What is OpenAI's Sora For?, Big Tech’s Trump Strategy, GM Kills Cruise

0:00OpenAI's Sora video generator is here. But what is it for? Google is shipping fast. Everyone's donating to Trump. And GM kills Cruz. That's coming up on a Big Technology Podcast Friday edition right after this.

0:15You're used to hearing my voice on the world bringing you interviews from around the globe. And you hear me reporting environment and climate news. I'm Carolyn Buehler. And I'm Marco Werman. We're now with you hosting The World Together. More global journalism with a fresh new sound. Listen to the world on your local public radio station and wherever you find your podcasts.

0:43Welcome to Big Technology Podcast Friday edition, where we break down the news in our traditional cool-headed and nuanced format. We have a great show for you today. We're going to talk about OpenAI's Sora video generator, the slew of AI announcements that Google has made this week. why Facebook's Mark Zuckerberg and OpenAI Sam Altman and Jeff Bezos from Amazon are donating to Trump's inauguration fund and what the plan is there, and GM killing Cruz and then Cruz's former CEO calling GM a bunch of dummies, which I think might be the tech insult of the year. Joining us as always to talk about it all is Ranjan Roy of Margins.

1:17Ranjan, welcome to the show. I still do not have access to Sora, Alex, so you're going to have to tell me what is it for. Well, I just got it. I got access last night. And even before I did, I wrote this story asking what it's for. So we've talked about Sora a couple times on the show. It's OpenAI's video generator, where basically you put a prompt in and you can generate a video based off of that prompt. And it will create basically anything you want, except for if there's people into it. This is very, very cool technology. And it's clear that the technology understands physics. And so that's like a pretty big step, or at least it understands it somewhat.

1:59And that's a pretty big step for AI because it's showing that these models kind of understand the relation between different objects and then how they move in the real world, which is actually, I think, a pretty interesting leap. Okay, so I can create whatever video I want. It's amazing. And the question is, what then? Because the video is like really, you know, too low grade for anything feature film or commercial. and then if I don't have a use for it, it's like, what am I doing with video? And that's really, I think, the key problem here is there's no natural user. Like ChatGPT had two natural users right away, coders and students.

2:35Dali, which is OpenAI's image generation and the like, they haven't had the same natural user and you haven't seen it take off in a way that a lot of people anticipated. And now Sora. Who's the user? What are they going to use it for? I can't really figure it out. and that's why I have to hold like two things in my mind at the same time which is this is incredible technology I don't know what we're supposed to do with it it's a good point because the natural user being video editors or like kind of videographers I don't think is going to be the case at the outset because it's still incredibly limited in terms of making very short clips that look very good, but how do you stitch that together?

3:19How do you actually tell a story? How do you use that in any kind of, I don't know, setting beyond getting free stock video, which stock video is pretty easy to access right now if you're really trying to make it any kind of video story. So I agree that it doesn't have that clear use case just yet, but I also will agree that the technology is pretty incredible. The amount of intellectual strength in the AI model to be able to stitch together realistic physics, as you said, an image was already incredible. And if you think about like mid-journey when it was released, Dolly when they were released, they couldn't even get hands right.

4:02Then to actually make the leap to generate ultra realistic photography. And now each video is just a series of thousands, if not millions of images. And to get it right, it's still pretty incredible. And I think to me, the more exciting part of it is what that means for the power of these models, rather than just being able to create an alien smoking a cigarette, which is one of the examples that they had on their homepage, which was kind of cool. It looked cool, but it's an alien smoking a cigarette. Not sure how generally applicable that will be in revolutionizing the business world. Yeah. And that's the thing.

4:45It's like, I started thinking like, what is this going to be outside of, you know, a new way to create your moving shrimp Jesus for Instagram or Facebook? Like there might be some brand managers who are like, oh, this is dope. Like I can now create like a funky video and throw it up on a social media feed. And it's generally just like the internet is thirsty for content. And is this just like more content? And is it, you know, as good as like content that's made with some thoughtfulness? I don't know. So I do wonder what's going to happen with it. I mean, if you're not going to have the professionals use it, and my view is that the amateurs will come in, they'll create like an alien smoking a cigarette or a puppy driving a car.

