Anthropic finds 'global workspace' akin to human consciousness in Claude models

Anthropic researchers have identified a region within their language models, including Claude Sonnet 4.5, that functions similarly to a "global workspace" in the human brain. This specialized area appears to hold and process concepts that the model is actively considering and could articulate. The discovery was made using a new interpretability tool called the Jacobian Lens (J-Lens), which analyzes the model's internal representations. AI

IMPACT This research could fundamentally change how we understand and interact with AI, potentially leading to more transparent and controllable models.

RANK_REASON The cluster describes a research paper detailing a new interpretability method and findings about internal model representations. [lever_c_demoted from research: ic=1 ai=1.0]

Read on dev.to — LLM tag →

• Anthropic
• Claude
• Claude Sonnet 4.5
• Jacobian Lens
• J-Lens
• Transformer Circuits Thread
• Verbalizable Representations Form a Global Workspace in Language Models

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →