dioscuri publishes 'Bad AI Consciousness Takes Bingo' rebutting common LLM consciousness arguments
dioscuri · x · 2026-10-12
dioscuri released a blog post systematically rebutting the most common dismissive takes on LLM consciousness:
- "Just matrix multiplication": any physical system reduces trivially at implementation level (Marr 1982); the question is emergent capacities — and LLMs interleave matmuls with nonlinear ops anyway.
- "Just next-word prediction": prediction is a deep cognitive capability (humans are theorized to be predictive engines too); post-training adds other objectives, and LLMs have been caught planning rhymes a line ahead.
- "No body": dreaming and locked-in syndrome show rich consciousness with minimal bodily agency; embodiment requires argument, not assumption.
- "I work in ML, trust me": the tech world itself is deeply split, and training models doesn't settle how features map onto cognition.
Retweeter FabienMikol agrees current LLMs likely aren't conscious but is appalled by the poor quality of confident counterarguments.
Related event: Researcher's 'Bad AI Consciousness Takes Bingo' Debunks Common Myths(3 posts)→
More from AGI Musings
- AI won't go mainstream until it cures something, argues one analysis — imjustnewatai · 2026-10-12
- Norbert Wiener's 1949 warning: the machine age is a struggle against our intelligence limits — pickover · 2026-10-12
- How Hard Is Stealing $1 From a Bank? A Probe of ASI's Real-World Attack Surface — AaronBergman18 · 2026-10-12
- How secure are power grids and banks at the limit of AI intelligence? — AaronBergman18 · 2026-10-12
- WSJ prints Zack Korman interview: "Why Did Technologists Become Doomers?" — max_paperclips · 2026-10-12
- A millennium of philosophy: the AI debate over measuring understanding — burny_tech · 2026-10-12