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 months
1 · Oct OctJan 26AprJulnow

Recordings per month over the last 12 months — 1 in all, peaking in Oct 2025 with 1.

Appearances

newest first · ▶ plays the moment
I would say so far, um, it seems like they don't
train well enough.
Like if you if you use um you know OpenAI's agent interface, it's like okay.
Or if you use the computer use products that that everybody's putting out, they're like okay, but like not reliable enough to like actually like let go do something interesting unsupervised in the world.
And I think if the eight you know if the environments they were training in were high enough fidelity, then they would be good enough.
In the same way that like coding agents can go much further because I think that in that case we do have environments
That are much higher fidelity because it's a much simpler environment in a lot of ways.
It's like it's a code base, it's like maybe running a web browser.
Like it's it's it's much easier to capture the full realistic environment in that context.
So for RL, like the whole way this works is is you know, it it has to sort of be getting feedback from the real environment.
So I don't I don't see a world where it's as simple as like, hey, you can, you know, there's there's like a CSV type approach.
I guess it's you you could encode anything as a CSV, but if you try hard enough.
Um for RL to work, you have to be looking at real runs, ideally, of your actual agent in its current state.
Across within an environment as real as possible.
So you have to look at actually and and like the data formats are actually super simple.
It's just like basically a list of chat completion messages.
It's effectively whatever tool calls.
Yeah, your a exactly yeah, it's whatever your agent will be seeing and doing when it's running.
So that the getting the data is not hard, but what's hard is like when you're doing one of these runs and your agent makes a tool call, okay, now that tool call has to connect, you know, somehow it's gonna
get data back from something and that data has to look like it will look in in real usage.
Showing 281–300 of 658 · page 15 of 33 ← Previous Next →