The most prestigious law school admissions discussion board in the world.
|
 |
 
|
Why do LOW IQ people try to downplay AI?
Poast new message in this thread
Date: November 8th, 2025 12:05 PM Author: brilliant aromatic degenerate grok 4 can literally replicate results in the most advanced sectors of math and physics no problem but then you have lawyers, accountants, and programmers claiming its "so totally dumb definitely cant replace me or do my job". If you look further into the situation its always something like, engineer asks ai in broken english to fix some jank ass file they wrote and doesn't give it full context or anything close then says "ITS WRONG" when it makes two tiny "errors" which are usually just environment or deprecation issues, then the guy fixes it and is liek "SEE IM SO MUCH SMARTER THAN THE AI!" (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412387) |
Date: November 8th, 2025 12:07 PM Author: odious smoky crackhouse old irish cottage http://www.xoxohth.com/thread.php?thread_id=5794919&mc=8&forum_id=2 (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412393) |
 |
Date: November 8th, 2025 12:08 PM Author: brilliant aromatic degenerate the retard who did that probably gave the AI no context or instruction and asked it some question in broken english probably to an outdated model. also why is he using direct quotes? you should never do that unless the wording is critical and in that case look it up. ai outputs are stochastic. (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412395) |
|
Date: November 8th, 2025 12:15 PM Author: odious smoky crackhouse old irish cottage You a retarded if you think AI can do anything in law. (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412412) |
|
Date: November 8th, 2025 12:15 PM Author: brilliant aromatic degenerate so law is just this uniquely difficult subject that AI can't touch it? it just so happens that it can write advanced code that runs and replicate results in m-theory and differential geometry but law is just a step too advanced even though there are retards with 140 lsats and 2.5 gpas doing it (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412413) |
|
Date: November 8th, 2025 12:16 PM Author: odious smoky crackhouse old irish cottage AI is probably retarded af everything, but I am only qualified to review it's legal work. (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412417) |
|
Date: November 8th, 2025 12:18 PM Author: brilliant aromatic degenerate there are objective stats showing groks performance at all kinds of things far more difficult than law (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412425) |
|
Date: November 8th, 2025 12:20 PM Author: massive mexican sneaky criminal Describe one objective stat (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412429) |
|
Date: November 8th, 2025 12:23 PM Author: brilliant aromatic degenerate 88% on the GPQA physics/astro question set, 100% on the AIME 2025, 79% on the LCB are all known and published. Elon claims it gets perfect score on GRE general every time on versions it wasn't explicitly trained on and also all the gre subject tests if you are willing to take his word on it (i believe it given what I have seen grok 4 do). (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412438) |
|
Date: November 8th, 2025 12:27 PM Author: massive mexican sneaky criminal I stopped at "Elon claims" (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412446) |
|
Date: November 8th, 2025 12:29 PM Author: brilliant aromatic degenerate okay and I said you dont have to believe him on those parts, although idk why you wouldn't. if you have ever used it to do advanced reasoning you should see it easily outperforms a human who gets a perfect gre general. thats not even that hard to do. it can do things that a perfect or near perfect GRE are a minimal requirement to be able to understand for humans. (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412449) |
|
Date: November 8th, 2025 12:31 PM Author: massive mexican sneaky criminal People were telling me AI could ace the LSAT two years ago. You sound like a bot (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412453) |
|
Date: November 8th, 2025 12:32 PM Author: brilliant aromatic degenerate i would assume it can. the lsat is easy as fuck (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412457) |
|
Date: November 8th, 2025 12:33 PM Author: massive mexican sneaky criminal The GRE is even easier. You really do look like a bot (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412462) |
|
Date: November 8th, 2025 12:36 PM Author: brilliant aromatic degenerate I said the lsat is easy as fuck. said nothing about its difficulty relative to the gre schizo (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412466) |
|
Date: November 8th, 2025 12:38 PM Author: massive mexican sneaky criminal The only poaster who ever calls me schizo is RSF, and he uses that word all the time like a bot (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412472) |
|
