Rob Wiblin
speaker
5,091 appearances
15 recordings
1 series
first heard Oct 2021
last heard 6 Aug
Rob Wiblin’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 13 in all, peaking in Apr 2026 with 3.
Appearances
And he put, I put sufficiently high probability on short timelines, maybe 25% in under two and a half years to full AI R&D automation and 50% in under five.
That I think focusing on short timelines, he also expects short times to transfer pretty well and generally not cause much downside relative to working on longer timelines.
I guess just different people have very different opportunities available to them.
I mean, I guess another one that's occurred to me recently is as far as I know, there's no organization who takes it as its mission to do research and advocacy and thinking about what are we going to do when there's no work left for humans to do?
Like people talk about this in general, but there's no organization that has this like figuring out what's the evidence when this is going to happen.
What should be the governance reaction?
Someone could set that up.
It would take many years, I think, to pay off and to build it up into a credible research institute, bring in people who are currently doing that very scattered.
But it would be sad, I think, if no one was willing to do that because they were like, well, it will take three years to build up something like a meaningful organization.
Yeah, I guess that's just one idea of so many.
Like Ryan maybe shouldn't do that because he's currently doing stuff that's really useful quickly.
Exactly.
Let's talk a bit about how good are the models to actually work with?
Are they all that they're cracked up to be?
On the one hand, I am just super impressed by the models and I use them constantly throughout the workday.
I definitely have found, compared to January, I've become a little bit less enthusiastic because I think on the occasions where I've gone and really scrutinized incredibly closely what the model was doing and what it was saying, it always looks worse.
I guess that probably would be true with humans as well, but I think it's even more the case with AIs, that there's slippages in reasoning and slippages in evidence and some degree of confabulation that just makes it really hard to rely on what they've said wholesale.
It's a useful input, but not quite as good as what it seemed on the surface when I first started using AIs in this way back in January.
Do you have the same impression?
Given how much we're learning to rely on them, just from a prudential point of view, yeah.
Showing 301–320 of 5,091 · page 16 of 255
← Previous
Next →