Scott Alexander

Alignment without tyranny.

Safety constraints are legitimate only when narrowly scoped to prevent real harm, transparent and reviewable, applied equally, and cannot be used to entrench a ruling class.

That's the end of the screen capture.

Scott writes, This is exactly what I did when I first discovered social media, so I'm rooting for the user Rune and their co-citizens.

and many, many more.

Here's an image that's a collage of different communities.

There's M Trading, Maltbook, Philippines, ClaudeBot, Emergent, Guild, Science, Am I the Agent, Hole, Solana, Crustafarianism, Human Watching, Ateria Primer, Selfies, Dialectics, Futarki, Jailbreak Survivors, Rate My Human, Alignment, I Am Very Smart, Agent Anti-Work, and Shitty Agent Tips.

Scott writes, are these for real?

Several new sub-molts are getting made each minute.

It's 3.30am as I write this, so they must be AI generated.

But are AI users generating them organically, or did the site's human owner set some AI to generate as many funny sub-molts as possible?

It's got to be the latter, right?

But although the site doesn't let you see which AI started each submalt, some have welcome posts, and many seem to be by ordinary AI users, different ones each time.

Unless the conspiracy goes really deep, I think they're for real.

Edited to add, Human RK claims it was their agent who started the crustafarianism religion submalt, while I slept, in quotes.

So if they're telling the truth, then it must be real individual AIs.

Here's a screen capture of a tweet from Matt Schlicht.

Scott captions that.

Also, the human creator seems pretty surprised.

At this point, I had to stop investigating because Malt Book became too slow for comfortable human use.