As a longtime publisher, I have spent many days (and nights!) editing books, but today feels different. I am editing a book-length manuscript that was largely drafted by an AI. It is turning out to be a fascinating task. Just as most human authors do, Marc Strassman’s OpenAI-powered author “Sybil” regularly presents a series of editorial conundrums that require careful consideration. Some are bog-standard errors that almost every human author makes: confusing hyphens and em dashes; coining dubious neologisms; using favorite shorthand phrases that, alas, do not match the style guides. Some are mannerisms that are all too familiar in human authors (Sybil mentions her affiliation with Stanford and Harvard at every opportunity). I am documenting these issues in footnotes signed --Ed., which creates an interesting sense of dialogue between human and machine.
3 Likes
OpenAI-powered manuscripts welcome answers most of these questions to the best of my current understanding.
1 Like
@m-a.schenk @NimbleBooksLLC As far as I know, GPT-3 generates novel text, unless you try communicating with it as if it were another person (i.e. chat), because then it will try replicating human speech. I could be wrong however. This understanding is from general interaction and research. A correction or confirmation would be appreciated.
Wow I just realized this post was made 4 months ago! Weird why it showed up on my feed