If the model is adequately FOSS, i.e. open weights, and can run on a single consumer GPU (or NPU), and the “author” (quotations because I’m personally undecided if one can claim code generated by an aforementioned model is theirs) understands it, then I really don’t see what the problem is…
Personally I have nothing against SLMs/LLMs as a technologically, to me my grievances against ChatGPT or Claude are mostly about their environmental impacts and selling us back our own art, also keeping knowledge behind a for-profit black-box - if those aren’t appropriate for a specific model, then I say using that model is fair, and good for productivity.
I’d take a model whose training data is open source and legitimately obtained. The only ones I know about are Apertus and OLMo, and they aren’t really competitive.
Unfortunately not yet, no, true FOSS models are likely many years away, but I would argue that that follows typical FOSS lifecycles. Emerging technology is typically outperformed by proprietary endeavors, which creates an audience, and then that audience undertakes a FOSS implementation that initially underperforms, then just about competes, then eventually overtakes (i.e, GNU/Linux).
The problem with open weight models is that they are still hard to deploy. For DeepSeek for example you require NVIDIA HGX B200, these things are expensive.
Yes, unfortunately they are, and I do think efficiency is going to be a significant research front for open-weight models. The nature of this topic is highly speculative as our compute capabilities have only recently reached what is required to consider running generative AI models, what we have today are very crude first implementations of what I personally believe will become an everyday tool for developers, and more.
And we have seen this, there are models now capable of running on an individual’s hardware (and not particularly expensive hardware either) that can outperform what ChatGPT initially launched with.
If the model is adequately FOSS, i.e. open weights, and can run on a single consumer GPU (or NPU), and the “author” (quotations because I’m personally undecided if one can claim code generated by an aforementioned model is theirs) understands it, then I really don’t see what the problem is…
Personally I have nothing against SLMs/LLMs as a technologically, to me my grievances against ChatGPT or Claude are mostly about their environmental impacts and selling us back our own art, also keeping knowledge behind a for-profit black-box - if those aren’t appropriate for a specific model, then I say using that model is fair, and good for productivity.
I’d take a model whose training data is open source and legitimately obtained. The only ones I know about are Apertus and OLMo, and they aren’t really competitive.
Unfortunately not yet, no, true FOSS models are likely many years away, but I would argue that that follows typical FOSS lifecycles. Emerging technology is typically outperformed by proprietary endeavors, which creates an audience, and then that audience undertakes a FOSS implementation that initially underperforms, then just about competes, then eventually overtakes (i.e, GNU/Linux).
The problem with open weight models is that they are still hard to deploy. For DeepSeek for example you require NVIDIA HGX B200, these things are expensive.
Yes, unfortunately they are, and I do think efficiency is going to be a significant research front for open-weight models. The nature of this topic is highly speculative as our compute capabilities have only recently reached what is required to consider running generative AI models, what we have today are very crude first implementations of what I personally believe will become an everyday tool for developers, and more.
And we have seen this, there are models now capable of running on an individual’s hardware (and not particularly expensive hardware either) that can outperform what ChatGPT initially launched with.