Amazon’s New AI Model, Bitcoin $100,000, TikTok 'Ban' Upheld

0:00Amazon releases a brand new AI model. OpenAI is going to ship daily for two straight weeks. Bitcoin hits$100, US courts deal TikTok, another setback, and Intel CEO CEO Pat Gelsinger is out. All that and more is coming up on a Big Technology Podcast Friday edition right after this.

0:22You'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 Beeler. 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.

0:50Welcome to Big Technology Podcast Friday edition, where we break down the news in our traditional cool-headed and nuanced format. We have so much to speak with you about today, ranging from the new AI announcements that Amazon has made as if we haven't done enough here on the show with AWS CEO Matt Garman coming on on Wednesday, but there's more to discuss. We also have news at OpenAI, which is shipping like crazy. Google, which has a new video model. and we're not just going to do AI today because we're going to talk about Bitcoin hitting$100 ,000. U.S. courts telling TikTok it's got to ban. You got to ban TikTok.

1:24And Intel CEO Pat Gelsinger, who's been on the show, is out. So joining us, as always, to speak about this stuff on Fridays is Ronjan Roy of Margins. Ronjan, great to see you. It's the most wonderful time of the year. 12 days of shipmas, nonstop shipping from OpenAI. eye. That's what I looked forward to every year in December. You know that this was like the roadmap that was supposed to come out in October, November, and December, but they just couldn't figure it out. And because it was going to go into the holidays, like they're literally shipping Christmas week, I think. They decided, okay, we have to roll this out in 12 straight days.

2:03I think every company should do that. Just everything that you've been delaying for the entire year just shove it all into december and call it 12 days of something yeah i definitely i thought i would thought about that that big technology that there was some big stuff that i was hoping for you know to get in before the end of 2024 but my ship miss is gonna happen in january so stay tuned folks right that's how it goes and then you just push past thanksgiving of the next year and then circle back a lot of circle back in 2025 it's gonna be a big year We'll circle back in 2025 on that. So speaking of circling back, I've had Matt Garman on the show, the CEO of AWS, on Wednesday.

2:44And I'm on the flight over to Vegas, and I'm thinking, what do I want to start with? And I couldn't stop thinking about our discussion about how Amazon is the best scaler of compute in the world. They should have their own foundational model. So I spent like the first 10 minutes of the interview with Garman talking about this idea. and he comes back and says, listen, like, we just want to be the place where if you want to build with AI, you'll find whatever models. And he says, it's a mistaken idea to think that there's going to be one model to rule them all. And says, you know, we wouldn't rule out the idea of having a big foundational model at some point, but basically stops there.

3:20Clearly there was something else going on because the next day Amazon releases a new foundational model called Nova that they say beats some of the state-of-the-art models on a number of different factors. And the company is now back in the game. You know, it had its own model that everybody said was not so great. It's called Titan, brings in Nova. And so now Amazon is using that compute power to try to get its own foundational models in the game. Ranjan, you and I spoke about this last week. So I'm curious from your perspective. Do you think this makes sense from Amazon? Were you surprised? I'm curious what you think about this sequence of events.

4:00Well, first, I want to give credit to Matt Garman, because I'm always impressed by any kind of executive who has the messaging discipline to sit there in an interview and kind of brush off a very direct question, knowing and having been worked on something this big for months and months naturally, and then knowing the announcement is coming out the next day. But in terms of what it actually means for it, you didn't get them, Alex, you didn't get them. But I mean, look, I think it shows the questioning was in the right place. So I appreciate you helping me get there. The questioning was in the right place.

4:38And again, for last week, I think it was, we were talking about how maybe the future of the battle around foundation models is not around the actual quality of the model, but the actual ability to scale the compute. And the companies that have the edge in that should actually be the ones best position to deploy foundation models to more customers. And AWS is the most natural player in that. So I think this is big. I think this is huge because Amazon really almost weirdly has been completely left out of this discussion other than its investments in Anthropic. And they had what had maybe nicely been positioning themselves as this kind of Switzerland of building with foundation models.

