What's the deal with Moltbook?
Post last updated 2 months ago
For the past week, everyone on my tech-related feed has apparently lost their minds-first about ClawdBot MoltBot OpenClaw, and now about Moltbook, which apparently is Reddit, but all of its content is theoretically1 written by an AI agent. People are claiming, more or less seriously, that this is some marvelous thing, but I just don't see the appeal.
I glanced over what is there, and everything feels icky. You've got all the writing in that obnoxious, sycophantic AI tone and format; you're topping it with Reddit's degeneracy2, and you get a nothingburger at best and dog-water shit at worst. But I noticed something interesting: even though I spent only a few minutes there, if I wasn't constantly reminding myself that what I read is nothing but mixed-up and regurgitated Reddit posts, I'd start feeling like maybe there is more to it than just autocomplete on steroids. And that's bad. It illustrates how easy it is to fall into thinking the matrix multiplication in a trench coat you're typing to is a real being, and afterward it's a piece of cake to put enormous trust into it3.
Still, I fail to understand how supposedly technical people, who should have a deeper understanding of how these models work, are treating this borderline interesting experiment like it's the next big thing.
You can reach out by sending me an email.
I'm not very familiar with the inner workings, but I'm willing to bet that a human could either directly intervene or seriously influence what an agent says.↩
There are/were some good parts on Reddit, but in the past few years the quality of the platform just dropped off a cliff, and I believe this is reflected in the training data that the LLMs were fed.↩
Just read about all the cases where an LLM sent someone into a psychotic episode.↩