Fighting last year's battle

Yeah it's just tough to predict how things play out in the short and long term. So you're willing to roll the dice if it won't impact national security that much if you're wrong. But if a terror group ends up on CNN in a video in front of a bunch of nukes...that's a huge miss for national security.

It increases the risk of attempting a regime change though. If nukes could get in the control of someone who will use them or disappear in the chaos, there's a lot more inherent risk to an op like this.

Cape Alava in ONP is one of my favorite places...the whole Olympic Coast is magical.

I'm not participating in whatever is going on over there...

Is this about meta and scale ai?

Are you on Mac? This sounds like something apple script could do for you.

Sqlite

Oh if anyone is reading this it's broken on Firefox on Android. I had to use chrome.

I tried it and it was completely unusable. Am I missing something?

Oh I hadn't heard that reporting. Really interesting then... Maybe hubris?

I don't really understand the assumption behind the question -- isn't Grok 3 a reasoning model and wouldn't have the same characteristics as a huge normal model?

I figure they just had to do a large pre training run to catch up with the other labs and then did the reasoning training on top.

Nevermind you were right

I'd actually be surprised by this one. It's one thing to release a new experimental feature to your most devoted fans.

But releasing gpt-4.5 to only pro users would be a vibe shift
"I can't use it so it's not real"
"Fake model"
Etc...

What do you mean by "in Paris" in this sentence?

Out of curiosity why are you deploying oauth for the tools site?

I'd even like some basic information from model providers on how to provide examples of tool calls in the prompt in a way that reflects how the model was trained to respond with tool calls.