AI Race Intensifies: Video, Robotics, and User Adoption Drive Innovation

Key Quotes

"The biggest thing that's going on right now is that there's a leaked memo that The Information has reported on where Sam Altman has declared, quote-unquote, a 'code red' at OpenAI. You and I have talked about for the last couple of weeks--we took last week off--we talked about for the last couple of weeks before that how good Gemini 3 Pro has been, and Nana Banana Pro. Both of these are very good, and there's been some stories out there that ChatGPT, particularly, might be losing a tiny bit of market share."

The author highlights the internal reaction at OpenAI to Google's advancements with Gemini 3 Pro, indicating a potential shift in market share. This suggests that even established leaders in the AI space are feeling pressure from competitors, necessitating strategic responses. The mention of "Nana Banana Pro" alongside Gemini 3 Pro implies these are significant benchmarks driving this competitive response.

"I guess, welcome to the club, Sam. We're all because, I mean, Google's new Gemini is incredible. It's my default driver for coding between that and Claude. I go to Gemini now to have videos analyzed or papers analyzed. Like, I lean on Gemini more as a default. I still give OpenAI my $20 but there's, you know, you mentioned the Pulse product. Have you used the Pulse product, Kevin, outside of trying it for the show? No, because I don't really know what it's for. It's not for me right now."

The speaker expresses that Google's Gemini has become their preferred tool for coding and analysis, even though they still pay for OpenAI's services. This indicates a user preference shift towards Gemini for specific tasks, questioning the utility of OpenAI's less integrated features like "Pulse" for everyday use. The speaker's reliance on Gemini suggests its effectiveness in practical applications is surpassing ChatGPT for their needs.

"So my question is really about, actually, we'll get into Mark Chen, who's the chief research officer at OpenAI, in just a second here. He did a very good podcast interview that I want to talk about when it gets into like what's coming next for them. But I think the big question I have is, we've talked about this on the show for a couple, a couple months really, this idea of like, how do you open the door to more people using things for a longer time, right? I think one of the interesting things is there seems to be an audience of creatives that kind of goes to where the mainstream kind of platform is. And right now, I think Nana Banana Pro, and we're going to talk about some other fun Nana Banana Pro stuff that people have done, is really kind of owning that creative ecosystem."

The speaker poses a critical question about user retention and engagement in the AI space, particularly concerning creative professionals. They observe that "Nana Banana Pro" is currently dominating the creative ecosystem, suggesting it offers a more compelling platform for this user group. This highlights the challenge for AI companies to not only attract users but also to provide sustained value that keeps them engaged over time.

"So, um, to speak to Gemini 3 specifically, you know, it's a pretty good model. Um, and I think one thing we do is try to build consensus. You know, the benchmarks only tell you so much. Um, and just looking purely at the benchmarks, you know, we actually felt quite confident. Um, you know, we have models internally that, uh, perform at the level of Gemini 3, and we're pretty confident that we will release them soon, and we can release successor models that are even better."

Mark Chen, Chief Research Officer at OpenAI, acknowledges Gemini 3 as a strong model but expresses confidence in OpenAI's internal capabilities. Chen suggests that benchmarks alone do not tell the full story of AI model performance and that OpenAI possesses internal models comparable to or exceeding Gemini 3. This statement indicates OpenAI's strategic approach to development, focusing on future releases that they believe will surpass current industry leaders.

"The big thing, the very special thing, the big thing about Shallet Pete, and also what Mark says in that interview, is that one of the things that came out with Gemini 3 Pro is that there were some advances that, that Google made in pre-training, which there was this kind of overall sense in the AI space that pre-training might be over, that you weren't going to be able to get anything else extra out of making bigger, bigger models when you first train on them."

The speaker explains that Google's Gemini 3 Pro has demonstrated significant advances in the pre-training phase of AI model development. This challenges the prevailing notion in the AI community that further substantial improvements from pre-training were unlikely. The mention of "Shallet Pete" as a related development suggests it might be an internal OpenAI project or a concept being discussed in response to Google's progress in this area.

"Kling 01 was the big thing that came out on Monday, and what this is is their multimodal kind of editing model, right? So you can use it to, to edit things. It's pretty cool. It's pretty cool. And now the thing I will say is if you followed the AI space at all, you know, like over the weekend, a bunch of the AI influencers were out there saying like, ah, it's going to, everything's going to change again. It's all changing again. And this is good. I will not say like, when I played around with it, it was like a massive step change, but I'm not going to crap on it because there's a lot of really cool things that happened with it."

The speaker introduces Kling 01 as a new multimodal editing model, noting its release and the general excitement surrounding it in the AI community. While acknowledging its capabilities and the positive reception from influencers, the speaker tempers expectations by stating their personal experience did not reveal a "massive step change." This provides a balanced perspective, recognizing the model's potential while offering a grounded assessment of its immediate impact.

"So instead of just like printing pixel by pixel, it's like a printing press. It stamps the frame, so it can go really, really fast. Um, the math that they use, though, that's what I was looking for, normalizing flows. If you want to deep dive into that, that's like, that's the mathematical, that's a concept that this is based off of. But here's why it matters, because it doesn't necessarily need to process video the same way as these diffusion models. It can work forward and backwards. Um, it can go text-to-video output or video-to-video or image-to-video without having to be retrained for those approaches. It just handles it all the same way."

Kevin explains Apple's Starflow V video model, highlighting its novel approach that differs from traditional diffusion models. The speaker describes its "printing press" method