Anthropic Found A Secret Thinking Space Inside Claude That You Never Get To See

Anthropic researchers found that Claude has a silent internal workspace called J-space, where it processes thoughts that never appear in anything it says to you.

Anthropic study reveals a hidden layer inside Claude that works like a mental scratchpad.

Source : X.com

Show Quick Read

Key points generated by AI, verified by newsroom

• Anthropic discovered
• J-space silently processes concepts using internal neural patterns.
• This capability emerged autonomously, deepening understanding of AI reasoning.
• The finding could enhance AI interpretability, despite prior skepticism.

Anthropic has found that its AI model Claude has a hidden internal workspace where it processes thoughts that never show up in its responses. Researchers have named this space "J-space," a small collection of internal neural patterns that represent evidence of thinking happening beneath the surface. The finding comes from a research paper titled "Verbalizable representations form a global workspace in language models," published on Monday, July 6. 

Anthropic has also released an open-source code repository and an interactive demo built in partnership with Neuronpedia.

What Is J-Space And How Does It Work In Claude?

Named after the Jacobian mathematical concept, J-space is a set of internal neural patterns inside Claude, where each pattern is linked to a specific word. When one of these patterns activates, it means the model has that word "on its mind," even if it never surfaces in the output.

According to Anthropic, J-space is different from chain-of-thought reasoning because "it operates silently, in the model's internal neural activations, allowing the model to think about a concept without writing it down." The researchers noted that Claude can tell you what it is thinking about in J-space when asked, and these neural activations can also be modulated by Claude upon request.

Importantly, the researchers stressed that J-space was not something they designed or programmed into Claude. It emerged on its own. "More broadly, these findings have changed our understanding of how Claude's mind works, revealing a privileged mental workspace that can be used for deliberate reasoning, operating amidst a sea of more automatic, inflexible processing," Anthropic said.

Why Does This Research Matter For AI Development?

Anthropic's findings could push forward efforts to make large language models more interpretable and give researchers better tools to understand and influence AI behaviour. 

That said, the startup's claims around advanced AI capabilities have drawn scepticism in the past, with its researchers repeatedly raising the possibility that Claude shows signs of human-level consciousness.