I knew inference compute could be expensive but this is crazy! (/s)

One issue with reliability is that you have to specify a reference class. 50% reliability at a 2hr time horizon means "if all you know about the task is that it takes humans 2 hours (and is in distribution), there's a 50% chance the model will succeed." But if you need

To the ~hundred of you who just followed me hoping for a lonely campaign to attract attention to AI safety, someone like Rob Wiblin is probably a better bet. But if you want a sober analysis of data and research related to AI capabilities and risks, I have lots in store!

If I was a hacker, I'd use a model with a lower refusal rate like GPT-5, and I'd get zero data retention (afaict this isn't hard, though you're still subject to the UserSafety classifier). I wonder if this is the tip of the iceberg, the only group sloppy enough to be caught.

Ultimately nearly all jobs will be automated, but here are 7 categories I think will stick around longest: 1) Anything involving complex physical labor that's sometimes unpredictable: construction, bodyguards, archeologists. 2) Overseeing the output of AIs and preventing AI

"no contractor, supplier, or partner that does business with the United States military may conduct any commercial activity with Anthropic" I'm confused does AWS needs to pick between working with Anthropic or DoW? That's broader than what 'supply chain risk' usually means.

It's interesting how disconnected the 4o-ers seem from reality. They post AI-generated protest pics, but I've seen 0 pics with real humans at the protest. They've also posted a bunch saying various kinds of uproar around DoW stuff are actually people being mad about 4o.