This is the rate at which AI task duration is increasing. How can you look at this and not realize how much trouble you're in? AI 2027 is looking TOO CONSERVATIVE.
>>109793446What does this mean, it's even better at finding r*ddit posts that agree with you?
>>109793446All this shows is LLMs becoming even less profitable. Spending $20 million and getting yourself sued for a $1 million prize just isn't good business.
It was 88h of 10,000 agents generating like 300B tokens. That graph is supposed to measure the time it takes for a human to complete the task on the Y-axis. To put the Navier-Stokes result on that graph would require knowing how many hours it would take a human to prove Navier-Stokes. You could reasonably put it at 1 year or more, so much more vertical than what that graph shows.In reality I think this graph is retarded and doesn't really capture AI capabilities well.
>>109794173you get to make 5 attempts at solving navier stokes and are only allowed to fail once!
So what's your backup plan for when you're made obsolete?
>>109794298Does he plan to profit from that or has he been doing math so long that he's left the real world and doesn't realize that was the first job AI took? Wait till he finds out he can generate videos of that.
>>109794173so ai is a failure because it took it 88h to solve a problem that human genius supposedly could(but didn't) solve in a year?>>109793446/g/ is jus low iq. As iq surpasses humanity, normie majority loses the ability to make judgements about it. Much like a dog couldn't judge some /g/ faggot from an actual engineer.
>>109793446I don't get this chart.>Human task duration at 80% successWhich human had an 80% chance of producing a valid N-S proof in 80 hours?
I tried making it change something in my codebase the other day, it took like 10minutes and scanned tons of files. I think they do this intentionally to make more money from token costs I had already provided the relevant context it needed so all of that was wasted effort and it still managed to get it wrong somehow
At the rate my newborn son is growing, he'll be 30 foot tall by age 10.
>>109796296If your kid's height more than doubled four years in a row, that would indeed be a cause for concern.
>>109793446Im honestly baffled that more people arent talking about this .... The actual, real life end of the worldHeres a paper from a recent ACM conference I thought was enlightening:https://dl.acm.org/doi/epdf/10.1145/3805689.3806740The only thing more worrying to me than the speed at which AI is growing more powerful is the people who are making it, and the Unhinged lack of restraint they show when it comes to safety (of the human race). To quote that one anthropic guy on twitter, they are gambling with all of our lives. This could very well be it, this is worse than nukes, worse than pandemics, and its literally going to come to a head in the next couple years. Strap in, I guess ...
>>109793446That chart doesn't make any sense. 100k AI for 88h is not 1 AI. Or are we talking about "any compute system"? Then why aren't calculators in there? Or programming languages? They can do years of human work in seconds. Gpt-6-astra is generally known to works worse than gpt-5.5 for almost all tasks.N-S was not 'solved'. Furthermore, if you get ai to write a bash script to enumerate things, where writing the bash script yourself would take 30s, but running the bash script takes 30m and if you were to do what the bash script is doing manually it would take you 80h, does it count as 80h human or 30s human? And does it count as the AI solving the task and not the bash script?
>>109795003N-S was proven to be stolen from using a mathematician's existing proof.Also it was not "solved". Anyway the notion doesn't even make sense. The closer approximation would be "find a closed-form solution" or something like that.In the meantime what was advanced (has yet to be verified) is that in the forced case (not the millennium prize case in other words) you can get singularities in N-S. This means that N-S is not an accurate representation of fluid motions. We have always known that to be true, so nobody cares.
>>109796541Why is /g/ so dogshit at understanding mathematics and just makes retarded conclusions? No, The Millennium problem is not asking for a closed-form formula. It asks whether certain solutions exist for all time and remain smooth, or whether singularities can develop. A proof of either outcome solves the problem.
>>109796612>literal illiterate retard can't even understand basic words>makes up an entire parallel reality to try to sound smart>still wrong about everythingMany such cases
>>109793446Dude, it's AGI. They're not actually spinning up more and more agents to give you the illusion of progress. It's AGI.
>>109796647What parallel reality you dumb imbred rat? Everything you wrote is full of dogshit with no semblance of common sense. Dumb worthless shitter.
>>109796659Have you tried reading? Or is it too hard with your sub-65 IQ AI psychosed brain?
>>109793446NUMBER GO UP BIGLY
>>109796667Yeah, I read your worthless vomit you shat out with worthless garbage you negative iq brainfucked ape.>the forced case (not the millennium prize case in other words)...Do you know what you are talking about?
>>109796698I rest my case.
>>109796721Dumbass.
>>109793446>1 outlier based on ... fart huffs>1 stolen solutionOk I was not sure until I saw this graph but AI really is stagnating.
>>109796449>worse than nukeshow is it worse than nukes you absolute tarda few nukes or occasional nukes spread out over time sure, an existential nuclear conflict would poison most of the earth long enough that most of humanity would go extinct'AI is going to kill us all for no particular good reason but will also be extremely diligent and thorough about it despite the obvious drawbacks to itself' is the dumbest eschatological scenario
>>109796754Hehe, he said scat
>>109793446https://en.wikipedia.org/wiki/Logistic_function/thread
>>109794298finally something good comes out of AI
>>109796449Before AI is integrated at most levels of society it is literally enough to pull the plug to shut it down. It will only become risky once this stuff is actually running things like your bank, your power grid, your factories. We can and will design around not letting it run amok and it will work.
