News for hackers. Stuff that matters.
Re:Intellect-2 Release: The First 32B Model Trained Through ... by refulgentis on Monday May 12, 2025 @11:35AM (#43964210)
> weights after the was done work have lower loss than the input weights

> (not sure about the magnitude of difficulty though)

Oh, my sweet summer child!

There are no comments on this comment yet.