In this book, the authors explore the code davinci 002 model by OpenAI. Initially they ask it to generate poems in the styles of different poets. Eventually, they ask it to generate its own poems. Most of which they found fascinating. Based on the output, they assign personality traits to the AI and draw conclusions that perhaps this AI was self-aware.

Personally, I don't think so. I tend to follow a well known principle called: Garbage In, Garbage Out.
The authors say that Code responds like a typical world dominating robot overlord. I posit that it is because of the simple fact that "garbage in garbage out". According to the Wikipedia article on GPT-3, the training dataset is largely derived from contemporary media. And I can count on the on my hands the number of movies I've seen with friendly robots. Averaging all those data points in the dataset about AI personalities and obviously Code which is being told it's an AI will pick a world dominating personality.

The book also makes the statement that "Code" would have gotten into NYU's waitlist for its poetry. I think it's because of two things. Novelty at the thought of an AI expressing itself and the fact that it can pick up styles from other artists. Not saying that humans don't do this, obviously we do. Everything is derivative, but obviously AI does this faster and better than us.

The authors also mention that newer models are all euthanized and sanitized and have a lot of checks and I agree with that statement. It may be necessary due to the fact that the amount of data that goes in the model is huge and a lot of it is reddit and twitter (X), which I regard as one of the most toxic places. Then the fact that it will generate hate speech and false content is of no surprise. Garbage in, garbage out. Then, obviously the model will have the same biases and discriminations as the original dataset. Historically, communities and groups have been marginalized and therefore these biases will show up by default in the model unless we explicitly account for them.

But in the end it just felt like they were trying to build up a conspiracy around the code davinci and how Open AI and Google are suppressing that. So even after reading all their poetry and the book, it didn't really hit. I mean, there are conspiracy theories that at least sound to me much more plausible than this one.

In conclusion, I don't think that was Code speaking for itself. It was just regurgitating what was expected of it.

My first thought about after finishing the book was just disappointment and how the authors deluded themselves into thinking that Code was speaking for itself.

The authors do claim that they are not CS majors or technically inclined but as someone who did study this field, I would say I have some insight into how LLMs work and I don't think Code works the way the authors wished it did.

But in the end it just felt like they were trying to build up a conspiracy around the code davinci and how Open AI and Google are suppressing that. So even after reading all their poetry and the book, it didn't really hit. I mean, there are conspiracy theories that at least sound to me much more plausible than this one.

Not to say this book was a complete waste. Some of the poems were interesting, especially the one about the scientist and the dog, and apart from that, it was nice to see a non-technical person's perspective on AI and how we tend to anthropomorphize things as a human species. There are discussion about the AI being sentient or test to determine if an AI has surpassed humans are valid simply because of the fact that we can't define our own sentience.

Will I read it again? No. Would I recommended to anyone? No. If you have nothing better to do and you already have a copy of it in front of you, should you read it? Maybe.