Are you getting better performance from an LLM than a Bayesian classifier?