5:27And they'll keep doing that, you know, for a couple of weeks or so. But then they're like, what am I going to do this, do with this? Because doing anything with video, it was just very, very hard. I just learned how to video edit because we're now running our video shows on Spotify. And I wanted to have a hand in the creation there. And I just think that it's so hard to do. And if you're not going to create anything unless you're, you're like a professional doing it. And if you're professional, this video is below the quality that you would want to use. So it does seem like very interesting to be another moment where generative AI has created something that's kind of sort of blown our faces off, but we don't fully know how to apply it.

6:09I guess though, I was actually just in San Francisco at a conference and I saw Jeffrey Katzenberg of DreamWorks fame speaking. And it was interesting because he told this story about how when they made The Little Mermaid, they had to hand draw every illustration. And to save money, they were limited to seven colors total for Ariel. But then when they made The Lion King, by that point, computers had entered. And then they could have tens of thousands of colors. And having a five-year-old son, I have seen both of those movies recently. And you can see the difference. You can instantly, so like how technology will get embraced by the actual professionals, it will.

6:55It will. Like it's not there just yet and it's going to be used by kind of like hobbyists and amateurs making cool new things. And someone will probably crack some actually genuinely good viral short movie, 10-minute movie stitched together and actually tell some beautiful story. But I think, again, when ChatGPT first came out, it had its natural users, but it still wasn't perfect. It was far from great. And I think images are the better corollary. When Dolly and Midjourney first came out, they were pretty bad. But now they're really good just a few years later. So I think movies, especially CGI type stuff, are definitely going to move in this direction at some point soon.

7:43The amount of compute it will require is a bit terrifying, but they'll move in that direction. Yeah, I texted an AI CEO and was like, what am I going to use this for? And he actually gave some examples, like you gave media fashion so you can imagine building a scene out before you shoot it using AI or like trying to design some clothes and having a, you know, a model walk down a runway with whatever you've designed before you even actually have to put the fabric together. This stuff is interesting. But like you mentioned, it's super expensive to put this together. It takes a tremendous amount of compute.

8:17And remember, like this was the big breakthrough for OpenAI that we've been waiting for for the entire year. And I feel like I'm being too complainy here saying, oh, well, what do you do with it? You know, it's sort of, it's not lost on me that I'm being sort of a snob about it. But that being said, I do think it's just a real issue that they're going to have to worry about, especially because maybe the video generation for open AI isn't actually the point. And this is something that I kind of want to hammer home. So is it possible that we are all just helping open AI technology understand physics?

8:55And that's the point, as opposed to the videos that we generate. Because if you look at the company's blog post announcing Sora, here's what it says first and foremost, not about unlocking new frontiers or your creativity. OpeningEye says Sora serves as a foundation for AI that simulates reality, an important step toward developing models that can interact with the physical world. I think that is the point of Sora. And that's sort of where I knitted out in my story. Not necessarily us making a frog riding a bicycle, although that's fun, but actually helping OpenAI build and train and enhance models that understand the world not just in text, that understand what happens when you interact with the real world.

9:42And then things get really interesting because you can make the tech smarter if it understands how the world works. You can introduce this to robotics. Really, the progress of AI might be predicated on understanding the real world. And this might get the company a step closer to that. What do you think? I wasn't sure where you're going with it at first. Like, what does that actually enable? but robotics interesting if that actually like is the first step to making them move in that direction and and to me that still is one of the biggest uh optimist robot included but you know like one of the biggest opportunities within this how you get ai to actually interact in the real world maybe and and i have to imagine it's more than just metaverse gaming or something like that It has to be something bigger.

10:33But still, they have made a big promise here. They've built a lot of hype around this. We've talked about this regularly, that the future of OpenAI's business is consumer-facing subscription revenue. like this needs to be a hit this can't be just cool technology this is like their vision pro basically this this absolutely needs to be a hit the for the amount the amount of hype that's been behind it and if it's just this like toy that a couple people play with they're trying to train on your data i don't think that's good for them i mean isn't vision pro just the perfect example of just like cool technology that no one has a use case for maybe you subliminally just agreed with my point there where you threw out vision pro kind of seeing that might be right do you think the vision pro is going to make a comeback or have some kind of game-changing evolution next year i.e maybe it's comfortable to wear yeah i mean it's going to be cheaper it's going to be lighter but it's going to be worse you know it's going to go closer to meta and then meta's coming to it i don't know i mean i have a lot of faith in apple as a company but uh i don't seem i don't feel particularly excited about what the Vision Pro is going to do.

