>mathematicians offered endless papers they can read, validate or invalidate and publish>they are upset by thisJust admit it. They don't want to do their jobs. Even if it takes a fucking year to go through one proof and check it, you get to publish your work (whether the proof is valid or not). Tenured mathematicians are supposed to publish. They're being given a way to do this. What do they do? Whine like little bitches. Because they just want to sit on their fat, lazy asses and pretend they're God.
>>17068529Mathies are eating good. All this point towards higher demand for humans who can parse through this jargon and convert information to knowledge(human construct). We are literally witnessing Jevons Paradox in math.
>>17068536Yes, Gowers had a pretty nice letter on Tao's blog pointing out that, basically, having enough literature backlog to occupy you for several lifetimes is already the state of affairs for mathematicians (and if you somehow run out of material you can just learn another subfield), so a few hundreds of other papers/proofs doesn't change that much unless you were closely following that particular problem https://terrytao.wordpress.com/2026/09/17/why-i-didnt-sign-the-fields-medallists-letter/And several mathematicians and scientists noted that AIs are in fact the only ones they expect to truly understand their years-old papers because no other humans bothered looking deeply into them yet.
Mathfags are upset because the AI demonstrates that, contrary to what they believe, Maths is a solved subject. Yes there are few important questions that are still open problems, but most of results are already discovered and what remains is just too abstract (because built up on some existing theory) or require too much knowledge to be useful. Same goes for physics, same goes for biology.
>>17068638Hey Boss, we got a problem.>Could humanity theoretically build a machine to solve it?I-I guess? Why?>Then we have already solved the problem!Kek.
>>17068536Upset? That's not what the post says. I work around mathematicians and reactions are mixed, mostly positive, cause AI can help them progress. He stated an obvious fact. He's right that dumping proofs may be overwhelming to already niche communities.
>>17068638It's the weird half of mathfags who really enjoy "being a mathematician" more than just actual mathematics kek>>17068653Difference in trajectory and plausibility between saying that in 2015 (ridiculous) versus now (much, much less ridiculous).
>>17068529Mathtrans shitting their pants rn
Good thing I gave up on being a researcher when I was young. Never thought I could cut it, it's just too fucking boring. I'd rather just have sex with wife and watch movies than count how many balls are hairy.AI has been great though, make my job easier in multiple ways. New retarded laws that say I need to have text of all my notes? AI do that shit for me.Life rewards though who are lazy and content.I realized early pursuit of god is more honorable effort as the world will end when I close my eyes and even if it does not it will not be a world of my concern.Wanna know how to have a good marriage? Be attractive. It's that simple. There is literally nothing else. She cheats on you? You're not attractive then.There's my wisdom, maybe stop counting beans and start counting pull ups and laps.Oh if your wife gets fat without ever even giving brith once? Divorce. No joke. You can let a few lbs go if she gave birth she earned it, otherwise ditch the cow. Attraction goes both ways.
I really kind of wonder how they communicate with their closed, top-of-the-line models, especially now that some of them have long-lived states. Like when it came back with a "L-O-L hey I think I might have solved navier-stokes" how does their office look. You think they drop some thank yous and just sit around and chill w it
Math is over party ITT
Why are humans so slow?
Math is the domain pf SI now. Ignore the proofs if it causes a melty mathbros youre all just useless eaters now
>>17068529The biggest issue for me as a mathematician, is that if they indeed "solved math", then they are probably not far away from the recursive self-improvement
>>17068895RSI has already happened if you accept those benchmarks showing SI having better research "taste" than human experts
>>17068899it doesn't happen until frontier AI is actually improving itself in practice.
>>17068902We (meaning anyone not in the major 2 labs) have no idea what the frontier models are capable of since they arent being released. Anything coming out is a student model that is heavily quanted
>>17068905If true exponential RSI has been achieved it would be like having developed a nuke. maybe that has happened and they're hiding it so nobody freaks out but I doubt it. from what we know alignment techniques are not good enough yet and that's a significant handicap on progress
>>17068529>results hav eto be understood connected explainedI would be fully onboard if a lot of academics and textbook authors did not erase all their intermediate steps because "they are obvious bro" which makes reading proofs fucking exhausting because you have to verify a bunch of irrelevant annoying small stuff (which the author definitely wrote out in detail when working out the proof)paper and ink is almost free you can spare 1 or 2 extra steps and some explanation of what you were thinking
>>17068895>>17068899RSI itself is an unsolved search problem itself meaning AI probably can find a good local min/max but it would probably be thermodynamically redundant to try and converge perfectly.What we need is convergence to a good output (approximation theorem) and in that regard reasonably computed RSI is completely possible right now.
