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
That's
I I do think you probably are gonna see some domain specialization.
I think uh my guess over the next twelve to twenty-four months is that you'd see some domain specialization on benchmark scores diverge because of how all these labs are starting to do the next evolution of training, which is they're using RL environments to generate synthetic COT traces, doing their sort of model trainings on that data.
And they're trying to go get it on a lot of different just different domains.
you know, I the O three, the original O three paper.
you know, I think was interesting are on the benchmark results where, you know, on this new sort of COT reasoning system, they had uh relatively high scores on math and coding.
Um, but r the the but the the the gap um or should say the step function increase in those scores that was much higher than the increase in like legal reasoning.
Which you would sort of maybe intuitively guess that or think suggest or uh expect that like legal reasoning would probably be one of the best like general domains for if you trained a reasoning model that was really good at math and coding, like it should be like and it's a language model that like that would directly transfer into like the legal domain because like, okay, it's symbolic reason you know, reasoning that's like self-consistent.
Um that wasn't the case.
So I think that's I I suspect that's what we'll see.
There, you know, that there's obviously the big scale news.
Um the thing that I'm seeing now is there's uh probably like I don't know, at le a handful that I know of uh these new uh startups and that have come up in the last several months, but are all getting founded to basically go build RL environments to generate synthetic or semi synthetic data uh and like selling them to to sort of the major labs or to the major frontier folks building these next gen systems.
Um I I think we're gonna see more of that.
I expect that's kind of what what's gonna drive a lot of areas.
Yeah, what do you what do you even more
was pretty good.
Uh I mean like look the the macro change here is uh is from a regime where like we're pre we're scaling and pre-training.
We want as much text, as much high quality label text as we can get our hands on to to scale these pr these foundational models into one where we're trying to train process models or you know make the foundation models really good for process thinking and COT generation.
Um that that is a complete shift in how you want to generate that data.
You want an RL environment that you can create lots and lots of COT traces, really very long traces over long running tasks as well.
Showing 481–500 of 927 · page 25 of 47
← Previous
Next →