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
It can't self-correct its own reasoning: https://arxiv.org/abs/2310.01798
No comments yet.