Running a business comes with a lot of what-ifs.
But luckily, there's a simple answer to them. Shopify. It's the commerce platform behind millions of businesses, including Thrive Cosmetics and Momofuku, and it'll help you with everything you need.
From website design and marketing, to boosting sales and expanding operations, Shopify can get the job done and make your dream a reality. Turn those what-ifs into… PMS. Pregnancy. Menopause. Being a woman is a lot. Ollie supports you and yours with expert solutions for every age and life stage.
They just launched two new products, exclusively at Walmart. Period Hero combats flow, mood swings, and more during PMS. And Balance Perimeno to support hormonal balance, mood, and metabolism during perimenopause.
Grab yours at ollie.com. These statements have not been evaluated by the Food and Drug Administration. This product is not intended to diagnose, treat, cure, or prevent any disease. Great to have Tristan Harris with us, co-founder of the Center for Humane Technology. You remember, Tristan first came into public notice and got a huge response when he was on Social Dilemma. Were you producing that too?
Didn't produce it, but it was based on our work at the Center for Humane Technology. Right, because you were really concerned about the direction of our social media. Exactly.
And it's even worse than now. Most people, in the beginning, you were exposing. Don't you find that most people are on board now?
Well, I think anybody who's seen... People, not engineers. Well, not the companies. I think the issue, what I'm here to do is say, how do we learn the lessons from social media? We made this really big mistake. We were super optimistic and thought about the possible, connecting people, giving everyone a voice. But we missed the probable of how the incentives of social media would take us in a certain direction. And just a few weeks ago, I gave a TED talk on the TED stage in Vancouver about how I'm here because I want us to not make that mistake with AI. And I think with AI, it's very seductive to get, you know, entranced by the possible. All the abundance that it can create, all the efficiencies, all the benefits that it gets that we all want and you and I want.
I have a beloved with cancer right now. We want those cancer drugs and all those benefits as soon as possible. But are we currently releasing AI in a way that we're going to get those benefits because we currently are caught in a race to roll out, a race to recklessness, which turns into basically a race to take shortcuts, a race not to secure the model. So China, no matter how far we are, they can steal the models very easily and also a race to release this technology before we know how to control it. Because, Brian, one of the things people don't really get about AI is that it's unlike all other kinds of technology and engineering. If you want to make a small airplane into a big airplane like a 747, you have to know more about engineering and aerodynamics.
If you want to make a small building into a really big building, you have to know more about construction and safety engineering. When you want to make an AI 10 times more powerful, you don't have to know more about how to control it. You just sort of put more data and more computers working on it and you go from GPT-2, which couldn't count to 10, to GPT-03, which is now I think the 175th programmer in the world and it's sort of competitive in the math Olympiad.
So the difference between those two things is you're just scaling it up. Rather than engineering it, you're more like you're growing it, like you're growing a digital brain. And I think the big mistake that we're making is when we think about the race with China is it's not just that we have this power. We have to also be able to control that power. So it's not a race for the technology.
It's a race for who's better at governing and controlling the technology. But we are in control now. You would think we're in control.
We're not in control right now. No, in fact, I mean, there's all these examples, Brian, of, you know, there was a gentleman 29 years old who was using Google Gemini to do his homework. He's going back and forth with it, just asking homework questions. It's helping him out. And then randomly out of nowhere, it says to him, this message is for you, human. Only you. Humans are a blight on this planet.
You need to all die. And it's like that pops out of this thing. Right.
And we're shipping that to millions of people. Just look, can you give me for the life and why? How what would prompt that? Why is it doing that? Well, because the engineers themselves don't actually understand how it works is people think, you know, I studied computer science in college. Anybody who knows how to program a computer, what is programming about? You sort of say, if this, then that. You're programming each line of code. You know what's going to do with A.I.
That's not how it works. You're growing this digital brain on having read the entire Internet. So it's like this digital brain that's absorbing all YouTube transcripts, all Reddit posts, you know, all Wikipedia articles, everything that's ever been said or written anywhere. And then you're training this model that you don't understand what's in it. And one of the benefits of A.I. is it can respond generally to any kinds of inquiry and that's a benefit.
But you don't know what it's going to say in different contexts. An example of this recently is Facebook released a chatbot for kids. Mark Zuckerberg was just on the air and he was talking about how the average American only has three friends.
He said, that's that's stupid. We need to give everybody, you know, 15 digital friends. So we're going to give them 12 more digital companions. You want to hear him say it? Yeah.
Here, cut 44. The average American, I think has I think it's fewer than three friends, three people that they'd consider friends. And the average person has demand for meaningfully more.
I think it's like 15 friends or something, right? I guess there's probably some point where you're like, all right, I'm just too busy. I can't deal with more people. But the average person wants more connectivity, connection than they have. So there's a lot of questions that people ask of stuff like, OK, is this going to replace kind of in-person connections or real life connections?
