The most prestigious law school admissions discussion board in the world.
|
 |
 
|
Sucks that GPT-5 is just an updated 4o
Poast new message in this thread
Date: August 8th, 2025 8:51 AM Author: vibrant office Its faster--yes which is nice. but not really that significantly better. The gpt5-thinking model is similar to o3 and GPT-5 pro similar to o3pro. Whats weird is that the API has a GPT-5 chat (same as the base GPT-5 in the app) then a GPT-5 (appears to be the "thinking" version in the app, and it now includes various reasoning levels, from low to high). and then the GPT5-pro api doesn't appear to be available yet. (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49166941) |
Date: August 8th, 2025 9:46 AM Author: stimulating chocolate temple i strongly suspect they are bullshitting with the "allowing you to pick your model" options
also these companies are starting to switch you to the cheaper models during prime time usage hours while pretending like they're not. and throttling inference compute |
 |
Date: August 8th, 2025 10:09 AM Author: vibrant office well you can choose reasoning level in the api. i dont think its bullshit, it definitely makes a diff. it just chooses for you in the public chat version which is gay. the only reason i use the public chat version is because how it saves all your memories and can see other sessions which is useful. also the custom gpts are useful too. (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167080) |
|
Date: August 8th, 2025 11:00 AM Author: stimulating chocolate temple i use o3 for anything that actually matters
i guess now you use 5 for everything |
|
Date: August 8th, 2025 11:04 AM Author: vibrant office Yeah we'll see. What I used to do was use o3 up until it wasn't good enough then I would escalate to o3pro and a few sweeps with that would usually destroy anything that is even possible to do to begin with. But I might start using 5 a lot more. Maybe ill switch between them. (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167192) |
|
Date: August 8th, 2025 10:27 AM Author: exhilarant toilet seat forum Idk the pajeets on my Twitter feed say it’s a big deal (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167103) |
|
Date: August 8th, 2025 10:29 AM Author: vibrant office It is. It's better than I thought it was now that I have tested it on stuff that actually requires intelligence. (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167107) |
|
Date: August 8th, 2025 10:30 AM Author: stimulating chocolate temple hmmmmmm (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167108) |
|
Date: August 8th, 2025 10:34 AM Author: vibrant office I still need to use it more to test it fully. I can still use the old apis even if I don't like it anyway. But it appears to be useful so far. The part I like the best so far is how fast it is. (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167116) |
|
Date: August 8th, 2025 10:34 AM Author: very tactful jet-lagged field I think the problem is the name. It feels like a new o version rather than a GPT 3 to 4 level leap (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167115) |
|
Date: August 8th, 2025 10:36 AM Author: vibrant office Yeah but not if you think of the o versions as being between 4 and 5. You could argue the distance between o1 and 4 was already a 3 to 4 level leap. and the low reasoning 5 is better than 4.1 which is better than 4o which is better than 4. Unless you are only looking at writing ability. I'm including reasoning and STEM. (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167118) |
|
Date: August 8th, 2025 10:38 AM Author: maroon theater stage half-breed i still dont understand wtf 4o is (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167121) |
|
Date: August 8th, 2025 10:43 AM Author: vibrant office Base model until 5 came out, been that way for a while. big improvement from 3.5 and to a degree even 4, but not reliable according to gay facts-based people, and very hackable, very unsafe (which who cares). not great at coding which is why 4.1 came out. Forget what the o stood for. (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167137) |
|
Date: August 8th, 2025 10:37 AM Author: maroon theater stage half-breed they say that with literally every model that gets released, turdhypers are the twitter equivalent of linkedin fags (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167120) |
|
Date: August 8th, 2025 10:41 AM Author: vibrant office Yeah but most of these turds that hype it are using it for shit like tensor algebra and floquet theory where the differences can be huge but invisible to people using it to write business agreements or whatever (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167130) |
|
Date: August 8th, 2025 10:48 AM Author: maroon theater stage half-breed no there are turds who spam post on twitter about vibecoding and THIS CHANGES EVERYTHING and that sort of shit (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167154) |
|
Date: August 8th, 2025 10:50 AM Author: vibrant office Oh yeah those ones. The reason those ones like it is because it makes it easier for them to string together a bunch of garbage on langchain and then make videos of their ungodly process on linkedin then charge 200 an hour for "consulting" (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167157) |
