The most prestigious law school admissions discussion board in the world.
|
 |
    6 posters qualify.
You have 0 points. Copying this list does not send funds. Paste it on Disperse, then connect your wallet there. |
The Penis is right. Chinese LLMs suck.
Poast new message in this thread
Date: November 6th, 2026 11:38 PM Author: Historical American (SQQQ) I went to DeepSeek one time and it was a bunch of unusable chinky garbage.
But I guess the AI gods at Anthropic and OpenAI won't let you run frontier models locally on your system huh? So you gotta do OpenClaw or the Chinaman dance? |
₿ Bitcoin
bc1qfk03xf3zekand03r6xut652mq049duj7yhw3rt Date: November 6th, 2026 11:49 PM Author: The Penis noone is running full frontier models locally on their computer anyway. even with a maxxed out 128 gb unified memory 16 inch m5 macbook pro, even at 2 bit quantization you aren't fitting that or running that on there. with ssd streaming you can do a little better, but still people are running the smaller versions locally. even those specialized nvidia dgx systems they sell now can't fit an entire frontier model. local frontier Chinese llms are for businesses with h100s and whatever to run them on.
agree I deep seek. still garbage every time I go to use it. like insanely bad. |
 |
Date: November 6th, 2026 11:55 PM Author: Historical American (SQQQ) https://youtu.be/qw4fDU18RcU
How about this? Big of small? 120 billion parameters? |
|
₿ Bitcoin
bc1qfk03xf3zekand03r6xut652mq049duj7yhw3rt Date: November 6th, 2026 11:58 PM Author: The Penis 180 setup. we should get one of these and run a dark net crime ring. (https://xoxohth.com/thread.php?thread_id=5919821&forum_id=2#50239756) |
|
Date: November 7th, 2026 12:01 AM Author: Historical American (SQQQ) How would that even work? (https://xoxohth.com/thread.php?thread_id=5919821&forum_id=2#50239762) |
|
₿ Bitcoin
bc1qfk03xf3zekand03r6xut652mq049duj7yhw3rt Date: November 7th, 2026 12:04 AM Author: The Penis you have a model not governed and monitored by open ai or anthropic running local on that thing, preferably an ablated heretic model, and can just do whatever the hell you want with it. stuff that would be way harder and slower if you didn't have ai. steal accounts, data, run darknet marketplaces. whatever you want. make easy millions. hypothetically of course, haha. (https://xoxohth.com/thread.php?thread_id=5919821&forum_id=2#50239775) |
|
Date: November 7th, 2026 12:06 AM Author: Historical American (SQQQ) ablated heretic model tp (https://xoxohth.com/thread.php?thread_id=5919821&forum_id=2#50239784) |
|
Date: November 7th, 2026 12:40 AM Author: json statham lol what does their website have to do with LLM
And why would you be using a Chinese inference engine or implementation for any open weight?
Let's start with how much RAM do you have, and what kind of agent do you want to build?
A 3-year old M2 64gb can easily run Qwen3.8-Flash-Next IQ3_XXS which benchmarks around GPT 5.6 and beats Sonnet 5 on Max
Lmao at "durrr can't do anything" with unlimited untethered no-guardrails GPT 5.6 level AI
|
|
Date: November 7th, 2026 12:47 AM Author: .Nicholas J. Fuentes A 64gb m2 costs how much
Also isn’t it the case Mac’s are dogshit compared to equivalent ram Nvidias on win machines |
|
₿ Bitcoin
bc1qfk03xf3zekand03r6xut652mq049duj7yhw3rt Date: November 7th, 2026 12:59 AM Author: The Penis Macs are arguably the best for running local models because you can run huge models due to the integrated memory. But it’s not as fast as Nvidia. (https://xoxohth.com/thread.php?thread_id=5919821&forum_id=2#50239876) |
|
₿ Bitcoin
bc1qfk03xf3zekand03r6xut652mq049duj7yhw3rt Date: November 7th, 2026 12:53 AM Author: The Penis That shit is not gpt 5.6 level (https://xoxohth.com/thread.php?thread_id=5919821&forum_id=2#50239871) |
|
Date: November 7th, 2026 12:58 AM Author: Historical American (SQQQ) (https://xoxohth.com/thread.php?thread_id=5919821&forum_id=2#50239875) |
|
₿ Bitcoin
bc1qfk03xf3zekand03r6xut652mq049duj7yhw3rt Date: November 7th, 2026 1:04 AM Author: The Penis Assuming you can even get that full thing working on that mac, which is plausible but I’d like to see it, not even on the benchmarks does that thing hit gpt 5.6 level unless you are comparing it to the low reasoning levels. But the benchmarks aren’t even a perfect measure of ability. You can score well on discrete tasks and overall suck at real world problem solving. Even the best open source Chinese models are ass. I tested them plenty through the APIs on azure and aws. Haven’t tried every single one but the ones I tried weren’t good. They are like gpt o1 level. There is more to general cognitive ability than those benchmarks. Claude has some of the highest benchmarks and open ai is blowing them away in terms of real open reaearch problems solved. (https://xoxohth.com/thread.php?thread_id=5919821&forum_id=2#50239887) |
|
Date: November 7th, 2026 1:07 AM Author: Emotionally + Physically Abusive Ex-Husband (https://xoxohth.com/thread.php?thread_id=5919821&forum_id=2#50239893) |
|
Date: November 7th, 2026 1:09 AM Author: AZNgirl asking Democrat Handsome to Pound Her Next DS doesnt suck benchod, i used it to program shit and it was good and doesnt have kike strict limits like gpt and claude and shitz
only thing its retarded making a file it will show u the entire generation and then have a CONTINUE button when its too long u have to keep pressing continue
other AIs just make the file in background and u download it
CHINA is 180 and u benchods shld not disparage china saars, they are using AI for actual good like manufacturing not for making dildo pics of animals being raped |
|
₿ Bitcoin
bc1qfk03xf3zekand03r6xut652mq049duj7yhw3rt Date: November 7th, 2026 1:12 AM Author: The Penis You could program shit with gpt 3, it doesnt mean it was good or reliable or you could do serious stuff. The levels go really high. Even Astra ultra and opus 5.5 and Fsble still get stuck on hard problems. (https://xoxohth.com/thread.php?thread_id=5919821&forum_id=2#50239900) |
|
Date: November 7th, 2026 1:38 AM Author: AZNgirl asking Democrat Handsome to Pound Her Next i use the free cpt but it has pretty low limits and then u cannot upload shit it even does this kike shit sometimes where it edits a program then u ask for more edits and it tells u to upload the file again, i think they just want u to use up alll ur uploads
and it does that annoying new chat bullshit
im not gonna pay a cent for shit, im jeet |
|
₿ Bitcoin
bc1qfk03xf3zekand03r6xut652mq049duj7yhw3rt Date: November 7th, 2026 1:42 AM Author: The Penis It’s not actually that bad I’m just being a benchod, but you can’t really use the chat apps for serious coding anymore you gotta use the api and run it through a harness. No one has time for glue code anymore. (https://xoxohth.com/thread.php?thread_id=5919821&forum_id=2#50239944) |