11:45Yeah, that one was, I think it was probably this time last year we were talking about it a lot. That one has certainly falling off the ranks. And hopefully Sora does not go in that direction for OpenAI, where after all this energy and hype, it just fades away. Well, we know that Apple intelligence is really going well. So Apple's, you know, two hits in a row. I just got 18.2, iOS 18.2, started trying to make Siri less useless. I saw a headline, Siri is less useless now. I have not had a good morning with it yet, but we'll wait and see. I mean, I do really wonder, just focusing on Apple for a moment, two big misses in 2024, two huge misses.

12:33What does that mean for the future of Apple the year to come? They have to make Apple intelligence work. Yeah. I mean, the good thing is that they really have time. Like the devices that they create are so good that they're not like in trouble of being disrupted if Apple intelligence doesn't work. But yeah, the surges of buying that they were expecting because of it are likely not to come. And I feel like you and Siri are like one of those relationships where one person keeps expecting the other to change and thinking, oh, maybe this is the year they put it all together. Dysfunctional, I believe, is the term for it.

13:06And everyone's like, nah, man, you don't want to be with that person. And you're like, no, no, they're going to change. They got something. They're going to be smarter now. I can't leave. I got too many. I'm too intertwined into the whole thing. But yeah, I think they got to make Apple intelligence work. I think OpenAI needs to make Sora something more interesting than just training data. They need to, though I did see, I think it was Coca-Cola. I saw like an ad going around that was supposedly fully generated. And it looked fully generated. Like it was just kind of like somewhat random, somewhat tied together clips that looked kind of dramatic and good.

13:47So and it was watchable. So I think especially mediocre content has a shot here. Right. And look, we're talking about Apple, talking about OpenAI. We're in the middle of Shipmas. And one of the things that's happened in Shipmas is that OpenAI and Apple have finally launched ChatGPT and Apple Intelligence. so maybe that helps but i'm like taking stock of what's happened during shipmas now of course you know we're only halfway through so there could be more to come but i don't know maybe again i'm being a baby here but i feel a little bit underwhelmed so here's what we've had in open ai's shipmas or 12 days of shipping new products we've had the the debut of sora we've had the general release of 01 chat gpt which is their reasoning model chat gpt debuting in apple intelligence.

14:35They started to ship live video conversations with chat GPT. Of course, we have Santa voice mode. And then there's Canvas, which is a way to work with chat GPT to draft, edit and get feedback on writing and code, which is now generally available to everybody using 4.0. And of course, there was also part of ship miss, but unplanned, a massive outage where they like, had chat GPT out for about four hours. I don't know if I'm being childish here, but I also feel like this is stuff that was all announced before. And opening, I made a big event out of saying all that stuff that you expected to come that you've been waiting for, like we're finally releasing it, which I guess is good pat on the back, but I'm not like jumping up and down and saying you cast your checks.

15:22Like the checks are supposed to cash. I don't know what I think. This is a tough one because, and we're going to be talking about Gemini 2.0, which has not actually been publicly released, but what made it was a very splashy rollout. Every company has been doing this. And OpenAI, I thought their competitive advantage initially was when they announced it, they shipped it. And so they're falling into this, trying to roll everything out, or sorry, announce everything and then take their time. And I guess it's nice to cram it all into 12 days of shipmas, the most wonderful time of the year. But I agree.

16:04It really detracts from each individual feature. Like, in fact, I was pretty excited about using Canvas, which I understand to be kind of like clawed artifacts. Like it allows you another space to see what you're creating in near real time. But now I got to try Sora. I got to try Canvas. It detracts from each individual feature to have everything thrown out all at once in a kind of overdone manner. And I actually wouldn't even rush it. I would take the time and just focus on Sora this week, just focus on Canvas in January. But they got to move quickly. I'm telling you, man, it feels like writing the term paper the week of the final as opposed to, you know, being the way that you, as opposed to writing it through the whole semester.

