RT @emollick: Because we are used to how a deterministic program works, we tend to judge LLM output as failure or success - a single error…