The most prestigious law school admissions discussion board in the world.
|
 |
 
|
Anti AI people are seemingly increasingly deluded
Poast new message in this thread
Date: May 31st, 2026 6:21 PM Author: .,.,...,..,.,.,:,,:,...,:::,...,:,.,.:..:. By that I mean people in the Bluesky/Ed Zitron sphere. The central point of their claims isn’t that AI is bad, but that it’s basically fake and the AI house of cards will collapse any day now. They have been saying for several years now that AI won’t get much better than it is now, the benchmarks don’t mean anything and AI companies will soon give up and admit defeat. This was amusing for a while even when it was obvious the scaling hypothesis was true, but it’s seemingly increasingly insane now that we are over 6 months past the release of Opus 4.5. Revenues are exploding, people are increasingly recognizing that agentic models actually work and model releases are still showing regular improvements. Mythos is going to be released in a few weeks (the model that found thousands of cyber vulnerabilities), and surely these people will find a way to dismiss its capabilities like everything before. (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49908982) |
 |
Date: May 31st, 2026 6:50 PM Author: ..,.,.,,,,.,.,..,.,,,.,..,,.,.,,, Counterpoint
If the LLMs really represented an imminent , disruptive force that will displace all white collar work, where are the killer apps?
They’ve been at these things for years, and it’s been nearly 4 years since chat gpt, nearly 2 years since performance gains from pre training have fallen off a cliff. We have yet to see any of this translate to functional, let alone vital, applications outside the area of computer code.
If AI is coming for white collar work, it’s a long, difficult, and uncertain road. That said, it won’t stop dumb companies from fooling themselves into “replacing” workers with AI slop. |
|
Date: May 31st, 2026 6:53 PM Author: Dario Amodei you dont get to see them. they're implementing them at the enterprise level and you will get floored by the giant wave when they decide its your time. hth (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909045) |
|
Date: May 31st, 2026 6:54 PM Author: ..,.,.,,,,.,.,..,.,,,.,..,,.,.,,, Right. It makes sense they would keep them a secret, what with the imminent IPOs. Don’t want to show your hand (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909047) |
|
Date: May 31st, 2026 6:54 PM Author: Dario Amodei its all being privately financed. don't worry about it (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909050) |
|
Date: May 31st, 2026 6:56 PM Author: ..,.,.,,,,.,.,..,.,,,.,..,,.,.,,, I will just take the good word of honest guys like Sam Altman and Dario amodei (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909059) |
|
Date: May 31st, 2026 7:04 PM Author: need tokens (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909072) |
|
Date: May 31st, 2026 7:15 PM Author: Pope Leo XXX (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909080) |
|
Date: May 31st, 2026 7:49 PM Author: need tokens (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909149) |
|
Date: May 31st, 2026 7:11 PM Author: 67 mythos actually is revolutionary but it is definitely not a turn key product, it still hallucinates, gets things wrong, needs heavy amount of prompting to get good outputs
all this stuff is just a (very effective) tool in the hands of the right people. but that’s really it |
|
Date: May 31st, 2026 7:12 PM Author: Dario Amodei That's it tp (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909078) |
|
Date: May 31st, 2026 7:17 PM Author: The Penis How much time do you think its actually saving though? Even modest amounts of time saved can quickly scale to decades or centuries across all human research teams. And the "wasted time exploring things" or whatever that I hear sometimes I think isn't an argument for not-saved time it's the opposite direction when you consider that someone can traverse god knows how much ground in like a week that would have taken years before eliminating bad options that would have been a stalled research program. It doesn't have to be effective in everyone when all of the smartest people have access it still is going to result in advancements. (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909082) |
|
Date: May 31st, 2026 7:25 PM Author: 67 oh absolutely, but the notion that you can just replace everyone with AI is retarded. there are very few skilled jobs (data entry, sending emails, and administrative tasks are not a skilled jobs) where it could actually replace people rather than just making competent people more effective
every time paul graham mentions “AGI” i roll my eyes. it is being massively oversold but that’s doesn’t mean it isn’t a revolutionary tool for certain tasks |
|
Date: May 31st, 2026 7:25 PM Author: The Penis Oh yeah I agree with that for sure (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909094) |
|
