What should be done?
I don't think that anything can be reasonably done, considering the tools (or lack thereof) that have been provided by Stack Overflow to address this specific issue.
@AaronBertrand commented on the question above:
An enhancement that might make sense for some technologies is to add a separate tagging system for answers
I agree that this is something that could help a great deal (along with trying to find other ways to incentivize keeping answers fresh while maintaining quality). However, I do not see it happening. (For context on the following: I was Tech Lead for and a Staff Engineer on the PubPlat team during events described.)
Throughout 2021, the Public Platform team (as it was called at the time) was working on a series of projects related to fixing the "Outdated Answers" issue. A number of changes were made to the site to address aspects of this, including trending sort and unpinning accepted answers.
However, the cornerstone of the effort (and in my estimation, the most important part) was answer versioning. TLDR on the project: allow for tags to have versions, and then allow answers to be given tag versions based on their question having the parent tag. Allow for easy filtering in a question based on the version that you are looking for.
A lot of thought was put into this. We went through multiple rounds of workflow revisions. I put together a plan for an MVP implementation that could be tested out on some of the most popular tags that would have taken the team (in my estimation) 2-3 months to get out. The project went to user research once. Then twice. The third time it went to use research, it never came out. Maybe a lot of people were to blame. Maybe no one was to blame. But it doesn't really matter anymore. What does matter is that some time in 2022 the product team declared "Mission Accomplished" on the Outdated Answers and we moved on to Staging Ground (for all the good it did us). And since then almost every single person who touched answer versioning has left the company.
I don't know why we didn't move forward with it. But we didn't. And I don't see it happening any time soon (note: I have been out of the company for over a month, and off the PubPlat team working on non-AI work for nearly a year, so I have absolutely no insight into the current roadmap, which to the best of my knowledge has yet to be made public).
And it is kind of sad, ya know? In my new job I am working on a completely new technology stack. I am searching stuff all the time, and see lots of links to Stack Overflow questions. And more often than not, it is for accepted answers from 2016. Or 2011. Or 2008. Yeah, it has 857 upvotes. However, I don't know about you, but canonical data from a decade and a half ago doesn't instill me with a lot of confidence in its veracity. And I know first-hand of the challenges that a new answer will face in being seen and promoted to the top of old incumbents for any popular question. I don't know if answer versioning would have solved anything by itself (all of the Outdated Answer efforts were seen as addressing different aspects of the multi-faceted issue). But it could have made a big step in the right direction.
The reliance on so many old answers, like the (in my belief mistaken) belief that Stack Overflow will forever own Google search results are two big signs to me that the site will continue to lose relevance in the future (note: I'd love for this to be reversed or to proven wrong, and I have written elsewhere about what I think needs to happen to achieve this).
None of the solutions seem to be working
Yeah, that about sums it up.