What Is ChatGPT Doing... and Why Does It Work?

I loved this quick little read. Its “typical time to read” is 3 hours on Amazon and it felt about that long. The first part of the book was a pretty thorough explanation of the inner workings of large language models. I’m not very well educated in the details, most of my experience is in directly interacting with artificial intelligence, not theory. A lot of it was over my head, but very interesting.

The second part seemed to be about the differences between ChatGPT and Wolfram|Alpha, one of Stephen’s longtime projects. I also don’t know much about this, but it’s very interesting. This reminded me that I had used his system a long time ago in high school, but not for very long. I should probably give the system another look.

At times it felt like he was trying to push this project, almost as if touting it better than AI, but this wasn’t the case. Instead, he makes an argument about the strengths and limitations of both systems, and how they can be integrated to create something bigger. I’d be surprised if nobody is already using the Wolfram|Alpha system interfacing with AI.

To top it all off, this book almost feels late. Dated. It was published March 10th, 2023. This experience is really fitting to me as it emphasizes how quickly things are moving. I’ve been using GPT-4 for over a year now (probably daily), and have since started using Claude 3 Opus. A year is not a long time, but so much has happened in the AI space since then.

I’d recommend this read to anyone who is curious about AI and isn’t scared of a little theory. It does get a bit heavy in places, but I liked his writing. It’s been a while since I have ready anything about AI so it’s nice to learn a bit more about this strange and magical topic.