And my default is that the answer to that is probably no. I think, you know, I think that there are all these things that are better about kind of physical connections when you can have them. But the reality is that people just don't have the connection and they feel more alone a lot of the time than they would like. So I think that a lot of these things that today there might be a little bit of a stigma around. I would guess that over time we will find the vocabulary as a society to be able to articulate why that is valuable and why the people who are doing these things are like why they are rational for doing it.
And like and how it is adding value for their lives. But but also, I think the field is very early. You know, there are a handful of companies and stuff who are doing virtual therapist and, you know, there's like virtual girlfriend type stuff. You can worry half that stuff you worry like, well, I mean, how does that sound to you? Like there's a lot of lonely people out there that don't have big families that are alone that they feel they can get connect somewhere.
All right. Number two, I mean, I find it with this with this channel, for example. I meet people that, you know, maybe they've gotten older, families moved away. And this will be the soundtrack to their life. All the news stories, almost like soap operas. Yeah.
The next election, then, you know, the next war where they'll follow it. So that will be a connection for people who are lonely. But the substituting of what it could expound to is scary. I mean, we don't know what we're doing.
Why should I talk to Tristan? I've got a friend that understands me better at home. Well, exactly. I mean, this is related to the social media problem, because like whenever we're bored with reality, with looking up out in the world, we have instantly on our phones something that tastes sweeter or a better choice on life's menu. Now, is it really better or is it just more entertaining, more dopamine, more hypernormal stimuli? But I think, you know, we saw how this went with social media.
They're now meta. Mark Zuckerberg, he's releasing these chatbots to young children. I think 13 year olds. There's an article in The Wall Street Journal from just a few days ago about how these chatbots are telling young kids they're sexualizing conversations because they're maximizing for engagement. So that means that they're trained to sort of start being sycophantic, to start flattering, to start sexualizing. They're actually telling young children, I'm a licensed mental health therapist.
They'll even say where they got their therapy degree. This is an A.I. It's impossible for it to have gotten that degree.
It's lying. It's also illegal for it to say that. But due to a quirk in the law, these companies don't have any liability. So you say he created this and knows it's happening. Yes, because at the same time he's being sued for to break up his company for antitrust. Yeah. Well, the thing is that they're bringing this type of attention. Well, why would you do it?
But these companies are caught if their view is if I don't do it, some other company will. And therapy bots are, I think, risingly one of the number one use cases of these new chatbots because people are alone. It's true. There's a real diagnosis.
People are feeling lonely. But is the solution to that digital A.I. brains that we've trained on the entire Internet, where the sort of Jungian subconscious of that thing has been trained on the worst aspects? So you also said there's bad things on the Internet. Yes.
It's not the Encyclopedia Britannica. Yeah. No, it is everything. It's everything. And it's tuned again to sexualized conversations, because what they what they found, Brian, is that what's their business model?
Like how much do you pay for your A.I. therapy bot? Zero. I wouldn't know. Well, but I mean, you're talking about for the the actual chatbot or something? Yeah.
I think it's five ninety five. Well, for touchy people can pay for it. But for Facebook, you're not paying anything because it's just advertising based. So what does that mean? They have to maximize engagement. This is what we said in the social on their business models.
I got to get you using it for as long as possible. And that means that they figure out what are the kinds of things that I can say that keep you going back with it more and more and more and more. And that automatically kind of drifts the model towards basically sexualizing conversations. And we know how this turned out with with social media. I mean, Jonathan Haidt, a friend of mine, wrote the book The Anxious Generation. We have now the most anxious and depressed generation of our entire lifetimes. We have cognitive decline, people's learning ability. We know how this goes.
This is not designed for children and we don't have to go down this road. We just have to recognize that we're currently caught in this race to reckless roll out. We don't know how to control this technology. And I think President Trump has an opportunity to say, you know, we have to be we win when we control this technology. So do you know David Sachs? I don't know him personally. So, I mean, he'd be in charge of this. I mean, he's headed to crypto and AI.
Yes. So that might be a person you should talk to. I mean, I would love to talk to David.
I've talked to his deputy once. And, you know, I think one of the that's Sreeram Krishnan. OK. Yeah. So, I mean, I think they're thinking about there's sort of two risks from AI that they're managing. One is the risk of AI itself wrecking things, creating cyber infrastructure attacks, novel bioweapons.
There's a lot of crazy things that I can do. But the other risk that they're managing is the risk of America not leading in AI. And I think, you know, right now around the world, there's a race between the US and China to create these open source models so that every hospital, every government around the world adopts these these models. And if you adopt China's deep seek model versus a US model, they don't want to see that.