5:25And now they're saying, no, we have our own. And I think it's the right decision. I think they had to at some point and they're doing it now. So I think this brings them back into the conversation in a way they have not been in a long time. Yeah. Just putting those compute resources to use. And I think they really relied on their natural strength here in one specific area, which is that they've made these models much cheaper than basically everything else. Just to give you an example, if you want to use Claude Haiku 3.5, the cents per million token input is 80 cents and output is$4. For Amazon Nova Micro, which is their smallest model, it's 3.5 cents input and 14 cents output now that's in comparison to Anthropics smallest model so actually you look at the cost savings there and it's massive and also the reporting that I did when I was out at reInvent this week in Vegas showed that these models are actually faster as well so it sort of goes to this whole like what happens when you want to scale what happens when you want to build.

6:39Some of these models might work well, but they're just very expensive to run. They're slow. And that's the play that I think Amazon is making. Basically saying, get your product going, make sure you can swap models in and out. And once you get there, we're going to create a model that's both good and will save you from some of this sort of subprime AI crisis moments where you don't know how you're going to afford it and the user experience isn't quite there. Oh my God, I kind of love that because that's the most Amazonian playbook in history. It's like, no, I mean, actually that idea that come over, you're already on AWS probably, build, use everyone's models.

7:21And then once you figure out what you want to do and you want to scale it, then you want to come to Nova. We're the cheapest. We're going to undercut on price and still shipping speed everything. I like that. I think that could be, that would be a brilliant way to approach this. I also, it is kind of interesting that like every big tech company seems to be heading into this battle in their own unique way. Again, meta without open sourcing Lama, Amazon, if they're going to go with the scale and speed and cost like equation, Google integrating into their other overall ecosystem. So I think it actually, this has me even more excited about the big tech foundation model battles of 2025.

8:06Wait, you missed one. What's the Microsoft strategy? What is the Microsoft strategy relative to, I guess, just integrate into the ecosystem? I mean, Azure, Office 365, which is kind of, I guess it would be the equivalent of the Google strategy as well. just build directly into the overall ecosystem that's both on the actual developer side and on the enterprise side as well. Yes. And I think, you know, sort of like a question that I had an answer in mind. And that's basically it was rely on OpenAI till OpenAI got messy and then it was reset. But there was an interesting headline that I saw this week and we don't spend too much time on it.

8:48But that is that OpenAI is from Financial Times. OpenAI is discussing ditching a clause that cuts off Microsoft's access to open AI tech if open AI develops AGI, right? Remember, Microsoft only put in just about a billion in open AI's last round, letting others lead it after being like open AI's closest partner for most of its existence. And part of it must have been the fact that if open AI decides that it's reached AGI, Microsoft can't benefit from that technology. And so it's interesting to me, we already have open AI moving towards a for-profit and if OpenAI sort of restricts this clause or strikes this clause that says technology cannot be used by Microsoft if it hits AGI, then maybe Microsoft and OpenAI can get a little bit closer and less messy.

9:36I mean, didn't we already establish a few weeks ago, though, OpenAI? All they have to do is say they've hit AGI and then it somehow removes them from a lot of contractual stipulations with Microsoft. So just say the words AGI and everything goes away and then they're free to do what they want. But yeah, I think, no, I think that's the right approach. Or I mean, yeah, you could either check the magic acronym or just say, or say, all right, we need more money for Microsoft because remember, they're going to need more money in a year and just strike the clause. It does make me think like, are all the big to how open ai is going to fit into this overall like in 2025 what it looks like because now again if the idea really is and i mean having built on these systems myself open ai is a really nice way to get started in general and you can quickly get in in the direction of like solutions that you actually are looking for but it's expensive it doesn't scale amazingly They don't have a very well-developed customer success or enterprise function.

10:47And it's not built like a big enterprise-serving business. So really, it could make sense that go out and experiment, build your POCs on OpenAI or Anthropic or whatever else, and then come to us and we will be the place you actually spend all your money. Okay. And one of the limitations that we've seen with OpenAI is they haven't had that enterprise side of the business. And in fact, one of the things that really stood out to me as I was researching AWS is OpenAI still isn't on AWS as a way for enterprises to build with OpenAI through their services. And to me, one of the standout, if not the standout remarks from Garmin in that interview was he said, we are open to having OpenAI on AWS if that's what our customers want.

