Astro Hacker News - Anthropic says a key AI safety alarm has been triggered, days before its IPO

vintagedave |next [-]

> "We made a determination that we have met this threshold," … It based the call on "what we expect from future models and our overall rate of progress, rather than on what we have observed from current models."

How can these two statements both be true?

cyanydeez |root |parent [-]

Because they're likely just parroting whatever their AI tells you.

These people are not in charge of their faculties; havne't developed the intelligence to rationalize their own actions; and lastly, are high on their own supplies.

That said, these models are as powerful as the security systems you bypass to test them. So the danger is more this is a private company well funded and could hold people hostage with demands.

denverllc |next |previous [-]

How can they plausibly reduce training spend without admitting they can’t finance it?

olyjohn |next |previous [-]

This shit is just getting old.

Havoc |next |previous [-]

> "We made a determination that we have met this threshold," the company wrote

We have made a determination that more hype is needed

hn993302 |root |parent [-]

Looks like "this weird trick feels illegal" marketing

> This one is triggered when AI is doing enough of Anthropic's research to roughly double how fast its models improve. The policy says that pace would pack two years' worth of AI progress, at its 2018-2024 rate, into a single year. The filing says AI "has already been accelerating our research and development." Crossing the line means more safeguards, not a shutdown.

cedws |root |parent [-]

If this were true why would they not simply train and keep the uber AI offline, and use it for themselves to improve safety of models given to the public?

It just makes no sense. If you physically airgap the machines it's not going to break free, put down the sci-fi books Anthropic.

fwn |root |parent [-]

It could superpersuade you to plug it into the network. Like.. It could just be very very charming.

Surac |next |previous [-]

All there models got dumber and slower days befor the Fable5.2 introduction

foweltschmerz |next |previous [-]

Ok, i like some of their products but they really need to get a new shtick.

q3k |next |previous [-]

Guys please buy our stock!!

unrented7977 |root |parent [-]

We're like, sooooo dangerous! It's scary! Give us money to make the scary AI stop!

ath3nd |previous [-]

[dead]