Olivier Godement: belief

Transcript context

…The model distillation and evals is definitely, like, the most interesting. Moving away from just being a model provider to being a platform provider. How should people think about being the source of truth? Like, do you want OpenAI to be, like, the system of record of all the prompting? Because people sometimes store it in, like, different data sources. And then, is that going to be the same as the models evolve? So you don't have to worry about, you know, refactoring the data, like, things like that, or like future model structures. The vision is if you want to be a source of truth, you have to earn it, right? Like, we're not going to force people, like, to pass us data. There is no value prop, like, you know, for us to store the data. The vision here is at the moment, like, most developers, like, use like a one size fits all model, like be off the shelf, like GP40 essentially. The vision we have is fast forward a couple of years. I think, like, most developers will essentially, like, have a. An automated, continuous, fine tuned model. The more, like, you use the model, the more data you pass to the model provider, like, the model is automatically, like, fine tuned, evaluated against some eval sets, and essentially, like, you don't have to every month, when there is a new snapshot, like, you know, to go online and, you know, try a few new things. That's a direction. We are pretty far away from it. But I think, like, that evaluation and decision product are essentially a first good step in that direction. It's like, hey, it's you. I set it by that direction, and you give us the evaluation data. We can actually log your completion data and start to do some automation on your behalf. And then you can do evals for free if you share data with OpenAI. How should people think about when it's worth it, when it's not? Sometimes people get overly protective of their data when it's actually not that useful. But how should developers think about when it's right to do it, when not, or…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.