To uzu lim @cutezu_ about papers like “Neural Tangent Kernel”
To uzu lim @cutezu_ about papers like “Neural Tangent Kernel” Richard Collins: It makes a huge mistake, starting from random. Much better to start with an open, verified and verifiable, accessible, and trustworthy core of weights and beginnings. Don’t get hung up on “linear” and “non-linear”. Just require whatever methods people or AIs use to be
Read More »