The most prestigious law school admissions discussion board in the world.
|
 |
 
|
OpenAI is crushing Grok and others in the LLM market
Poast new message in this thread
Date: April 24th, 2025 10:06 AM Author: ,.,.,.,.,.,.,..,:,,:,,.,:::,.,,.,:.,,.:.,:.,:.::,. Number of paid subscribers according to Grok:
ChatGPT - 15 million Claude - 2 million Gemini - 1 million Baidu - 500k Grok - 100k Llama - 50k Deepseek - 30k
. . Active monthly users:
ChatGPT - 600 million Claude - 50 million Gemini - 350 million Baidu - 20 million Grok - 10 million Llama - 10 million Deepseek - 20 million |
Date: April 24th, 2025 10:09 AM Author: Cyberpunk Age Pervert OpenAI and Anthropic are ahead of everyone else. Google has infinite resources but their models just aren't as smart. (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876648) |
 |
Date: April 24th, 2025 10:10 AM Author: ,.,.,.,.,.,.,..,:,,:,,.,:::,.,,.,:.,,.:.,:.,:.::,. People are saying the new Gemini is really good but I haven’t used it (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876651) |
|
Date: April 24th, 2025 10:14 AM Author: scholarship It did recently get one question right about a new law that all the others I asked got wrong but their free version is so stingy I didn’t get to test it more than 3 prompts. (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876667) |
|
Date: April 24th, 2025 10:43 AM Author: aqua jeet (🧐) gemini pro 2.5 is considered by many to be the most well rounded model. it's great for coding and has the longest context window
AI studio is also the best user interface |
|
Date: April 24th, 2025 10:57 AM Author: Cyberpunk Age Pervert It is a complete Joke compared to o3 and isn't really better than Claude 3.7 except for long contexts (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876797) |
|
Date: April 24th, 2025 11:18 AM Author: aqua jeet (🧐) claude 3.7 is horrible, it will often make wild assumptions and comically overengineer things
and "except for long contexts" is pretty huge. it's not just that gemini has a larger context window, it's that it is currently the only model which doesn't suffer extreme degradation when you actually utilize a very large context. every other model, even if technically supports a 500k/1m context window or whatever, will start choking on its vomit well before you approach the limit. |
|
Date: April 24th, 2025 11:20 AM Author: Cyberpunk Age Pervert I've used it for coding a lot and think it's great, this is just something people started saying on X. Gemini is also good at code, but neither is even close to as smart as o3. o3 honestly feels close to AGI. (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876873) |
|
Date: April 24th, 2025 11:21 AM Author: aqua jeet (🧐) I've had claude code hallucinate an entire library's APIs. "Hey man this sample looks great but where is the library that actually lets you do this?" "Oh oops lol sorry that doesn't actually exist I made it up out of thin air based on an analagous library in Java." It's not a meme, it fucking sucks ass if you actually even halfway know how to code. (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876876) |
|
Date: April 24th, 2025 11:25 AM Author: Cyberpunk Age Pervert I can't speak to your personal experiences, but it's used by actual developers far more than Gemini. You can see this quite clearly in the API usage on OpenRouter. Claude Code is a great tool as well. (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876888) |
|
Date: April 24th, 2025 11:21 AM Author: .....,,,,,...,,,,,..,.,,,,,.,,. Gemini is currently at top of the lmsys leaderboard: https://lmarena.ai/leaderboard (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876878) |
|
Date: April 24th, 2025 11:23 AM Author: Cyberpunk Age Pervert Community voted shit is fake and gay. Gemini does do well in many benchmarks but this isn't the sort of evidence I would use. It's better to look at hard benchmarks and the use on sites like OpenRouter excluding models auto-added. (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876884) |
|
Date: April 24th, 2025 5:13 PM Author: ,.,.,.,....,.,..,.,.,. o3/o4 mini and 2.5 pro cluster pretty closely together on most SWE benchmarks (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48877841) |
|
Date: April 24th, 2025 10:02 PM Author: Cyberpunk Age Pervert Having used both Gemini 2.5 and o3 a lot there's no comparison between them, o3 is far smarter. Benchmarks are sort of useful but can't tell the whole story. o3 is clearly ahead in them though (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48878629) |
|
Date: April 25th, 2025 12:24 AM Author: Mainlining the Secret Truth of the Mahchine (You = Privy to The Great Becumming™ & Yet You Recognize Nothing) (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48878899) |
|
Date: April 24th, 2025 9:20 PM Author: https://imgur.com/a/o2g8xYK The offline variant is tits (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48878520) |
|
Date: April 24th, 2025 9:21 PM Author: Pierbattista Pizzaballa I was using Meta today and it was straight up retarded. Redid the conversation with Claude and it was miles better. (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48878525) |
|
Date: April 24th, 2025 10:10 AM Author: scholarship Surprised claude has so many. Claude is the only one I pay for but it’s more about text and code and it doesn’t really do images at all. Gpt makes sense because it’s so much better at images than everyone and kids can pick up a sub from the iphone app when they really need to tweak a meme (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876652) |
|
Date: April 24th, 2025 10:15 AM Author: UhOh and turn their friends into porn (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876671) |
|
Date: April 24th, 2025 10:17 AM Author: scholarship I heard from some parents with kids in my neighborhood that one of their friends got suspended for distributing AI nudes of the girls in school. Apparently this is really common now. (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876684) |
|
Date: April 24th, 2025 10:15 AM Author: Bar Mitzvah Money CLAUDE is 180 as a data analyst but they 100% steal everything you put in (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876672) |
|
Date: April 24th, 2025 10:11 AM Author: peter thiel thoughtfully nodding and stuttering the newest claude and gemini models are both good and way more cost effective. they are going to eat into openAI's market share over time (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876654) |
|
Date: April 24th, 2025 5:03 PM Author: Bar Mitzvah Money Claude has no space and expensive if you need it for something legit with like years of data ime
|
|
Date: April 24th, 2025 10:18 AM Author: internet g0y how did all these companies come out at around the same time
are they all using govt funded public research or something?
