[Flavius]

Open Source isn't a tech stack or a specific way of typing syntax, it’s an ideology. It’s the belief that knowledge and tools should be free to share, study and modify. You cannot kill an idea. Whether I write a function by hand or 'vibe' it into existence with an LLM, the act of liberating that code for others to use remains the same.

[AlexandrB]

What's not the same is that the LLMs used to create the code are highly centralized and controlled. I suspect it's only a matter of time until the content industries start trying to restrict what code LLMs are allowed to produce so that you can't use an LLM to bypass DRM.

[georgemcbay]

> I suspect it's only a matter of time until the content industries start trying to restrict what code LLMs are allowed to produce so that you can't use an LLM to bypass DRM.

I don't think this is a possibility anymore for multiple reasons. As others have already pointed out there are already "open models" available to use and that genie can't be put back in the bottle, restricting the commercial models wouldn't fix the issue.

And secondly, I think the state of commercial LLMs show that the big tech companies behind LLMs have already become far more politically powerful than the traditional content industries. (I don't think this is good thing, but I think it is a thing).

If you had explained the LLM situation to 15-years-ago me in terms of how they are trained (on almost entirely copyrighted material) and what kind of output they could generate and told me Disney hadn't managed (or really even tried) to sue various players out of existence I wouldn't have believed it, yet here we are.

[AlexandrB]

I hope you're right, but the content industry is politically savvy and very persistent.

[charcircuit]

There are competent open source LLMs out today. They are not highly centralized.

[pocksuppet]

There's one at the top of Hacker News right now, Qwen3-Coder-Next: https://news.ycombinator.com/item?id=46872706

[int_19h]

A 80B MoE model with 3B params per activation is not a competent model regardless of what their cherry-picked benchmarks say. This reminds me of back when every other llama-7b finetune was claiming to be "GPT-4 quality".