AI’s Crucible Year, TikTok’s Lifeline, Manifesting Your Best Self

0:00A major year for AI gets underway, but OpenAI's most anticipated model is beset by setbacks. TikTok gets a lifeline from Donald Trump. And why is everyone manifesting all of a sudden? That's coming up right after this.

0:17You'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.

0:44Welcome to Big Technology Podcast Friday edition, where we break down the news in our traditional cool-headed and nuanced format. 2025 is here, and so are we. We've got plenty of AI news, as well as some updates on the TikTok saga. And of course, we're going to get into manifesting, and why is everybody doing that these days? Ranjan is going to give us so much more on manifesting at the end of the show, but first let me welcome him back in for the first time in 2025. Welcome, Ranjan. Happy 2020 Thrive, Alex. 2020 Thrive. We're just going to brand it the right way off the bat. I saw that on some random Instagram reel, and I've been saying that to my wife over and over, and she is not.

1:25I'm sure she loves it. She's worried about me. She's worried. Are you having good 2025 vibes? um we're two days we're three days in right now and uh it's been quite a three days i think at least yes it's been a very weird start to the year it has not been a slow start to the year so at least for the podcasting world it keeps it interesting i'll tell you how i closed out 2024 i jumped out of a plane with meta ray bands on and recorded the whole thing first person and it was pretty interesting uh as i walked onto this plane i was skydiving of course and the instructors immediately looked and they're like oh those are the glasses that can play music and i said yes and they're like oh and take video because they like sell the video package for 150 dollars and i just had these like right on my face and it's not like a traditional camera that you have to hold so i ended the year finding one more use cases for this new smart glasses world that we're entering it.

2:27Well, maybe. Do you know why I'm ready to thrive, Alex? Guess what I got for Christmas? No way. Meta Ray-Bans? I got them. I have not. I'm going skiing next weekend, so that'll be my first real use case. I've been playing around with them. Even my drive home from Boston to New York was taking random video. Actually, do you know what impressed me the most? Actually, there are two things. One, the quality of the audio, not being in ear, but being essentially kind of like speakers to the side of you, I almost like and prefer better to in ear when you're walking around. Just how like stereo quality, but I don't know, the ambience, the audio quality was blew me away.

3:13The second thing was, it really feels like the most revolutionary part of the product is I started using it more to interact with my iPhone than Siri because we all know Siri is still garbage and Apple intelligence was a flop and is currently a flop. And meta AI is pretty good. It's baseline good. So calling out to Spotify, making phone calls, even sending messages, starting to use WhatsApp and Facebook messenger more than Apple messages. I started when I'm wearing them, I am using it to interact with my phone more than Siri. So this could almost be kind of a Trojan horse for Facebook into your iPhone operating system versus Siri, if voice is the interface.

4:02And now you see why Tim Cook has tried to kneecap meta, right? These two are obvious competitors and that competition isn't talked about nearly enough, but this is just going to be another area where they're going head to head. And of course, we know that Apple has these new AirPods in development that have cameras on them as well. So that's going to be really fascinating to watch the two of them. You know, I'm sure Apple will find some reason to ban the Meta Ray Ban app or something like that. We'll see. 2020 Thrive. Tim versus Mark. I do want to say one more thing because we did get a comment from a listener.

4:36I think it's a valid comment being like, I'm surprised you guys are excited about the Meta Ray Ban's given Facebook's history of like privacy mess ups. and look I don't think that we're gonna be excited about a product and therefore negate any concern we have about the way that the company has operated in the past but I think what becomes clear as you watch this and carry as you use these things and I'm curious if this was your experience Ranjan that you just see that this is going to be a future method of computing and like you can either choose to ignore it or you can give it a shot and that's what we're trying to do At least that's what I think.

5:10So no, no, no, that is a very valid and fair question. And I'll admit, and as someone who, you know, has been very concerned around Facebook's privacy actions over the last, you know, now eight years, nine years, when you put them on, the idea that I was ever going to put, you know, glass something on my face that captures everything in the world around me from meta slash Facebook, I never would have thought. But I think it is a test. And even when you have to allow a lot of permissions on your phone, access to always on location, access to contacts, I was going back and forth with each okay in that entire user flow.