ChatGPT came out a couple years ago then every other company was like 'yea we are basically there too heres our product' |
|
Date: April 24th, 2025 10:28 AM Author: ,.,.,.,.,.,.,..,:,,:,,.,:::,.,,.,:.,,.:.,:.,:.::,. the method they used (transformers + a ton of compute) wasn’t a secret and was published for everyone to read back in 2017. Open AI was just the first company to take it seriously and try it. in fact the architecture was first conceived of at Google, but Geoffrey Hinton was running AI at Google at the time and he thought transformers were a dead end, so it want pursued. Once OpenAI proved that they weren’t everyone started doing transformers. (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876710) |
|
Date: April 24th, 2025 10:40 AM Author: .,..,.,.;,.,,:,.,.,::,...,..,:,..;,.., lol at you pedo internet addicted dorks who care about this shit. it's all just repackaged google searches/wikipedia crap. (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876755) |
|
Date: April 24th, 2025 11:19 AM Author: aqua jeet (🧐) (retard lawyer pumo) (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876871) |
|
Date: April 24th, 2025 11:20 AM Author: I'm white (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876875) |
|
Date: April 24th, 2025 4:55 PM Author: Ass Sunstein It's all based on algorithms developed by Myspace Tom. (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48877774) |
|
Date: April 24th, 2025 4:58 PM Author: metaphysics is fallow i heard google kept it internal and was using it in search because of the ethical issues of making it available to prompt. then once openAI made it public, basically not caring about the ethical side, everyone had to do it. thats why you saw goofy ethics shit from google at first like black nazis or whatever (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48877790) |
|
Date: April 24th, 2025 6:53 PM Author: tsar booba It eats into Google Ad revenue, so Google didn’t really want to release it until they had to (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48878054) |
|
Date: April 24th, 2025 11:10 PM Author: bowlcut template Cr. Google needs intentionally shitty search to steer people into ads. AI fucks that up completely. So google’s search “AI” is modified to be completely garbage (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48878799) |
|
Date: April 24th, 2025 9:14 PM Author: ,.,.,.,....,.,..,.,.,. Geoff Hinton wasn’t running Google AI and never really said anything indicating transformers were a dead end. He spent a lot of his career trying to find a better algorithm than back propagation for training them, but that’s different. The continued success of backprop caused him to reverse his opinion on it and think it’s potentially better than what the brain does since chatgpt knows orders of magnitude more than any human despite having a small number of parameters (relative to the brain).
