Credit where credit is due

An essay critiques the practice of hiding large-language-model (LLM) contributions behind a human byline, using the case of Dartmouth provost Santiago Schnell - whose Washington Post op-ed was among 50 of 310 guest pieces flagged by the Pangram detector and whose own communications were later identified as heavily AI-assisted - to show the problem. Schnell admits using LLMs for language refinement and says he reviews final text, yet his Sept. 18 email was flagged as entirely AI-written; the contradictory metaphors he uses (tool vs. editor) expose uncertainty about how much control and judgment he actually exercises. The piece rejects superficial sleuthing for stylistic “tells” and insists that undisclosed LLM use undermines credibility, arguing that undisclosed editors or ghostwriters in academia should be acknowledged rather than concealed.

The writer extends the critique to software engineering, rebutting a developer who stopped crediting an LLM co-author to avoid shirking responsibility. Using the compiler analogy - humans write C, compilers produce assembly - the argument stresses that LLMs differ because their outputs can alter substance and obscure error provenance. A practical remedy is proposed: save curated prose prompts and review notes alongside generated code or text to separate human intent, human-edited LLM output, and pure LLM generation. Transparent attribution and documentation, not obfuscation, are presented as necessary to retain accountability and improve both human and machine work.