MIT Technology Review · AIJul 10, 2026, 04:22 UTC+8

Anthropic found a hidden space where Claude puzzles over concepts

Anthropic built a tool called the Jacobian lens (J-lens) and used it to uncover a hidden region—dubbed J-space—inside Claude Opus 4.6. J-space surfaces words related to what the model is about to say, but those words may not appear in the final output. Anthropic claims monitoring these words offers a new way to understand and control its models. The company published a paper and released a public demo with Neuronpedia. Goodfire chief scientist Tom McGrath called the work “very good and interesting.” The post does not disclose J-lens false-positive rates or any impact on model performance.

Why it matters: Anthropic published a new interpretability paper using J-lens to find a 'J-space' in Claude Opus 4.6's middle layers where the model pre-processes concepts before output. MIT Tech Review broke the story with paper and Neuronpedia collaboration details. Not scored higher becaus...

Read the original ↗Export Markdown

Take

This one's worth opening because it moves interpretability from "which neurons fire" to "what words is the model chewing on." Anthropic built a tool called J-lens and used it on Claude Opus 4.6 to find a hidden region—J-space—in the middle layers. It surfaces words related to what the model is about to say, even if those words never appear in the final output. Think of it as catching the model's internal scratchpad. They published a paper and shipped a public demo with Neuronpedia. Goodfire chief scientist Tom McGrath called it "very good and interesting."

I'd discount this a bit for now. The post doesn't disclose false-positive rates or whether J-lens adds inference overhead. Right now it reads like a research team's debugging tool, not something you'd run on a production model in real time. But the direction matters—if you can reliably catch moments where the model's internal draft diverges from what it actually says, that's useful for safety alignment.

Source: MIT Technology Review · AI · technologyreview.com
Published: Jul 10, 2026, 04:22 UTC+8 · Jul 10, 2026