Okay, good morning everybody, and Shou, thanks for taking the time to be with us. Before we go into the chat, I thought I'd like to have the audience give some of their views. So there is a question for the poll. Can we bring the poll question up, please?
Are you all going to bring the poll question up?
Okay. If not, I tell you what I'm going to do. All right. Just, ah, okay. Is it up? Okay. Good. So we asked this question earlier on of many in the audience, and we asked: in the age of AI, what will matter more and not less?
And so there were a few answers we had, and we can still see some answers early on: human creativity, human judgment, human curiosity, human relationships, or human purpose. So for those of you who have not put in your answers and you'd still like to put them in, you can try and scan and put your answers in now. I think some of you had done it a bit earlier.
Okay, let's give it, still moving. Still moving. Let me be respectful of those who are scanning and putting your answers in.
Okay, let's give it another few seconds or so.
Oh, we're still moving.
Okay. All right. I think from here we can see from these poll results. So, thank you very much for your participation. Looks like human judgment has got 36.9%. Now, 36.6. So it's between human judgment, human relationships, and then kind of human creativity and human purpose kind of on the edge. So thank you very much for your participation. Now I've given all of you one answer to give, so it might have been a bit difficult, but for sure, since it's our special guest, I thought I'll give him a little bit more of a kalong, you know, in our Singapore terms, a bit of kong thing, right? So Shou, you can have more than one answer, but what do you think about this?
I think it's such an interesting survey, and first of all, thank you DBS for having me. Thank you, Susan, for the introduction. I am from Singapore, and I'm quite proud of that.
In transparency, I am too, born, bred, true blue.
I think what is really interesting about this survey is we don't have to question the definition of human here, right? I think human is obviously a very broad term, and between human to human, even all of us in this room, I think it is fair to say our judgment is different. We're going to have a varied judgment between myself, you know, people in various corners of the room. So the fact that we are able to, you know, very quickly come to a consensus that there's such a thing called human judgment, human relationships, human creativity, I think it juxtaposes with what we are dealing with, which is something that's not innately biologically like us and may behave in a way that's very different from us. I do agree with basically, you know, all of the above. I think all of these five things will matter a lot. I think for judgment, you know, there's a lot of talk about AI alignment at this moment in time, and of course there are some very extreme examples that have been given out there. I think there's a paperclip example that's very famous, and I may be butchering this analogy, but it's something along the lines of you ask the machines to produce more paperclips, and it ends up doing that one thing only, which is to completely overweight on production of paperclips at the expense of everything else. So that, you know, I think we're grappling a lot with how do we, as we're dealing with something that's so powerful, as Susan alluded to, things that can solve very complex mathematical problems. Now, I was chatting with some other entrepreneurs over the weekend. My view is that in the next two to three years, we're going to have a real scientific breakthrough in physics, biotech, chemistry, material science because of AI. It's very, very, very possible. And if that happens, then, you know, the question will be: how do you align the models? How do you align the machines to be doing things that are obvious to us as human beings? Like, this is very obvious that these are the boundaries of what you should do morally and ethically, and these are the boundaries of what you should not do. I don't think we have an answer, to be honest, at this moment in time. I think it really is a very complex issue, but it will matter in the age of AI. So that's on judgment.
On creativity, for TikTok, it is very relevant. Just as a way of introduction for those who don't know, TikTok is a short video app. If you don't have it, there's Wi-Fi. You can download it. It's free to use. Apart from TikTok, you know, at ByteDance, we also produce our own AI models. Right now, for our video generation model, it's called Seedance. It's probably one of the better ones today. Our industry term is SOTA. It's state-of-the-art. And our large language model is not bad. It's pretty good. It's pretty up there. And we have built those models into two businesses. One, which is to power our internal consumer apps like TikTok, like Douyin, like Topa in China. And the other is to sell it as an API through our AI team. So if anyone's interested, you know, of course, Susan, I'm going to use this opportunity to do some advertising for our model-as-a-service business.
