Anthropic’s Claude Is Good at Poetry—and Bullshitting

The researchers of Anthropic’s interpretability group know that Claude, the company’s large language model, is not…