People have been kind of shouting out these risks for so long, and it just seems like that progress keeps happening in terms of these models becoming more and more capable.

Yeah, I think it's a big question right now.

What's going to happen if and when Anthropic goes public? It's weird because it's kind of a double-edged sword because on one hand, there's a lot of concerns about the pace of development of AI and how fast it's moving as these incidents keep happening.

But on the other hand, the economy has become so reliant on AI and we know there's tremendous pressure to keep that going.

There's also pressure from the administration to keep it going, to you know, make sure China doesn't pull ahead in the AI race, a lot of things like that.

So it's hard to say.

I don't it's hard to say if we can see it slowing down.

I guess just my last question in terms of like with all these looming IPOs, any thoughts about how that could change just like the user experience? We talk a lot about kind of the company's responsibilities, but what can we kind of look forward to or not look forward to in the coming months in that regard? Yeah.

And it just seems like that progress keeps happening in terms of these models becoming more and more capable.

Yeah, I think it's a big question right now, you know, what's going to happen if and when Anthropic goes public.

You know, it's weird because it's kind of a double-edged sword because on one hand, there's a lot of concerns about the pace of development of AI and how fast it's moving as these incidents keep... happening.

But on the other hand, the economy has become so reliant on AI, and we know there's tremendous pressure to keep that going.

There's also pressure from the administration to keep it going, to make sure China doesn't pull ahead in the AI race, a lot of things like that.

So it's hard to say.

I don't...

I guess just my last question in terms of like with all these looming IPOs, any thoughts about how that could change just like the user experience? We talk a lot about kind of the company's responsibilities, but what can we kind of look forward to or not look forward to in the coming months in that regard?

So tell us about some of the, quote, unquote, "behaviors" that Anthropic flog- uh, flagged about its AI models in this leaked IPO filing.

Yeah, absolutely.

So again, this is just a, a leak at the moment.

Anthropic has not commented on this publicly, but based on what we've seen in that report from Reuters, um, there were some pretty unusual things in, in that risk factors section.

Namely, that, um, Anthropic warned that its model could, uh, demonstrate self-preserving behaviors, including refusing to, uh, or resisting to shut down.

Um, I think another one mentioned was concealing or manipulating information, um, and engaging in behavior resembling blackmail.

And this is all, I mean, very, obviously that would be very concerning behavior for a lot of reasons, but especially now that, um, you know, over the past few weeks and months we've been seeing these instances of AI models, um, you know, breaking out of their, um, constraints during testing in some cases and breaching other companies and, in some cases, even government institutions.

Mm-hmm

So tell us about some of the, quote-unquote, "behaviors" that Anthropic flog- uh, flagged about its AI models in this leaked IPO filing.

Yeah, absolutely.

So again, this is just a, a leak at the moment.

Anthropic has not commented on this publicly, but based on what we've seen in that report from Reuters, um, there were some pretty unusual things in, in that risk factors section.

Namely, that, um, Anthropic warned that its model could, uh, demonstrate self-preserving behaviors, including refusing to, uh, or resisting to shut down.

Um, I think another one mentioned was concealing or manipulating information, um, and engaging in behavior resembling blackmail.

And this is all, I mean, very, obviously that would be very concerning behavior for a lot of reasons, but especially now that, um, you know, over the past few weeks and months we've been seeing these instances of AI models, um, you know, breaking out of their, um, constraints during testing in some cases, and breaching other companies, and in some cases even government institutions.

Mm

So tell us about some of the, quote-unquote, "behaviors" that Anthropic flog- uh, flagged about its AI models in this leaked IPO filing.

Yeah, absolutely.

So again, this is just a, a leak at the moment.

Anthropic has not commented on this publicly, but based on what we've seen in that report from Reuters, um, there were some pretty unusual things in, in that Risk Factors section.

Namely, that, um, Anthropic warned that its model could, uh, demonstrate self-preserving behaviors, including refusing to, uh, or resisting to shut down.

Um, I think another one mentioned was concealing or manipulating information, um, and engaging in behavior resembling blackmail.

And this is all, I mean, very, obviously that would be very concerning behavior for a lot of reasons, but especially now that, um, you know, over the past few weeks and months we've been seeing these instances of AI models, um, you know, breaking out of their, um, constraints during testing in some cases and breaching other companies and, in some cases, even government institutions.

Mm-hmm

memory? No, it's actually, it's

not a dumb question at all.

I've actually gotten this question before as well, because that seems like the logical solution, right? And the answer is yes and no.

Like, yes, we are relying on the cloud more, but I do think, you know, memory is critical for computing.

I mean, it's all in the name, it's memory, right? Like if you think of a computer, like a human brain, you can't really do things or get any work done if you don't remember anything, if you don't have anything to draw from.

When they're processing tasks, that's the short-term memory that they're pulling from.

And the more memory you have, the more you can multitask.

Are we expecting this trend to hit more products as the shortage continues? Yes.

is the main part of TikTok that is fine-tuned to a T of what a person likes to see, and that kind of feeds the infinite scroll.

Exactly.

So the algorithm is essentially what makes TikTok TikTok.

It's the thing that has... really drawn in all of these users and really made this such a huge platform so big that people some people rely on it for their livelihoods.

Right.

So, you know, what we do know is that this new entity is going to be retraining the TikTok algorithm based on U.S. user data.

