But what if you could outsource the writing of a book to an algorithm, and the book could be drafted within a matter of minutes?
At some point every week, when I get a moment in between meeting the regular publishing deadlines that allow me to make a living, I tell myself that I really should be working on another book.
This is the promise of the Generative Pre-Trained Transformer 3, an artificial intelligence application focused on language that has reached a level where, if given a prompt, it can use its access to existing information and self-training on English syntax and grammar to produce paragraphs or even pages that seem indistinguishable from a competent human writer.
GPT-3 works by having the algorithm predict the next word in sequence, essentially asking itself over and over, given what word was just put on the page, what word makes sense next. When given a prompt, this results in passages that have the appearance of sentences strung together in the service of expressing a larger idea.
This got me thinking. Doing a little back-of-the-envelope math, I write 600 words per week for this column, 52 weeks a year, times 10 years. That’s over 300,000 words of me writing about books and reading, just in this spot alone. Add in the rest of my shaggy dog oeuvre, and you’re looking at maybe a couple million published words straight from this noggin, probably more than enough for GPT-3 to capture my style.