Yes and no. LLMs are definitely struggling with long horizon planning in complex situations, and can hallucinate/misunderstand environmental states such that th

Yes and no. LLMs are definitely struggling with long horizon planning in complex situations, and can hallucinate/misunderstand environmental states such that they can recommend nonsensical or impossible steps. But we are starting to see more research that "grounds" LLM answers in a set state. Where they shine in the planning stage is they provide a sense of contextual heuristics we didn't have available before without a lot of expert domain programming. LLMS in their current state are certainly not the answer, but point to some exciting possibilities that we couldn't do before because of sheer complexity.