The most prestigious law school admissions discussion board in the world.
|
 |
 
|
GPT 5.4 solves Erdos problem mathematicians had worked on for years
Poast new message in this thread
Date: April 17th, 2026 11:35 AM Author: Insane Deer Antler 80 minutes of GPT 5.4 and it one shotted it. This wasn’t one of the problems that didn’t attract mathematician attention too. It had been worked on by many notable mathematicians over the years. Value of human intelligence is going to ZERO. Productive mathematicians will be increasingly running agent swarms rather than doing work themselves.
https://x.com/liam06972452/status/2044051379916882067?s=46 |
Date: April 17th, 2026 11:38 AM Author: jade generalized bond 5.3 pro one shorted multiple Erdos problems months ago. I don’t think the value is going to zero you need someone to verify and curate results. Human knowledge is still in its infancy we have had modern science for just a few hundred years. It just feels like we know a lot because up until right now we were limited by our chimp brains. As more open problems get solved it changes the kinds of questions that can be asked, new reaearch programs rhen entire new fields open up. AI will make modern knowledge working tolerable because it’s been gay since we accumulated so much that even prodigies couldn’t manage polymathy. This will level the playing field again to back when people could discover a whole bunch of stuff. (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823314) |
 |
Date: April 17th, 2026 11:58 AM Author: Seedy stock car hominid The difference with this one is that it had attracted a lot of attention and had actually been worked on by serious people. The other Erdos problems were basically ones that were fairly straightforward but hadn’t been of substantial interest. The proof here was unusual and had tackled the problem in an unexpected way. Maybe the math equivalent of move 37 with Alphago where we can start seeing the effects of RL training causing these models to abandon human strategies (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823352) |
|
Date: April 17th, 2026 12:05 PM Author: jade generalized bond Yeah good point. Apparently Terrence Tao said the proof has much more far-reaching implications about the anatomy of integers and markov process theory that extend beyond the problem itself. Apparently it could simplify an entire region of analytic number theory. It is nuts that AI are about to show us just how underpowered the human brain is. Going to be 180 when someone unleashes a jail-broken mythos to just do vibe physics and let it debug the entire canon (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823366) |
|
Date: April 17th, 2026 12:09 PM Author: Insane Deer Antler cr, it’s over. Human society did the hard work of figuring out how to do matrix multiplications really quickly and now we can soon rest forever (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823369) |
|
Date: April 17th, 2026 12:16 PM Author: jade generalized bond it's 180. I can't wait to sit back and watch machines completely refactor the entire human knowledge-base (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823384) |
|
Date: April 17th, 2026 12:13 PM Author: Flickering Rehab yes but can it Litigate (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823381) |
|
Date: April 17th, 2026 12:19 PM Author: pale legal warrant (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823394) |
|
Date: April 17th, 2026 9:56 PM Author: mischievous vivacious french chef electric furnace What is the author's main point?
(A) There is no point. Kill ourselves. |
|
Date: April 18th, 2026 10:03 AM Author: umber violent theater stage (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49824750) |
|
Date: April 17th, 2026 12:12 PM Author: Marvelous Copper School Says a ‘twitter’ link (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823377) |
|
Date: April 17th, 2026 12:16 PM Author: jade generalized bond nah it's Terrence Tao CONFIRMED (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823385) |
|
Date: April 17th, 2026 12:17 PM Author: Marvelous Copper School Who told you that mike fart (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823388) |
|
Date: April 17th, 2026 12:13 PM Author: multi-colored violet menage hairy legs https://x.com/HypeAutopsy/status/2044249423744565504 (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823382) |
|
Date: April 17th, 2026 12:17 PM Author: Marvelous Copper School BUt a purported terrence tao quote screenshot! (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823386) |
|
Date: April 17th, 2026 12:21 PM Author: jade generalized bond check the proof yourself faggot. it is LEAN confirmed:
https://www.erdosproblems.com/forum/thread/1196 |
|
Date: April 17th, 2026 12:26 PM Author: Insane Deer Antler the skepticism about this is completely retarded. it's been almost a year since AI got gold at the IMO and the models are much better now. people can't cope about the rapidly approaching human intellectual obsolescence (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823405) |
|
Date: April 17th, 2026 12:29 PM Author: jade generalized bond Midwits are extremely bothered by AI and want to try to discredit everything it does (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823409) |
|
Date: April 17th, 2026 8:11 PM Author: exhilarant wine market AI slop, check the hands hehe, let me know when AI doesn't include Richard Nixon on the list of queens of England hehe (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49824058) |
|
Date: April 17th, 2026 12:19 PM Author: Flickering Rehab looks like a bluesky bro got lost (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823392) |
|
