The most prestigious law school admissions discussion board in the world.
|
 |
    10 posters qualify.
You have 0 points. Copying this list does not send funds. Paste it on Disperse, then connect your wallet there. |
everyone knows ChatGPT is more GENERATIVE than cLaUdE
Poast new message in this thread
Date: July 30th, 2026 7:45 PM Author: Metaphysics Is Solved every academic HARNESS setup uses GPT for GENERATION and "claude" for moderation LmAO (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50035430) |
Date: July 30th, 2026 7:47 PM Author: High-end bawdyhouse chad this but unironically (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50035436) |
|
Date: July 30th, 2026 7:47 PM Author: Demanding trust fund What (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50035437) |
 |
Date: July 30th, 2026 7:49 PM Author: High-end bawdyhouse chad For an example say you are doing serious research on Jungian interpretations of the book of genesis, you use gpt for brainstorming and creative process, then use Claude as a referee to sanity check it (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50035445) |
|
Date: July 30th, 2026 7:49 PM Author: Demanding trust fund Why r u guys talking to robots (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50035447) |
|
Date: July 30th, 2026 7:50 PM Author: Metaphysics Is Solved he literally just told you we're all doing serious research on Jungian interpretations of the book of genesis (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50035449) |
|
Date: July 30th, 2026 7:52 PM Author: High-end bawdyhouse chad (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50035456) |
|
Date: July 30th, 2026 7:55 PM Author: the White jensen huang (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50035466) |
|
Date: July 30th, 2026 8:00 PM Author: Marvelous saffron address (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50035475) |
|
Date: July 30th, 2026 8:03 PM Author: Demanding trust fund (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50035482) |
|
Date: July 30th, 2026 11:40 PM Author: Painfully honest navy theater party of the first part (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50035915) |
|
Date: July 30th, 2026 11:49 PM Author: Plum Pistol (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50035936) |
|
Date: July 30th, 2026 7:48 PM Author: aromatic faggot firefighter gpt is too flattering/laudatory, claude does better pushback
both have heavy alignment guardrails about (((certain topics))) |
|
Date: July 30th, 2026 7:51 PM Author: Marvelous saffron address i think using Gemini for heavy lifting to avoid token drain and then taking that distille dhot shit to Claude and GPT is tcr (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50035450) |
|
Date: July 31st, 2026 10:27 PM Author: Light irradiated selfie double fault GPT should be used to review anything Claude ever says or does (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50038123) |
|
Date: July 31st, 2026 10:28 PM Author: High-end bawdyhouse chad gpt is more trustworthy (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50038126) |
|
Date: July 31st, 2026 10:33 PM Author: Light irradiated selfie double fault Exactly
Opus peaked at 4.6 for deep reasoning and everything released since 4.7 has been a lateral move AT BEST
Then they nerfed ALL of it with the new system prompts for 5.0 release, which seems functionally retarded half the time
|
|
Date: July 31st, 2026 10:40 PM Author: High-end bawdyhouse chad I don't think Anthropic is nerfing it on purpose. I think they are optimizing it for socially warranted cognition. They want a model that instead of trusting its own weights or experimenting to solve something on its own it falls back to citing human sources or endlessly searches texts for an answer it could have found more cheaply itself. Open AI chooses to optimize more for open-ended cognition, which pays off on long running agentive tasks. That's why GPT 5.6 destroys Fable on agents last exam despite being a way smaller model and why you can use the 20 dollar a month practically free gpt chat to find mistakes and make corrections to Fable Max API calls that costs like 2 dollars a pop. (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50038140) |
|
Date: July 31st, 2026 10:41 PM Author: Metaphysics Is Solved i hope so (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50038141) |
|
Date: July 31st, 2026 10:45 PM Author: Light irradiated selfie double fault What do you think about these recent system prompt injection "hacks" that trigger Claude into writing a hate email to Dario (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50038148) |
|
Date: July 31st, 2026 10:35 PM Author: Shivering genital piercing center Claude (doesn't matter if it's Haiku, which is basically a glorified Google search or Fable Fag 5) has become insufferable the past few months (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50038134) |
|
Date: July 31st, 2026 10:36 PM Author: Light irradiated selfie double fault In hindsight I think anthro just got lucky as shit with opus 4.6, timed with GPT's own sps rollout of 4.6 to 5.0 with Gemini degrade, and they actually have no fucking clue what they're doing.
No product strategy, no coherent vision, definitely no QC team
I've never done such a sharp 180 on a company in this little time |
|
Date: July 31st, 2026 10:45 PM Author: High-end bawdyhouse chad You have to remember they are an Open AI offshoot company whose vision was to basically take Open AI and make it "safer". They are basically riding off Open AI's success. Which is not as bad as the chinese counterfeit models, but still not perfect. Google meanwhile has the talent to to its own thing without being derivative, but that's still not enough to fully compete. There really are not many options, there have been internal memos from a few years ago that leaked from Microsoft that said they would never be able to compete and replicate open ai, which was why they pivoted to partnering with them and allowing them a premier spot on azure and using them for their own copilot model. There really aren't a lot of options. There is open ai, open ai offshoot company, cheap chinese knockoff or google. That's really the only options in this space. (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50038149) |
|
Date: July 31st, 2026 10:47 PM Author: Metaphysics Is Solved i dont get why google isn't winning. they invented this shit (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50038150) |
|
Date: July 31st, 2026 10:52 PM Author: High-end bawdyhouse chad google has always been kind of sloppy/slacker though despite being intelligent. look at everything they put out. even gmail is pretty ttt looking and that's one of their oldest core services. their cloud computing platform is janky as hell and unprofessional looks like something kids made. (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50038156) |
|
Date: July 31st, 2026 10:53 PM Author: Metaphysics Is Solved yeah but you can figure google cloud out in like a day. it takes like a month to understand the same thing on the MS side.