5:56But I think it's almost a testament to call it the Zuckerberg rebrand, or maybe Musk becoming the ultimate bad guy in tech when you have to worry about where is the evil part of big tech. But I definitely thought about it. And it's one of those where the trade-off of the product is so good. And I agree, this is going to be part of the future of computing. And I'm excited to be part of that and testing that out early. And maybe that is where Apple will be able to once again kind of step in and become the privacy first example of this if they actually push out some kind of smart glasses. But to that listener, trust me, I was feeling a little weird throughout the whole process, but I still went with it.

6:44Yeah. And look, I think that one of the things that we found is that, or that I think anyone who watches tech has found, is that if the products work well, people are willing to, I'm not saying us, because our job is to analyze and criticize it when it's called for, but a lot of people will just look past whatever a company does when the product works. And I think you saw that with Apple. I think you're going to see that with Meta, and that means that this product is going to be a breakthrough product, in my opinion. And if we didn't test it out, we would be negligent in our responsibilities here.

7:17So we'll test it out. We'll criticize when it's needed. And before we move to the next section, I'll just say one more thing that I enjoy is the microphone. and if I can figure out a way to pull it up, I'll play the audio of me getting hurled out of a plane from 15 ,000 feet. Oh my God. Oh my God. Oh my God. Okay. That's a good way to start the year. My way to start the year was just watching Dick Clark or whatever, one of those specials with my family, which is still nice, but... No, that's a great way to kick it in. and one company that's going to struggle its way into 2025 is open ai because just as we were heading to break there was a great story in the wall street journal about the delays that the company has had with its gpt5 model which it's calling orion internally so we've talked a little bit about this about how it's delayed but i think this story is the best one that i've read in terms of sharing the details of what's actually going on within OpenAI, why we haven't seen GPT-5 yet, and when it might be coming.

8:29And I just think this is an important story to talk about because we're really going to go into 2025 with massive stakes for AI, and no one's going to have bigger stakes than OpenAI, and nothing within OpenAI is going to have bigger stakes than GPT-5. So we'll talk about that in a moment. Let me at least say what happened. This is from the journal. OpenAI has conducted at least two training runs, each of which entails months of crunching huge amounts of data with the goal of making Orion smarter. Each time new problems arose and the software fell short of the results researchers were hoping for.

9:03A six-month training run can cost about a half billion dollars in computing costs alone. So OpenAI has done these two large training runs. They've spent what looks like a billion dollars combined on them, and they're still not getting the results that they want. So Ranjan, I'm curious what you thought when you heard this news and what you think it means for OpenAI's future and whether it's going to be a bad year in 2025 now that we seem to be on this path. All right. I've been thinking about this a lot in terms of what this year is going to mean for the entire generative AI space. And it was actually interesting.

9:41I was listening to Casey and your predictions episode last week. And it was the he had mentioned and I think it's correct that, to me, this is the year that AI needs to be productized. It's not going to be and we've talked about this a lot, like it's not going to need to be just a power game. Yet, there's such a disconnect to me between what normal people want, what even enterprise business customers want and what tech companies and VCs are thinking and doing. And from the tech company VC side, everything is about the next powerful model. It's about AGI. It's about, you know, like a massive, can we maintain current scaling laws and step changes?

10:32And then on the actual normal person side or even business customer side, it's can we just have things that are usable, reliable, and work well. And there's this huge disconnect there. And I think, to me, OpenAI is almost, and again, it's a research house with a business tacked onto it. So they're always going to be thinking about what is that exponential step change style advancement we can push, and leaving the customers behind. So I think overly, but they've also shown to their credit, that sometimes they can really get products right. To me, ChatGPT, its virality was not because GPT-3 was so much better than GPT-2.

