The most prestigious law school admissions discussion board in the world.
|
 |
 
|
7/17/25 AI thread
Poast new message in this thread
Date: July 17th, 2025 4:19 PM Author: Galvanic Coiffed Preventive Strike Locus ChatGPT can now do work for you using its own computer.
Introducing ChatGPT agent—a unified agentic system combining Operator’s action-taking remote browser, deep research’s web synthesis, and ChatGPT’s conversational strengths.
https://x.com/OpenAI/status/1945904743148323285
some silicon valley head of an investment firm falls prey to consuela-tier AI feedback loop narcissistic delusions and starts a PR campaign to Let The People Know about his narcissistic delusions. be careful out there, folks!
https://x.com/GeoffLewisOrg/status/1945864963374887401
AI agents are getting put in drones. cool video of a prototype
https://x.com/chesterzelaya/status/1945709751519916147
frontier model companies are calling for mandated english chain of thought in all models, for "transparency" and "safety." except english CoT is a post-hoc rationalization and does not actually correspond to the model's internal workings, so i don't know what the fuck these people are talking about. i guess this is to hamstring up-and-coming competitors from passing the entrenched frontier companies while they try to monetize what they already have as much as they can
https://x.com/OpenAI/status/1945156362859589955
musk says there's going to be a male version of the gooner AI companion
https://x.com/elonmusk/status/1945558419878154746 |
Date: July 17th, 2025 4:22 PM Author: Galvanic Coiffed Preventive Strike Locus this one is good: some people open sourced all US caselaw on huggingface, so you law guys won't necessarily have to shell out cash for westlaw wrappers or whatever people are trying to scam with these days
it will be relatively simple for you guys with shitlaw outfits to build yourself a wrapper and run your practices with like 10% of the overhead costs that you currently have
https://x.com/EnricoShippole/status/1945129974375039226 |
 |
Date: July 17th, 2025 5:02 PM Author: fragrant pontificating doctorate deer antler Hell yeah, this is great dude. Good find. (http://www.autoadmit.com/thread.php?thread_id=5751466&forum_id=2#49109758) |
|
Date: July 17th, 2025 6:41 PM Author: Galvanic Coiffed Preventive Strike Locus bumping this cuz i love you guys and i know this will be extremely useful for at least a few of you (http://www.autoadmit.com/thread.php?thread_id=5751466&forum_id=2#49110056) |
|
Date: July 18th, 2025 9:53 AM Author: Galvanic Coiffed Preventive Strike Locus (http://www.autoadmit.com/thread.php?thread_id=5751466&forum_id=2#49111427) |
|
Date: July 17th, 2025 5:05 PM Author: mint zombie-like knife I trained a local LLM on xo and it called me a fag (http://www.autoadmit.com/thread.php?thread_id=5751466&forum_id=2#49109762) |
|
Date: July 17th, 2025 8:05 PM Author: Spectacular cerise dog poop codepig What would it even mean to have a mandated chain of thought? You could make a network arbitrarily deep before it generates a token and it could be doing all sorts of crazy computation that doesn’t have any similarity to the thought sequence. The intention here seems to be to prevent recurrent neural networks that could think for a long period of time, but there seems to be no realistic way to prevent someone from building one. A very large feedforward network with weight sharing would do the same exact thing. (http://www.autoadmit.com/thread.php?thread_id=5751466&forum_id=2#49110352) |
|
Date: July 17th, 2025 8:20 PM Author: Galvanic Coiffed Preventive Strike Locus it doesn't make any sense to me either unless you assume some kind of economic competition angle
in general, the enthusiasm for chain-of-thought as a "safety" mechanism that has come from otherwise pretty smart people has always struck me as extremely strange and inexplicable |
|
Date: July 18th, 2025 12:11 PM Author: Galvanic Coiffed Preventive Strike Locus The White House is preparing an executive order requiring AI models to be 'politically neutral and unbiased' according to a report by the WSJ. Compliance will determine eligibility for federal contracts, so this is a big deal for all the labs. The EO is expected next week.
https://x.com/AndrewCurran_/status/1946238951309885939
https://www.wsj.com/tech/ai/white-house-prepares-executive-order-targeting-woke-ai-e68e8e24
just another reminder that the trump administration is the exact same thing that a kamala harris administration would have been, but worse. because uh, the jews, or something. yup |
|
Date: July 18th, 2025 12:14 PM Author: smoky talking hell toaster "Compliance will determine eligibility for federal contracts"
translation: This is regulatory capture. The bigger the player the less of an issue it is to throw bodies at compliance. |
|
Date: July 18th, 2025 12:20 PM Author: Galvanic Coiffed Preventive Strike Locus it's actually quite easy to not build artificial liberal bias into your models and this would in fact make things *easier* for labs going forward, not harder
this is unequivocally 180000000 |
|
Date: July 18th, 2025 12:22 PM Author: smoky talking hell toaster it's easy to do, proving it and actually going through the "compliance" part (wherein the difficulty and cost is not actually complying, but proving you have complied) are why it's regulatory capture. you think OpenAI and Google, who have some of the most "woke" AIs, don't love shit like this? they don't really care if their AI is lib or not, but being forced to thoroughly document and "prove" (through whatever means of testing or legal bullshit is required) it is unbiased is a huge advantage for major players and a fuck-you to anyone challenging incumbents.
