Anthropic developed a new interpretability tool called the JLens, which can read the internal con..., Sonic AI
“Anthropic developed a new interpretability tool called the JLens, which can read the internal concepts a model is disposed to verbalize even when they do not appear in the output.”