Knight Lee

A Solu­tion to Sand­bag­ging and other Self-Prov­able Misal­ign­ment: Con­sti­tu­tional AI Detectives

Com­mit­ment Races are a tech­ni­cal prob­lem ASI can eas­ily solve

Think­ing Machines

An idea for avoid­ing neu­ralese architectures

Cy­cles (a short story by Claude 3.7 and me)

De­tailed Ideal World Benchmark

Scan­less Whole Brain Emulation

[Question] Why do fu­tur­ists care about the cul­ture war?

The “Every­one Can’t Be Wrong” Prior causes AI risk de­nial but helped pre­his­toric people

Re­duce AI Self-Alle­giance by say­ing “he” in­stead of “I”

Knight Lee’s Shortform

ARC-AGI is a gen­uine AGI test but o3 cheated :(

Why em­piri­cists should be­lieve in AI risk

The first AGI may be a good en­g­ineer but bad strategist

Keep­ing self-repli­cat­ing nanobots in check

Hope to live or fear to die?

Should you in­crease AI al­ign­ment fund­ing, or in­crease AI reg­u­la­tion?

A bet­ter “State­ment on AI Risk?”