François CholletVerified account

The most remarkable thing about the brain is that it wasn't designed. As a result, how the brain works is largely an irrelevant question.

Brain architecture is the result of an optimization process. So what really matters for reproducibility is the optimization problem itself.

The hypothesis space, the objective function(s). Everything else is noise.

Essentially: it doesn't matter what the answer is, bc one can mechanically derive it once the question has been asked. What matters is the Q

unless running the optimization again takes 1B years of evolution, in which case inspecting a solution might be a decent shortcut.