Rob Wiblin

speaker

1,787 appearances 3 recordings 1 series first heard Sep 2024 last heard May 2025

Rob Wiblin’s voice in public audio — every appearance, attributed to the second.

Trend

recordings per month · last 12 months

No recordings in the last 12 months.Older appearances are listed below; set an alert to hear about the next one.

Appearances

newest first · ▶ plays the moment

"The Cognitive Revolution" · AI Control: Using Untrusted Systems Safely with Buck Shlegeris of Redwood Research, from the 80,000 Hours Podcast · 4 May 2025 podcast

Well I y I mean you you th then I've like effectively sabotaged the the atta attack, which I really didn't want to.

And I guess and it might be a very rare attack, so I might have prevents the like very rare occasion which it might happen.

Okay.

The the monitor always has to uh

direction.

Okay.

Yeah, yeah, yeah.

I I guess and and again we think that we should be able to put them in situations that are indistinguishable between like one where we're setting them up versus not.

Okay.

Well that's the big problem, right?

then the second problem is But then we have to generate a realistic um like decoy thing.

So we have to ask the model to like to create a setup.

And that's like how do we assess whether the setup would be realistic to the

like they have some encoded message in the thing that you you ask it, I would like to test whether you are trustworthy.

Can you make a thing that would be indistinguishable between a decoy and a real thing to to you?

And it's like obviously you could just stick a code in there saying this is a decoy.

Okay.

Right, right, right.

Showing 181–200 of 1,787 · page 10 of 90