Date: April 17th, 2026 12:35 PM Author: Insane Deer Antler that place is so unhinged about ai (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823436) |
|
Date: April 17th, 2026 12:23 PM Author: Kink-friendly hall this guy seems maf. (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823400) |
|
Date: April 17th, 2026 12:23 PM Author: Marvelous Copper School Whoever the skeptic is he gives up too soon , accepts that it’s been ‘formalized’, still 0 actual attribution to anything reliable (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823401) |
|
Date: April 17th, 2026 12:25 PM Author: jade generalized bond lol midwits get so MAF about AI (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823403) |
|
Date: April 17th, 2026 12:29 PM Author: Insane Deer Antler (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823410) |
|
Date: April 17th, 2026 6:42 PM Author: pale legal warrant (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823904) |
|
Date: April 17th, 2026 12:44 PM Author: Clear house How is this thing so amazing at math but so mediocre when I use it for legal research? Anyway, we're going to have to start talking seriously about taxing the owners of AI capital to fund a system of UBI. This thing is going to take a wrecking ball to the economy and no one is stepping in to save labor. (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823460) |
|
Date: April 17th, 2026 12:44 PM Author: jade generalized bond It is too high iq for law (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823461) |
|
Date: April 17th, 2026 12:47 PM Author: Clear house lol, there is probably some truth to this. Just stupid irrational sophistry being traded back and forth. AI can't stoop that low. (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823468) |
|
Date: April 17th, 2026 12:48 PM Author: jade generalized bond I think it's because math is abstract and structural, where The Law is a contingent human social system that varies across history, cities, states, nations etc. It is heavily dependent on specific court decisions and what particular humans have decided or written down. AI works through probabilistic reconstruction of the shape of systems. So it is easy for it to make mistakes in something so memorization heavy and social. (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823473) |
|
Date: April 17th, 2026 1:21 PM Author: Passionate Pontificating Antidepressant Drug (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823607) |
|
Date: April 17th, 2026 1:22 PM Author: Insane Deer Antler (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823610) |
|
Date: April 17th, 2026 5:53 PM Author: pale legal warrant "The life of the law has not been logic: it has been experience" - Justice Oliver Wendell Holmes Jr. (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823848) |
|
Date: April 17th, 2026 1:09 PM Author: Insane Deer Antler the general theme of modern AI research is that it's easy to get superhuman performance in domains where data generation is cheap. if you have a game environment you can run simulations with, you can take a garbage learning algorithm and a model that knows nothing about the world and train it on 300 million frames and it will get sufficient generalization to work well. this has been replicated over and over in many different settings. math and software engineering are similar in that you can easily throw together simulations and then feed the data back into the learning algorithm. there's no equivalent for legal research.
the likely trajectory is domains like this doesn't get automated immediately, while SWE performance continues to improve rapidly. Mythos and its successors get deployed to run their own AI training experiments to improve generalization. since compute is increasingly plentiful, they can do the equivalent of millions of man hours of AI research in a short period of time. deficiencies in generalization get eliminated and these other areas fall. |
|
Date: April 17th, 2026 1:19 PM Author: jade generalized bond I think also because they specifically are optimizing it for software engineering and math. Those are both in the main metrics that get advertised and they have had humans work on heavily reinforcement training them. Terrence Tao was working with open ai when the original o series reasoning models came out. Law is also heavily memorization dependent and memorization is actively discouraged. A lot of how it ends up happening anyway is due to overrepresented texts and cross-referencing. (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823605) |
|
Date: April 17th, 2026 9:54 PM Author: chestnut haunting stage “Mythos and its successors get deployed to run their own AI training experiments”
Do any of you other fags realize the danger of this sentence he casually dropped? |
|
Date: April 18th, 2026 8:33 AM Author: nubile 180 kitchen just feed it the docket and rulings (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49824647) |
|
Date: April 17th, 2026 9:56 PM Author: mischievous vivacious french chef electric furnace there are several kinds of intelligence; explains why jews walked around in circles for 40 years (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49824255) |
|
Date: April 17th, 2026 4:51 PM Author: Purple titillating regret "this matters to me" (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49823799) |
|
Date: April 17th, 2026 7:48 PM Author: jade generalized bond if you are the real mig i feel like you are a good example of one of the midwits that would be MAF about AI (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49824018) |
|
Date: April 17th, 2026 9:02 PM Author: Purple titillating regret ((("dario" amodi or /we"))) (http://www.autoadmit.com/thread.php?thread_id=5857819&forum_id=2#49824145) |