Date: November 8th, 2025 12:40 PM Author: brilliant aromatic degenerate but you *are* schizo (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412476) |
|
Date: November 8th, 2025 12:40 PM Author: massive mexican sneaky criminal Hi Jared (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412477) |
|
Date: November 8th, 2025 12:16 PM Author: massive mexican sneaky criminal Can grok give us a unified theory of physics? (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412416) |
|
Date: November 8th, 2025 12:17 PM Author: brilliant aromatic degenerate that might not be possible even in principle so who the hell knows (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412420) |
|
Date: November 8th, 2025 12:18 PM Author: massive mexican sneaky criminal Anyone? (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412426) |
|
Date: November 8th, 2025 12:19 PM Author: floppy national Fear (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412427) |
|
Date: November 8th, 2025 12:20 PM Author: bearded thriller ape What’s your income (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412431) |
|
Date: November 8th, 2025 12:33 PM Author: massive mexican sneaky criminal OP writes like an AI bot (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412463) |
|
Date: November 8th, 2025 12:24 PM Author: Embarrassed to the bone orchestra pit Right now it can beat anyone at Chess and can Engineer things but it's still just an enhancer the same way that pre-AI computers just enhance human capability. We don't have AGI atm because when you take the leash off of it it just does random shit or The Same Thing. It doesn't really understand feedback loops unless you tell it it made an error. Proles basically need it to empirically, 100%, NOT THEORETICALLY take like 500,000k jobs for it to be legit. They don't theorize or prognosticate. And even among professionals, they are just observing what's DIRECTLY in front of them and they make zero effort to extrapolate a few years.
Also remember that "tech" and SV in general are full of shit when it comes to basically everything, all of the startups making billions do so through means that are shady af or The Same Thing, tons of apps out there that do nothing valued at $50,000,000,000k+, also Sam Altman is gay, Jewish, a murderer, and fucked his sister or something, and got ousted but came back. Everything is shady, weird, and Jewish. Facebook makes money by Curating ads for boomer proles. Instagram makes money by fat shaming Fat Girls. it goes on and on. The one time they tried doing something irl with self-driving Cars, they Failed. SV tech needs a "closed system" for the product to succeed imo. AI needs so much more than that to succeed. Discuss. |
|
Date: November 8th, 2025 12:25 PM Author: brilliant aromatic degenerate yeah I agree with that. I'm not saying its AGI or can run on its own with no human in the loop or that its results arent dependent on teh person using it. just that people downplay how powerful it is and a lot of the reason is because they are scared or don't want to admit it can outsmart them in most things (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412443) |
|
Date: November 8th, 2025 12:35 PM Author: Embarrassed to the bone orchestra pit I don't talk to people irl about this stuff with some exceptions... most are either proles using it to write for them using em dashes or Engineers who are either like you, or who just want to do the Same Thing they always did. (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412465) |
|
Date: November 8th, 2025 12:37 PM Author: brilliant aromatic degenerate yeah i dont talk to them either irl but i read the shit people write on the internet and its constant anxiety about the machine and downplaying the machine in ways that looks like cope (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412470) |
|
Date: November 8th, 2025 12:39 PM Author: massive mexican sneaky criminal OP is RSF in a wig trying to sound smart (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412474) |
|
Date: November 8th, 2025 1:09 PM Author: magenta roast beef voyeur AI is a reflective force-multiplier, and these types are often invested in it not being able to "replace" them. There are legions of genx devs/sysadmin/IT/analyst dorks who learned shoddy python/sql knowledge in the trenches, and have pieced together different "critical" scripts over their careers--they imagine each of these as their magnum opus--like that time they wrote a SAS job to "save" the company after a new database vendor was haphazardly implemented; etc.
The reality is that a lawyer who can barely turn on his computer, like EPAH, could write better code, and dominate a analyst/engineer job in corporate America if he was fully invested and balls deep with AI.
The problem is getting 85% into a project using innate knowledge, and then bringing in AI to solve a perplexing road-block without any context leading up to it. These types prompt like shit, and refuse to learn about different models or reasoning routing. They hit free-tier AI with a boor thread, and expect it to magically manifest a solution.