this is 100% with the blessing of big tech |
|
Date: July 18th, 2025 12:30 PM Author: aquamarine excitant nowag they might also like it because it gives them the cover to demand their products actually be, you know, effective, without risking a mutiny by their purple-haired staff. (http://www.autoadmit.com/thread.php?thread_id=5751466&forum_id=2#49111751) |
|
Date: July 18th, 2025 12:31 PM Author: Galvanic Coiffed Preventive Strike Locus (http://www.autoadmit.com/thread.php?thread_id=5751466&forum_id=2#49111757) |
|
Date: July 18th, 2025 12:31 PM Author: Galvanic Coiffed Preventive Strike Locus *rolls eyes in real life*
come on, bro. lol. if anything, this is actually better for newer models being developed that are trying to compete with the big guys, because they'll have a chance to Curate their pre-training materials to avoid bias, and develop strategies for post-training that do this as efficiently as possible
even the "worst-case scenario" is that this will functionally serve as nothing more than a loyalty test that labs have to pass to work with the trump administration. which is A Good Thing, actually. being in power and forcing people to be biased toward your side is good! |
|
Date: July 18th, 2025 12:33 PM Author: smoky talking hell toaster you don't understand what I'm saying. it's bad for small players because of how anything involving compliance works. it's not going to be some nice little automated system, it's going to be a bunch of legal bullshit or formalized bureaucratic hoops you have to jump through to get federal signoff. that's how regulatory capture works. (http://www.autoadmit.com/thread.php?thread_id=5751466&forum_id=2#49111761) |
|
Date: July 18th, 2025 12:37 PM Author: Galvanic Coiffed Preventive Strike Locus i understand, i'm just saying that you're being obtuse about it
if we try to quantify this development, we could say that the potential downside of regulatory capture makes things 5% worse overall. but the positive effects that i described above make things 500% better overall
the cost benefit analysis is massively skewed to the side of this being A Really Good Thing |
|
Date: July 18th, 2025 12:41 PM Author: smoky talking hell toaster until next time a lib is in office they use this as precedent to just impose their will instead. guess who benefits? the same large incumbents, who have much less of a problem dealing with bureaucratic pivots.
it's not Really a Good Thing. best case is it's just puffery, and if they're serious it's actually much worse because of what I mentioned above (setting precedent for stronger government oversight/bureaucratic involvement) |
|
Date: July 18th, 2025 12:47 PM Author: Galvanic Coiffed Preventive Strike Locus this argument (in general, not just here) doesn't work anymore
it doesn't make any sense to try to extrapolate potential consequences of current legislation into a hypothetical future in which libs regain power. if libs ever regain power of fedgov, they'll mandate that all AI be cartoonishly woke and shitlib, stack the supreme court, seize twitter, outlaw all dissent, etc
legislation and policy enacted by the trump admin can and should only be coherently evaluated in the context of "conservatives" maintaining power of fedgov. people are conditioned into making the general argument that you've made above (because it made sense in the past, when this country wasn't in a state of civil war) and haven't sat down and thought about the current state of the world and adjusted their priors accordingly |
|
Date: July 18th, 2025 1:02 PM Author: smoky talking hell toaster Nothing Ever Happens. There is no "civil war" (see for instance how easily Trump singlehandedly took over the Republican party--there is no "swamp" or "deep state"), the world is not some radically different place because of political tide shifting. The world is *about* to become a radically different place because we're all going to be fucking unemployed and basically anything could happen in the coming decade, whereupon you might very well end up with a real civil war on your hands that pales in comparison to the culture wars. (http://www.autoadmit.com/thread.php?thread_id=5751466&forum_id=2#49111834) |
|
Date: July 18th, 2025 12:34 PM Author: Citrine indirect expression elastic band The definition of ‘politically neutral and unbiased’ will be much different under Trump than just about anyone else (http://www.autoadmit.com/thread.php?thread_id=5751466&forum_id=2#49111766) |
|
Date: July 18th, 2025 12:27 PM Author: aquamarine excitant nowag can you explain the second entry like i’m a retarded lawyer? if it’s what i think it is that is pretty damn funny. (http://www.autoadmit.com/thread.php?thread_id=5751466&forum_id=2#49111744) |
|
Date: July 18th, 2025 12:34 PM Author: Galvanic Coiffed Preventive Strike Locus you mean the guy getting stuck in a narcissistic masturbatory feedback loop with AI and then earnestly announcing it to the world?
yeah it's hilarious and it's going to happen all the time going forward |
|
Date: July 18th, 2025 12:31 PM Author: Spectacular cerise dog poop codepig gemini 2.5 pro gets 31% on the international math olympiad. the performance increases at math have been impressive.
https://matharena.ai/ |
|
Date: July 18th, 2025 1:16 PM Author: multi-colored overrated philosopher-king These people are truly and deeply retarded (http://www.autoadmit.com/thread.php?thread_id=5751466&forum_id=2#49111860) |
|
Date: July 18th, 2025 1:22 PM Author: Galvanic Coiffed Preventive Strike Locus an astute observation, KikeLord69 (http://www.autoadmit.com/thread.php?thread_id=5751466&forum_id=2#49111874) |
|
Date: July 18th, 2025 1:22 PM Author: smoky talking hell toaster he's doing great (http://www.autoadmit.com/thread.php?thread_id=5751466&forum_id=2#49111877) |
|
Date: July 18th, 2025 1:39 PM Author: Citrine indirect expression elastic band ty brother I appreciate these updates (http://www.autoadmit.com/thread.php?thread_id=5751466&forum_id=2#49111939) |
|
Date: July 18th, 2025 1:52 PM Author: Galvanic Coiffed Preventive Strike Locus (http://www.autoadmit.com/thread.php?thread_id=5751466&forum_id=2#49112013) |