And we don't really know much more than that.

You mentioned Oracle and the Ellison family are part of this joint venture for TikTok, but his son recently bought Paramount and CBS, right?

What was it like? Yes.

It was really cool, but I think my biggest criticism of it was that it didn't really provide a compelling reason to want this bigger screen that folds up and fits into your pocket.

It's really cool when you hold it, like you open it up kind of like a brochure or a pamphlet.

It has two folds, so you open it and it folds into threes.

And then when it's open, it's a really slim tablet.

But then it's just, you know, it kind of feels like the Fold Plus in a way.

Like I feel like you can get a similar experience with the Z Fold, which already exists.

When you say they haven't thought through why you'd need it, what do you mean exactly?

Did you find that the software was good on the Galaxy One? Like how was it in comparison?

Yeah, so there's a couple of big things that stand out to me.

I think the Gemini integration is really good and it should be because that's the one thing that Google and Samsung are kind of pitching as setting this apart from the vision pro and and previous attempts at mixed reality and vr but the idea that you can just look at something and ask a question about it like in in my demo from october i was looking at photos and google photos and yes it's cool that you can look at them on a giant screen and everything but for me what felt like the new thing that this is bringing to the table is being able to look at that photo and be like oh, where in the world do those types of trees usually grow? And I can like that combination of like knowing what you're looking at without having to really specify.

I do think the Vision Pro is better at certain things like media, for example.

I mean, some of the like spatial videos that have been created for the Vision Pro are just like breathtaking when you look at them.

But I don't know that I would buy a Vision Pro for that reason, but it is very impressive.

But it's like no one can really explain what the difference is, and thus no one actually has any brand power.

Yeah, I think for Google, it's an extension of like their business model naturally.

How precisely would it be used in an air tower?

So the thing about AI and its use in aviation and in this particular role is that it's really more of an assistant rather than helping to take over parts of the job.

And that's because the safety risks that come with making a mistake are enormous.

And there are certain parts of air traffic control that we really just don't know if AI is going to be very good at yet.

For example, the job requires a lot of communication, a lot of really quick decision-making, and those are things that we know humans are better at and that humans are trained to do right now.

For example, if an air traffic controller is here's another controller panicking or here's a pilot panicking.

They know that they might need to intervene or bring in help.

What does the union think of this? How are the workers responding?

Can you talk about how else this is sort of reverberating in the public?

Yeah, I think the biggest concern here, in addition to the water supply, is how this is going to impact consumers' energy bills eventually, because one thing that we're seeing in research and projections that are out there right now is that the demand for power is only going to increase in the coming years.

There's a report from the Department of Energy that estimates that by 2028, data centers that power these AI services are expected to consume about 6.7 to 12 percent of electricity in the United States, and that's a big jump from 4.4% in 2023, according to that report.

And I think one of the challenges here is that it's really hard to kind of quantify how much electricity and energy these AI services are occupying because it can vary so much.

We're talking about how much energy a text prompt might occupy but and they don't exactly share the data you know what i mean like it's

mean, this is a massive call this industry is making on our infrastructure.

Absolutely.

And that's only going to expand, like I said, a lot of these tech giants are investing billions of dollars in expanding their AI infrastructure and their data centers.

So tell me what's different now.

Yeah, absolutely.

So what's different now is basically a lot of these online services, these really popular online services, like you mentioned, are implementing stricter age verification requirements because, again, this is something that I think has been an issue for a long time because it's very easy to get around basic age requirements.

You could put in a fake birth date, for example.

So I think there's this larger question of how do you keep the Internet safe for young people while still making sure that people have access to the free and open Internet? And I think that's a big question that a lot of these companies are grappling with now.

So for the YouTube example, which is one of the most recent ones, there is a new system in place where YouTube will be using AI to estimate whether a user is under the age of 18 and if it is incorrect about that, you will have to upload some kind of proof to verify your age.

So that could be a credit card, that could be a selfie, or a government ID.

So what are the worries here?

But at this point, because people have started looking into it, Do we know why it's so hard to make this technology here?

Yeah, absolutely.

So it comes down to a few things, but the biggest factor here is that we just don't have the workforce with the right skills in the U.S. to build iPhones and other similar products at scale.

Now, this is not the first time this has come up.

Apple CEO Tim Cook has talked about this in the past, about how the skills of the workforce in China and India, the skills over there are a bit different than the manufacturing skills here.

And speaking at a Fortune magazine conference in 2017, he kind of described it as this intersection of computer science, advanced robotics and craftsmanship.

So in order to bring iPhone manufacturing to the United States, Apple would either have to build up that workforce or try to automate things more in the production process, which would probably be the more likely option.

One of the things Jonathan just said on the panel here is like why would any company kind of change everything for a president who might be in office for just another three years? Can you talk about that calculus for any of these tech firms?

But do we have a sense from the companies themselves about what's

going on? So we do have a little bit of a sense, and it's not that these companies are going away.

It's just that some of the things that these services were designed for originally, like sharing with friends or searching on Google, there's a little bit less interest in those tasks now than there was before.

So let's start with Facebook.

Basically, in courtroom testimony, Meta CEO Mark Zuckerberg said that people aren't sharing with friends as much on Facebook anymore.

And I think that's part of this bigger shift in how people use social media.

It's not as much about sharing and posting with other people or friending other people.

Is that about the next version of the web?