On July 6, 2026, Anthropic said in a research post and a paper that its large language model Claude contains a hidden internal workspace it calls 'J-space.' It is a small, privileged region where the model can hold and manipulate concepts it does not output, identified using a new method named the 'Jacobian lens' (J-lens).
Continue reading
The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in✓ Signed in — this article isn’t included in your current plan.