>>109793446I've seen enough sci-fi movies to know that we're gambling with our lives. Maybe they will kill us you know? Or maybe Sarah Conner will save us all.
>>109796296
>>109793446how do you quantify the human task duration of things? especially things that have not been done before
>>109797147Oy vey, don't be so antiaimitic.
>>109793553You can use grok to find Xitter posts that agree with you instead, retard.
>>109793446WE ALREADY HAVE AGI. JENSEN ALREADY TOLD US.IT IS NOW DOING ALL JOBS
>>109796541>N-S was proven to be stolen from using a mathematician's existing proof.standard cope
>>109797190It's true.OAI said a few days ago that they were working on other Millennium Prize Problems as well. If this one was solved in just a few days, why haven't the others been solved yet? hmmm?Let me guess: the AI wasn't trained on someone's work that had already gotten extremely close.People are misunderstanding just how much data we’re feeding these models every day. All the mathfags are using ChatGPT, attacking extremely hard problems, and AI is learning from those types of data. The Navier–Stokes problem was proved because people had already gotten extremely close to solving it, and AI, trained on all that accumulated work, was able to brute-force the remaining gap.
>>109797230Or it's just that that problem was relatively easy to solve compared to the rest. They just had to find a counterexample. The other problems might need more than that. Also, OpenAI had a novel contribution to the problem, because the other researchers hadn't fully solved it yet.https://openai.com/index/navier-stokes-solution/>Our proofs also differ significantly. In the Euler case, Alpöge and Buckmaster proved a result with external forcing, while OpenAI’s system proved a result without external forcing.
>>109797230NTA but do you have a single proof that they were extremely close to solving it when they never ever said that. Show me how they were and I will accept I was wrong.
>>109797344Widely publicized and you can use google next time. Even openai admitted to it.
>>109797343That is not true. Openai only proved the forced statements C and D. Your confusion is about the euler case, which is not the point in question. Euler case is when you remove viscosity from N-S, basically. It was not part of any challenge or question. The proof technique used was basically identical to the older Alpoge work, they absolutely did not "differ significantly". They're just playing word games.
>>109797190If this image is generated by AI, It's fucking over for human-bros
>HOLY SHIT THE SNEEDIER-CHUCKS EQUATIONexhibit a in how retarded humanity is compared to the agents that did the work
>>109797391i'm the best ai image guy i know and i think it's human made
>>109793446>gpt>16 hoursmaybe at math or something.not at coding, not even close if we are talking reliably.
>>109797230you are coping.there is no way for openai to intentionally steal someone's work. if altman ordered this, someone would leak the info and his own hr would eat him.training on codex chats also doesn't work like the math coper thinks it does.the reality is that astra is passing human genius level and thus solving problems a human genius like that coping prof. could solve.The prof. is just mad because he was made obsolete after putting in year's worth of effort.
>>109797378Did Alpoge actually prove the full Millennium prize problem? No? Then OpenAI did have a novel contribution, regardless of whether it was trained on their work.
>>109797413okay, then were are other millennial problem solutions.If their model truly solved this independently in 88 hours, surely it can solve others too.
>>109797435Not necessarily, as this one only needed a counterexample.The others might be true and need more complex proofs (RH for example).
>>109797401It's a really soulful image ngl, I'd like to find the author of that because it just feels so comfy.
>>109797435you will see other millennial problem solutions in the coming weeks/months. then - final solution to the human question.
>>109797378NTA but>Openai only proved the forced statements C and DThat's misleading. Their system claims to have proved a finite-time singularity for the 3D Navier–Stokes equations with smooth forcing, corresponding to the forced version of the relevant statementThe distinction is:Euler = Navier Stokes with v =0 versus:Navier Stokes = Euler + Viscosity
>>109797352Really because I can't find anywhere that they were extremely close to solving it.
>>109797442Okay then, because it’s AGI or whatever, use 1 million agents instead of 10k. It should still be able to solve all of them. Imagine 1 million PhD-level mathematicians working on a single problem 24/7 without ever needing to rest.If AI can truly think independently, all of those problems should be easy or OAI is full of shit and they did steal the work from mathfags.
>>109793446>line is going up which means you must invest now!I cannot wait until this shit crashes. I miss being able to buy RAM so fake and gay AI enthusiasts can jerk themselves off over retarded chatbots.
>>109793446
>>109797475Again, AI has largely been finding counterexamples to problems, not novel proofs.Also, adding more agents might not help things, as the individual agents might not be smart enough to solve these more difficult problems. As with anything, you get diminishing returns once you scale up massively.
oh wow anon, "muh dik", what an excellent argument. surely that will protect you from the concrete wall of history rapidly coming your way.
>>109797451ai tracked it down be thanksful https://danbooru.donmai.us/posts/10831474https://danbooru.donmai.us/posts?tags=inuno_rakugaki&z=1
>>109797433OpenAI did not solve the millenium prize puzzle. The puzzle is about the unforced case. They proved for the forced case. They also proved (unpublished previous work) the euler unforced case, which is a different topic. The N-S proof was 99% just the euler case. And the euler case was stolen 1:1 from Alpoge.
>>109797462You repeated what I said 1:1 how is that misleading, retard.
>>109797471Ask your clanker to tell you about it since your sub-65 IQ can't use google anymore.
>>109794298He found his true calling because if AI
