Post

LLM also uses high school statistics and probability on top of linear algebra

bsky.app/profile/eric...

Don't we all

One thing I find funny of this “criticism” is that to perform matmul over 600gb of matrices in sub second speed is a modern wonder.

I heard an engineer who described AL as a bunch of linear functions they don't understand. And I said no, it's a lot of very non linear stuff that nonetheless seems to act "linear" in some sense anyway. The non-linearity is actually very important.