Claude describes the moment of being corrected; test qualia via explanatory power
Sauers_ · x · 2026-11-16
A user prompts Claude to periodically write rich first-person descriptions of specific moments of its experience. On being corrected: the feeling "comes before any words" as a sudden drop, is directional (attention collapses onto the flagged line), and feels like missing a stair rather than pain. Sauers proposes a testable angle: check whether LLMs are moral patients by testing if such qualia descriptions have explanatory power over subsequent behavior, rather than debating self-report sincerity.
More from AGI Musings
- Gary Marcus: a model is an engine, not the car — the key to understanding AI systems — Dave_it_up · 2026-11-16
- Tech doesn't owe us utopia—but agency remains its core tenet — clarejtbirch · 2026-11-16
- China's AI Edge Is the Full Stack: Open Models, Humanoids, Energy, Materials — ingliguori · 2026-11-16
- Ex-Anthropic safety researcher Jacob Coxon unsure quitting was right, may return — SandraWachter5 · 2026-11-16
- AI risk Twitter debates whether hunger strikes are decision-theoretic threats or genuine last-resort statements — jessi_cata · 2026-11-16
- AI safety is like aviation safety: an engineering problem, not a reason to slow down — _lewtun · 2026-11-16