An interesting article by Stephen Wolfram about how ChatGPT works under the hood (and AI in general). A funny attempt to explain several courses of machine learning from linear functions to transformers in one article.
What's interesting is that part of the GPT architecture is such that no one understands why it works, but they just found out that it does.
While reading the article, especially the part where it describes how our language and knowledge are encoded in a set of fingerprints sized 768x768, I was reminded of our attempts to send messages to extraterrestrial civilizations. Besides radio signals, we twice sent physical objects (discs with information). They mainly contained sounds/music/images. It would be very logical and much more informative to send the source code and weights of the ChatGPT model.
More about the message https://ru.wikipedia.org/wiki/METI
Article about GPT https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-doing-and-why-does-it-work/
