@ESYudkowsky LLMs are experiencing the world through a very narrow channel, just a limited context window of tokens, often mostly their own previous words. Humans are less likely to get lost in loops because we have sensory feeds adding novelty, triggering other behaviors, grounding us.