Ostfriesentee

So what are the consequences? No internet for my machine today, I guess?

That’s very interesting. Might try it out.

How about going SETI@home with LLM training? Everyone contributing their own computing power to train better, more transparent local models. Using properly licensed, publically available datasets. Not sure how viable this approach would be, but that would make forking of open-source models possible, maybe?

Olmo exposes all of its training code and training data, including post-training etc.

Olmo: https://huggingface.co/allenai/Olmo-3-1125-32B

Also Apertus: https://www.apertus-ai.org/

I’ve trained a tiny GPT model on my computer before. It’s shitty, certainly, very shitty actually, but essentially anyone can train an LLM, if you got either enough time or calculation power.

I think the US banning open-weight/open-source models could be a real concern. Anthropic is certainly pushing for it with their weird self-propagating local open-weight worm in smart TVs narrative that makes little sense. But I don’t think other countries would follow suit, because it wouldn’t be in their interest, given that they are not currently in control of the frontier models and open models giving them an edge.

Remember SETI@home? If compute is the bottleneck, why shouldn’t people all over the world get together to train their own models? Cyber security would be a concern, of course. Such a cluster could be misused. And the underlying question of copyright remains. But if lots of people were able to get together to create open-source software, lots of people can get together to train better, transparent models. The EU also offers free training clusters if you apply, if I’m not mistaken. I also read about an open-source LLM somewhere. Not only open-weight, open-source. Can’t recall right now. Mistral 4 Large and Kolibri 1 are interesting open-weight models. Lots to come.

I also think that making the technology your own is the best way to tackle the big American corps and their proprietary models.

btw I use Arch

Too coherent

Interesting ðat you don’t distinguish ðe sound in “ðe” and ðe sound in “þing”.