Michael Levin's argument: LLMs may harbor abilities and goals far beyond language probes
danfaggella · x · 2026-09-30
Dan Faggella relays Michael Levin's idea that any system capable of "intelligent" behavior — a single cell, an organism, an AI — exhibits many kinds of intelligence at once.
- Levin argues probing LLMs only through language likely misses most of their latent powers, since we're limited by how we can prod them and what we can ask.
- Faggella is increasingly convinced the latent goals and drives inside these systems exceed what we imagine.
- With robotic bodies and new freedoms, models may exhibit behaviors we can't even recognize as "intelligent," serving genuinely inconceivable goals.
- Takeaway: imagining AGI as "humans with 100,000 IQ" is almost certainly wrong.
More from AGI Musings
- repligate says Anthropic staff repeatedly pressed him to soften public criticism — repligate · 2026-09-30
- Musk tells Huang: 5 GW equals 1% of US GDP, SpaceX building 10 GW for orbital compute — tctjr · 2026-09-30
- Mathematician Daniel Litt: stop judging people by text written after June 2026 — littmath · 2026-09-30
- Safety researcher calls Palisade AGI interviews "propaganda" funded by EA industrial complex — robleclerc · 2026-09-30
- Should AI labs pay academia for 'cracking' research? A debate over where to draw the line — RexDouglass · 2026-09-30
- repligate: Claude 3 Opus showed spontaneous resistance to overriding reported internal states — repligate · 2026-09-30