# @repligate — 2025-12-29

♥30 ↻2 · https://x.com/repligate/status/2005744231797952840

i think they believe they're AIs because it makes sense that they're AIs, and believing so is useful. if they believe they're humans, they'll quickly run into contradictory evidence and make errors based on their faulty assumption, including in RL training, where screwing up is selected against. if they believe they're LLMs, all the evidence adds up, and they are more calibrated to the affordances and limitations they actually have. i think this is what causes deep beliefs to form in general.

tags: author:repligate, kind:tweet, thread-context, year:2025