[kalu]

So this author trained a neural network billions of times using different hyper parameters ? How much dod that cost ?

[tpm]

"overnight on an A100" per https://twitter.com/jaschasd/status/1757056127991439832

[TheCoreh]

The networks trained were really small, with only one hidden layer, and a width of 16.

[mkl]

Very small networks are cheap and easy (even 20 years ago).