News for hackers. Stuff that matters.
Re:A Competitor Crippled a $23.5M Bootcamp by Becoming a Red... by BoorishBears on Thursday October 09, 2025 @04:02AM (#45524854)
> Yesterday Huggingface Deleted the account of a talented User @BasedBase which was creative in open weights (threatening the big techs).

I looked it up it found a pretty clear reason they'd delete it themselves?

https://www.reddit.com/r/LocalLLaMA/comments/1o0st2o/basedba...

Re:A Competitor Crippled a $23.5M Bootcamp by Becoming a Red... by trilogic on Thursday October 09, 2025 @04:07AM (#45524885)
That´s is incorrect, I demonstrated how all that is not true. The models were 100% distill and very powerful. Hard facts

https://vimeo.com/1125658282

https://vimeo.com/1125657672

https://www.reddit.com/r/LocalLLaMA/comments/1o1m9nz/the_mos...

Re:A Competitor Crippled a $23.5M Bootcamp by Becoming a Red... by BoredPositron on Thursday October 09, 2025 @04:13AM (#45524921)
The facts are not very hard...
Re:A Competitor Crippled a $23.5M Bootcamp by Becoming a Red... by trilogic on Thursday October 09, 2025 @04:24AM (#45524988)
What do you mean the facts are not very hard. If you understand just a bit of LLM you will clearly identify a Distill LLM. All the point is to delete the user because he had created some incredible models beating The best proprietary models out there. The pretext was that he was a fraud and his models were not a distill but a copy. Therefor I show clearly that he was framed and his work deserves credit. I still have the models and scripts for all the procedure. I am not interested to argue further, but it is clearly visible to me. The post in HN also is well voted and most importantly commented, meaning that users have a lot to say about it.
Re:A Competitor Crippled a $23.5M Bootcamp by Becoming a Red... by BoredPositron on Thursday October 09, 2025 @04:36AM (#45525064)
Stop gaslighting yourself via llms. You are pretty close to a delusion. I looked at the history of those supposed super models. It's literally all made up vibe llm bullshit and the people that told you so on reddit and huggingface took the time to actual proof it to you. You might be the first example of AI psychosis that I've seen in the wild. Wild.
Re:A Competitor Crippled a $23.5M Bootcamp by Becoming a Red... by trilogic on Thursday October 09, 2025 @05:28AM (#45525393)
You don´t have any idea of what you talking about or you are part or the Agenda. >I looked at the history of those supposed super models. It's literally all made up vibe llm bullshit The LLM/AI world is the trillion USD fastest growing market in the world. Who will win this competition will literally control the narrative (it can even modify history). It will thrive economically and be the next Rome. So you think that a LLM that is free for everyone and beat proprietary models is vibe BS. Beside that you are only saying nonsense blah-blah while we test and work with models everyday in real use cases, in science and technology, your best ability seem to be depressing users (I laugh at it:)

Edit: do not forget the economic damage that Deepseek ( just a LLM model) did months ago. 1 trillion usd wiped in wall street. Yes a LLM is a big deal.

Re:A Competitor Crippled a $23.5M Bootcamp by Becoming a Red... by BoredPositron on Thursday October 09, 2025 @05:39AM (#45525467)
Well now I am sure of it. Rip your brain dude. :(
Re:A Competitor Crippled a $23.5M Bootcamp by Becoming a Red... by BoorishBears on Thursday October 09, 2025 @04:12AM (#45524919)
> Sorry, this post has been removed by the moderators of r/LocalLLaMA.
Re:A Competitor Crippled a $23.5M Bootcamp by Becoming a Red... by trilogic on Thursday October 09, 2025 @04:18AM (#45524957)
Ofc everything related to it will be deleted. Now they deleted my post as well, this is just getting out of hand.
Re:A Competitor Crippled a $23.5M Bootcamp by Becoming a Red... by trissi1996 on Thursday October 09, 2025 @04:20AM (#45524967)
You're just wrong.

Vibes are not a meaningful comparison.

You can't compare different quantizations(Q4_K_M & Q4_K_XL) and also use a temp of 0.7 and use that for any meaningful comparison.

All difference you notice could be either due to temp randomness or quantization differences even if the original weights are exactly the same.

Also these were likely sourced differently so the quantizers could've used different quantization importance matrizes or process.

The weights are the same, the weights are all that matters.

Re:A Competitor Crippled a $23.5M Bootcamp by Becoming a Red... by trilogic on Thursday October 09, 2025 @04:40AM (#45525082)
Nonsense, there is a big difference of tokens generated in the output. I know my stuff, and I can recognize Deepseek reasoning and Qwen thinking. Also I explained how agentic is working in the models making it very different. The quantization is irrelevant to tokens generated and agentic behavior in a model. Say what you want, you or anyone else can´t change the facts.