Anthropic’s global-workspace paper hints at a fixed semantic coordinate system in LLMs
tszzl · x · 2026-07-26
- The author argues that Anthropic’s recent paper on global workspaces in LLMs sheds light on the Chinese Room puzzle.
- The most striking result is that averaging Jacobians across contexts can still produce a meaningful readout for which word a token would generate at a given layer.
- That suggests intermediate LLM representations may live in a shared, fixed coordinate system for meaning, rather than changing arbitrarily with context.
- In this view, different layers communicate in a common semantic language before producing the final output.
More from AGI Musings
- François Fleuret: Only Two Long-Term Futures — No Super AI, or Staying Fully Human With It — francoisfleuret · 2026-09-11
- IG reel debunking the 'winning the AI race against China' fallacy hits 500k likes — louisvarge · 2026-09-11
- Post-AI World Leaves No Room for Learning on the Job — rachittshah · 2026-09-11
- Researcher questions AI safety eval firm, citing 'blatantly sloppy' security and monitoring — Kyrannio · 2026-09-11
- AI researcher memes agent-swarm tinkering with He Jiankui's embryo-editing quote — dejavucoder · 2026-09-11
- nabla_theta: happy to be wrong if the AI utopia arrives with little ex ante risk — nabla_theta · 2026-09-11