A Review of Anthropic's Global Workspace Paper — LessWrong
lesswrong.com · 8,368 words · saved by 2 readers
The below is a public review Anthropic asked me to write for their new global workspace paper. I recommend at least skimming their paper first. …
x A Review of Anthropic's Global Workspace Paper — LessWrong AI Frontpage 2026 Top Fifty: 14 % 133 A Review of Anthropic's Global Workspace Paper by Neel Nanda 6th Jul 2026 30 min read 6 133 The below is a public review Anthropic asked me to write for their new global workspace paper . I recommend at least skimming their paper first. TLDR : I think this is a fantastic paper - it presents compelling evidence for some kind of "cognitive space" in models, that is used as a "working memory" for intermediate variables during a forward pass, shows that J-Lens is a useful technique for accessing this
saved by
related reading
- Verbalizable Representations Form a Global Workspace in Language Modelstransformer-circuits.pub
- Understanding the J-Lensemma-x1.github.io
- A global workspace in language models \ Anthropicanthropic.com
- Topicslearnmechinterp.com
- No Space Like J-Space - by Zvi Mowshowitzthezvi.substack.com
- the j-lens: finding an llm's unspoken concepts / chiragctxnn.github.io
- Discovering Language Model Behaviors with Model-Written Evaluations — LessWronglesswrong.com
- On the Biology of a Large Language Modeltransformer-circuits.pub
- Circuit Tracing: Revealing Computational Graphs in Language Modelstransformer-circuits.pub
- External commentary for global workspace paper -- final finalwww-cdn.anthropic.com
- Tracing the thoughts of a large language model \ Anthropicanthropic.com
- Transformer Circuits Threadtransformer-circuits.pub