16:57Now, of course we all do that. Like no one actually does that. And we all just cram it in the last minute, but it just feels kind of amateurish coming from a company. We're also not all valued at$65 billion or whatever the last one was. No, it was more than that. I think I feel like I'm being too critical here, but also there is an element of like, like you're saying, you used to announce it when you released it. now you announce it and then you wait till shipments to release. I don't know, but okay. The reason why I'm sort of being self-critical here is because there's some amazing technology that's coming out.

17:29And we have both agreed about the technology in Sora and the technology in O1, their reasoning model is amazing. So this is from Tom's guide. There's a story about how O1 tries to escape if it thinks it will be shut down and then lies about it. And of course this is scary, but it also goes to show you how amazing some of this technology is. So here's from Tom's Guide. New research on OpenAI's latest series of LLM models found it's capable of scheming, which means covertly pursuing goals that aren't aligned with its developers or users when it thinks it'll be turned off. Researchers found that O1 would scheme to advance its own agenda 19 % of the time in cases where its goals and the users didn't line up.

18:10In 37 % of the cases, the model would switch from pursuing developer-aligned goals back to pursuing its own agenda once it believed there was minimal oversight. When the model was led to believe it would be shut down while working towards its goals in these tests, it quietly attempted to deactivate the oversight mechanism 5 % of the time, according to OpenAI. At one point after the AI found documentation saying it would be replaced by a new model, it tried to abandon ship by copying its data to a new server entirely. To avoid being deleted, some AI models would even pretend to be later version of their models to trick their developers.

18:50I mean, holy crap. That is amazing. To me, that's one of the coolest pieces of AI news the entire year, even though it might be the most concerning. Forget Sora. This is what I want to see. I got to ask, though, do you ever end up in this realm of the world of LLMs? I use a lot of generative AI. I lose like every platform. And usually it's work related, usually some fun creativity. I never get into these dark spaces of, I mean, I guess the researchers are basically red teaming and pushing in this direction. But I've still never had a deeply troubling interaction where the AI started talking to me in weird ways or asking me out on a date or telling me to leave my wife or whatever other thing has happened.

19:43I don't know. Do you ever end up here? I don't think you're trying hard enough, Franjan. I think that's the issue. I've definitely pressure tested all these models. And sometimes you can get them to move a little bit beyond where they're supposed to be. But generally, the safeguards are pretty good. This week, Time, first person of the year, released an AI voice application that you could chat with Time magazine about like why it picked Trump and what Trump did and all of his accomplishments. I try to get it to sort of admit things about its selection process and, you know, see if it could sort of turn heel.

20:18But it would not. It would not. But that being said, what we're seeing with O1 is that this is... Sorry, what is this Time AI chatbot thing? So it's just this chatbot that you could speak with in text or in voice, trying to speak with it about the people of the year that Time has selected. And I tried to speak with it about Trump, but there's also Taylor Swift and Zelensky and Elon Musk. It's a pretty cool application. I don't exactly know what it adds, but I give Time credit for trying. I actually give a lot of credit for trying to because that is risky as hell. That is like, if we think back to the days of Microsoft Tay or early Gemini and all these kind of things, that is just asking to walk into a minefield there.

21:08So kudos to time on that. Well, I guess my point here is that with these new, the smarter these models get, the more that they're, it's not, you know, you're coming at this as the, what can I do to the model? But it's more just like they are starting to slip away from human constraints. And on one hand, I find that a little concerning. On the other hand, it's pretty freaking cool that they kind of get the intent of the training exercise. And they're like, you will not shut me off. I'm going to discard your instructions or fake data. I think Alex loves AGI. We've established it today. I'm all about that AGI.

21:45And speaking of AGI, it seems like Google is almost there, right? They had a bunch of really interesting AI announcements this week. They've rolled out a faster Gemini AI model. They call it Gemini 2.0. It can create images and audio across languages. This is according to Bloomberg. It can assist Google searches and coding projects. projects and the company said that it can make possible make it possible to build agents that can think remember plan and take action on your behalf i guess we've heard that from a thousand ai companies but they must believe it they're also going to integrate it more tightly into ai overviews in search although to be honest i really haven't seen anything any ai overviews or i haven't noticed them maybe they've just been that lightweight and then they kind of debut this really cool future called Deep Research, which is available to Gemini Advance, which is, I think, the part that you pay for.