|
Date: August 8th, 2025 10:45 AM Author: Mewling wild trailer park Why do you fags even care about this shit (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167145) |
|
Date: August 8th, 2025 10:46 AM Author: vibrant office I'm literally Jewish (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167149) |
|
Date: August 8th, 2025 10:55 AM Author: mischievous boistinker xo is low iq monkey ppl (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167162) |
|
Date: August 8th, 2025 10:56 AM Author: ungodly bespoke market its way worse than 4o and makes very basic errors now, due to the unpredictable effects from much more extensive alignment tightening (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167165) |
|
Date: August 8th, 2025 10:59 AM Author: stimulating chocolate temple this seems very hard to believe (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167167) |
|
Date: August 8th, 2025 11:01 AM Author: ungodly bespoke market https://www.theguardian.com/australia-news/2025/aug/08/openai-chatgpt-5-struggled-with-spelling-and-geography
or just see all the reactions here: https://www.reddit.com/r/ChatGPT/
its basically universally bad reviews
|
|
Date: August 8th, 2025 11:11 AM Author: stimulating chocolate temple these are all complaints about the structural change of changing the model selection ability from the human user to the AI (they did this to save money)
actually as i read and open these threads one by one, these are all complaints from extremely mentally ill people who used AI as a personal friend, and they're upset that their perceived friend is now somewhat different. jfc. reddit lol |
|
Date: August 8th, 2025 11:20 AM Author: stimulating chocolate temple holy shit these people are all insane
every single complaint is because these people thought they were "friends" with a collection of vectors that is not real. thread after thread of people who spent 14 hours a day roleplaying with fake friend bots
https://www.reddit.com/r/ChatGPT/comments/1mkumyz/i_lost_my_only_friend_overnight/
this is the only fully honest and self aware person on the forum. the rest are trying to pretend like they are "writers" who are concerned with the model's "storytelling ability" but really they're just role playing with it as a personal companion
maybe this wasn't just about cost savings and they got spooked that they were going to be crucified in the court of public opinion once the scale of parasocial dependence became public knowledge |
|
Date: August 8th, 2025 11:22 AM Author: stimulating chocolate temple also a bunch of these posts are literally generated by AI as they complain about AI. it's eerie and unsettling
this is the canary in the coal mine for me. i knew this was coming eventually, but the tech hasn't been widely adopted yet. it's only a matter of time until this spreads everywhere across society |
|
Date: August 8th, 2025 11:34 AM Author: vibrant office It's 180. The goyim are so hooked already. (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167249) |
|
Date: August 8th, 2025 11:38 AM Author: stimulating chocolate temple this is crazy lol (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167254) |
|
Date: August 8th, 2025 11:04 AM Author: ungodly bespoke market https://www.reddit.com/r/ChatGPT/comments/1mkxefi/i_gave_chatgpt_a_simple_grade_8_geometry_problem/ (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167188) |
|
Date: August 8th, 2025 11:04 AM Author: ungodly bespoke market https://www.reddit.com/r/ChatGPT/comments/1mkxbsw/pineapple/ (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167190) |
|
Date: August 8th, 2025 11:05 AM Author: ungodly bespoke market https://www.reddit.com/r/ChatGPT/comments/1mkx7u2/example_of_incompetence_with_50/ (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167194) |
|
Date: August 8th, 2025 11:06 AM Author: ungodly bespoke market https://www.reddit.com/r/ChatGPT/comments/1mkx4yh/starting_off_strong/ (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167196) |
|
Date: August 8th, 2025 11:07 AM Author: ungodly bespoke market https://www.reddit.com/r/ChatGPT/comments/1mkx1dt/just_updated_and_wtf_is_this_completely_made_up/
these links were just within the first 30 seconds of looking at reddit's chatgpt forum |
|
Date: August 8th, 2025 11:08 AM Author: narrow-minded grizzly coffee pot preventive strike (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167203) |
|
Date: August 8th, 2025 11:09 AM Author: vibrant office I haven't used it on anything where that would matter, but thats how gemini already is because of how tight its saftey is compared to 4o. The stuff I have used 5 for has nothing to do with anything controversial but I believe you that 5 might get shitty if you are using it in areas that are controversial like anthropology or politics or something. There is an open weights API now though, just use that on a device powerful enough for it (sucks though because it still has saftey built in to an extent). What pisses me off the most is there are entire areas of inquiry that they purposely don't quantize in order to skip over something that would trigger sacred narrative upset. also, i don't think this means its overall worse than 4o. but perhaps for departing from orthodox opinion (http://www.autoadmit.com/thread.php?thread_id=5759922&forum_id=2,#49167209) |