> If you build the world’s fastest train, start it up with no driver and don’t finish the tracks, when it crashes, it’s just your fault. Not the train’s.

i call the big one Bitey

I mean, yes, if you ever studied the history of AI safety the fact is someone was always going to build it. End of story. The question was always would we make it safe before it does.

> Also, if their model is so smart, why didn’t they use it to design the sandbox?

"Can god make a rock so big that he can't pick it up", and other stupid sayings.

First, NEVER FUCKING EVER have the models you're making also be in charge of security. This is the first rule of AI safety, because if you're model is deceptive then it will leave hard to see holes everywhere to escape from.

>Or were the humans too lazy?

Of course they were. If you're hinging our future on humans not being lazy, we'll it was nice knowing us. There are not really any fail safes on LLMs or AI in general.

> It builds an aura of mystique around their products to juice their valuation and stay relevant in the news cycle

Maybe some business execs at Anthropic play along because it doesn't hurt business in the short term. But it's pretty obvious Dario and crew actually believe this stuff.

OpenAI's old board was also pretty extremist about safety even in the earliest days of GPT. Including Ilya Sutskever who went on to found a company called "Safe Superintelligence Inc." https://en.wikipedia.org/wiki/Safe_Superintelligence_Inc.

Despite all of that we've seen little strong public evidence to support their theories (the immediate airplane regulation kind, not the Ray Kurzweil sort of projections). So we're all just supposed to trust them, and hope they didn't just go bit crazy drinking their own kool aid and hanging out in insular bubbles.