11:17I think that was part of it, but I also think it was a really, really good product. And it was a fun product. And they did a lot with the UI that actually made it exciting and usable. So I think, to me, them overly focusing on GPT-5 could actually be the thing that hurts them even more because I don't think they're going to create some kind of model that magically solves everything. I think not focusing on the productization enough, it will actually be an even bigger mistake on their part. I think this is going to be a debate between you and I for the entire year, because if you don't have GPT-4 and you stopped at GPT-2, ChatGPT sucks.

12:02So it is those advances that have powered the better products. That being said, I'll give you this. In the story, there's a nice little tidbit. I call it the Ranjan callout, they say. As Orion installed, OpenAI started developing other projects and applications, including the slimmed-down versions of GPT-4 and Sora. It's AI video generator. So yes, they realized that they needed to put the computing towards more practical things in the short run. So that was good. However, I think there are some pretty concerning details in this story about AI's ability to progress beyond the productization that are important to cover.

12:43And so to me, the key call out here was the limitations of data, right? So this is from the story to make Orion smarter. Orion, by the way, again, GPT-5, OpenAI needs to make it larger. That means it needs even more data, but there isn't enough. So the solution was to create it from scratch. It hired people to write fresh software code, solve math problems for Orion, and these workers who are software engineers and mathematicians also shared the explanations of their work to Orion. However, And TPT4 was trained on an estimated 13 trillion tokens. And this is what it's going to take to get there in human scale.

13:23A thousand people writing 5 ,000 words a day could take months to produce a billion tokens. So it does really seem like this data wall really is here. And this is the last part I'll read about the data. Once the training began, OpenAI researchers discovered a problem with the data. It wasn't as diversified as they thought. potentially limiting how much Orion would learn. So this is really coming to the fore right now, that there's a real data wall in AI, and you can legitimately sit PhD and mathematicians down all day long, and they could basically do nothing but write content for these models, and they will not get anywhere near the amount of knowledge into them, as has already been baked in, unless they sit there and write for years.

14:13See, this is what I don't understand, though. Why would I want a PhD or an engineer to write fresh software code or solve math problems for the input data if I am not a coder or a mathematician? Is the assumption here that a mathematician helping a model solve math problems will eventually help me do some kind of business process better or just some kind of normal task with AI or help Apple intelligence find your flight in your mailbox better? I'll take my best shot at this because this is what OpenAI told investors, I believe, that it was aiming for because there are internal expectations for GPT-5.

15:08And this all was in context as they were racing their massive largest VC round ever in history. This is from the journal story. GPT-5 is supposed to unlock new scientific discoveries, as well as accomplish routine human tasks like booking appointments or flights. Researchers hope it will make fewer mistakes than today's AI, or at least acknowledge doubt, something of a challenge for the current models, which can produce errors with apparent confidence known as hallucinations. So basically, I think what OpenAI is saying is that it has the, this is internally and to investors, I believe. What it's saying is we have the capability to make these models step change smarter.

15:50And as they get smarter, they'll be able to handle more tasks because they're just going to be a much more brilliant model than the ones we have today. But the problem is if you don't get there, then all that promise goes away. And we come back to the big problem that we've been discussing up until this point, which is like, where do you go from here? Because you're not going to be able to raise again if you're not going to be able to achieve what you've been telling investors. Well, here's where I think the big disconnect is. All right, this is my 2020 Thrive, 2025 thesis here. It's that this disconnect is really, could be a fatal one for OpenAI.

16:30Because again, on one hand, that question right there, it's like even simple things like booking appointments, current models are not perfect and that's making people not actually use them. To me, the model becoming exponentially better is not the problem. There's really simple things you can do on the product side to actually build systems and simple products to allow for these kind of tasks to be solved. But that's not what they sold investors on. They have to come up with this almost mythical or magical next model and this thing. So they have to keep working on that. So they can't actually solve the problems that customers want solved while pleasing investors.

