I Wrote a Book with GPT-3 AI in 24 Hours — And Got It Published

From Medium:

In early 2021 I signed up for the GPT-3 beta program to see how good it is. A few days later I had co-authored Aum Golly — a book of AI poems on humanity. A few months later it was published. This is what it means for writers and publishers.

On January 30, 2021, I realized I was the weak link.

I had been working with GPT-3, the autoregressive language model from OpenAI for 2 hours. I was tired. My creative juices were running low. We had maybe 5 poems ready — out of the 60 or so poems we needed for the book.

I stared at the blinking cursor. GPT-3 was patiently waiting for my input.

To finish the project in the 24 hours I had given myself, I realized I had to change the way I wrote. I had to lean more into GPT-3. Let it do the heavy lifting.

Let go of my ego.

And that’s when things started to get a lot easier.

. . . .

AI for writers: the hype and the reality

Every hype cycle someone says: “This time it’s different.”

Aum Golly, co-authored by GPT-3 and myself, was published in Finland in April 2021. GPT-3 came up with the themes, the title, and the 55 poems themselves.

Having seen what GPT-3, the latest in generative language models, can do, I too am inclined to say: “This time it’s different.”

GPT-3 has been hailed as the newest generation of language models capable of generating text that you can’t tell from something written by a human. For the first 5 minutes of using GPT-3 I was hyped: it really was eerily good, most of the outputs really could have been written by a human.

But what struck me most was how versatile GPT-3 was: it could summarize text, come up with title variants, write introductory paragraphs based on a title… and it could write poetry.

Link to the rest at Medium

4 thoughts on “I Wrote a Book with GPT-3 AI in 24 Hours — And Got It Published”

  1. Lots of modern poetry written solely by humans reads like random nonsense. It’s no big deal when a computer helps generate it.

    • Close to what I was about to say – the output might be indistinguishable from that produced by a human writer, but is it indistinguishable from that produced by a good human writer?

      • And how much is on the reader’s interpretation? In other words, how much “intelligence” is there on the AI “creator” side, and how much on the human reading it? I think modern art (painting, James Joyce, etc) offers some parallels.

Comments are closed.