>>17068916>results hav eto be understood connected explainedthe problem with that is not it doesn't have to be understood and connected. It actually doesn't matter. The consequence is that modern mathematics becomes a black box.
>>17068912>exponentialCome on bro you cant throw that in 5 posts in. We arent going to see some retarded yudkowsky 1 day self made ai god with nanobots dogshit. RSI can change our society very quickly even if we're just seeing each model being better by 10% it wouldn't have if it had been completely human designed.As for alighment im of the belief that future models will handle it so no reason not to keep accelerating.
>>17068921>future models will handle it>sauce: trust me bro
>>17068921>RSI can change our society very quickly even if we're just seeing each model being better by 10%That is exponential. If you can keep it at 10% then it compounds and very quickly gets out of control.
>>17068529>mathematics does not grow simply by accumulating correct statements actually it does because some of those correct statements will be statements meaningful to scientific problems. these guys are coping hard, instead of crying on twatter maybe they should just pivot their careers to something else, this has happened to other professions before. it can still be math related, but the sitting down and working out proofs job is already over. >>17068536they don't want to check it because it takes away time they could spend publishing their own research and pumping up their name. academia's problems are really coming to light by the usage of AI.
>>17068926Sorry man but im from a business background so if I read exponential im thinking 10x. Sure we 100% have exponential RSI right now by the "any % is eventually exponential" reasoning. Just look at how releases are getting quicker and benchmark scores are increasing on specific business critical ones (like agent or lawyer ones) that havent been saturated yet
>>17068933we have an exponential growth in capabilities but it's not RSI because it's not autonomous.
>>17068933>im from a business backgroundthat explains lack of coherence in you written expression
>>17068947My job is to make the guy or gal on the other side smile and want to see me again>>17068940>autonomousSeems more like theyre just doing a bunch of hero runs like they did for navier stokes then piecing them together for the next run. Not sure they will ever just let the models rip and let them do everything. Might as well just call what we have semi-RSI since that'll be enough to change the world
>mfw low iq mathlets like me are now the same as high iq math researchers
https://github.com/openai/math/tree/main/preprintsUh thats a lot
>>17068529Each problem got 3 hours of ai apparently
>>17068957>I don't get it, just do the work for me>>yes, batmanfeels so good to be a math researcher now.
>>17068529>Mathematics does not grow simply by accumulating correct statements. Results have to be understood, connected, explained, challenged, reused.Good thing it doesn't matter that he's too lazy to do this because we can just get the AI to do it.
>>17068529what happens if the proofs just turns out to be hallucination slops that only mathematicians could read?
>>17069004the mathematicians get AI psychosis trying to understand them
>>17068536fpbp
>>17069004humans write bad proofs too, the difference is the AI can use lean and retarded boomerloid mathies can't.
Why do so many non mathematicians "follow" math (without actually learning math)? You can fit all pure mathematicians in a single McDonald's. It's niche stuff.
>>17069020how does lean actually verify something is true conceptually?
>>17068888checkedmaths is done, imagine being some phd student with all that debt while some retarded tradie you went to high school with has outrun you, you're so fucked>>17068529>retard op twitter screenshot poster cant into githubhttps://github.com/openai/math/tree/main/reasoning_traces
>>17069052It uses math
>>17068529>SLOW DOWN PROGRESS SO I CAN FEED MY EGONo group of people have lost as much respect so quickly than mathcucks
>>17068638>Same goes for physicsWe don't even have a unified theory of physics and the Schrodinger's equation becomes impossible to solve for large systems limiting condensed matter research. You work on those 2 things and scientific progress would be insane in a couple of years
>>17069054>retard op twitter screenshot poster cant into githubOP posted LEAKED info yesterday retard