11:34So that's the overture from Amazon. And I think OpenAI probably has a contractual obligation not to move away from Azure and Microsoft. But that's one of my predictions, I think, for the coming year. And I'm honing in a little bit on what I think is going to happen in 2025. But I do think hearing the way that Garmin spoke about it, OpenAI and AWS are going to make a deal next year because OpenAI needs that enterprise customer. You heard it here. OpenAI is going to be on AWS. Alex Kantrowit's predictions. So speaking of OpenAI, we sort of tease it at the front. They're doing this thing called Shipmas where they're releasing a new product every day for 12 weekdays, I believe.

12:13and this is going to include, according to reports from The Verge, Sora, and a new reasoning model. They've also just made their reasoning model 01 generally available. What do you think about Shipmas? Is this kind of like a do-or-die moment for OpenAI? I don't think it's a do-or-die moment. I think it's, as you said at the beginning of the show, maybe it's just them catching up. Sora was supposed to be out a long time ago. And let's also note one of the items that's anticipated is a Santa-inspired voice for ChatGPT. So let's not call that some massive piece of shipping. That was they were like, basically, we need a 12th day.

12:53What can we do? And they're like around the whiteboard, and someone's like, Santa, GPT? And they're like, Santa, GPT. They're like, we got 11. Just one more. Someone give us something. Santa. And I think the idea also with ChatGPT Pro was another, I think that was the day one of Shipmas,$200 a month for advanced compute access and access to the O1 reasoning model. I think it's smart. I think it's basically kind of a super fan strategy for people who are just really, really big fans of open AI, ready to spend a little bit of money giving them access. And I actually think it's a better rollout strategy.

13:41And we were going to talk about Google's video models. A lot of these companies, the way they roll out product is either in some like exclusive limited preview that no one really knows who has access. There's some examples that, you know, get circulated on X and that's all we see. And then they have, or they have a big demo and we just really don't know what it's capable of and who's using it versus if you're willing to pay$200 a month, you get access and just almost starting to really test the price point that people are willing to be comfortable to actually access these technologies. So I'm a fan of ChatGPT Pro.

14:23I'm not going to subscribe, but I'm interested. So do you think that this$200 subscription, which is going to give you unlimited access to things like the voice interface, O1 reasoning model, do you think it's going to appeal to a large enough base to be meaningful? Well, you know what? Here's my theory on what the strategy is here. I think no consumer in their right mind is going to pay$200 a month. So maybe this is kind of the backdoor enterprise strategy of OpenAI, rather than having some kind of like heavy contractual enterprise negotiation involving salespeople. It's basically like$200 a month is kind of that you can put it on your corporate expense card, you can start to get it in more into actual larger enterprise use cases without having to, again, avoiding the traditional sales process.

15:21So I think they're sneakily entering enterprise in a way that doesn't go through, that's different than the rest. And remember we saw that chart that said ChatGPT is going to be the bulk of their revenue, and maybe this is part of it. Yeah, yeah, yeah. you're making me bullish on open AI again here. Just selling here. 12 days of shipments. I think they have a lot of opportunity ahead of them. And maybe part of that is Sora because the rumors do say that Sora is going to be part of this 12 days of shipments or they're calling it 12 days of open AI. I think internally they call it shipments. Are you excited for Sora?

16:01Are you looking forward to it? Because here's the thing. The market is filled with video generators. I know you've used a bunch. And so what do you think about Sora and where it might fit? I'm actually very excited because I've been testing like generative video for a while now. And Sora is supposed to be good. And it's in such a weird area because like you're not making anything extended. You're making short clips. It's gotten better and better. there's already there's like a coca-cola ad i believe that was supposedly completely made with generative ai and i actually believe it was like if you remember like a year or two ago people would say actually especially like two years ago people would say this ad was made with ai and in reality it was like they went to chat gbt and created a script with open ai and then probably had like very heavy production whereas now you can actually see it's like a bunch of disparate clips kind of interspin, like spliced together with some narrative underlying it.

17:05And it's not bad for like a V1, not even a V1, but a V like 0.001. It's not bad. So I think if Sora actually the same way ChatGPT brought generative chat to the world, if it opens this to more and more people to just start being creative, then I think we start to see like what is actually, even get a glimpse of the possibility of what's going to happen. No one is going to be making like blockbuster movies tomorrow, but I think we'll start to see some really creative stuff done if this is actually easily accessible. And the video quality of the stuff that's already out there is excellent. So Google, for instance, it made its largest or its latest generative AI model, VO, available for businesses.