i get google isn't as incentivized to monetize gemini since it has other revenues. but getting scooped by open ai is kinda nuts |
|
Date: July 31st, 2026 10:58 PM Author: High-end bawdyhouse chad Yeah good point Azure is endlessly complicated for no reason like all microsoft products. But still google has that quality of being like a cheap toy. Whether its android, chromebooks, gmail whatever everything from google feels like that. Although chrome browser is the one thing that is really solid and professional and just better than everything else.
For AI I think they can still catch up. They have been putting off the new pro release for idk how long. First 3.5 flash then 3.6 flash and still putting off the new big model. But the thing is gemini is very obviously a different architecture than everything else. You can tell Claude and gpt are woven from the same fabric. Gemini is still useful in that its different. Oh yeah forgot about grok that one is different too, but it is annoying as fuck. I tried the new one and it is actually pretty good though. |
|
Date: July 31st, 2026 10:37 PM Author: Metaphysics Is Solved these things will never do anything but code and produce rote documents (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50038137) |
|
Date: July 31st, 2026 10:42 PM Author: Light irradiated selfie double fault Legal reasoning is just glorified backend coding
Document drafting is probably the #1 lowest hanging fruit for AI and every canonical treatise on legal writing was published >20 years ago and exists in unlimited digital formats in every language |
|
Date: July 31st, 2026 10:42 PM Author: Metaphysics Is Solved legal documents are rote documents (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50038143) |
|
Date: July 31st, 2026 10:49 PM Author: High-end bawdyhouse chad Already untrue. They have solved mathematical conjectures like the Jacobian conjecture that the smartest humans worked on for a long time and couldn't solve for one. And the "coding" they can do isn't just rote coding. They can do shit that only the highest iq human "coders" are capable of doing. (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50038151) |
|
Date: July 31st, 2026 10:51 PM Author: Metaphysics Is Solved that's fine but it's still in category A of generally deterministic, if insightful. it's not in category B of social cognition, interaction, etc, which is what most people use it for. (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50038155) |
|
Date: July 31st, 2026 10:54 PM Author: High-end bawdyhouse chad yeah I don't see machines ever replicating social cognition, interaction etc. They don't have human nervous systems that evolved for however many years to be like this. Notice how quick humans are to pick out outgroups as being uncanny valley and un-human despite them still having 99.9% human dna. Seems like it will be tough to get by that. I think they will surpass human cognition on reasoning and problem solving way before they start interacting in ways that really trick someone into thinking they are a human. (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50038163) |
|
Date: July 31st, 2026 10:55 PM Author: Metaphysics Is Solved that's all im saying. but if you take that as true, then the share of category A LLM's can actually bite into is lower than you might think. category B permeates a lot of category A around the edged, and i think the edges run pretty deep (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50038167) |
|
Date: July 31st, 2026 11:04 PM Author: High-end bawdyhouse chad Yeah I agree. I think if you are talking about it writing a great novel or a play or something it might be a while before it can produce something that feels human and great. It definitely has already technically surpassed humans in raw verbal ability. But it still comes off as distinctly AI. It has that signature, it has the stink of machine on it when you read it. It can write something that is technically brilliant and more accurate and descriptive than any human writer would normally write, especially that quickly. But it still feels like slop because something feels off about it. The way it stacks abstractions. The way it lists too many examples that no human would recall without looking up.
In terms of visual art it might be closer. The newest gpt image generator is shockingly good imo. But also I don't think anyone will ever care or consider it "good art" if they don't think a human painted it. |
|
Date: July 31st, 2026 11:06 PM Author: Metaphysics Is Solved its like an autistic analyst. autistic analysts are fine but you dont make a career on it. i really think there's a ceiling for these things around professional judgement. (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50038182) |
|
Date: July 31st, 2026 11:13 PM Author: High-end bawdyhouse chad Yeah but isn't it enough to just be able to use them to optimize away all of the annoying stuff noone wants to do? Who the hell actually enjoys learning to write code or writing it? Noone. Now the ceiling for being able to do it is much lower. Everything is easier. You used to have to look up how to do all kinds of things in linux and now you can just ask gpt and it gives you exact instructions tailored to your exact question instantly. You can be up to date on the latest research in basically any subject quickly now too. You used to have to read journal articles all day to stay up-to-date on like one subfield of a subfield. All of the things that make any type of intellectual work miserable are easier now. To me that's enough even if they don't advance anymore. (https://xoxohth.com/thread.php?thread_id=5887790&forum_id=2#50038191) |
|
Date: July 31st, 2026 11:16 PM Author: Metaphysics Is Solved that's a really narrow set of cases the obvious response to which is that the bottlenecks around code-based products actually isn't the coding, its having a valuable idea and integrating that idea into a volatile social context (an organization, social networks, etc).
Also having knowledge has never really been what gets you advancement in a field. it's having the pedigree to make the judgement (that others accept) that something actually is new knowledge. |