But for creativity itself, it's quite interesting because we have allowed our users to produce a lot of very realistic AI videos using our models. I have produced a few myself. We have published it. There is still an imbalance between the percentage of videos that are being created that are AI-generated and the number of views and the percentage of views that are being seen, and it's lopsided in the way that there are more people, as a percentage, creating AI content than there are views of it. So I think there's still a very strong desire because the views is powered also by AI, by recommendation algorithm, and it's a very democratic experience. So it really shows people what they want to see. So what we took away from this is I think there's a lot of experiment with using AI to create more content, and a lot of it is a lot more creative than you can ever imagine. But at the same time, I think human adoption of just watching AI content is not the same at this moment in time. It's still low single-digit percentage in terms of views. So I think people still value a lot of authenticity. When they see a video and, you know, it's very obviously not real, it's very obviously artificial, like sort of artificially generated, I think there's still a little bit of pushback in terms of consumption. People still want to see very real, very authentic things. So I agree that human creativity will continue to be important. I'll pause here. Yeah.
Thank you, Shou. You've been very generous with commenting on that. So as we have heard from what Shou has said, actually all of these various characteristics here, which have got some people voting for, are indeed certain aspects that we human beings, I think, can and will have to work with the evolution of this technology. So in fact, actually, Shou has answered one of the questions I was going to ask him about whether in this world where AI is increasingly being used to recommend, to influence, and to make decisions, what is meaningfully human, or what remains meaningfully human. I think Shou has given quite a good kind of explanation to this. But if I may just build on that a little bit more, therefore, in light of this, given that we're at this stage of AI evolution where, as you've mentioned, you know, things are still evolving, authenticity is still very important in content creation. What, therefore, should human beings actually work on in developing our capabilities in order to remain successful in this age of AI?
I think there is increasingly the belief that this should be a very, it should be a human-AI sort of combination rather than letting the machines take complete full rein of all decisions. Now, of course, increasingly as technology gets better, I think, and as we build the infrastructure to let the agents do more things for us, and we build that trust and the guardrails, probably what will happen in the long term is we will allow the agents to do more and more things autonomously. I think this is probably the trend. But in the short term, I think, short meaning at least for the next two, three years, I think it's going to matter a lot, the skills that you develop to be able to use AI very effectively. I think that's probably the most important thing for us internally, like what we are encouraging, not just our engineers, but also, you know, our lawyers and people working in corporate functions, do is to have more firsthand experience. Because if you don't use these, and this is changing so quickly, if you don't use these models and you don't use these products yourself, you cannot figure out what exactly it can or cannot do just by reading about it. So I think one of the key characteristics, you know, at least for the next couple of years, that will be really important is being hands-on, like being curious, being very hands-on. You know, OpenClaw came out, you know, we encourage people to try it, obviously understand the boundaries of, you know, the security aspects of it, but try it for yourself so you understand what the hype is about, you understand, you know, where the boundaries of this technology can be, and it's not some mythical thing that can do everything at this moment in time. So I think a lot of that, at least for internal employees, we have developed a lot of programs to allow them to have access to frontier models. We give them some programs that we pay for some of their token consumption, not only for work use, but also a little bit of personal consumption as well. And the encouragement is to make sure people are being very hands-on and really using these new tools as they come into the market to stay on top of it. I think that's probably the most important thing right now.
Yeah.
Thank you. Thank you. I can't agree more with many of these things you're saying, right? About still having really humans at the center, making the judgment, deciding what to do, what not to do. And in this case, about really embracing AI. I think sometimes in life people take a very extreme stand. It's either humans or machines. Sometimes people take a therefore even more extreme stand to totally reject it, protest against it, and some are overly hyped up with the embracing of it. So I think that ability for us to balance, I agree with you, is actually very, very important.
Yeah, on that, I think it was, I can't remember, somebody else said this. I think short-term we are probably overestimating it, but long-term we're probably underestimating it. The trend is very, very clear. We are only at the beginning of what, you know, artificial intelligence can probably do. And if you look at, you know, where the trend is, the progress, particularly in areas of science, it's really trending on a very steep curve at this moment in time. So I think long-term, you know, we probably don't fully understand the profound impact this is going to have across our human lives, you know, over the long term. But in the short term, it's, you know, human beings need time to adopt these kind of technology. We talk about AI coding in our industry. We are probably one of the first to get disrupted by AI, especially in terms of coding itself. But coding really only takes up 30% of most of the engineers' time. A lot of the other stuff is still very human, you know, writing your engineering plan, planning, you know, talking to your product manager, and a lot of that coordination. Yes, you can see how as AI tools get better, that also goes away, but it will probably take longer than people expect. There's still a lot of, you want to let, you know, you want to double check a little bit. You're not going to deploy your code without checking at all. That's really not, you know, advisable. So short-term is probably this kind of, you know, sense in the gear is probably going to slow down progress.
