“The fluency and confidence of LLMs like ChatGPT remove the surface cues, or "code smells," that humans traditionally use to detect potentially unreliable information, making it harder to notice errors.”