Ep 76: Sora Creators Bill Peebles, Rohan Sahai & Thomas Dimson on Their Unexpected Viral Success

**Jacob Effron** (0:00)
I feel like I have the main characters of the AI ecosystem.

**Thomas Dimson** (0:02)
People are very creative. It's a social network. So social networks are about people and relationships.

**Rohan Sahai** (0:07)
Even if you change the ratio of creator to consumer by 1%, that is a massive impact on the world.

**Jacob Effron** (0:12)
Celebrities and rights holders are going through a very fast evolution of understanding this technology and how to use it.

**Bill Peebles** (0:17)
Shag is posting bangers, by the way. I think the first scientific breakthrough that comes as a result of simulating some phenomenon in video models will be an insane milestone, right?

**Jacob Effron** (0:26)
They have a prediction of when that might be.

**Bill Peebles** (0:27)
I would be shocked if we are not making breakthroughs like this by...

**Jacob Effron** (0:31)
Sora has taken the world by storm this past month. It's number one in the App Store, has spawned endless hilarious videos, and I had the privilege today of speaking with Bill, Rohan and Thomas, three of the main folks at Sora behind this magic at OpenAI. It was just a really fun conversation. I got to ask them everything that I've been thinking about that Sora spawned. We talked about their reactions to Ben Thompson's take on Sora, as well as kind of the key product decisions that factored into how they thought about making it into a social experience. We talked about the state of video models, how they progressed over the past years, and what's next, future milestones, including novel physics discoveries and their predictions for timelines on those. We also hit on how video models in general will fit into the broader OpenAI ecosystem, including ChatGPT over time. Just a ton of fun to get to talk with these three brilliant folks about all these things in the video ecosystem. I think folks will really enjoy it. Without further ado, here's the Sora team.
Guys, I'm so excited to do this. Thank you so much for coming in here.

**Bill Peebles** (1:26)
Thanks for having us.

**Jacob Effron** (1:27)
I feel like I have the main characters of the AI ecosystem. As a podcast host, it's a fun thing to get. So thank you, guys. I feel like a bunch of different things we'll want to hit on today. Maybe to start, I feel like there's been all these stories told of the ChatGPT launch and how it was not expected to be anything big and out of nowhere turned into this just massively used consumer product. I'm curious whether the same was true of Sora. Did you guys expect this reaction? Any fun stories from that first day?

**Rohan Sahai** (1:53)
I mean, I didn't expect to be number one in the App Store for like a month, but it's one of those things you look at in retrospect, like the research team absolutely cooked. We have a knack at opening the eye of creating these viral moments really well. And so, you know, happy to have done it, but I think it surpassed my expectations.

**Thomas Dimson** (2:12)
I think Bill also has a hopeless optimism about him.

**Rohan Sahai** (2:15)
Yeah.

**Thomas Dimson** (2:15)
It's not hopeless, actually.
I think it was like, there was a meme, I saw it the first, it was a couple of days after launch, where it was like number three in the App Store, and then Bill was like demanding number one. I think Sam was putting you in the cage or something. It's like until you were number one.

**Rohan Sahai** (2:31)
It's the one most like, 25K, willed it to life.

**Thomas Dimson** (2:38)
I've worked a lot of different social media products, Instagram, some successful, some failures within that ecosystem. I tend to be very grounded and realistic and all this stuff, but I have to say that optimism is actually very important. We ended up being number one on the App Store, which was wildly crazy for my expectations, but it's delightful to see.

**Jacob Effron** (3:00)
I imagine there's an aspect of this where you have to reserve some GPU capacity based on what you expect. I was wondering, how in the world do you figure that out?

**Bill Peebles** (3:08)
It's not really a science. It's a lot of... I think everyone feels compute constrained across the entire industry, not just with an OpenAI. Certainly, when we launch an entirely new product surface that is as compute-intensive as video, it requires some pain elsewhere in the company. I think one thing that's really nice about OpenAI and to Sam's credit, everyone feels invested and everybody's success across the board. Certainly, when ChatGPT releases amazing new visual generation features, we want to make sure those are great and people are having an awesome experience in chat and likewise, the company really feels invested in Sora success. People will see that we need to take these new bets and it's important for the company's longevity, not just to have a super successful LM facing consumer product, but also the winning one in video, and everyone's willing to chip in, do your part in wartime.