Fucking hell

For the first time I have a real appreciation of the scope of params, tokens, and scaling for compute costs

I had a very rough understanding but trying to parse exactly what loss is (quantified) and where the theoretical min is thought to exist, and what that gets you.

Just mind boggling.

But at least now I understand this thread and why even using every word spoken or written by a single human is effectively inconsequential for a model like this.

This graph just cost me two hours of my life.

People like living in places that aren’t total shitholes

go on the TV and say that Donald Trump rapes children. go do it now.

You look at *least* ten years younger.

When did the press know about the press’ pedophilia cover-ups and when did they know it?

I mean, you’re probably right. After seeing Natanz, maybe we should use more than one.

I’d have just left it at “is Lindsay Halligan?”

(Yes, outlook calendar search is atrocious.)

God no

Fair. I just setup specialty search macros.

Find by person, find by chain, find by category

Easy buttons to switch between inbox vs folder vs entire mailbox. All are single clicks.

Wait, what sucks about it? Not trying to dunk, I’m genuinely surprised.

To be fair, haven’t you also been wearing a tweed tie for like ten years now?

Look man, I didn’t examine them closely. I had a shitpost to make.