That works if the LLM has adequate external feedback from a terminal and browser in context with the past trial etc. It can't self-correct its own reasoning: h

That works if the LLM has adequate external feedback from a terminal and browser in context with the past trial etc. It can't self-correct its own reasoning: https://arxiv.org/abs/2310.01798 https://arxiv.org/abs/2310.01798