This is a clever trick, making out corporations to be the beneficiary - who is the real beneficiary? The one who prompts, because they can use the output of the model to solve their problems. In the meantime AI developers make cents on 100K tokens and are mostly in the red.

Why not put the issue where it is - it's a debate about public empowerment with AI vs incumbents protecting their copyrights. Especially for open source models - this situation is not about corporations. Yes, corporations are the ones with deep pockets and easier to sue, but that's just how lawsuits go, you don't waste money suing someone who can't pay up.

What is more important in the next decades - public empowerment or extending copyright to block training AI models? Can we agree to limit copyright to exact reproductions or should it cover all content generated from a model, even if when it looks different from all training examples?

I think going the copyright maximalist way will indirectly hurt creatives because all their works will be checked with the same tools we use to check AI. Anyone could be secretly using generative models. The AI attribution methods will reveal all sorts of things we don't like to see.

The outrage as packaged by the media really does mirror the anger against Napster from 25 years ago. I seem to recall the small-time musicians were more excited about Napster, though.

I remember when BitTorrent users were getting sued left and right, I was thinking about how can a P2P network allow downloading without revealing what was downloaded. Now we got the answer - the LLM - you can download the same LLM like anyone else and nobody knows what you're getting from it.