NewsBite analysis · AI, Data
ProPakistani
Anthropic Unveils AI ‘Thinking Space’ Inside Claude
Anthropic has unveiled an internal activation subspace named J-space in its AI model Claude, which functions similarly to the human brain’s global workspace. This J-space exhibits cognitive traits such as internal reasoning and flexible generalization, enabling Claude to perform more advanced reasoning tasks while routing simpler actions elsewhere. The new interpretability tool, J-lens, allows researchers to examine the processes within J-space, aiding AI safety and alignment efforts. This discovery prompts valuable discussions about the relationship between AI models and human cognition.
ExperimentConfidence: HighShort term horizon
Published 10 Jul 2026, 12:56 UTC2 min read
Visit original article ↗Lead image — ProPakistani
Summary
TL;DR
- J-space enhances complex reasoning capabilities in Claude
- J-lens tool provides insights into Claude's internal processes
- Finding suggests parallels between AI reasoning and human cognition
- Disabling J-space hampers multi-step reasoning performance
- Research aims to improve AI transparency and safety
Interesting facts
- J-space emerged naturally during Claude's training process
- J-lens can help detect hidden malicious intentions in AI models
- Anthropic does not claim Claude is conscious with these findings