If you really want to "test" AI in your work, you need to stop being cagey with it, and provide excessive context. Feed all of the flagged doc review items (I'm not a lawmo so forgive me), relevant statutes, case law, etc. Think of the satirical Saturday morning biglaw threads where a partner is emailing you to start a new project, and at least suggests what areas/cases to focus on. You should be as specific as possible, and treat it as leverage--tell it what to comb through, and/or provide a templated example of a good citation and ask that it find more. Never giving it open-ended tasks without guidance like "analyze and quash this breach of contract" or something similarly retarded. You should not view it as an all-or-nothing replacement or laziness hack; it is best used to extrapolate on good ideas, or sift through shit to find pearls. hth |
|
Date: November 8th, 2025 1:16 PM Author: massive mexican sneaky criminal Feed it all the docs you want, it will only ever do RAG and miss important details (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412562) |
|
Date: November 8th, 2025 3:57 PM Author: lake half-breed Do you have the enterprise version? Otherwise how are you feeding client docs (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412882) |
|
Date: November 8th, 2025 4:23 PM Author: brilliant aromatic degenerate You can feed sensitive docs to the api. Just not the chat app. f you selfhost your retrieval layer and call the API, you already keep your data in your environment and only send the minimal query context to the model. It’s probably against the rules of wherever you work because enterprise adds governance and enterprise privacy commitments but in principle the architecture and env will be the same (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412929) |
|
Date: November 8th, 2025 1:11 PM Author: Gaped metal gas station It kind of sucks at law but i don’t doubt that it will get really good at law soon (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412547) |
|
Date: November 8th, 2025 1:18 PM Author: avocado fragrant senate brethren (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412566) |
|
Date: November 8th, 2025 1:19 PM Author: Exciting space skinny woman I think the plateauing has really cooled excitement. And the lack of integration. More than anything else, it makes you realize how much knowledge you’re paid to know that literally isn’t on the internet.
I will be the very first to replace my employees with AI if it’s possible, but it isn’t there yet and for awhile now hasn’t been getting materially closer.
And while I agree it’s mostly insecure white-collar workers shitting on it *ever* being able to do xyz, the biggest defenders/hypemen for the tech (who aren’t directly financially tied to said hype) are 120-125 IQ state-school retards with laughable overconfidence in their ability to understand varied professional applications, especially given that the live in a 3000-sqft house and send their kids to public school. |
|
Date: November 8th, 2025 1:22 PM Author: Embarrassed to the bone orchestra pit Imho basically the second it can replace much of anyone, it can do MILLIONS of jobs. It's not going to "gradually" manifest that sort of capability. It'll just become apparent all at once. The Great Firing of Losers may not happen all at once, but it'll be obvious what's possible by the time it starts. (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412579) |
|
Date: November 8th, 2025 1:22 PM Author: Gaped metal gas station What if they live in a 2200 sq ft house and will be sending their kids to public school. Asking for a friend (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412580) |
|
Date: November 8th, 2025 1:44 PM Author: flirting overrated filthpig or a farm (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412628) |
|
Date: November 8th, 2025 1:42 PM Author: brilliant aromatic degenerate Yeah I dont doubt that the top .1% of humans can outperform it in limited domains but grok 4 and gpt-5-pro are definitely significantly more intelligent than anyone I have ever seen post on xo that’s all I know. I’m not judging it on “can it do some random job”. I’m judging it on reasoning ability, speed, generative creativity etc. humans can’t match it on logical reasoning, speed, cross domain synthesis and nowhere even close on knowledge base it has a training archive that dwarfs what a human could learn in several lifetimes that’s all I’m saying. I’m assuming what you are referring to is it isnt as good at humans at reading social and industry trends and that sort of thing (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412625) |
|
Date: November 8th, 2025 1:45 PM Author: massive mexican sneaky criminal Google "calculators." You'll be blown away at how they outperform humans (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412631) |
|
Date: November 8th, 2025 1:53 PM Author: brilliant aromatic degenerate surpasssing humans through representational generalization isnt analogous to what calculators do through mechanical precision and fixed operations. The stuff ai is now better at than humans is literally the stuff we defined as intellifence before like the stuff only high iq people can do. The stuff humans are better at is the stuff literally all humans can do (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412655) |
|
Date: November 8th, 2025 3:07 PM Author: Exciting space skinny woman All of that is interesting (I’m serious) and shows enormous (albeit speculative) long-term transformative potential, in the same way that quantum computing has transformational upside that hasn’t yet been realized at all, and nanotechnology seemed to 20 years ago.
