Evidence receipt / preference
Published · transcript-backedSuhail Doshi: preference
20 Dec 2023 Latent Space The AI-First Graphics Editor - with Suhail Doshi of Playground AI
“Maybe this is the best reason why it can be helpful to not be a researcher also sometimes like I'm not trained as a researcher, I don't have a PhD in anything AI related, for example. But I think if you care about products and you care about your users, then the most important thing that you want to figure out is like everyone has to acknowledge that Midjourney is very good.”
Source trail
Everything needed to verify it.
- Speaker
- Suhail Doshi
- Attribution
- Verified speaker
- Claim type
- preference
- Recorded
- 20 Dec 2023
- Publisher
- Latent Space
Transcript context
…You also released a new benchmark called MJHQ30K for automatic evaluation of a model's aesthetic quality. I have one question. The data set that you use for the benchmark is from Midjourney. Yes. You have 10 categories. How do you think about the Playground model, Midjourney, like, are you competitors? There are a lot of people, a lot of people in research, they like to compare themselves to something they know they can beat, right? Maybe this is the best reason why it can be helpful to not be a researcher also sometimes like I'm not trained as a researcher, I don't have a PhD in anything AI related, for example. But I think if you care about products and you care about your users, then the most important thing that you want to figure out is like everyone has to acknowledge that Midjourney is very good. They are the best at this thing. I'm happy to admit that. I have no problem admitting that. Just easy. It's very visual to tell. So I think it's incumbent on us to try to compare ourselves to the thing that's best, even if we lose, even if we're not the best. At some point, if we are able to surpass Midjourney, then we only have ourselves to compare ourselves to. But on First Blush, I think it's worth comparing yourself to maybe the best thing and try to find like a really fair way of doing that. So I think more people should try to do that. I definitely don't think you should be kind of comparing yourself on like some Google model or some old SD, Stable Diffusion model and be like, look, we beat Stable Diffusion 1.5. I think users ultimately want care, how close are you getting to the thing that people mostly agree with? So we put out that benchmark for no other reason to say like, this seems like a worthy thing for us to at least try, for people to try to get to. And then if we surpass it, great, we'll come up with another one. Yeah, no, that's awesome. And you killed Stable Diffusion Excel and everything. In the benchmark chart, it says Playground V2 1024 pixel dash aesthetic. Do you have kind of like, yeah, style fine tunes or like what's the dash aesthetic for?…
Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.