A few days ago, something really interesting happened: For the first time, an AI generated a mathematical proof that was not yet known: https://www.erdosproblems.com/728In most cases where this has happened, it was later discovered that a solution already existed in the literature, but this problem had been formulated incorrectly and was only corrected a few months ago, which means there was no prior literature on it.Sometimes of you say that AI will be limited by "not being able to create new things," but I think this case shows that it's not quite like that at all, and that most people are still underestimating AIs.
AI is a paraphrasing machine, like 99% of humans.
>>107832560Well a broken clock is sometimes right.So, this isn't much evidence.
>>107832588t.
>>107832560would not have generated the proof if it wasn't prompted by a talented mathematician. people underestimate how much it matters that the person who is prompting the AI is competent.
>>107832560>more or less autonomously>after some feedback>"in the spirit of" the problem>to the best of our knowledge>t. i take money from commie chinks obsessed with off-brand technocracy
>>107832560>For the first time, an AI generated a mathematical proof that was not yet known: https://www.erdosproblems.com/728IIRC few novel solutions to math problems were already achieved
>>107832579>AI is a paraphrasing machine, like 99% of humans.Duh, where do you think the idea for neural networks comes form?
>>107832601The prompt is nothing but the math problem
>>107832601>people underestimate how much it matters that the person using the Ouija board is competent
>>107832601cont.Often for puzzles like this there is a key "hint" or phrasing that allows you to more efficiently sample the latent space in a way that makes getting the right solution from the AI more likely.
>>107832609>where do you think the idea for neural networks comes form?A completely delusional jew academic who published scam papers about inventing an "artificial brain" with impunity until another jew god tired of it and published criticism that his invention can't do so much as learn a simple xor.
>>107832611I don't think that's true. >after some feedback from an initial attempt
>>107832560>a problem that was solved 50 years agoWonderful.
>>107832620>allows you to more efficiently sample the latent space Complete pseudbabble.
>>107832629Sauce?
>>107832626when it fucks up you call it a retard and say look it up and you get a better answer
>>107832629It wasn't solved, nigger. The problem was literally formulated incorrectly and got fixed last year. There was no solution for that. Stop being a coping nigger and just accept that even mathematicians are leveraging AI to solve complex problems.
the promise of completely autonomous ai agents changing everything last year was coming from a place of misplaced confidencethose completely autonomous agents will mostly not exist this year eitherhowever, this year the the ai assistant agents will arrive on people's desktops and phones and will just do things on requestthey will fundamentally change how work is done on computers and be a rude awakening for a lot of people
>>107832631it made perfect sense to me and matches my experiences with using llms
>>107832560I like how these niggers presume to know the 100 TB of data it was trained on does not contain that output, or the scraps needed to generate it.
>>107832669>my experiences with using llmsYou're a retarded golem who clearly uses LLMs a lot, which selects for a target audience prone to mistaking meaningless babble for intelligible output.
>>107832683oh you are one of those people
>>107832560>Meme problem that was allegedly misstated>guy argued with AI for a few months>other program does basic search for "valid" lean4 proof allegedly parsing the user's English proof into lemmas to search with. I mean, it's cool I guess?
>>107832686>oh you are one of those peopleNo. You're one of those unpeople. In your next post, explain what """allows you to more efficiently sample the latent space""" means, in concrete technical terms. What "latent space"? How do you sample it "more efficiently"? Notice your subhuman urge to ask your shartbot to answer these questions for you. Notice your blood pressure rising as you realize I will immediately recognize you did this.
>>107832631>>107832683you can just look up the terms instead of getting defensive. aren't you people fans of the 'stochastic parrot' framing anyway? this aligns with that
>>107832702cringe
>>107832702you prime its context so it gives more probability to sequences of symbols in the sample space that contains the desired outputs and avoid getting stuck in hallucinatory reddit loops
>>107832706See >>107832702>In your next post, explain what """allows you to more efficiently sample the latent space""" means, in concrete technical terms. What "latent space"? How do you sample it "more efficiently"? Notice your subhuman urge to ask your shartbot to answer these questions for you. Notice your blood pressure rising as you realize I will immediately recognize you did this.
