Kyle Corbitt
speaker
658 appearances
1 recordings
1 series
first heard Oct 2025
last heard 16 Oct
Kyle Corbitt’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 Oct 2025 with 1.
Appearances
So you say, okay, this is the task I'm trying to achieve.
Here's four different runs of an H.
agent trying to achieve it, which of these did best and it it stack rank some.
And it turns out that works phenomenally well with GRPO.
Like way better than I expected, way better than you know, anyone who kind of like I talked to before we actually tried this expected.
Because it's sort of in in in the the LM Urginus judge, it it can have sort of like self-ground because it's it's just getting these relative ranks, right?
So it doesn't have to like have like a an omniscient view of like what good
So that has worked at basically everything we threw it at.
Um we've done it with a bunch of client projects, we've done a bunch of our own customers.
It basically just works.
Like it's basic like I I honestly kind of feel like the reward assignment problem is like fairly solved.
Yeah, w which is it's fantastic.
Just any LMS judge off the hook, like we've tried it with so many things.
Like so one of one of the results we published was we used Quen two point five fourteen B as the model we're training, and as the judge we used Quen two point five thirty-two B, which is like not I mean it's fine, but it's like not a it's it's much worse than any frontier model, right?
And even with that combination, we were able to get our our agent doing like state of the art better than any frontier model.
model on on the tasks we tried it on, even with like an extremely weak judge model.
So it's it really doesn't depend on having like a really great judge model um in practice.
So yeah, it's it's just like it's just not something we've had to worry about since then at all.
So that's kind of like checked off.
So that's sort of like got me like a significant increase in like, okay, this is actually something people can apply.
Showing 461–480 of 658 · page 24 of 33
← Previous
Next →