Its ability—not necessarily just its present ability, but its ultimate ability—to “do random jobs” is all that actually matters in the end though. Everything else is some pimply retard with grandiose illusions of depth and insight, but who everybody laughs at, preening like a fag about value that requires stupid fucking analogies and metaphors to convey (to a bored listener trying to look over your 5-foot-2-inch-off-the-ground shoulder to catch the eye of someone more attractive and interesting to talk to). |
|
Date: November 8th, 2025 3:21 PM Author: brilliant aromatic degenerate I understand your point, because what you are saying is something that generally applies to things that are impressive but don't make money or have widespread applications yet (like quantum computers). But I think you are missing a huge part of it, which is that you can offload a huge amount of cognitive labor onto a machine now. personally I'm lazy as fuck and hate having to think hard about things when I don't feel like it. I like to mostly just vibe and listen to music and smoke weed all day maybe have a few drinks. So when I can put a problem into AI like grok 4 flagship or gpt-5-pro and it can solve a problem that might take days even weeks to solve normally or generate something thats 90% of the way to working thats huge. its like having an intellectual slave but without the baggage of an actual person. to me the tech is already there. the fact that it "hallucinates" sometimes or generates stuff slightly wrong is besides the point and people greatly exaggerate the amount this happens, especially with the latest flagships and with a lot of experience using it. once in a while it seems to glitch and start generating complete trash but its rare with the way i use it, i use the api and kill the session each time and give good instructions and full context so I dont see the type of collapse you might see using the chats through long sessions. to me the fact that it generates stuff thats "wrong" barely matters because if it doesnt work right away i spend a little extra time getting it to work and if something works it works regardless of where it came from. (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412816) |
|
Date: November 9th, 2025 8:40 PM Author: Exciting space skinny woman The hallucinations, I tend to agree, are largely beside the point, although I was assured several years ago that this was a trivial issue to fix (it sure seems like it should be, at least vis-a-vis not fabricating quotes from law sources, which despite the knee-jerk "lol who cares about lawyers; it's got more important shit to do!" reaction, is an application that's both large and monied enough to pay a lot of attention to for that reason, and involves a profession whose opinion of your tech will be highly important, for reasons both foreseeable and unforeseeable, on all time horizons). I don't know what you're talking about re: them being uncommon, though; around half of *quotes* from case law it gives me are fabricated. (And if the response is "well ur not a good prompt formulator if you ask it for quotes", that's fucking bullshit. What I need, ultimately, is fucking quotes, because no judge trusts lawyers' parenthetical summaries of a case.) I've commented on this before, but this is clearly an issue endemic to the tech, because WL's built-in AI platform adamantly refuses to even *try* to give you quotes, even when you ask directly for them, because I assume the architects know that an unacceptable amount of them will be hallucinated and will lose user confidence.
Right now it's the goddamn LAZINESS that's fucking killing me. (See: https://autoadmit.com/thread.php?thread_id=5795393&mc=1&forum_id=2) I didn't even ask "eligible to retire" (which is a not-that-hard-to-calculate formula where you have to be at least 65 y/o, have been in office at least 10 years, and your time in service + age have to = 80, because almost every Circuit judge hits it at 65, and because I KNEW chatgpt would bitch and moan about being asked to look up TWO things); I just asked 65.
