As announced a little while back, MathOverflow moderators are participating in a network-wide moderation strike (currently at 121 moderators striking; see the background post for more details). The short explanation is that the company has been making life difficult for moderators in rolling out changes to policy that binds how moderators can operate around dealing with AI-generated content, after many sites on the network set up their own policies (which is how it's supposed to work) banning it. The policy as announced privately to mods and what got announced publicly were slightly different, and the private one essentially tied all our hands.
Stack Exchange the company released internal data analytics that it felt showed that moderators were taking too much action on allegedly AI-content, leading to many false positives, but that analysis is by all accounts not that solid. The 35 responses at this post pull it apart in many ways. The biggest failing of the analysis is that it proceeds via a proxy count that by now demonstrably doesn't agree with what is seen on the ground by moderators, as the behaviour of users changed since the release of ChatGPT (this is the best guess among moderators, it's really unclear).
One concerning matter that arose since the start of the strike was that it was discovered that the latest quarterly data dump of the whole network Q&A data (which is done under a Creative Commons license, and is a way to ensure the knowledge generated by users remains free) was missing. This was secretly turned off, and people were very much not happy about it. Thankfully, as a result of the resulting pressure, the data dump was restarted and is now done.
To add insult to injury, Stack Exchange had announced it was going to roll out a tool for formatting code help on StackOverflow, the biggest site on the network. It turns out this was a simplistic wrapper for a query to some generative large language model from OpenAI (the makers of ChatGPT), which a user extracted in ten minutes—complete with typos and instructions the bot happily ignores, as many users found out as they jailbroke the tool within half a day and made it do all kinds of fun things (write a My Little Pony story, answer the question outright, play a game of chess, invent questions out of whole cloth, remove crucial parts of the question). Thankfully, Stack Exchange listened to the criticism, is turning off the experiment and going back to the drawing board. Though, to temper this relief, it's unlikely to go away forever. There's no hint that it will come to MathOverflow, since it's touted as a "code formatting assistant", but the smell on the air is that the company wants to get more out of the AI pie as it's being passed from hand to hand. Watch this space.
You can read the official strike update post here. The biggest move so far is that moderators elected representatives for negotiation and the negotiations have started. There is some movement: see the data dump point above, and another one that I'm not sure I can mention, but it's fairly minor. No one wants the network to fall over, but we might be in this phase for a while.
In the meantime, please be patient with the decrease in moderation here on MO. Keep flagging content, because after the strike this can be returned to and cleaned up. Blatant spam/vandalism should absolutely be flagged as such, as with enough flags it will be destroyed automatically. Downvoting to make it less visible is not enough.
