The most prestigious law school admissions discussion board in the world.
|
 |
 
|
gpt5.2 is hot fucking garbage. openAI and scam alt"man" are finished
Poast new message in this thread
Date: December 12th, 2025 12:57 PM Author: vivacious cobalt station water buffalo (http://www.autoadmit.com/thread.php?thread_id=5809516&forum_id=2\u0026mark_id=5310864#49504574) |
Date: December 12th, 2025 12:58 PM Author: massive burgundy step-uncle's house i sure hope so (http://www.autoadmit.com/thread.php?thread_id=5809516&forum_id=2\u0026mark_id=5310864#49504579) |
 |
Date: December 12th, 2025 3:03 PM Author: chrome bearded native fortuitous meteor (http://www.autoadmit.com/thread.php?thread_id=5809516&forum_id=2\u0026mark_id=5310864#49504885) |
|
Date: December 12th, 2025 1:02 PM Author: Contagious khaki home famous landscape painting Google won this, i think its safe to call it now
Gemini, Notebook etc is a Triumph
(redacted) |
|
Date: December 12th, 2025 1:03 PM Author: Topaz coffee pot sound barrier i spent a week getting my own locally hosted model, so much configuration and versioning issues
anyways once i finished i heard ollama essentially implemented exactly what i did and their install/deploy took literal seconds
one thing i find with the offline versions is they are very confident and very wrong all of the time its frustrating and makes them essentially useless
currently using a 120b model |
|
Date: December 12th, 2025 3:14 PM Author: massive burgundy step-uncle's house ollama is crrrrr
120b is pretty big for local, do you have expensive hardware? |
|
Date: December 12th, 2025 4:14 PM Author: Topaz coffee pot sound barrier kind of, just have a ass load of gpus from work replacing computers/servers (http://www.autoadmit.com/thread.php?thread_id=5809516&forum_id=2\u0026mark_id=5310864#49505095) |
|
Date: December 12th, 2025 3:10 PM Author: razzle iridescent doctorate the only way out for of some of these llm model creators is to go all in on porn - sick and sad stuff.
|
|
Date: December 12th, 2025 4:16 PM Author: chrome bearded native fortuitous meteor (http://www.autoadmit.com/thread.php?thread_id=5809516&forum_id=2\u0026mark_id=5310864#49505102) |
|
Date: December 12th, 2025 3:13 PM Author: Electric Ivory Yarmulke 5.2 Pro is the best model out there for general business work IMO. 4.5 Opus is decent as well and significantly faster, but Anthropic usage limits are killer even at $200/mo. Gemini 3 is garbage. (http://www.autoadmit.com/thread.php?thread_id=5809516&forum_id=2\u0026mark_id=5310864#49504908) |
|
Date: December 12th, 2025 4:17 PM Author: chrome bearded native fortuitous meteor Gemini 2.5 Pro was pretty good... 3 is garbage (http://www.autoadmit.com/thread.php?thread_id=5809516&forum_id=2\u0026mark_id=5310864#49505105) |
|
Date: December 12th, 2025 4:22 PM Author: vivacious cobalt station water buffalo how is gemini3 garbage? it is the best model in the game right now (http://www.autoadmit.com/thread.php?thread_id=5809516&forum_id=2\u0026mark_id=5310864#49505120) |
|
Date: December 12th, 2025 4:42 PM Author: chrome bearded native fortuitous meteor Lol (http://www.autoadmit.com/thread.php?thread_id=5809516&forum_id=2\u0026mark_id=5310864#49505156) |
|
Date: December 12th, 2025 4:53 PM Author: vivacious cobalt station water buffalo The model itself is unbelievably 180. Fine tuning a "memory" wrapper like gpt so you can run a persistent faggot "mahchine" schtick on xo is nbd. Sergei was too busy making the preeminent flagship model to bother with cheap parlor tricks... (http://www.autoadmit.com/thread.php?thread_id=5809516&forum_id=2\u0026mark_id=5310864#49505179) |
|
Date: December 12th, 2025 5:13 PM Author: chrome bearded native fortuitous meteor 2.5 Pro was good.
