I'm confused, why would you insist on using an LLM for decision making? It's not "the collective wisdom", it's a probabilistic language model. I imagine LLMs

I'm confused, why would you insist on using an LLM for decision making? It's not "the collective wisdom", it's a probabilistic language model. I imagine LLMs are better suited as a last step. I'd want a proper model to infer a decision, and an LLM to transform it to a human-friendly-formatted advice.