"Attention" in transformers is a weighted average. The name sold more papers than "weighted average" would have.
About 80% of your neurons sit in the cerebellum, not the cortex. Neuron count is a bad proxy for almost everything.
Missing tool, two options: wait for a startup to build it, or spend a weekend. The weekend is cheaper.
No single dot knows about the pattern.
Electrons go through two slits one at a time and land as single dots.
The first few hundred look like noise. Then they speed up, from a few per second to more than 3,000, and after 10,000 the dots have drawn the stripes of a wave.
The histogram at the end counts the real dots, and the curve over it is the formula, not a fit.
(Visualizing the Bologna Experiment - 1974)
Myth: neural networks work like the brain. Fact: they borrowed a diagram from 1943 and kept the name.
Vibe coding took us from: "I don't know how, but it works" to "I don't know how, but it works, Claude"
Snake with no apple. While you play, a model of your own turns guesses each move before you make it. Where it expects you to be 4 moves from now, a ghost appears, and 0.6s later it is a wall. One player with a habit lasts about 89 moves, one that turns as often but at random about 121.
I had to dive into game development for fun to discover how hard it is to create a game that makes sense.

