Mike Knoop
speaker
927 appearances
4 recordings
1 series
first heard Mar 2025
last heard 18 Nov
Mike Knoop’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 — 1 in all, peaking in Nov 2025 with 1.
Appearances
We can have AI systems that can go try to do this stuff, have humans try to do it, and measure the difference and actually look.
And like I've always found that that's a going straight to the truth is like a much faster way to get to the frontier of knowledge.
Uh, if you really want to know what's true or not, you kind of just have to look for things like that that measure it as opposed to like, you know, relying on proxies like like other folks.
But yeah, I would say Marcus has generally been more right than wrong.
Uh and for what it's worth, I think actually the
Better lesson is somewhat under like uh I I think often gets misinterpreted, you know, makes a statement about search and learning as these general four methods of of scaling.
Yeah.
Um, but uh Sun also makes the key point in the in the paper that that hey, like these are the the the thing that we are actually applying search and learning on top of is an architecture that was invented by a human in the first place.
Mm-hmm.
The core idea of the thing that we are scaling finally came from.
From from a person, from a human.
And uh that's still to do true today, still.
And I think that is very inspiring, even in the current regime we find ourselves here in March 2025, where um yeah, I think we actually do need some idea changes.
I think we need some structural changes in terms of how the architecture, how the algorithms work here, um, in order to, you know, certainly beat something like RPJ2, the high degree of efficiency.
And yeah, there's gonna be a scaling component to it, but like uh don't miss that, like, ah, yes, there's actually.
actually an idea component too that uh often gets like kind of brushed over.
Well, a few individuals that I think are sort of doing really interesting work in and around um program synthesis, which is a sort of a th a parallel paradigm AI paradigm to deep learning.
Um actually don't think either is sufficient.
I think some merger of the two is what's necessary to get to AGI.
Top story for another day.
Showing 861–880 of 927 · page 44 of 47
← Previous
Next →