right, getting incorrect output from an LLM deterministically might be challenging