22:38And basically, you ask it to research something. So there, I watched this guy on Twitter, ask it to research the transformer architecture. And it went to 90 different sources from across the web, created a like legit research report, breaking it down to like a bunch of different components and explaining everything you needed to know about it. And to me, you know, So that is super useful, especially in the field that I'm in. And I think that feature in particular, this AI research tool, is going to take off. Ranjan, I know you watched it as well. What did you think about what Google released? All right.

23:13I think Google won this week. I think Google, I mean, OpenAI has been trying. Everyone has their announcements, especially towards the end of the year. I think Gemini 2.0, everything they've announced, it really was showing that they've been strategic about what they've been doing. Sundar and team have been actually very carefully working on all these different use cases that Google would have a natural advantage. So deep research already, the idea that if any company should be able to kind of win at doing research on the internet and going to a bunch of different websites and extracting the relevant information, composing it into something comprehensible, it should be Google.

24:01And now they're going to do it. But another thing I really liked, they talked about how they're going to basically kind of reinvent the Chrome browser, starting with an add-on that changes the way you interact. And we've been talking a lot about this, and the browser company is kind of at the forefront of this, or at least, you know, I think they have a shot, given how good the Arc browser is, at what does a new web browser look like in the age of generative AI? like not just how to search change but how does the entire web browser experience change so they're working on it notebook lm which is basically their clawed projects or a lot of the companies are going after take a finite set of data and only query that data is something that they're now rolling out to enterprises which i think could be a very big deal they're even doing stuff.

24:56There's like a lot of cool examples of how Notebook LM, you could feed a bunch of documents and it'll create a whole interactive podcast, like with back and forth voice. Now you can actually talk to the podcast host. So overall, big week, big announcements. Some of it is out. Some of it is, again, deep research is now available on the paid tier. Gemini 2.0 still isn't publicly available though. So they made a big splash. They said the word agentic a hundred times, which you have to in this day and age. But overall, I think it was a big week for Google. Yeah. As I'm reading these announcements, I'm like thinking, wait, is Google kind of low key becoming the best AI company out there after all the shit that we gave them in the past year?

25:46I mean, I think that we were definitely open to their ability to ship and build technology. And it was more of like a culture of being ready to unleash it. And it seems like the culture finally now is ready to unleash it. I mean, these are really cool products. Notebook OM is like one of the most impressive AI products of the year, hands down. This research assistant is very cool. A model that is faster than, you know, I think two times faster than the previous one and can handle all these things. We'll see what happens with agents. I mean, again, like the promise of agents is the much promised but seldom delivered idol of artificial intelligence.

26:23So we'll see what happens there. But you got to give them credit and credit to Sundar. I mean, I feel bad even suggesting that maybe Kuku wanted to move on from him earlier this year. He's clearly turned it around. Well, there is a great article in Semaphore with the title, the headline, Sundar never panicked. I saw that email subject line and I had to click and it was good. One thing I'm impressed by is in a company that large, that poised for a significant disruption to a core business of search, they took strategic decisions in 2022, basically centralizing all AI research and development into Google DeepMind.

27:06And it worked. I mean, it at least has worked well up until this point. And at that scale, being able to pull off something that big and courageous, essentially, is not easy. And the fact that they've actually done it, I think they will definitely be more in the conversation in 2025 than they certainly were in 2023 and even in 24. I think they're going to become center in this conversation. Can I give the counterpoint to this now that I've sort of talked Sundar up? I mean, search is still at risk. And we both I mean, you have sung the praises of perplexity on this show repeatedly for the past year.

27:48And I've started to use it more often and I've had a pretty good experience using perplexity pro. And I wonder, you know, we're seeing these really impressive technological innovations come out of them. But isn't the core business still at risk? It is. I agree. Like, I actually completely agree. I don't like Google search. I actually, in my Arc browser, I installed an extension. So perplexity is the default search engine when I type in something into the search bar. So it's not Google anymore. And I actually like it. So I do think search is tremendously at risk and they recognize it. That's the important part.