Eliezer Yudkowsky
Eliezer Yudkowsky is an AI safety researcher and writer, best known as the founder of the Machine Intelligence Research Institute (MIRI) and the primary author of the LessWrong sequences — a multi-year series of essays on rationality, epistemics, and decision theory that became foundational reading in the rationalist and EA communities. He has been writing and thinking about AI alignment since the early 2000s, before the field had that name, and is one of the earliest and most influential voices arguing that misaligned superintelligent AI poses an existential risk.
He is largely self-taught and has no academic affiliation. His writing style is unusually direct: he names things that most people avoid naming, and he expects readers to update on arguments rather than defer to credentials.
The "Shut Up and Do the Impossible" framework
The one ingested essay captures a core piece of his practical philosophy: the distinction between making an effort and achieving an outcome. Most people flee the tension between trying hard and failing by redefining success as effort. Yudkowsky insists the goal is the outcome — "get out of the box," not "try to get out of the box."
His levels of effort: tsuyoku naritai (keep improving) → isshokenmei (all-out desperate effort) → extraordinary effort (non-default, leave comfort zone, take real risks) → "shut up and do the impossible."
"Impossible" almost never means logically impossible. It usually means "I see no way" or "beyond current human ability." Studying something can make it seem less impossible. See shut-up-and-do-the-impossible.
The AI-Box Experiment (winning twice by convincing a skeptical human to "release" an AI through text chat alone) is his personal proof-of-concept: the seemingly impossible was accomplished, and people's refusal to believe it happened is itself an example of the denial he describes.
Connections
- shut-up-and-do-the-impossible
- humans-are-not-automatically-strategic
- forward-chaining-vs-backchaining