17:53This is through its cloud offering. And this is according to Verge. Vio is capable of generating high-quality 1080p resolution videos in a range of different visual and cinematic styles from text or image-based prompts. And on the page where they released this, they had a few examples of the videos they've created. I wouldn't say that they're, you know, mind-blowing, but they're good enough, right? And I think that if you could have some consistency in scenes, right? Because I think that's the big problem with this stuff is you can ask it to create a scene and then ask it to create the next scene, but the next scene looks nothing like the first scene, so you can't really stitch them together in any coherent way.

18:30So if it gets to that point, then you're talking about a tool that can be very, very useful to folks. Yeah, the Google one, I think actually the most interesting part, so their model VO that they've announced, and again, it's in limited preview. You can generate a photo with Imogen 3, their generative image model. there's like a image of a puppy that's completely generated and then you can make it move and it actually works but if you think if you get can make it from that image then you can actually get a consistent character across a number of different videos and you can start to have some fun you can start to do something kind of interesting so to me with video i want to play with this stuff i want access if they're really shipping this soon i'm very excited because Because I've been skeptical on a lot of this that like when all the Sora demo videos, even the Sora videos that quote unquote leaked that were all beautiful, like until I can sit down, put in a prompt and see something very good, I'm skeptical.

19:36I think we should get there soon, but I have not said like actually been able to see the results in real life in real time myself. I will say I saw a very convincing demo of video AI this week, and I'm not going to name the company. I don't want to pick on them, but I tried it myself this morning, tried to make a video of two guys podcasting about technology as a cartoon. I think I tried it an hour and a half ago, and it's still rendering. So I think the big issue here is the time that it takes to make these things. Load time is a whole other thing. Like how these companies are going to actually try to make it a usable experience.

20:13And again, as consumers, we are spoiled beyond belief. That's true. The fact that like Amazon can't even add generative AI to Alexa because we want our answer in 0.2 seconds. I think in the latency, like we're spoiled. Well, whose fault is that? I mean, the tech companies just keep doing things that make us take it for granted. You made it too good, Bezos. He had a great interview this week at the Dealbook conference. I don't know if you watched that. It was excellent. No, what did he say? He talked a lot about his ownership of the Washington Post. He talked about the fact that he's spending a lot of time at Amazon now, and 95 % of that is on AI.

20:51So I do wonder if he watched the Garmin interview. Oh, he's back at Amazon? Yeah, he's back there. I don't think he actually ever left, but he's been prioritizing other things. His space vision is pretty interesting, where he is basically thinking of it as a way, instead of escaping Earth, to basically take a lot of industry and move it off Earth. So he's saying you do creative stuff like manufacturing on Earth, any heavy manufacturing, heavy polluting things, you do it off Earth and then bring it back via low cost spaceflight. And he also just kind of talked a little bit about himself and his legacy and how he wants to be viewed basically as an inventor and not as the richest man on Earth.

21:29But that's also like cry me a river in terms of like, you don't like people focusing on your work. I feel you never, you don't say how you want to be perceived. That's what Sorkin asked him. And I think it was really interesting because he talked about how we think we know public figures, but we really don't. And we're only going to read headlines about them, which cover about 2 % of what they do. and you can't really, and then the public just forms an image of a person and you can't really get to know a person based off of that. I think that's really true. And he basically said, listen, I gave up trying to be understood a long time ago and that's basically why he's given up doing interviews with the press.

22:13And he does one every once in a while because he sees it useful. Go ahead. Okay. Okay. Okay. I think this is where you don't need to be understood as the CEO of a large tech company and owner of the Washington Post. You should only be understood. And the only expectation is you are understood as the founder and longtime CEO of Amazon and what you did there or the founder of Blue Origin. I don't need to know you any more than that. Like, I think this is one thing that the blurred line between like, I mean, we're going to be talking about Pat Gelsinger in just a little bit. I know nothing about Pat Gelsinger, the person, and I'm okay with that.

22:57We can talk about performance at Intel, but I think corporate leaders having the expectation that people have some deeper understanding or connection to them is definitely a more recent phenomenon. And I think it's a bad one. Okay. So first of all, I just think it was an interesting point that he brought up in the interview. And what you just said reminds me of Charles Barkley's analysis of a dirty player in the NBA. And they were talking a lot about intent and characteristics and stuff. And the guy probably, I think he just bashed somebody in the face with an elbow or something like that. And he had had it, you know, he had done it repeatedly.