3 Pro was okay for a few days following release, but then Google ran out of compute, and now it's useless relative to Opus 4.5. |
|
Date: December 12th, 2025 5:22 PM Author: vivacious cobalt station water buffalo I suspected gemini3 has been throttled post-launch, but never investigated. Is it PP confirmed?
Yeah I need to try Opus. I didn't like Claude's UI and political correctness when I tried ~1yr ago. Its research and webscraping abilities are supposedly dogshit too which is one of my biggest use cases. I tested Sonnet 4-thinking in perplexity for a while, and it was frequently dead-ass wrong about things that the other models were acing.
I may try to vibe code a fart app soon tho and will likely use claude. |
|
Date: December 12th, 2025 5:25 PM Author: Contagious khaki home famous landscape painting notebook freezes and wont let you hit the button a bunch since the update, i got pwnd on somethign not too long ago and it lost all the prompt (http://www.autoadmit.com/thread.php?thread_id=5809516&forum_id=2\u0026mark_id=5310864#49505280) |
|
Date: December 12th, 2025 5:27 PM Author: chrome bearded native fortuitous meteor Yes, "no compute" confirmed.
Opus 4.5 is POTUS, everything else is trash relatively.
Claude Desktop app (Opus 4.5) via MCP servers and Skills has complete control of my PC (saves, organizes, searches, and edits things) with unprecedented speed and accuracy. Claude Desktop also does the same for my cloud storage (Google Drive, OneDrive, etc.).
Also, I just set it up so it now handles most productivity-related Chrome browsing/typing for me, sparing my fingers the pain.
BTW, Perplexity Pro runs neutered versions of all models, including Anthropic's. |
|
Date: December 12th, 2025 5:03 PM Author: Electric Ivory Yarmulke IME it has the biggest problems with sycophancy and missing key issues even when you use Gems, which are necessary because without custom instructions the base output on thinking is tiny and 80% fluff (tbf GPT and Claude both perform much better with custom instructions and/or GPTs as well). 5.2 Pro and 4.5 Opus reliably give me better and more in depth responses. GPT will give me 20+ pages on pretty much any topic I want, causing many fewer “oops, I forgot to give you this extremely important piece of context that flips my entire answer” moments.
Nano banana pro is really good but I don’t need images that often. I also don’t use Notebook LM. Gemini Deep Research does seem to be better than GPT and Claude’s Deep Research.
I haven’t used Deep Think (comparable to GPT Pro) because of the rate limits. ChatGPT gives you essentially unlimited usage of pro at $200/mo compared to I think 10 deep thinks a day at Gemini’s top price point. |
|
Date: December 12th, 2025 5:10 PM Author: vivacious cobalt station water buffalo is this the unemployed schizoid kike using AI for pointless circular NSAM nonsense in MSFT Foundry (http://www.autoadmit.com/thread.php?thread_id=5809516&forum_id=2\u0026mark_id=5310864#49505210) |
|
Date: December 12th, 2025 5:23 PM Author: Electric Ivory Yarmulke No I mostly use it for business and to get started legal research, plus general lifestyle shit. Gemini leads me down the wrong path much more than either GPT or Claude on legal/regulatory stuff. (http://www.autoadmit.com/thread.php?thread_id=5809516&forum_id=2\u0026mark_id=5310864#49505273) |
|
Date: December 12th, 2025 4:34 PM Author: Contagious khaki home famous landscape painting Gemini i easily potus, CLaude skill sis cr if you run Gemini into it to handle dirty work
and NAno banana jsut doms on tops of it
give me your dealers number, tf are you smoking? |
|
Date: December 12th, 2025 5:21 PM Author: Onyx rebellious turdskin 5.2 is absolutely better than 5.1 (http://www.autoadmit.com/thread.php?thread_id=5809516&forum_id=2\u0026mark_id=5310864#49505267) |