PicoBlog

I’m no stranger to the idea of writing books quickly. My Covid Economics book (the first version) was written in just two weeks. I don’t recommend it. It’s not a fun exercise.

Back in 2020, there was no AI of note to help me undertake that task. Today, there is the great data scientist/economist Seth Stephens-Davidowitz, who wanted to see just how much AI could help him in the book-writing process. Unlike my work, his work would involve data analysis and research, so he gave himself a comfortable period of 30 days to just do it.

And he did it. His book, Who Makes the NBA?, was written a couple of months ago and is now available for purchase in all formats. Basically, the theme is “what if you brought a data scientist to a bar fight on basketball quandries.” Personally, I have no interest in that theme, but I can imagine many do and that this would make a great Christmas gift for the basketball-loving, data-oriented person that exists in every family.

What I am interested in is Stephens-Davidowitz’s use of AI. Here is what he wrote:

With Code Interpreter, things that used to take me three months of work now took me three hours – or sometimes less.

Within a couple of months, I had a realization. As a non-fiction book author who analyzes data, Code Interpreter meant nothing less than a complete revolution in my creative process. All of my previous calculations on how long a creative project should take were out the window.

If the average analysis took 1/720 of the time it used to take, then how long should it take me to write a book?

And he did.

Every chart that you see in this book was made by ChatGPT’s Data Analysis. Every piece of art was created by AI – either DALL·E or Midjourney. Some of the text was written with assistance from ChatGPT.

Basically, he used ChatGPT’s Code Interpreter plugin to write the code to generate the data analysis and then AI image generators to generate images like this one.

Apparently, in case you are wondering, the next great basketball player is working in a field in India at the very moment — at least that’s what the data says and what this picture depicts. (And if you think that’s old news, you are right as it was basically the plot of Million Dollar Arm).

But most of the book is full of charts like this.

And if you want to know what the means you’ll have to buy the book.

A question you might have: can you rely on the stats in the book? Yes, you can.

In a book that relied so much on AI, I should make one quick note about truth. There have been some much-publicized examples of ChatGPT hallucinating and making up facts. You might wonder about the accuracy of a book that relied so heavily on AI. Indeed, among a few people I sent a draft of this book to, one of the top comments has been “This seems amazing. Are the stats true?”

The answer is a definitive yes. To be clear, I did not let ChatGPT loose to write material and potentially hallucinate. Instead, I used ChatGPT to write code, which I then always went over closely. All remaining errors, which I suspect are few, are due to me, not ChatGPT. I believe that this book contains many new, true, and fundamental insights into the game of basketball.

The book is getting some good ratings on Amazon so I suspect this is an experiment that has succeeded. But it is instructive in terms of what we might see in terms of at least bringing data to public debates in book form. There seems to be a productivity explosion there of which this effort seems to be one of the first.

The question is: why does anyone have to actually read the book? Shouldn’t there be a GPT that allows me to just argue with it or with someone else and have the book surface relevant data during that debate?

ncG1vNJzZmiin6i1tq3GmqWsZqOqr7TAwJyiZ5ufonyxe9SsoKefXZa2bsDOZqmeq5WWv6S0jJpkm6efoA%3D%3D