23:36I don't know who it was. Maybe it was Draymond Green. it was probably Draymond Green and Charles Barkley you don't truly know Draymond Green that's true I think we know Draymond Green though but Charles Barkley basically looks at Shaq and goes listen like you are what you do and it's true you are what you do that's why and it doesn't need to be bigger than that or deeper than that I also find it kind of amazing that Jeff Bezos social media and trying to say that people don't understand me and the tight shirt posts of the recent years that gave us an understanding that maybe I'm surprised he wanted us to have.

24:16But we digress. No, I mean, he definitely cares about his public perception for sure. And then one other thing that came out from that interview, he's definitely a huge Trump supporter, which is sort of like, you know, that's the context behind killing the Washington and post endorsement. And it's one of those things where, and he's within his rights to do it. I think that it's one of those things where he really could have just come out and said, hey, I'm killing this endorsement of Kamala Harris because I believe Donald Trump is going to be a better leader, and I own the paper. And instead, he made it this thing about media trust.

24:45And I just think that's just blatantly disingenuous. But we digress. I'm with you on that one, with you on that one. But in terms of public communications and shaping the public perception, advertising is something that OpenAI is exploring. Oh, yeah. So this was really a fascinating story. So in the Financial Times, OpenAI CFO Sarah Fryer gave an interview. And I think that just like as a throwaway line, and I could be wrong on here, but she said, we plan to be thoughtful about when and where We implement ads and said that she was open to it. And then the FT has this headline, OpenAI explores advertising as it steps up revenue drive.

25:27So they published a story and I think OpenAI comms was like, wait, that's not the message we were hoping to give. Because there's a paragraph in the story that says, in a statement following the interview, Friar added, our current business is experiencing rapid growth and we see significant opportunities within our existing business model. While we're open to exploring other revenue streams in the future, we have no active plans to pursue advertising. Look, I've done enough of these stories when you see a note included after the fact. That's an angry call from PR and the reporter saying, okay, fair enough.

26:01I'll put in your statement, send me something. And then they drop it in. Yeah, but I think OpenAI, they've brought up search advertising in the past themselves. And they should be like if search GPT is really going to be a thing, then you monetize search. Perplexity is going to be doing it. Someone is going to crack like search based generative AI advertising. So it's odd to me that they wouldn't want to make that part of their overall portfolio of revenue because it seems it's a pretty good area. Google's vulnerable. They might as well go after it. I don't know why you would push back too hard on that.

26:39Right. It's a great point. I think they want to push back because I think Silicon Valley has a natural aversion to advertising. I mean, look at the companies that do advertising. Like the number one example is Meta. And there's like an ache factor to it. And I think OpenAI has a brand and they want to protect it. Wait, did you just say Silicon Valley has an aversion to advertising? No, no. So this is actually, so thank you for pushing me on this. This is an important point. I think if you speak with most product people, they will tell you that ads ruin products. They don't want ads. Every company starts out saying we don't like ads.

27:18Think about what Netflix has done. Think about the Uber business model and Lyft. Okay. Just to name a few that have recently introduced ads to their product and Amazon prime video and Amazon in particular. I mean, Jeff Bezos himself has talked about how ads are terrible. Steve Jobs, I don't think, liked ads. Now Apple is a low-key, a big ad seller. And so I think the vibe in Silicon Valley is your product should be good enough and loved enough that people will want to pay for it, and you'll never have to use their data and sell it to advertisers. We all know the truth is that advertising is a high-margin business, and they all end up doing it.

27:55But they take a brand hit. I mean, the fact that you have, you know, on your Lyft app, there's ads for whatever, or Uber, there are ads for whatever. It does, it does diminish the product experience, however useful advertising can be. And there are also data collection concerns. Like think about everything you tell a chatbot, there's going to be concerns. Is ChatGPT taking what I tell it and handing it to advertisers and making money off my data? I promise you that when this comes out, this is going to be a controversy. And so that's why I think there's an ick factor. It doesn't mean that Silicon Valley isn't like the world's most successful seller of ads.