that AI's horizon and memory (in terms of RAM, not storage) at this point is already superior to one of a human.
>>16965952>goal post movingGot it.
>>16965957wdym
>>16965938True, but it's not all that expensive. I mean sure, 5.4 pro is very expensive, but 5.5 on high can replicate the performance for a fraction of the cost, and in 6 months you will have chinese models doing it for 1/100th of the cost. Even if you use gpt pro that's not going to cost you much more than, say, DYI tinkering as a hobby, which plenty of nerds in Eastern Europe do. Far cheaper than a 2 months experiment in a biolab.
>>16965934fuckin grok is gonna solve reeeeeman and all internet encryption will get shafted overnight
It's over, mathbros.All our skills will become obsolete and all the achievements devaluated.We were bravely ascending the mountain of knowledge, but goofs who stayed behind just took a ride on a helicopter that appeared out of thin air.Life is pain, God is evil socialist.
>>16965940You’re not going to get to the level where that level of contribution is possible in the first place of access to frontier-level models is required to do serious mathematics. The absolute best case scenario here is that if you’re born poor, you become a top level Olympiad performer as a child where AI access is a minimal influence, but even that is heavily biased in favour of the children of wealthy families. It’s quite depressing to consider but I don’t see how mathematics is not going to gradually head in this direction now that there’s a real possibility it becomes a pay-to-play field.
>>16965983>>16965970>>16965938Science and education had always been for the ones who can afford it.Too many degrees, time for a correction.
What are you afraid of?
>>16965983What you are talking about happened a while ago in ML. You have two facts:1) Huge LLMs are state of the art method that makes everything else obsolete2) In academia, you don't have resources to train your own LLMs at relevant scaleAll the current ML research is very uncertain in its value. It has to be "verified" by the commercial labs in their own "real" training runs (as opposed to toy-size models in the papers), but even if they do that it's all in secret. Maybe they do their own research inside that we will not know about that makes public studies irrelevant.I think similar things will happen in math soon.
>>16966001Why shouldn't people have a PhD?
>>16966009Because he has a PhD and thinks that the more rare it is, the more valuable it will be.
>>16965934You can panic when it can solve these problems without any human prompting it to compel it to. Otherwise, we're essentially describing humans solving problems with a fancier calculator, not the AIs themselves solving anything.
>>16965934The merit for solving the problem goes to Liam Price, Jack Lichtman and Terence Tao; not the calculator they used.I suspect Open AI is astroturfing this to regain some degree of relevance against Anthropic and Google
>>16966101>Terence Taolmao, Ramanujan for chinks
>>16965947This.
>>16965934Get fucked mathcucks. How about you study something useful and get a real job.
>>16965934As a midwit normie with no math experience past high school calculus but some money to blow I kind of feel like paying for these research-level subscriptions of different chatbots so I can solve some obscure unsolved shit and steal away the pride from actual smart people. I like how it’s different from making ai art since normies will bitch and refuse to acknowledge the bot art quality, but when it comes to math you can’t just refuse to acknowledge a definitive proof lmao>>16965979god it would be so fucking funny if we all had to admit Elon Musk solved the Riemann Hypothesis and hand him the millennium prize and forever record that down in history
>>16965934The future is not so bleak. Proving theorems is arguably a byproduct of doing what math is about, which is advancing our understanding of mathematics. If some LLM formalised the Riemann hypothesis in an unintelligible monster of a document in a proof assistant language, we wouldn't care that much in the long term. We would still try to come up with a human proof to further our understanding.Math is not a logical game of symbol pushing where you just need to find the right order of logical statements. Formally it is, but this logical game is a small part of what makes math valuable and interesting.
>>16965938>What bothers me more than this is that mathematics used to be very egalitarian in the sense that if you were smart and worked hard enough you could make something of yourself in itNot really. Connections, one's background and knowing the right people to vouch for you and willing to mentor you are all extremely important factors. It's like saying sports are "egalitarian".
>>16966224Yes, we would care. The topic will be dead instantly.No one with any ambition will work on clarification of AI's results.Sure there will be some people trying to "popularize" it, but as long as math is not completely dead the mathematicians who actually work on advancing the field will move on.When all that is left is desperately trying to catch up with AI, human math will be dead.
>>16965947Don't forget some solutions by AI being more or less pursuing a particular path to a proof that was long neglected by everyone becauae no one wanted to waste time being a trailblazer and would rather already start off with the work people already established. It's like going through a maze and then your friend going.>"if your lost why not back track to the first branching path in the maze and pick path B?
If AI can do novel mathematics, some very useful and suffering-reducing technology is very near in the futureThe past decade or so has been interesting and there is more to comeHopefully the future system is not perverse
>>16966246>No one with any ambition will work on clarification of AI's results.It doesn't have to be clarification. An AI could prove the RH, which could spur on some mathematicians to come up with the theory and understanding that leads to a proof. And knowing that it is true could encourage some because there's no risk of trying to prove something which might be false. Also, there are brilliant mathematicians who have the luxury of not having to be "ambitious". I guarantee if an AI proved RH there would be well-established mathematicians jumping on it.Sure, LLM or other AI proving and formalising theorems would be helpful for those theorems where the theorem itself is a nice tool and there's no deeper meaning required from a proof..But human math would still be necessary and far from dead. To advance math you need to build theories, come up with the right frameworks and definitions and make conceptual leaps or connections that an AI could not make, at least in the near future.I'm not saying it's all going to be positive for mathematics, but I'm not so pessimistic either.
>>16966239No shit, but if the statement is math is the most accessible of the STEM fields that’s absolutely true and the same for sports. Sports you can similarly come from a working class background and make it if you’re that good. Obviously wealthy people have advantages in everything but that’s not the point
>>16965934From looking at the actual chat and following the "reasoning/chain of thought" it looks like this was fairly straightforward just that some of the techniques used are recent/recently named and standardized for those specific substeps. I don't believe, with a training cutoff at the 2010s that this model would solve it. >>16965938>mathematics used to be very egalitarianHoly kek. Yep so egalitarian if you could afford to waste your entire life on pointless shit that you tricked some other fag into paying for.
