Comment on Voxtral Transcribe 2: transcribes at the speed of sound

• Tetsuo@jlai.lu ⁨1⁩ ⁨day⁩ ago

  My understanding was that the weights were the “essence” of the training.

  I think it’s a bit misleading to present them as “executables”.

  I agree that open weights is a bit of marketing mumbo jumbo but I wouldn’t say they are akin to a closed source binary.

  That being said I was just reading an article about LLM sleeper agents and their trigger words… So you can hide stuff on training and it’s fairly hard to spot with just the weights.

  But again it’s not really like a black box executable. And I’ve seen many great models that successfully builds on top of an open weights model.

  In the end I much prefer open weights to the very popular “‘’'openAI”“” that have opaque training and weights…

  source

  • XLE@piefed.social ⁨1⁩ ⁨day⁩ ago

    For the purpose of simplification, calling it a closed as an executable is close enough. Or a closed-source freeware ROM that you can download and run on an emulator (since you can just download models and run them via ollama or something similar). Or a closed-source game that supports modding and extension like Minecraft. Or a closed-source DLL with documentation…

    Anyway, the point is, it’s closed. If it’s not closed source, I’d beg you to link the source, both code and data, that compiles to the output.

    source

    • Tetsuo@jlai.lu ⁨1⁩ ⁨day⁩ ago

      But a modeled isn’t “compiled”.

      The weights are fully readable. Every single one of them.

      In a binary you have to use special software to get to the source code. The weights are the source and can be freely used to create a model with. The weights are used “as is” no transformation has to be done like when reversing a closed source binary that could also use obfuscation to make it more difficult.

      That’s why I would like to insist that open weights are not like a binary as they are usable as is essentially. When I use a model like that my computer is no executing the weights like instructions. They are aptly named “weights” for a reason and are a mere reflection of what the model learned through training.

      ---

      Disclaimer, below this is the explanation between a closed source binary and open weights by an LLM :

      A closed-source binary is like getting a sealed, opaque machine. You push a button, it gives an output. You have no idea how it works inside.

      Open weights are like getting the complete, labeled blueprints (architecture) and the exact, measured specifications for every single component—every gear, wire, and circuit (weights). You don’t have the designer’s engineering notebooks (training data/recipe) or the factory that built it (training compute), but you can absolutely build the machine yourself, measure every part, swap out components, and try to improve it.

      source

      • XLE@piefed.social ⁨1⁩ ⁨day⁩ ago

        The source for creating the model is closed, locked, heavily guarded as a corporate secret. But unlike code for software, this data might be illegally or unethically gained, and it might even mandate Mistral to publish it.

        You can “read” the assembly language of a freeware EXE program just as easily as you can “read” the open model of a closed source LLM blob: not very easily. That’s why companies freak out over potential hidden training data: not even they understand their compiled outputs.

        source

        • -> View More Comments
• Tetsuo@jlai.lu ⁨1⁩ ⁨day⁩ ago

  What about the mention of the Apache 2 licence?

  source

  • Peruvian_Skies@sh.itjust.works ⁨1⁩ ⁨day⁩ ago

    Open weights under the Apache license. The model weights are open but not the training process.

    source