Pattern comments on Google’s new 540 billion parameter language model

[Pattern]

I think that higher precision isn’t always needed (or used efficiently).