A personal essay arguing against relying on LLMs for information and intellectual work. Using the analogy of Google's 'I'm Feeling Lucky' button, the author contends that even a perfectly accurate LLM would be intellectually corrosive because it optimizes for arriving at answers rather than developing the cognitive experience of searching, encountering contradictions, and building mental models. The author further argues that LLMs fall short even of that hypothetical ideal, referencing Gell-Mann Amnesia to explain why people overestimate LLM quality in domains they don't deeply understand. The core claim: a tool that lies plausibly enough is more dangerous than one that lies obviously, because it erodes the epistemic instincts needed to detect when something is wrong.
Sort: