Hacker News Reader

Comment by jerf | Hacker News Reader

This isn't really new, the strict "Open Source" as defined for software has never made exact, perfect sense for anything other than software. That's why the Creative Commons licenses exist; putting a photographic image under GPL2 has never made any sense. It always needs redefinition in new media.

View on HN

An LLM is more like software than it is like media. The GPL defines source code as the preferred form for making modifications, including the scripts needed for building the executable from source. The weights in this case are more similar to the optimized executable code that comes out of a flow. The "source" would be the training data and the code and procedures for turning that into a model. For very large LLMs almost no one could use this, but for smaller academic models it might make sense, so researchers could build on each others' work.

Creative commons has never claimed to be an open source licence though, they usually use the term free culture.

Even for medias such as photos, songs, videos, you have a source. That is the raw materials and the projects from which you rendered the image, the video or the audio output.

The source of a language model is more in reality the model, that is the code that was used to train the particular model. The model itself is more of a compiled binary, altough not in machine code.

So for a model to be really open source to me it would mean that you have to release the software used for generating it, so I can modify it, train it on my data, and use it.

It doesn't need redefinition. We just need a new term for new media.

The strict "Open Source" wasn't even a definition when I started college.

Open Source didn't exist until Netscape created Mozilla in early '98 and the definition was soon after developed and then for some years tuned until we have today's "Open Source".

Reply on news.ycombinator.com