The continued success of dumb algorithms has been a source of surprise for most ML researchers |
|
Date: April 24th, 2025 10:36 AM Author: ,.,,.,.,,,,,,..................... Where is NYU's tax school on that list? (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876739) |
|
Date: April 24th, 2025 10:40 AM Author: The Mercantilist Policy April (No Future) (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876758) |
|
Date: April 24th, 2025 5:04 PM Author: Bar Mitzvah Money (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48877810) |
|
Date: April 24th, 2025 10:42 AM Author: scholarship (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876759) |
|
Date: April 24th, 2025 5:04 PM Author: Bar Mitzvah Money (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48877811) |
|
Date: April 24th, 2025 10:40 AM Author: shitlib shibboleth Grok is pretty good even in the free version. I switched from GPT because it was too PC and wouldn’t always give you a clear answer. Grok is better at walking you through the relevant information if it can’t find an answer. (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876754) |
|
Date: April 24th, 2025 10:57 AM Author: ,.,.,.,.,.,.,..,:,,:,,.,:::,.,,.,:.,,.:.,:.,:.::,. They’re all commoditized at this point and largely have the same capabilities for 90% of users. But OpenAI got first mover advantage which is everything in tech (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876794) |
|
Date: April 25th, 2025 2:26 AM Author: ,.,.,.,........,....,,,.. You’d have to be crazy to use Grok for serious work applications if only due to the stigma of it being associated with X/Elon/instability. At least if there’s some big data hack at ChatGPT it won’t look as bad that you were using it. (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48879065) |
|
Date: April 24th, 2025 11:24 AM Author: AZNgirl Nominating Her Jewish Handsome as New Pope ChatGPT has gotten a lot better, it doesnt make as many stupid mistakes and its easier to use than Grok which talks too much or DeepSeek which doesnt have updated data. I wld prefer to use gorgeous Chink product but somehow this fag Altman is a bit ahead right now at least for my queries like "how big is a lizards pussy" etc (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48876885) |
|
Date: April 24th, 2025 4:53 PM Author: ,.,,.,.,,,,,,..................... lol (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48877769) |
|
Date: April 24th, 2025 4:56 PM Author: .,.,...,..,.,.,:,,:,.,.,:::,...,:,...:..:.,:.::,. Grok is always going to kill ChatGPT and Gemini because both are aged. ChatGPT is aged what, 18 months? I recently asked grok for the Pope's cause of death the day after he died and got a great and comprehensive answer. (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48877778) |
|
Date: April 24th, 2025 4:57 PM Author: peter thiel thoughtfully nodding and stuttering yeah but its actual performance sucks ass because it's literally made by indians (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48877785) |
|
Date: April 24th, 2025 5:00 PM Author: .,.,...,..,.,.,:,,:,.,.,:::,...,:,...:..:.,:.::,. DESCRIBE what you mean by "actual performance sucks ass." Different LLMs have different pros and cons but I just listed a way in which grok is superior to all the others.
By the way, all of them are coded by Indians except for DeepSeek which is coded by Chinese people. |
|
Date: April 24th, 2025 5:05 PM Author: metaphysics is fallow is grok doing constant training? i think they all stay up to date by googling for info (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48877815) |
|
Date: April 24th, 2025 5:12 PM Author: peter thiel thoughtfully nodding and stuttering they do, it's just "RAG" off google or twitter posts. this guy doesn't know what he's talking about (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48877838) |
|
Date: April 24th, 2025 5:13 PM Author: metaphysics is fallow yeah they all do that (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48877842) |
|
Date: April 24th, 2025 6:11 PM Author: .,.,...,..,.,.,:,,:,.,.,:::,...,:,...:..:.,:.::,. Grok - yes. It is constantly working off of X and search engine content.
ChatGPT - it does it in waves but is generally 18 months behind.
Gemini is less specific about the aging but it is definitely aged and doesn't have open access to the internet.
DeepSeek - I just asked it and it said it is only current up to stuff that happened in or before July 2024.
This is a massive advantage for grok. |
|
Date: April 24th, 2025 6:11 PM Author: metaphysics is fallow ur dumb (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48877969) |
|
Date: April 24th, 2025 6:12 PM Author: .,.,...,..,.,.,:,,:,.,.,:::,...,:,...:..:.,:.::,. Which part confuses you? (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48877970) |
|
Date: April 24th, 2025 6:13 PM Author: metaphysics is fallow lmao (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48877973) |
|
Date: April 24th, 2025 4:55 PM Author: .,.,...,..,.,.,:,,:,.,.,:::,...,:,...:..:.,:.::,. Are they not counting a paid X sub which includes paid Grok as a paid sub? (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48877773) |
|
Date: April 24th, 2025 5:01 PM Author: .;:..;:.;.:.;.,,,..,.:,.;....;,;;;..;,..,,.,,...., Cq. Number seems way too low (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48877799) |
|
Date: April 24th, 2025 6:07 PM Author: .,.,...,..,.,.,:,,:,.,.,:::,...,:,...:..:.,:.::,. Exactly. I also know several people IRL that have premium subs because they're X subscribers. (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48877958) |
|
Date: April 24th, 2025 5:33 PM Author: ,.,.,.,.,.,.,..,:,,:,,.,:::,.,,.,:.,,.:.,:.,:.::,. *farts cum* (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48877887) |
|
Date: April 25th, 2025 2:27 AM Author: Mainlining the Secret Truth of the Mahchine (You = Privy to The Great Becumming™ & Yet You Recognize Nothing) SuperGrok is 180. (ChatGPT Plus and Claude Pro are also essential) (http://www.autoadmit.com/thread.php?thread_id=5715526&forum_id=2#48879068) |