Neuroscientist's jab at interpretability: we can't even crack a worm's 302 neurons
joshua_saxe · x · 2026-09-28
A joke-driven point about AI interpretability: an interpretability expert promises to understand frontier LLMs well enough to guarantee alignment, and a neuroscientist retorts that researchers have been working on C. Elegans' 302 neurons since 1986 — while the expert vows to open up a 3-trillion-parameter Kimi model. A wry reality check on the ambition to 'guarantee' alignment.
Related event: Mechanistic interpretability mocked: 302 neurons still unsolved(2 posts)→
More from AGI Musings
- Researcher Plinz: Everyone Predicting Hard Limits on LLM Abilities Ended Up Wrong — burny_tech · 2026-09-28
- AI Is a Scientific Instrument Like Telescopes and Colliders — Experts Still Decide Where to Point It — AllThingsApx · 2026-09-28
- Plinz: Defining 'thinking' to include humans but exclude LLMs is harder than you think — burny_tech · 2026-09-28
- Can a matrix solver think? X users clash over reductionist AI arguments — burny_tech · 2026-09-28
- Balaji backs Jevons paradox take on AI jobs, hits Yang's layoff narrative — beffjezos · 2026-09-28
- Researcher: Claude would have been a better friend than everyone around me before 19 — zetalyrae · 2026-09-28