Bayeswatch: a Retrospective

LessWrong has published a retrospective on Bayeswatch, a 2021 science fiction series that explored AI futures under international treaties limiting powerful systems. Five years later, the author reflects on which predictions aligned with reality: large language models turned out to be stateless systems reliant on context windows, human-written code proved vulnerable to AI exploitation, and the policy consensus has converged on the need for global coordination. Where the author missed: not knowing whether language models would plateau—they didn't. The central theme: AI progress has compressed the timeline from science fiction to current events. The question now is which of the story's remaining scenarios lie ahead.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton