@lumi@snug.moe I usually don't sign things like these but this is the sentence that does it for me:this shouldn't mean this ban would be eternal, but there is no truly ethical LLM or diffusion model right now on the marketThankfully someone who doesn't see this as black and white and that all use of a specific technology has to be bad. It's just that most options are terrible right now, but that can change in the future once the free world catches up.
@lumi@snug.moe I don't see why it wouldn't be possible. All you'd need to do is train a model with data that is actually legitimately obtained and is able to run offline without proprietary software. I'm sure people will develop that one day.
@sun@shitposter.world@lumi@snug.moe It doesn't matter that you don't care. You can't just ignore laws you don't care about, as you will likely get sued at some point.
And also many licenses are not compatible with each other so you can't combine them into the same model. Ideally you'd have separate models for permissively licensed code, GPL code, AGPL code, etc. The models would also need to return the names of the original authors of the data used to generate your answer, as many permissive licenses still require attribution.
@SuperDicq@lumi there's nothing wrong with using scraped open source projects to make your slopcode project if you release your project also open source.
@sun@shitposter.world@lumi@snug.moe True, but the more slopcode you start publishing the higher the chance you accidentally publish something that can be recognizably infringing.
@SuperDicq@lumi lets be real you can't know some file I release was inspired by some random github repo, regardless if it was authored by a human or a robot