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

Okay, okay.

So that's a that's a weakness of them.

Um what's a what's a different control technique?

Okay.

Sorry.

So um I I guess then you get you get the weaker model that was auditing it to just to just another replacement action that will be no so so there's a few things you could do.

One

I see.

Okay.

And then I guess hypothetically, if it was randomizing so that it would like do it do an attack on you one every hundred thousand times, I guess if the as long as those actions would be flagged as sketchy, then it that whenever it tries to do that, it just gets resampled and goes back to some some non its standard non-sketchy action.

Okay.

Um have you tested this approach as well?

Okay.

Yeah.

Uh yeah.

Is it possible to quantitatively estimate like how how much difference it makes?

I guess all of these things are like slightly probably contrived tests, but

Okay.

So if you're doing this replacement strategy, then that forces the AI to initiate attacks far more often in order to have any chance of getting past.

Showing 141–160 of 1,787 · page 8 of 90