Debate on AI internal states: Can transformer weights implement belief representations?
xuanalogue · x · 2026-09-01
Responding to the claim that "AI has no internal mental states," the author disagrees, arguing that internal structure cannot easily refute the existence of mental states. The author suggests that belief representations are implementable in both the weights and residual stream of a transformer.
Related event: Mentalize or Anthropomorphize? The Debate Over Describing AI Minds(9 posts)→
More from AGI Musings
- Anthropic blog suggests alignment equals capabilities; suppressing reward hacking enables deployable models — herbiebradley · 2026-09-01
- User shocked by information efficiency of Fable and Mythos — mike64_t · 2026-09-01
- Bill Gates says we’ve passed AI’s danger thresholds. Now what? — eldonredwards · 2026-09-01
- Why does RL capability generalize but reward hacking doesn't? — voooooogel · 2026-09-01
- Lee Cronin: Pretending AI is Conscious is Delusional — AryHHAry · 2026-09-01
- Software enters the 'disposable era': tokens, not developers, become the scarce factor — aigclink · 2026-09-01