I asked ChatGPT about the Circle of Fifths, and it gave me a good general description. Then I asked it to illustrate (using DALL-E, also from OpenAI) and it produced this colorful monstrosity. This captures some superficial graphic similarities to other representations of the Circle of Fifths but completely misses on every level of detail … and it doesn’t know that anything is wrong.

To me this illustrates the vast current differences between how people think and what current generative models are doing. We tend to think it’s almost human-like until it does something like this.