Date: May 31st, 2026 7:29 PM Author: need tokens ???? but if it makes competent people more effective, then some of the competent people lose their jobs as companies realize they don't need as many. so that's replacing people, no? maybe "displacing" is more accurate? (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909100) |
|
Date: May 31st, 2026 7:43 PM Author: 67 maybe, but in my admittedly narrow experience every job i’ve had we have been constantly hiring even when the company is laying off thousands of fake job people (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909134) |
|
Date: May 31st, 2026 7:46 PM Author: need tokens probably depends how the company is doing (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909138) |
|
Date: May 31st, 2026 7:26 PM Author: cowgod You clearly know nothing about the State of Gaming. “Killer apps” require Huge Teams, immense budgets. And a 6-12 year Runway. (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909095) |
|
Date: May 31st, 2026 7:30 PM Author: ,.,....,..,.,.,,,,..,..,.,..,.,.,.,... labs are still hardware constrained and there are easier gains from post-training. post-training gains tend to be narrower and more domain specific, but they are large and especially important for the tasks that matter the most currently for enterprise customers (agentic tasks, especially coding). Mythos and Spud are apparently larger models (especially Mythos), so pre-training gains are happening. by 2030, they'll likely be able to afford training runs in the quadrillions of parameters (although I have to wonder whether that will be justified - recurrent/looped transformers are probably the most logical way to exploit greater compute). there are many ideas out there that could be exploited if labs had orders of magnitude more compute. the compute ceiling will be lifted substantially over the coming years, so large performance improvements even with no new ideas are inevitable.
for agentic coding tasks, the performance increases over the last year have been very large. guess what task matters the most for recursive self-improvement loops? |
Date: May 31st, 2026 6:22 PM Author: Dario Amodei (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49908985) |
|
Date: May 31st, 2026 6:34 PM Author: everything is biology the inherent technical limitations of LLMs are not ever going to magically disappear
they can't think, they can't plan, they don't do anything but predict text. they are going to continue to get better and better at computer stuff, because computer stuff is all based on text. and computers can do a lot of stuff, so they will be increasingly useful as long as energy costs stay below whatever threshold is needed
the problem with Computer People - among many problems with Computer People - is that they think that computers are the real world. when a Computer Person says that "LLMs are going to do everything," they're not lying, because for Computer People, "everything" is computers. it's kind of like how women don't even perceive unattractive men as existing, so whenever they are commenting on or thinking about "men," they are only thinking of attractive chad men
this leads to a lot of tedious semantic confusion surrounding LLMs and AI because Computer People are unable to perceive that there is a world outside of Computers, and that LLMs will never be capable of navigating this world, so they are constantly saying bizarre and ridiculous things about LLM future capabilities and trajectories that confuse and frighten non-Computer People |
|
Date: May 31st, 2026 6:35 PM Author: need tokens 180 and cr (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909015) |
|
Date: May 31st, 2026 6:55 PM Author: The Penis Can't think, can't plan is factually false. and "computer people" doesn't mean much when computers are used in most 2026 jobs either directly or indirectly. (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909054) |
|
Date: May 31st, 2026 7:03 PM Author: need tokens I didn't read "Computer People" as people with jobs that use computers
but I disagree with his assumption that LLMs will never be able to navigate the world, as embodied AI is and will presumably continue to be built on LLMs. I just agree about where it is now and why people talk past each other. |
|
Date: May 31st, 2026 7:07 PM Author: The Penis But llms apply to everything people use a computer for. Researchers in chemistry or biology aren't computer engineers, but alpha fold still compressed decades or more research into like 3 years. Protein folding had been an open problem for what like 50 years? (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909074) |
|
Date: May 31st, 2026 7:12 PM Author: need tokens yeah it's possible he's underestimating how widely applicable LLMs are, I just like his point about people talking past each other because some people overvalue Computer (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909077) |
|
Date: May 31st, 2026 7:15 PM Author: everything is biology LLMs can approximate "thinking" and "planning" when the environment is text-only. it's what a contemporary LLM is doing when it is producing conversational or coding output
