Seems like he must have really had some kind of innate physics ability to fall in love with gears like that. Just casually imagining a gear train with multiple

Seems like he must have really had some kind of innate physics ability to fall in love with gears like that. Just casually imagining a gear train with multiple parts in any meaningful way is not a concept that makes any sense to me. I can imagine a picture of something like a gear train, but I couldn't even hold a consistent picture of a bunch of gears in my mind long enough to reason about it. To be able to imagine a whole system with multiple parts is amazing to me. Computers make much more sense. The context window you need is tiny. Supposedly, it's all about precise specifications, but it's really far more about vague guesses than almost anything else. We think we understand everything, but it's generally accepted nearly all software has bugs, and that most software is never "done", and that it's likely understands the whole of large a modern program. Some HTML and CSS barely specifies anything, and what it does specify you might not even fully understand as a web dev. Nobody at all understands what ML is "thinking". Even more algorithmic stuff is rarely understood. It's more like an impressionist sketch, iteratively refined. If someone really knew every part and every detail, then random trivial bugs that almost resemble LLM hallucinations might be less common. To actually see gears in your head that have any meaning in reality is completely different. There's no external storage, let alone any external validation, the action is truly in the mind, not a collaboration where the machine is doing as much or more than the person.