They need a Q&A site. A place where they can ask the questions they still can't find the answers to.
This need is one that existed before chatgpt, and one that would have continued to exist without it, but it will become more and more necessary as less and less people are willing to provide those answers.
I would like to see SO, and the network as a whole, move more in the direction of being a Q&A site first, and a knowledgebase second. What I mean by this is, the goal of curation should expressly not be preventing answers to "bad" or "not useful" questions that won't contribute to the knowledgebase, rather, the goal of curation should be ensuring the posts that do contribute to the knowledgebase are as good as they can be and can cut through the fog.
I don't like the normal ideas that are associated with this goal. I don't like the idea of removing downvotes, or close votes, or increasing barriers or any of that, I'd rather this be approached from more of a toolingcosmetic, or presentation direction.
One example I've been thinking about for a good while:
People find asking questions daunting and likely to fail.
Change the way curation affects users. Don't remove it, just make it less antagonizing. An example would be more or less doing away with reputation and focusing on "People helped" (not to be confused with the horrendous "People Reached" metric.) Make upvotes give 1 rep and downvotes remove 1 rep, and instead of showing a post score, show "N people found this post useful. Y people found issues with this post.". This could be done without even changing how reputation works, meaning it could be tested on a willing network in a way that has no permanent side effects and could be completely rolled back; we would primarily need to hide reputation and instead show upvotes received minus downvotes received in place of reputation. I would also, in this scenario, remove the existing voting buttons/display and have a more standard "This was helpful" and "there are problems with this post" buttons, with the second optionally leading to a comment (hopefully more often than not.)
The benefits of this are people can directly see, without needing any reputation or a browser extension, if a significant number of people have found a post to be problematic even if it was heavily upvoted, and of course the opposite is true, people can see how many people were helped by a post regardless of whether or not others found issues with it. It prevents one form of voting from hiding the other and provides a direct meaning to the votes such that there is far less confusion in what voting should mean and what people should do if they receive downvotes.
Obviously, if something like this was tried and found to be a good way forward... the way we give people access to curation tools would need to change to match the new system, ideally by making them based on positive actions rather than just flat reputation. For example, after x approved suggested edits, you can make y edits without review per day, increasing as more suggested edits are approved should you ever go over y. This creates a sort-of dynamic throttle that prevents users from making a lot of edits in a short period of time unless they're a user that consistently does so in a positive way. Honestly this should just be how we handle access to curation tools in general regardless of if we go this route.