>>107832732See >>107832702Specifically:>What "latent space"? How do you sample it "more efficiently"?
>>107832741more efficient in that you dont have to wrangle it out of misconceptions as hard
>>107832757What "latent space"? How do you sample it "more efficiently"?Notice how you will repeatedly fail to answer these questions.
>>107832560this post was written by ai btw, kys clanker
>>107832766already been answered you dummy
>>107832560Mfer misstates an easier version of a solved problem and the slopbot regurgitates its datasetSlopbros...
>>107832560>>107832579>>107832601>>107832669>Looked into the details>It was written beforeStop lying anon
>>107832775Thanks for demonstrating my point so perfectly. You're a mindless biological token stringer complete with hallucinations.
>>107832676>I like how these niggers presume to know the 100 TB of data it was trained on does not contain that output, or the scraps needed to generate it./threadthis scam works so well because it involves scales beyond any individual human
It's a bit funny that most of the comments posting an LLM solution include the line "it's not very readable".
>>107832676>>107832854ayo deez retarded meatbags malfunctioning, dey be needin more training data fr
>>107832954Look at yourself slopbro
>>107832560>similar results proven by similar methods were locatedThey literally did the "can I copy your homework" meme lmao
>>107832560>Sometimes of you say that AI will be limited by "not being able to create new things," but I think this case shows that it's not quite like that at allThis doesn't establish anything about AI being able to create new things except in the same trivial sense that an interpolation creates new data points in between existing ones.
>>107832854>>107832676>>107833062>>107833133>>107833186>data exists>data involves scales beyond any individual human>data is extrapolated from>NOOOOOOOOOOOOOO THAT ISN'T FAIR THAT ISN'T LEARNING THAT'S CHEATING
>>107832560More like Merdos lol
>>107833236Whom are you quoting? What aspect of what post are you responding to? You sound legit mentally ill.
>>107833251ask grok to summarize
>>107833256>>107833256I accept your concession. Your irrational spergout doesn't address anything in any of those posts. Ask grok to diagnose your psychotic illness.
>>107833265nigger meme replythe irrational spergout works so well because it involves scales beyond any individual human.
>>107832854Pic, BBC, NOW
>>107833302ayo Wiz wya
>>107833293I accept your concession. Your irrational spergout doesn't address anything in any of those posts. Ask grok to diagnose your psychotic illness.
>>107833313ayo that nigger clanker posted it reddit spacingAGAIN
>>107833320I accept your concession. Your irrational spergout doesn't address anything in any of those posts. Ask grok to diagnose your psychotic illness.
>>107833344>the data this nigger meatbag was trained on only contains that one output, and none of the scraps needed to generate anything else.
>>107832560Is this the same kind of "proof" them two black school girls found in Chicago that turned out to be utter nonsense?
>>107833302
>>107833369I accept your concession. Your irrational spergout doesn't address anything in any of those posts. Ask grok to diagnose your psychotic illness.>>107832676>>107832854>pointing out why novelty claims about "AI" are inherently dubious>>107833186>putting Tao's own admission, that there is nothing novel here even as far as he knows, in ML termsThese are 100% correct.
>>107833421Name one (1) single instance of true novelty>he posted it againoh wait
>>107833433>Name one (1) single instance of true noveltyOk. Since we're on the subject of mathematics, and given that your likes are all mathematically illiterate, we can start with something simple: whoever came up with the positional number system didn't have an internet's worth of "similar results" and "similar methods" to go by.
>>107832601True and real. The average retard has no chance to give it the right prompts, and wouldn't even understand the answer.
>>107833475>True and real. The average retard has no chance to use the Ouija board correctly and wouldn't even understand the answer.
