The New York Times just cut ties with one of its book reviewers because it was discovered that he'd used an LLM to help him write a review.

The AI had lifted phrases directly out of a Guardian review of the same book. There's lots of talk in the culture right now about how it is and how it is not okay for writers to use AI. My thoughts...

An LLM is emphatically not a person. But as a rough guide, if a writer would ask another person to perform a task and we'd all think that was okay, then it is okay to ask an LLM to perform the same task.

If you were writing a book, you'd probably think it was okay to ask another person to do some research for you. You'd read the research, and that would inform the book you were writing. It is okay to ask an LLM to do research.

You'd probably think it was okay to talk to other people about the ideas in your book, and hear their POV. It is okay to talk to an LLM about the ideas in your book.

You'd think it was okay to show your first draft to another person, and get suggestions for changes that you can either accept or refuse. That person is called your 'editor'. It is okay to use an LLM in this way.

But you probably wouldn't say to your human research assistant or editor, 'okay, now you've given me research and ideas, if you could just write all that up into a chapter, I'll put it straight into my book'. Everyone would think that is absurd.

In the same way, it is not okay for a writer to ask an LLM to write the book, or parts of the book, for them.

This all seems pretty straightforward to me. But it feels as though the conversation around it right now is pretty tangled.