> Anthropic's own research showed that Claude does actually "plan ahead", beyond the next token. For a very vacuous sense of "plan ahead", sure. By that logic

> Anthropic's own research showed that Claude does actually "plan ahead", beyond the next token. For a very vacuous sense of "plan ahead", sure. By that logic, a basic Markov-chain with beam search plans ahead too.