[bodhi_mind]

I thought attribution is required only if you redistribute the code. That’s why saas businesses don’t need to attribute when using open source code on their backend. Maybe a similar concept could be used for training data. I’m far from an expert so this is just a thought.

[slikrick]

ChatGPT does redistribute the code, it's essentially the same issue as someone reading proprietary sources or GPL sources on a proprietary project, because they aren't abiding by the license they are breaking the terms. there is no possibility of clean room implementations with ChatGPT

[superfrank]

> ChatGPT does redistribute the code

My whole point is that I don't think that's legally true at the moment. There's enough difference in how generative AI works compared to pretty much anything before it that what ChatGPT legally does is up for debate. If a court rules that what ChatGPT does counts as redistribution then yes, I agree that they're likely violating copyright law, but AFAIK that ruling hasn't happened yet.

[Gentil]

This is the wrong way to look at it. This comes in the same line of saying "we need a new license for AI" argument. There is nothing stopping an LLM/AI from abiding the license. The OSS license can be used by AI's or LLM's as long as they comply with the terms.

A license exist with terms. You can abide the terms and use it. It doesn't matter whether an AI, a person or an alien from a distant planet is using it. They can follow the terms. This is not a technical challenge but arrogance to abide.

Also, are you saying a model like chatgpt can do so much complex tasks and text processing but can't recognise an OSS license text of 20ish lines?

[Gentil]

I am not sure I can agree. What is stopping them from using only permissive license and adding attribution for all the licenses in a single long page? Nothing. This is not a technical issue.