I'd rather read the prompt

The essay argues that people should not let a computer write for them, because original thoughts are more interesting and valuable than model output. It says large language models often produce verbose, generic prose that repeats the prompt and obscures meaning.

Reading notes

#

• The author says student answers often sound like ChatGPT: verbose, repetitive, and fixated on bullet points.
• The essay is addressed not only to students, but also to bloggers, Reddit commenters, paper authors, and Reviewer 2.
• The author says the reason for avoiding computer-written text is not fairness or honesty, but the value of original thoughts.
• One reason people use language models is that they think the task does not matter.
• Another reason is the belief that the model will produce better writing than they can.
• A third case is writing with “skin in the game,” such as astroturfing, customer service chatbots, and recipe prologues.
• The author defines writing as a way to communicate original thoughts.
• Copying a human’s words is still different from publishing model output, because a model has no original thoughts.
• When something does not matter, using a model can waste everyone’s time.
• When something does matter, the model may produce superficial work that does not help the real task.
• In professional communication, model output often adds fluff, changes meaning, or invents details.
• For programming, the author cites Peter Naur and says code needs a theory behind it, not only an artifact.
• Vibe coding is described as producing code with no theory, which makes it practically useless.
• The essay includes a Gemini completion exercise that the author says is boring, generic, and mostly a rehash of the prompt.
• The final claim is that the author would rather read the prompt than the generated output, because creative work should share human experience.