Searching...
14 results for “personalization”
And then you're kind of delegating everything to the agent. You don't really know how things work versus for cursors. Like, if you prefer something minimal, I think it's fine. You can do that. But you can start digging into more things. You can custo
42:45 / 43:50
from transcript42:45 – 43:50
But I I I hope the day is not far when it's all on real time API. It's It's gonna happen at some point. Right. Right. Right. Right. And, actually, maybe that's a good segue into talking about model customization because I suspect that you have such a
38:03 / 39:17
from transcript38:03 – 39:17
you know, like, they keep their own data, and there's kind of no interest in that. Because because it feels to me like the next level of scaling, this is kind of where we're at. And so I'm just kinda curious how Yeah. So, I mean, maybe even, like, ta
20:13 / 21:26
from transcript20:13 – 21:26
So basically, everyone used to talk about fine tuning the in the context of SFT, supervised fine tuning, where you would have inputs and outputs for a model and the model would, learn from those input output pairs. But the main issue and supervised f
31:34 / 32:40
from transcript31:34 – 32:40
these are the buttons where they are. And then it's like a fixed thing. But rather it's like there's, like, shared concepts and shared mechanisms of the same thing. But I could say take different forms where you can kinda expose ways for people to cu
41:15 / 42:16
from transcript41:15 – 42:16
API. So so so I I've I've noticed that you are moving towards kinda more sophisticated use of things like, you know, like fine tuning, which, you know, in a way, you could read that as a bit of a capitulation that, like, you know, there is product sp
19:24 / 20:58
from transcript19:24 – 20:58
AI is gonna be just a part of the fabric of everything we do, but I think there are gonna be a lot of fine tune models because why would you not want, to to more specifically customize a model against a particular use case? And so I think you're gonn
58:26 / 59:39
from transcript58:26 – 59:39
it's definitely adding some more complexity, but there's a lot of things we can do to reduce that and simplify it. But there's always a hand hovering over the button to can we get rid of this? Like, can we shed weight? Like, can we get rid of the lik
18:43 / 19:55
from transcript18:43 – 19:55
I guess I'm curious. Was that something you guys were really optimizing for when you trained this one, or how did you think about that? I mean, yeah, definitely for cuss customizability and character consistency are things that we closely monitor dur
8:55 / 10:00
from transcript8:55 – 10:00
...personalization of it. And then, Oliver, I know you used to be at Adobe, and then there's also software where we used to manually edit things. How do you see the knobs evolve now on the model there versus what we used to do? Yeah. I mean, I think tha
9:43 / 10:59
from transcript9:43 – 10:59
and they make that decision based on a lot of research. What we do basically is say, well, we're gonna do all that, but you don't you don't just get to pick, like, option a, b, or c. Instead, you get to design exactly the strategy that you want that'
38:56 / 40:10
from transcript38:56 – 40:10
We see it as a tool in the toolset. So we generally, if we're working with a startup and we're advising them technically, we don't recommend starting with fine tuning. We say, you know, these models are very steerable, especially GPT five and some of
11:58 / 13:01
from transcript11:58 – 13:01
bring in your own custom data and create your own custom version of, you know, o three or o four mini or something, or or GPD five even, suited to their own needs. It's very attractive, but the the most recent development I think is very exciting is
38:59 / 40:09
from transcript38:59 – 40:09
And I think that will be true into the future as well. Yeah. Actually, I'm glad you mentioned fine tuning, because that reminds me of something. I heard that a sort of under discussed strength of OpenAI specifically was the fine tuning. That because
11:12 / 12:35
from transcript11:12 – 12:35
Have a podcast?
Run your podcast through Clypt
Get ranked clips, hooks, and ready-to-post copy from your own episodes. Free to try.