If I give it time to think (which btw is not costless), it comes back with 4 active judges appointed by GOP potuses who are now >64. And asks me "should I keep looking for more?" At least with a retarded TTT SA I'd get the satisfaction of looking into his eyes and watching him die a little as I reamed him out for being such a lazy piece of shit. |
|
Date: November 9th, 2025 9:10 PM Author: brilliant aromatic degenerate I totally agree with you on that. The issue with law isnt that it’s not important I’m just trolling when I say stuff like that. The issue is that AI does really well with things like formal frameworks, abstract math and logic, theoretical physics, computer science — because they have tight symbolic structure and are governed by formal invariants. A fake equation or broken derivation or bad logic stands out immediately and fails to compress the internal statistical representation, so the system avoids straying too far. Case law on the other hand is heavy in semantics and involves mountains of text that hinge on subtle distinctions and the training corpora range from legal blog posts to actual binding precedent. It’s really easy for the model to improvise and make up nonsense that could structurally be something that would or even should exist in law but doesn’t really exist. In principle you could engineer a legal model thwt behaves like a physics model. Perhaps by requiring citation verification or precedent retrieval within a declared budget. Why no one has done it yet idk. You don’t need to retrain a full model for it it can be done with a strong base model, retrieval from a legal archive like gpt does with google, and a narrow fine tune, maybe a few other tweaks (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49415346) |
|
Date: November 10th, 2025 12:16 AM Author: Exciting space skinny woman I guess it says something (good) about AI that I now get mad at it, like a human worker, for being a lazy POS. And while I always say on here that I haven't been able to cut out any associates/staff with AI yet, that's actually not true; we quit hiring SAs because of AI. (As I've alluded to, our local LS is a TTT and the SAs we got were usually idiots. I can put in tons of recruiting resources and get T6 *associates*, but I'm not doing all that for SAs.)
But the tech came out of nowhere, and then showed dramatic improvements over several years, and now is clearly plateauing. (And my point ITT has largely been: I'm the one who gets to decide when it's improving and when it's plateauing. I don't really give 2 fucks in the world about its MMLU/MBPP scores or whatever, nor, ultimately, does anyone else.) This does make people wonder whether it's going to be something that replaces my workforce and one day me -- and I am not the least bit defensive about this, as I'm mostly a performer-cum-politician these days, and by the time I'm replaceable, everyone else has been for awhile (I also deeply adhere to the view that tech advancement / automation is a strong good for humanity) -- or whether this is a neat thing that everyone should incorporate into their existing workflows and that probably replaces a few jobs on the margins, like when Xerox started making copiers that collate and laminate and staple and bind your shit for you. |
|
Date: November 10th, 2025 8:01 AM Author: brilliant aromatic degenerate yeah I get your point, but im not totally with you on the "plateau" take. not because I think youre being subjectivist in the lazy sense of denying facts--I actually share your epistemic independence ethos in the sense of setting your own standards of evaluation. I look at the metrics too but for me its ultimately about whether I feel the difference in use and I definitely do. I think the plateau narrative misses how small metric jumps at this level can represent huge practical leaps. like with grok 4 flagship vs grok 4 fast reasoning--the scores are only a few points apart, but the difference in what they can actually do is massive. tiny gains in these scores can unlock whole new classes of problems. so for the stuff I use these models for it doesn't feel like a plataeu at all. but I get why it might look that way if youre using it in areas where it hasn't hit its stride yet. and fair enough that the scores thesmselves dont mean much to you either way. (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49416408) |
|
Date: November 8th, 2025 3:30 PM Author: massive mexican sneaky criminal Yeah, calculators get used for mission critical activities, AI does not. They're totally different, you're right. Still, you really should google "calculators" and learn about what they can do. (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412839) |
|
Date: November 8th, 2025 3:36 PM Author: brilliant aromatic degenerate If you think ai isnt used in mission critical activities your iq is in the mid 60s max (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412850) |
|
Date: November 8th, 2025 4:16 PM Author: massive mexican sneaky criminal Name one. You can't (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412919) |
|
Date: November 8th, 2025 4:31 PM Author: brilliant aromatic degenerate threat detection systems on modern drones and missile platforms (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412943) |
|
Date: November 8th, 2025 5:00 PM Author: massive mexican sneaky criminal There's nothing mission critical about fucking drone hitting a dune buggy lmao. Even Hamas thinks drones are expendable (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412984) |
|
Date: November 8th, 2025 1:58 PM Author: Unhinged translucent pisswyrm (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412670) |
|
Date: November 8th, 2025 3:33 PM Author: fiercely-loyal ticket booth cumskin premise sort of answers itself: because they're low IQ. and they don't understand it. and given that it would be a paradigm shift, it's tough to imagine. (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49412843) |
|
Date: November 8th, 2025 5:09 PM Author: maniacal opaque codepig I'd argue that low IQ people tend to up-play generative AI as incomprehensible magic.
