The problem is not websites, it's the entire network and their privileged access points ... including developer computers, a cesspit of intrusion vectors, pretty much each and every one.

First the regenerative agriculture people manage to create cows with internal fusion, now plants can do it too?

Ntrogen and carbon can come from the air, the rest not so much.

Giving out work visas for some vague supply gap was always a bad idea with perverse incentives. Compounding it with the PERM fiction for green cards, even worse.

Democrats should have implemented immigration registry reform while they could.

Nah, most people forget about it when they turn off the news. People are just not convinced the juice is worth the squeeze any more, marriage and kids are too much bother. Both birth rates and global warming are examples of human intelligence not being aligned with long term survival of human civilization.

Maybe AI can manage better ... as long as we don't manage to align it to our preferences, because as apparent those are fucked.

They are always gamed. The moment a "secret" test set is sent to a closed model, it's no longer secret. So every benchmark is compromised.

I suspect they even train adapters daily on github just for something hard to game like SWE-rebench. If I was launching a trillion dollar IPO I certainly would.

Since the courts fucked them any way, they took the reach around.

A nation can obviously make it traceable if it wants, it can also make it untraceable if it wants.

The app can just check your physical ID with NFC offline, set a flag "yep, older than 15" and be done with it.

"QR" code on screen (a RGB using code works better) mobile phone app sends confirmation with data from the code.

The Claude part wasn't autonomous anything. Just using Claude to help find an exploit in open source software to breach OpenAI due to shit network segmentation and login re-use (amateur hour).

It's awfully convenient they used Claude to throw shade at OpenAI.

Amazing the kind of PR campaigns a trillion dollar IPO can create.

Who is paying those independent evaluators?

Any way, best way to slow down at the moment is not to IPO.

That gives Anthropic the right to suspend service.

Thus its perfectly legal to train on your exfiltrated data.

PS. as long as you stay inside the Apple ecosystem I think you always need one device left registered to the account (directly or through recovery contact) to restore the backup. I don't think the customer service based account recovery (last ditch, prove identity through passport/etc) recovers the passkeys from icloud backup. Every ecosystem will differ.

With Android, iOS and MacOS they aren't on disk exactly. The database is encrypted with a hardware bound key and only opened inside a secure enclave. Windows Hello probably too.

I never expected Google and Apple to export to third party vaults, or the user hostile FIDO to create a standard to export to non manufacturer attested endpoints.