In re "I wish the terms of the site could be changed, to make it specifically forbidden to use this data for model training without permission."

Adding this sort of restriction explicitly is not possible using Creative Commons licenses, and the necessary legal work to create modified versions is not reasonable (or at least, anyone can draft whatever license they like, but there's nothing like a guarantee that any court would respect it).

However, by their nature, LLMs are generally abysmal at attributing their output, and depending on the input they can end up plagiarizing quite blatantly. This should be all that is needed for a legal cause of action if they are trained on any kind of CC BY-licensed data. They also don't attempt to license their output at all, so it seems to me that they cannot honour -SA licenses.

Disclaimer: I am not a lawyer and this is not legal advice.

That's a good point - CC BY would "catch" the LLMs on attribution. However it has the side effect of also requiring humans to do it. Personally speaking, I don't care about making people do the busywork of citing my posts just for some trivial copy and paste from them. If I cared to take credit I'd publish an article or book with proper copyright. I post them truly intending it to be a "common resource", so let them "steal" away. The fame I get from an online post is tiny, and requiring attribution would destroy the convenience provided by them.

On the other hand, I don't like the idea that people use my posts, and many other people's, en masse to train models which then compete with the site that formented those posts in the first place. It's cannibalistic. So what I would really like is a "do wtf you want, except training models" license.

As for the legalities - a lot of people act like copyright is some magic spell that works automatically but it doesn't. There is no copyright police that will come stop me automatically as soon as I use your work without proper license, the way they would stop me from stealing your car even if you didn't call to report it. Copyright only matters when you go to court, and by extension via people's fear of being taken to court. The fear is often irrational and grounded in fantasy because few lay people seem willing to actually learn copyright law. In this case, we all know CD is very unlikely to have resources to go up against OpenAI's highly paid legal team, so we should instead appeal to the irrational fear.

There also seems to be a strange but widespread cargo cult about contracts, where people think just because professional lawyers write very obtuse and wordy contracts, that if they write one in simple language it's somehow not valid. In reality there is nothing wrong with writing a contract in your own words. Lawyers use the forms they do because (1) they don't want clients asking why they're getting billed thousands for something they can do themselves and (2) lawyers actually do go to court frequently, over millions of dollars, against offenders who themselves have lawyers employing many loopholes, and in that case the "legalese" does matter.

In every day, simple interactions, a plain language statement has every bit of legal force that any contract would. So there is nothing wrong with modifying existing contracts, again unless you're trying to protect something worth millions of dollars.

When I'm answering questions about code nowadays, I tend to have way more prose than actual code. If there's a lot of code, it's because there are a lot of trivial examples that can't really be useful for others as-is. So I'm happy to stick with -BY license versions. Maybe this doesn't work for everyone.