But I think it's risky to think that just because it's slow in the short term that it will not become very important in the long term. That's the wrong conclusion to take away, I think.
I agree, and that's why it's very important for us to actually start embracing, building it, because at some point we believe that there will be an inflection, and when the inflection comes, it's very important for us to be ready personally as well as in the businesses that we run. If I may then just move on a little bit more, because actually, as you've mentioned, you know, it's probably something which human beings would take a bit more time to learn and adopt. And therefore, from what you have seen, not just in the business you run, but in your interactions with people, where have you really seen AI enable just ordinary people to achieve things that they previously never thought was possible?
Oh, many, many aspects. Internally, we try to simplify this into a two-by-two matrix. You know, if you take things that are routine and very non-routine, you know, very sort of differentiated as one axis, and you take things that are kind of high stakes and low stakes as the other axis, you very quickly find that things that are routine and low stakes, really honestly, the agents can automate and do very, very quickly. Anything that's higher stakes, you probably want to be a bit more careful, but the frontier models can probably do most of that too. Anything that's very high stakes and very, very differentiated and very high touch, I think those are the things that you probably still need a lot of human involvement. Let me give you one example. One thing about TikTok is we have a lot of users around the world today. Susan mentioned youth safety. You know, it's obviously a very big thing for us. Now, in order for us to understand, say, a 14-year-old's experience in Nigeria, you know, the French-speaking experience in Nigeria for a 14-year-old, you know, in the past, yes, you know, we can have our trust and safety human beings go set up an account, you know, test the account, watch a hundred videos, write a report about it. Now I can just deploy a thousand agents and do it overnight, and then next morning it will just generate its own report. What is the content you saw? You know, whether it's, how much of it is violative against our guidelines? I mean, this whole thing is automated. When we talk about content moderation, we have a very significant live streaming platform, and live, by definition, is quite difficult to moderate because it's live, and there are tens and tens of millions of live streams that's happening at any moment in time. In the past, you know, without a lot of the new AI tools in the market, it is quite difficult. You do some random sampling for live streaming rooms, but it's quite difficult to be very, very precise and very accurate at the same time. But right now with these video recognition VLM models, you're able to just very quickly take a live streaming video, convert it into a text narrative, and then match that against your community guidelines. Things like, you know, somebody's driving a car. The AI is able to see that the speedometer is tracking the car at 160 km/h. A human being might miss that, you know, if it's at 160 kilometers per hour, it's dangerous driving. You know, it's things like that that AI has tremendously improved our capability to do a lot of our work. So, I mean, I can give you a lot more examples like that. So on the one hand, you know, we obviously, like everybody else, are mindful of the new risk that it could bring to the market, the safeguards we need, you know, how do we understand the cybersecurity risk it brings. On the other hand, we're tremendously excited by how much productivity it enables within the company.
On a personal note, how has AI actually helped you already kind of amplify your own capabilities, your own creativity?
I use AI a lot as a sounding board. For example, you know, recently became a US taxpayer, and the US tax regime is practically impossible to understand, you know, without the help of AI. So with AI, it's really helped me a lot. I've also become, AI has become my personal nutritionist. I take a picture of, you know, food that I eat, and it tells me, you know, don't eat so much of this, eat more of that. It just feels like it's a very nice companion to have on a personal level, like a super smart person whom you can always have a conversation with. And on the health side, for example, I broke my shoulder earlier this year, and I maybe had 200 questions for my surgeon, but AI answered 90% of it. So I saved the surgeon 90% of questioning time. So use cases like this, I think it's just tremendously helpful.
Yeah, actually, I just met Shou yesterday at the paddock, right? And to me, my impression of Shou every time I see him, he looks pretty much very fit like that, right? And I hadn't realized he's lost some weight, he's gotten a lot fitter. And I said, how? He said AI, and exactly what he's just shared about how his agent will stop him from eating what he should have stopped eating.