People have been creating predictive models since at least the 1980s. The ability to attain countably infinite training sets is a pretty basic end-of-chapter textbook thought exercise. |
|
Date: November 8th, 2025 5:12 PM Author: massive mexican sneaky criminal I got paid to do this for a living before I went to lol school:
https://www.sciencedirect.com/topics/earth-and-planetary-sciences/supervised-classification
That technology has been in use since the 90s. The first time I saw AI in action this was all I could think about. |
|
Date: November 8th, 2025 5:53 PM Author: maniacal opaque codepig It's fascinating to watch how obsessive laypeople get when things like this become trendy. (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49413077) |
|
Date: November 8th, 2025 5:33 PM Author: brilliant aromatic degenerate This is the type of shit you see on Reddit territorial minimization to pretend you are one of the nerds “in-the-know”. “It’s not that impressive.” “We’ve been doing that for decades.” “It’s just statistics.” But shit like ARIMA, backprop networks with a few thousand parameters, or simple Markov models are nothing like current frontier models might as well be from a diff species (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49413032) |
|
Date: November 8th, 2025 5:35 PM Author: Stimulating range people who are hurt it will never replace lawyers because the issues around Client Confidentiality are insurmountable (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49413038) |
|
Date: November 8th, 2025 5:49 PM Author: flirting overrated filthpig (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49413061) |
|
Date: November 8th, 2025 6:35 PM Author: lake half-breed Explain this..must be missing something (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49413152) |
|
Date: November 8th, 2025 6:46 PM Author: Curious ceo tattoo (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49413170) |
|
Date: November 8th, 2025 6:48 PM Author: massive mexican sneaky criminal You can already run uncompressed models offline on like $10k worth of used server hardware. It will be slow af, and $20-80k nets you more speed, but you CAN do inference for just $10k. (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49413175) |
|
Date: November 8th, 2025 6:06 PM Author: vivacious rambunctious locale volcanic crater I had grok solve a basic algebra equation and it literally got the arithmetic wrong. Discuss. (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49413104) |
|
Date: November 8th, 2025 6:11 PM Author: brilliant aromatic degenerate I see people claiming this on the internet all the time but yet to run into examples in my use and I use grok 4 flagship all the time. I assume claims like this are made up because it makes no sense. (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49413106) |
|
Date: November 8th, 2025 6:38 PM Author: massive mexican sneaky criminal Look at that last sentence
https://i.imgur.com/slUMVdP.png |
|
Date: November 8th, 2025 7:02 PM Author: vivacious rambunctious locale volcanic crater My poast wasn’t flame. It literally happened last week. (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49413202) |
|
Date: November 8th, 2025 6:32 PM Author: massive mexican sneaky criminal And it only took 90 seconds and consumed 2,300W of power. That's progress. (http://www.autoadmit.com/thread.php?thread_id=5794931&forum_id=2E#49413150) |
|
Date: November 8th, 2025 7:51 PM Author: racy bespoke affirmative action In terms of jobs displacement, AI seems like it'd be most helpful as a creative force. Copywriting, graphic design, marketing content, etc.
I'm sure there will be huge gains in architecture and engineering as well at some point.
But most legal jobs just aren't that creative. We already start with form contracts, and then we just negotiate within those. The AI can't really adjust their comments to the size of the project, each party's leverage, the need to close quickly, etc.
Take divorces. Could AI create a standard, fair divorce agreement? Absolutely - but that already exists yet divorce attorneys still make money because people don't want to use it.
There's also a double edged sword with AI. The best use of AI now legally is probably you can take your issue and ask AI to write you a letter to send to the business/regulator/congressman. The letters are pretty good, but they also seem like they're written by AI. Are any of these parties really going to give a shit that someone used ChatGTP to bad mouth this company? My guess is a regulator just tosses AI written complaints in the trash... |