As opposed to outputting debugging information, which I wouldnt be surprised if LLMs do output "debug" output blurbs which could include model specific information.