Training a neural network to compress elementary functions into only two numbers (x,y) on an Archimedean spiral.
From those two numbers, the original function can be recovered. Each family of functions is eventually beautifully clustered in the embedded space of the spiral.
How to make Greece a neural network. Input enters from the north border. Each city is a neuron. Output from the south. Two bottleneck points make this an auto encoder. Map those two points onto the Greek coastline. You then have a Greek Inception Autoencoder. Somehow this thing trains. Neural networks with gradient descent can basically learn anything.
@synthwavedd Seems like the direction is breadth than depth cause depth starts slowing down. Just pick some new verticals that were ignored so far and hillclimb, citing huge improvements. More evidence that there’s no robust generalization.
@elonmusk@rauchg Prediction: this is in fact a lot harder than autonomous driving and won’t happen by end of 2027 unless we have an architectural breakthrough. The reason is the long tail.