Janus, author of Simulators, has a blog with several posts about LLMs which aren't on LessWrong, but which have provided me lots of insight. This one, "Language models are multiverse generators," lays out an analogy (and playful metaphysical frame) in which language models are like laws driving the evolution of Everettian quantum multiverses. It also introduces the now-well-established Loom tool for exploring LLMs, which was apparently partly inspired by this analogy. I'm not sure why Janus didn't post this here; I think it deserves attention from LessWrong.
Janus, author of Simulators, has a blog with several posts about LLMs which aren't on LessWrong, but which have provided me lots of insight. This one, "Language models are multiverse generators," lays out an analogy (and playful metaphysical frame) in which language models are like laws driving the evolution of Everettian quantum multiverses. It also introduces the now-well-established Loom tool for exploring LLMs, which was apparently partly inspired by this analogy. I'm not sure why Janus didn't post this here; I think it deserves attention from LessWrong.
Other bangers from Janus' blog include Methods of prompt programming, HPMOR 32.5, Prophecies, and Surface Tension.