I think that the Nooscope https://nooscope.ai/ has probably the most legible explanation of how ai (as machine 'learning') works today. It is a sober look that really cuts through the hype and all the crap that people are talking about - 'dreaming', 'hallucination' etc, which is a propaganda by which bad ai art gets funded. The focus in the text is on perception and statistical modeling, and once you understand the argument, the whole field becomes clear. Of interest is also how they explain pattern recognition one the one and prediction on the other side as two aspects of the same process. I also like the word **nooscope** - an instrument to see and navigate the space of knowledge. This is somehow what I always wanted my software to do. Some really important diagrams:      [[Ai Sandbox]]