Robert Wiblin

3 weeks ago | 80000hours.org | Robert Wiblin

There’s memes out there in the press that this was a big shift. I don’t think [that’s] the right way to be thinking about this situation…You’re taking the attorneys general out of their oversight position and replacing them with shareholders who may or may not have any power. …There’s still a lot of work to be done — and I think that work needs to be done by the board, and it needs to be done by the AGs, and it needs to be done by the public advocates.

1 month ago | 80000hours.org | Robert Wiblin

I do think that there is something to celebrate here. … But we’re far from done. … There’s an opportunity to actually really make this work — I think that’s an optimistic way of looking at it. Now that they have said the nonprofit will be in control, what we’re hoping for is to see details that will show us how the nonprofit will actually be in control. Like, the words are great… but what’s going to back that up?

1 month ago | 80000hours.org | Robert Wiblin

TranscriptIan Dunt: Britain can tend to be a bit lost in the opium dreams of its own past. Whenever you hear a prime minister talk about Downing Street, the thing they always talk about is the staircase with all of the portraits of the various prime ministers. And I get it: some of those prime ministers were people that were crucial in the history of liberal democracy, or some of them fought off Nazism. I get that that is obviously a powerful thing.

1 month ago | 80000hours.org | Robert Wiblin |Lukas Finnveden

Once we get to a world where it is technologically possible to replace those researchers with AI systems — which could just be fully obedient, instruction-following AI systems — then you could feasibly have a situation where there’s just one person at the top of the organisation that gives a command: “This is how I want the next AI system to be developed.” And then this army of loyal, obedient AIs will then do all of the technical work.

2 months ago | 80000hours.org | Robert Wiblin

Another thing you might do once you’ve caught your AI trying to escape is let the AI think it succeeded, and see what it does. Presumably when the AI got caught, it was in the midst of taking some actions that it thought would let it launch a rogue deployment. If your model knows a bunch of security vulnerabilities in your software, it might start deploying all those things. — Buck ShlegerisMost AI safety conversations centre on alignment: ensuring AI systems share our values and goals.