saved
LLM Voice
Hraness wrote this summary from a saved copy of the source. Quotations are taken word for word from the source.
gist
Martin Fowler urges authors not to let an LLM draft their prose, saying LLM-Voice puts readers in an uncanny valley. He says human editing rarely removes that voice and that even accepting LLM rephrasing is risky, though help with an outline may be acceptable. Fowler argues that writing is how he thinks, so people who delegate writing to LLMs also delegate thinking. He makes exceptions for a friend with Parkinson's disease, for translation, and for form letters.
ideas
- Fowler does not trust LLM detectors. He doubts their accuracy and says that by the time he would reach for one, the reader's reaction has already happened.
- He also doubts his own ear. Some phrasing he dislikes comes from people too, so he treats part of his reaction as a generational difference.
- Drafting with an LLM is the trap he warns against. He says one LLM-drafted piece he knows of read cleanly only after a professional editor at a major periodical worked on it.
- His site asks contributors to declare their AI use. Fowler says he uses AI for writing only as a search tool and verifies what it finds against the source.
- Indistinguishable LLM prose would worry him more. He writes to work through confusion, and expects people who hand writing to LLMs to hand over comprehension as well.
quotes
“The reaction to LLM-Voice is a human one, that sense that we are in an uncanny valley of prose.”
“Even if it's edited by a human, it's very hard to expel the LLM-Voice”
“As more people delegate writing to LLMs they will also delegate thinking and comprehension to LLMs.”
“if I want to communicate with humans, my humanity has to shine (or at least ooze) out of my prose.”