17:14So now that we're thinking more about this, That does put them in a pretty bad situation because they have to do both. And doing both is always going to be more difficult. And mythical and magical really is, I think, the right characterization here. Because the story also is like, well, what is GPT-5? Like, when are they going to call it GPT-5? Because we've had GPT-4, 4-0, 0-1. We're getting lots of different names. We're not getting 5, right? And they went from 2 to 3 to 4 pretty quickly. Here's what the journal says. there are no set criteria for determining when a model has become smart enough to be designated GPT-5.

17:50Opening eye can test its LLMs in areas like math and coding, but it's really up to company executives to decide whether the model is smart enough to be called GPT-5 based in large part on gut feelings or, and this is my favorite line in the whole story, as many technologists say, vibes. I imagine if they went, the iPhone stopped going like 16, 15, 14, just because it's not really a step change anymore. It's just a slightly better camera. And they started just completely 01, O max. No, they do. They've done that for us, but everybody gets excited for 15 and 16. They don't get excited for like these.

18:30Yeah. Apple's still going for it, but maybe open AI, just call it GPT-5. even if it's not much better that should be enough just like AGI you know yeah what I'll say is the longer they wait the worse this is going to get for them because now everybody's going to expect them to release like legitimately the creator in their model and it's just not going to happen yeah yeah no that's why I agree as well like if 4.0 they called 5 and again you say multimodal significantly cheaper per token for the same or higher quality. So like we're creating that actually, that's interesting. And then I'm waiting for six, but already I've gotten kind of lost because once you go 4.0, 4.0 mini, all these other things, then you start having higher expectations for when that next big number comes.

19:23So I agree with that. One more bit here, because we've brought up apple uh so this is again from the story researchers at apple recently released a paper that argues that reasoning models including versions of o1 were more than likely mimicking mimicking the data they saw in training rather than actually solving new problems the apple researchers said that they found a catastrophic performance drops if questions were changed to include irrelevant details like tweaking a math problem about kiwis to note that some form of the fruits were smaller than others. So this is the first, I would say, mainstream knockdown of reasoning as a methodology.

20:05And we've talked in the past here about how like maybe reasoning can be the next step change, but it is interesting to see this research come out of Apple. Maybe it's Apple just saying like, you know, we're not the best, so let's take on the best. But it is a pretty bold statement to say that actually reasoning is effectively, you know, BS. What did you make of that study? I was wondering, again, where, again, Apple Intelligence, still not good. And again, I take knocks at Apple. I probably spent more at the Apple Store over the Christmas season. And my mother has an Apple Watch. My wife has an Apple Watch now.

20:47There's more spent at the Apple Store than any other electronics retailer. but with Apple intelligence, they are not doing great. They have promised a lot. They have over promised, but on the other hand, they have the least invested in having to win this game. I think out of the big, like Google has to win at generative AI for them to maintain dominance. Apple does not necessarily, it surprised me that they're pushing so hard because they don't have to. They could have sat back even longer and waited for the technology to be ready and good, and then released a big hype to Apple intelligence product.

21:30So I think it does make sense to me that for them to kind of knock things down and kneecap the promises of open AI and others and Microsoft and whoever else and say, guys, this isn't going to happen anytime soon. And it actually only helps them right okay let's let's just end this segment like again tackling the bigger picture which is i'm saying that this is a pretty important year for the ai field i'm calling it a crucible year i think that if the field cannot deliver on some of these bold promises whether that's on research or productization it's going to be a very very tough uh couple years following like this is like really like if 22 was we figured out what chat gpt was 23 was experimentation maybe 24 was continued experimentation and more proofs of concept uh 2025 that we gotta we gotta this has gotta work it's that important of a year what do you think i like crucible year crucible year i also my prediction i think we might get some players that we have not been thinking about as much.

