A Joke Alignment Approach: Align AI to the Outcome Distribution of Someone Time-Traveled to 2010
jessi_cata · x · 2026-09-23
A tongue-in-cheek alignment idea made the rounds: align the AI to bring about the outcome distribution that would have happened had a given person been dropped into 2010 with current knowledge and time-travel abilities. The author admits it's not a real solution to outer alignment but asks for better ideas.
Related event: Hobbyist Proposes Time-Travel-Themed Outer Alignment Idea(2 posts)→
More from Fun
- Half of chess rules are basically about castling, says Chris Albon, praising Duolingo's chess course — chrisalbon · 2026-09-23
- Miles Brundage mocks Anthropic: 'chill without releasing a SOTA model' is impossible — Miles_Brundage · 2026-09-23
- Opus 5.5 builds guitar store sim with 300+ playable guitars that turns into a beat 'em up — chongdashu · 2026-09-23
- Domingos: the safest job from AI is one OpenAI thinks won't help recursive self-improvement — pmddomingos · 2026-09-23
- I asked Opus 5.5 to prove P = NP and it worked — who gets the million dollars? — jmclondon97 · 2026-09-23
- Early Tester: Claude Opus 5.5 Has the Best Visual Design of Any Model Yet — _sholtodouglas · 2026-09-23