I know you think I can see this post, but I can't.

Or shadow banning

Also possible the changes haven’t kicked in yet?

I thought the fact-checkers worked for third-party org and were not internal moderators. Find anything on Meta ending partnerships with third-party org?

I don’t know the culture at Open AI, but I see your point. But doesn’t that mean we need to be more careful about who we allow to develop powerful models? Are we not really in trouble if Open AI doesn’t care that much?

Thanks! Two questions: (1) Does open source not entail the risk that people who don’t seriously care about safety will develop models too fast? (e.g. DeepSeek). (2) Can you not fight the concentration of private power through gov. regulation (public AI enterprises, antitrust, etc.)?

This is how I’ve always understood Aristotelians’ point about practical wisdom.

"We don’t want the teenage kids to encounter bullying content only then to have to report it; we want them to be spared the burdens of encountering it in the first place."

Contrary to most takes I've read, Jeff argues that increasing the confidence an AI classifier requires to flag speech as hate speech is a defensible option, but ceasing to use classifiers to detect hate speech (as Meta is doing) is not.

I think that’s a flex. Dadcred.

And just to try to keep BlueSky a positive space: I think that debates about decentralized moderation are fascinating, and I'm grateful for advocates of decentralization to have proposed a way to diminish the power of Musk, Zuckerberg, etc. I see this as public service.