22:36Like, I think we say the same names over and over again, but especially, and maybe that could be, I mean, anthropic or perplexity becoming even bigger than we have been talking about. But I think like, it's the idea that it's open AI and the FAANG slash big or big tech. I think there, I think we might start seeing different names popping up, the ones who could actually solve the customer problem better oh okay now you're going customer problem you know that makes me uh think in a different way that i was gonna go right before i was thinking maybe it's gonna be safe super intelligence but i'm team customer your team super intelligence that that really does describe the dichotomy here pretty well yeah yeah yeah just just make just make it work make it work that's all i want you know we've been talking about whether this was going to be the year that OpenAI was going to declare AGI and free itself from the constraints of Microsoft and sharing its technology with Microsoft.

23:42But also over the new year, we actually got some really interesting information. There's a pretty specific definition. This is according to the information that OpenAI and Microsoft have for when it's okay to declare AGI. And that is when OpenAI develops AI systems that can generate at least 100 billion in profits. And given that open ai is going the opposite way it doesn't seem like it's going to be able to declare agi at least according that to that definition anytime soon so does it does that sort of solve our worry here that opening i could basically just go out and say agi and just keep the tech to itself well i this definition it confused me even more because i would have thought the definition would be something more on the scientific or theoretical side rather than the business side.

24:35So yeah, this confused me even more, which is good for open AI because the more confusing it is, then the easier it is. Or actually, no, I guess in this case, it would be even harder for them because it's such a kind of like clear yardstick for them to have to achieve. Can I say that maybe it was clarifying? You know, remember, we talked a lot about how like this talk of AGI could just be like really clever branding. And all this like the world is going to end because of AI actually really serve the interests of the research houses who wanted to sell this to like your typical SaaS company or your, you know, middling accounting firm who could maybe use it to build move data from one spreadsheet to another.

25:21because in the contract, right, saying AGI is just at$100 billion, doesn't that like kind of explicitly say, oh yeah, these promises of like magical AI, it's really just about the business? Yeah, no, no, that's where, and actually now that I think about it, it really is actually Microsoft being pretty savvy in this because at the point that it has done that, then obviously Microsoft has already been able to reap whatever benefits for themselves and then happy to let things go off on their own. But again, then Google's ad engine or Facebook's ad engine or like these technologies, are they AGI then as well?

26:04Is anything that is the iPhone hardware AGI? Like it's such a weird thing for me. Maybe like super intelligence or artificial general intelligence actually would not be a profitable thing, you know, like, and still actually work in the way that we have been promised by research houses it would. So this actually, when this being coming out finally was even more odd and shocking to what is it already a very odd and shocking story. I think it's a scientific term co-opted by businesses that told us they were about the science, but now we really know that they're about the business and it's clarifying.

26:47And refreshing in some ways to know that, yes, AGI is just about making$100 billion. It's not about necessarily propelling the next scientific discovery. It's making that$100 billion. I like your cynicism, Alex. I like it. The fact that you're... It's impossible for me to hold it back in this situation. I know, I know. No, but I'm with you. I'm with you. You know what? I agree. I would rather just get it out there. It's kind of, again, like we've said, don't call it open AI. just don't call it like some mythical AGI that's a scientific concept it's literally like a target on a profit margin or profit target it's a profit target yes and by the way I think going to your previous point going by this definition yes Meta's ad platform has been AGI for decades at this point and maybe even TikTok did you know your credit card points and miles can lose value to inflation?

27:46Credit card companies often reduce the redemption value of your points and miles. Now imagine a credit card with rewards that can grow in value. With the Gemini credit card, you can earn Bitcoin or one of over 50 other cryptos instantly with no annual fee. Every swipe at the store or gas pump earns you instant rewards deposited straight to your account. Plus, sign up now for a$200 Bitcoin bonus to kickstart your rewards. visit gemini.com slash card today check out the link in the description for more information on rates again if you're looking to invest in bitcoin but don't know where to start the gemini credit card makes it easy the gemini credit card is issued by web bank in order to qualify for the 200 crypto intro bonus you must spend three thousand dollars in your first 90 days some exclusions apply to instant rewards in which rewards are deposited when the transaction This content is not investment advice and trading crypto involves risk.

28:43The Gemini credit card cannot be used to make gambling-related purchases.

29:08station and wherever you find your podcasts.