but the real world is not text-based. there is no way to create an "irl LLM." the real world is essentially infinite compared to the very narrow environment of text and cannot be "tokenized" the way that text can be
every time this is pointed out to LLM maximalists they avoid addressing it, handwave it away, or just make up ridiculous lies claiming that "it will just Happen, okay?!??!" real-world AI is definitely possible, and imo even inevitable on a long enough time scale with a stable human society that can maintain an environment that can support research into it. but it won't look anything like LLMs and we are not even *remotely close* to it |
|
Date: May 31st, 2026 7:19 PM Author: The Penis Well yeah I agree with this if you are talking "real world" as in like sentient superhuman robots walking around everywhere. But I think you are underestimating how powerful "text based" AI is and the degree to which it is coupled to the real world when humans are constantly interacting with it, even if it isn't a primary coupling. (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909084) |
|
Date: May 31st, 2026 7:28 PM Author: everything is biology the danger here is that Computer People and their capitalist enablers will attempt to warp IRL into an environment that is maximally divorced from physical reality and maximally matches digital/text-based LLM reality - because that is what would be convenient and self-serving for them
that is: if you cannot create an AI that can dominate the real world, you instead change the real world into something that AI can dominate
this has, of course, already been happening for some time. historically, pre-AI technology and capitalism have already caused society to follow this pattern and AI will only accelerate it |
|
Date: May 31st, 2026 7:31 PM Author: The Penis Yeah I can see this happening. I mean it already has to an enormous extent just with pre-ai information tech. (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909104) |
|
Date: May 31st, 2026 7:33 PM Author: need tokens I think you're right. should we destroy all the data centers with an electromagnetic weapon? |
|
Date: May 31st, 2026 7:41 PM Author: everything is biology unironically, yes, but we are going to have to do so much more than that too (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909130) |
|
Date: May 31st, 2026 7:43 PM Author: need tokens cr, what else (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909133) |
|
Date: May 31st, 2026 7:53 PM Author: everything is biology have you seen the mission impossible movie with the villain who wants to trigger Ragnarok by engineering a worldwide nuclear holocaust (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909159) |
|
Date: May 31st, 2026 6:42 PM Author: Juan Eighty Anti AI chad here
can any bort AI bros explain to me how creating a hyperintelligent, all-powerful being ends well for humanity? |
|
Date: May 31st, 2026 6:49 PM Author: Book1.xlsx It's helping computer people jobs be easy for a couple years until everything collapses (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909041) |
|
Date: May 31st, 2026 6:53 PM Author: Dario Amodei (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909046) |
|
Date: May 31st, 2026 7:18 PM Author: the walter white of this generation (walt jr.) I use the shit constantly and it’s still a long way from replacing anyone. Definitely not yet an invention on the order of the internet. I used the thing to ‘help’ me do like 4 home improvement projects this weekend and it was wrong 2/3rds of the time. Took a snapshot of my hot tub innards and asked where the UV bulb goes, and it confidently identified every other goddamn thing first before the actual UV unit. (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909083) |
|
Date: May 31st, 2026 7:20 PM Author: need tokens "a long way from replacing anyone" is crazy. even the fact that it can't FULLY replace a given PROFESSION doesn't mean it isn't actively replacing people in that field as companies realize they can downsize their teams now that people are increasing productivity with AI. (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909085) |
|
Date: May 31st, 2026 7:32 PM Author: cowgod And yet as we speak, AAA Studios still have Huge Teams… (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909105) |
|
Date: May 31st, 2026 7:47 PM Author: The Penis Games aren't just "engineering" though, they are a Lifestyle (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909140) |
|
Date: May 31st, 2026 7:48 PM Author: need tokens (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909142) |
|
Date: May 31st, 2026 7:23 PM Author: The Penis I'm sure it has already displaced like a million jobs maybe several million worldwide in the past 5 years. Who the hell needs a research assistant or a junior engineer or a paralegal if they have current frontier AI? Replacing full humans in the sense of having "fully autonomous systems" in full place of a human I agree with you is further out. I think more likely will be semi-autonomous human-in-the-loop systems displacing massive amounts of jobs. (http://www.autoadmit.com/thread.php?thread_id=5870287&forum_id=2...id#49909090) |