Josh Purtell tears apart Yudkowsky's 'LLMs are conscious' argument: functionalism doesn't get you there
JoshPurtell · x · 2026-10-04
Josh Purtell criticizes Eliezer Yudkowsky's argument that LLMs are conscious as deeply flawed. He reconstructs the chain: functionalism, plus the claim that the complexity needed for cognition is a superset of that needed for consciousness, plus the claim that math is an upper bound on cognition, therefore LLMs are conscious.
He argues this is 'a really, really bad argument': functionalism doesn't support the superset claim (step 2), and both steps 2 and 3 represent massive logical leaps. A pointed debate on AI consciousness engaging a prominent alignment figure.
Related event: AI Researchers Clash Over Whether LLMs Could Be Conscious(14 posts)→
More from AGI Musings
- Is the GPU conscious? X banter over where AI consciousness could live — inductionheads · 2026-10-04
- Hinton admits Ilya was right: scale was the answer, as compute grew a billionfold — Chris_Armstrong · 2026-10-04
- Neuroscientist David Eagleman: the brain isn't a computer, it's a constantly rewiring 'livewire' — Chris_Armstrong · 2026-10-04
- Redditor proposes a trust that owns an autonomous AI agent to speak for you after death — snowdrone · 2026-10-04
- AI revenue must hit $3.5T by 2032, ~9% of GDP, to justify the buildout: Columbia paper — asusarla · 2026-10-04
- A year-old model intelligence is already a commodity, argues AI founder — ns123abc · 2026-10-04