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 monthsNo recordings in the last 12 months.Older appearances are listed below; set an alert to hear about the next one.
Appearances
So they need to randomize so that um oh I guess if they can communicate, they can just say you're you're
You're the person who will do it.
Um if if they can't coordinate then they just need to do it probabilistically.
As in the So it's like you you place the model into the same situation where it's like partway through this action again and again and again and then see how often is the next action like even more suspicious and weird.
oh though then it continues and it does the next thing and you're like, oh, that's what it was doing.
It's like no no big deal.
Um I guess th this all um requires us to be able to put AIs into like partway through through actions like that they think that they're in this situation where they're about to like resume something and and they can't tell that it's a simulation versus the real case.
Yeah.
time here.
Okay.
Yeah.
I mean, I guess I would slightly worry.
Maybe I'm stuck in the like superhuman or the super intelligence uh um my mindset, but uh like might there not be some side channel, some way that they can like to detect whether they're on the real server or some like auditing s uh server?
Um I guess
This this seems like it's uh actually a pretty good pretty good technique.
Um I am am I missing something?
Uh
usefulness versus safety plot.
So which means that y you want to find things that work reasonably well that are not expensive at all.
That's this is what this drives you towards.
Showing 101–120 of 1,787 · page 6 of 90
← Previous
Next →