We ask the editor in charge of AI if you can really see what's inside Claude's head.
⚡ Quick Summary
Anthropic has announced a new way to peer into the "inner thinking" of its models. In the ``J space'' inside LLM, there are words that do not appear in the final output but seem to influence the inference.
Anthropic has announced a new way to peer into the "inner thinking" of its models. In the ``J space'' inside LLM, there are words that do not appear in the final output but seem to influence the inference. But to what extent can this be called "thinking"? We asked the magazine's editor, who was in charge of the interview, how to read this unusual research.
← Back