I usually don’t even understand the lingo they use. “Open-weighted” is the most recent one, then it usually goes down to specific “models” that everybody is supposed to know about.

These are my thoughts (I will stick to the vague “it” for now, but of course therein lies another question: “and how does all this apply to various specialised AIs”):

  • Is it really feasible to run it 100% locally? I know there’s plenty of people with very powerful rigs indeed, but still. Or are 99% of these people really saying “it would, in theory, be possible to run that locally, therefore your concerns are invalid”?
  • If yes to the previous: the software doesn’t come from nowhere and ultimately still relies on gas-turbine-powered datacenters and stolen IP and stolen personal data, no?

If what I wrote above is true, what exactly are people arguing when they say it’s still possible to use LLMs ethically or true to FOSS philosophy, because … ???

edit

Thanks to all who answered.

I guess it’s my fault for asking several questions in one, but this thread has attracted exactly the type of people I’m writing about; several even used the term “open-weighted models” without explaining it.

Asking to get arguments explained, I got more arguments instead.

  • A_norny_mousse@piefed.socialOPEnglish
    2·
    22 hours ago

    Just as I suspected…

    Thanks for taking the time.

    there are very few that can be considered ethical - Olmo, Apertus, Starcoder

    This is software meant to be run always and completely locally?

    Sorry to whine, but so far nobody has eli5’d what “open-weighted” means, or “model” at that… please?

    • iForgotSpells@sopuli.xyz
      1·
      16 hours ago

      This is software meant to be run always and completely locally?

      That’s what is claimed, I haven’t run them locally since I don’t have a good system.

      To be honest, I’m not sure if I can eli5 weights and models, but I’ll try. Think of a model like the base - for example, OpenAI has different models like Astra, Sol, etc. These are different models, like different versions of a software or operating system like macOS, but for AI stuff. Like one would download a software, you download a model to perform tasks.

      Weights are vales that can influence inputs of these models to get a desired/better result. What most of these models are doing is mostly predicting what might be the next appropriate text/data to the question you asked. When you ask these AI models what 2+2 is, it is not performing a math operation like a normal program, it is looking at its training data to see what the closest option might be. It is doing pattern matching.

      These AI models inside can be thought of like an interconnected network, like neurons in our body, that keep passing information to the next neuron and to the brain to make a decision. (Before understanding LLMs it would help to understand Neural Networks first). These AI networks need weights and biases. These networks perform calculations and weights are used to determine how much importance/weight each input can have on the output. Bias on the other hand, is used to shift/change the output so the AI model can ‘learn’ to pattern match better.

      What open-weight models, do is they make the model available for download along with the weights. No information is given on training data. Like with ads, ones with most data emerges victorious i.e, has a better model. So these companies do theft, don’t list their training data afraid of getting caught. I forgot which one, but either Deepseek or Qwen (both open-weight) was caught ‘stealing’ from Claude (not open weight). You can probably guess how much these companies value ethics.

      I’m not sure if this entire thing goes away, but local models might be the ones left standing when this bubble pops.

      Open weight is different to open source. Open Source AI as it stands, the definition requires a model to have entire thing made public - so the weights, biases, training data used, the model. Apertus, Olmo etc are mostly meeting open source AI definition.

      If you need to know more, this is what we’d use to refresh our memory before exams :)

      I probably might have made mistakes here, English isn’t my first language either. But I hope you get an idea about these terms

      If you really need to understand this tech more, I recommend watching ‘AI for Everyone’ course on Coursera from Andrew Ng. It is free to audit, my friends who took his course were hyped (I wasn’t really interested in AI)

      • A_norny_mousse@piefed.socialOPEnglish
        1·
        7 hours ago

        Thanks a lot.

        I did not know it was possible to influence AI software/models and nudge them in a certain direction.

        These AI companies have even more power than I thought for a long time.