>>107832601This has to be the reason I find Copilot amazing yet most people complain about it
>>107833485
the levels of coping when google announces the navier stokes proof in a few weeks is going to be incredible
>>107833457But they still had, of course, necessarily, the scraps needed to generate it.>b-but muh volume of dataI accept your concession.
>two more weeks and the token guesser will win a heckin' nobel prize in physics and all you deniers will seethe so badly then ha-ha
>>107833530>nonsensical token string>>107833433>Name one (1) single instance of true noveltyOk. Since we're on the subject of mathematics, and given that your likes are all mathematically illiterate, we can start with something simple: whoever came up with the positional number system didn't have an internet's worth of "similar results" and "similar methods" to go by.
>>107832620You're literally just saying that the machines are statistically random and with enough trial and error you can cause it to randomly produce a correct answerYou just wanted to sound smart when you said it, because wording it plainly makes it sound not very special
>>107833520If you believe is that good then why you don't prompt that right now? What are you waiting for?
>>107833552He's not saying anything at all beyond "if you steer a LLM towards your desired output in the prompt you're more likely to get that output". It's made up technobabble.
>>107833547>whoever came up with the positional number system did so with no familiarity with numbers or writing systems>he didn't even have THE INTERNET>he just, like, invented it, mannnn
>>107833601Notice how your psychotic illness causes you to hallucinate greentexts unrelated to the post you attempted to refute.
>>107833556lmao, it's not one of the public llms. google's had a team working the problem with external researchers for a few years and about a year ago they let slip that they were ~1-1.5 years off from a solution. they put out something related in sept.so you will see an ai assisted proof of a millennium prize problem very soon.
>>107833622I accept your concession.
/g/ is full of Luddites for what it seems
Friendly reminder: adding "luddites" to the filter is a good way to avoid corporate spam.
^luddite tranny above me
I'm still pushing perfectly functional llm code to productionI'm still working 2 jobs at onceI'm still getting paid (handsomely) for itTrannies lost. Cope and seethe.
>thread bumped>no new posts
all the money and resources wasted on a mechanical parrot
>>107833858a mechanical parrot that is now smarter than every human
>>107833630>not one of the public llmsThen is a fraud, there will be no evidence the llm did it.
>>107833869
>>107832603yep lolTao's statement has so many qualifications, it could be licensed to practice medicine, law, and engineering in all 50 states
>>107834118>picaverage human compared to ai
Say what you want, but it's amazing how much AI has improved. It was just last year that it was telling people to put glue in the pizza because it could only copy Reddit posts, but now it can copy mathematical proofs too. What will AI be able to copy next month? Only time will tell!
>>107834497maybe we will have ai dungeon back
>>107832560Reminder, his son trooned out completely
>>107832560>it was later discovered that a solution already existed in the literatureThis will be the same
>>107832954Who wrote all that "training data"? Books don't exist in nature. Math doesn't exist in nature. Two apples have no idea that there are two of them, only a human does.
>>107832609me on the left
>>107837241>Math doesn't exist in nature.i seriously hope you guys don't do this>only a human can count to two, information only exists if there's a fully developed semiotic system to describe itthis is getting pathetic
>>107837583Humans and animals have the ability generate knowledge on their own from interacting with the world, and from the physical structure of their being. Like math, which exists in animals that don't even have the ability to acquire knowledge in any other way. In other words, humans and animals have intelligence.AI does not generate knowledge on its own and cannot do so, only copying things that other people have written down. As it is unable to naturally generate knowledge, it does not even know these things that it copies, as even copying requires genesis of knowledge. In other words, it does not have intelligence.
>>107832560https://www.erdosproblems.com/forum/thread/728#post-2785Looks like it was still building on prior literature.If the problem really was corrected only a few months ago, Pomerance may have gotten the solution back in 2017.
Here's your 2200 line python proof.Good luck verifying it, meatbag.