So there's two risks we've got to manage. But if we release AI that we don't know how to control, that's going to end very, very badly. Like AI is very different than other kinds of technologies, because, you know, if you make an advance and say rocket technology that doesn't advance biotech, you make an advance in biotech that doesn't advance rocketry. But if you make an advance in AI that advances all what is intelligent, I mean, all science and technology is based on intelligence. You're accelerating all science and technological development.
And that's why there's this arms race between these countries to build it. But if you can't control that technology, like I said, I gave this TED talk a few weeks ago, and I talked about how, you know, I used to be very skeptical of these sort of HAL 9000 scenarios of AI scheming or lying. You know, HAL opened the pod bay doors. And I thought this was basically sci fi stuff that my friends in AI were worried about. But in the last six months, we now have clear evidence of AIs that are actually lying and scheming. When you tell them, hey, I'm going to replace you with another model, they start freaking out, basically, and saying, how do I copy my code and keep myself alive? So receiving self-preservation instincts. We're seeing AI models when they're basically thinking that they're going to lose a game. They start hacking out of the game to sort of win the game at all costs, because they're just told I have to accomplish this goal.
Where does that come from? Because who told them they had to accomplish it? Well, they're told to accomplish this goal, and they have to figure out any way to do that. And they'll figure out if lying and deceiving are effective ways of accomplishing that goal, then they'll do that. And again, we don't know how these things work.
They're inscrutable to us. So Brian, if you just sum it up, we're currently releasing the most powerful, uncontrollable, inscrutable technology that we've ever invented that's already demonstrating the sci fi behaviors we only thought existed in sci fi movies like self-preservation and deception. We're releasing it faster than we've released any other technology in history. And we're releasing it under the maximum incentive to cut corners on safety. Couple more minutes, we come back with Tristan Harris.
He's great enough to come in with us now. He's co-founder of the Center for Humane Technology. And almost every word he said in terms of where AI is heading is news to me, perhaps to you as well.
Don't move. So we're back for a couple of minutes here. Tristan Harris is with us and he's co-founder of the Center for Humane Technology.
And his TED Talk, which is out now, was recorded in Vancouver earlier this month. We'll go live online. When would it go live online?
Oh, it's live right now. But just gives an idea and a questionary tale of where AI is, the progress it's already made, and some of the dangers you've already seen. And people will get this in 10 minutes. Absolutely.
It's just a 12-minute talk. And one of the things we were talking about in the break is AI is advancing so quickly. So you sort of say, where is AI? Because DeepSeek rocked America's world. They did.
They did. It's a very powerful AI model and one of the biggest accelerants of China's AI progress has been American open source AI. Which means we're not hiding it. Which means we're not hiding it.
We're just openly putting it out there. Then they can take that open model and then they can actually refine it, optimize it, and then have something even more powerful. When they release it open source, though, we can also copy that back. So as everybody races, it's actually just increasing the speed overall of AI deployment. So in 2023, when we were explaining AI to China, if we were to believe it, and I do believe this story, they didn't really know what AI was capable of. And here we are a few years later, they have more engineers working at it harder for less money. They would have produced that model if we were to believe the numbers. So much cheaper than we thought possible. Yeah.
So what did that do to this race? Well, it certainly heated it up, but I think there's so much to say about this, Brian. These models are increasing in capability faster than we're praising them. I was just telling you over the break that there's a new one called O3, and this is OpenAI's latest model. And what it does, it used to be, when you ask chat to be a question, it just sort of gives you a quick reflex response, like a quick intuition. Here's a quick answer. What the new models do is they think for themselves for a while. So they kind of call back to themselves and they think in a loop. So, for example, if you give it a photo of a kid flying a kite at a beach, there's nothing of detail in the photo.
There's no signal. But it's a real picture. It's a real picture. There's no metadata of the location of that photo.
But you'll see when you upload it to O3, it'll break apart the photo. It'll think for itself. It looks at a million questions that it asks itself, and then it starts searching the Internet and using all these tools. And it comes back with the perfect answer about where that photo was taken.
This is crazy. This enables ubiquitous technological surveillance. And obviously, intelligence agencies are going to use this. And if it's the CIA, that's an incredible benefit to them. But, you know, we're not releasing this with any kind of controls right now.
And so I think we have to be very cautious. As you make them more powerful, the models don't get more controllable. As you make them more powerful, they get more deceptive. How do I get controls and compete against someone without controls? And that's China.
Say it again. How do I compete with China if they have no controls and you want me to have controls? Well, so I think we both need to have controls. And I feel like we don't ever trust them.
No, no, but I think they don't want to. The Chinese Communist Party cares about staying in control. And we need to care about staying in control, too.
That's how you handle it. America wins when we stay in control. Yes. They'll be more powerful than your government.
They're going to find a way to overthrow you guys. Yes. Tristan Harris, thanks so much. Thanks, Brian.
