Skip to main content
Eric Xing

Eric Xing

American computer scientist

Oct 7, 2026

20:58
Yeah.
20:58
So there is a, uh, a boundary where a book can, can, uh, reach and then cannot go beyond, and that's basically where world model starts.
21:07
It is about the physical intelligence that you can do reasoning and planning and also all sorts of a cognitive thing in a physical environment using sensory signals.
21:20
So that's world model in general.
21:21
And the function of the world model is to allow you, using that information, to do a thought experiment.
21:28
Thought experiment is basically a simulation or reasoning exercise that you simulate, you know, uh, all the possibilities given actions.
21:37
Because, uh, whenever we decide to, you know, uh... what to act, you need to first know what will happen next after you act on this, before you act, and then accordingly, you choose the best action.

33 MINS LATER

55:15
Mm-hmm.
15:18
The debate, is it a legitimate one? Should we be concerned that this is the origin story of Skynet playing out in real time?
15:29
Well, of course, we need to be concerned about safety issues from any sources, including AI.
15:34
But on the other hand, when we talk about threat to humanity, we need to be a little more rigorous.
15:39
For example, we can say that one day the sun will run out of its fuel and extinct.
15:43
But there should come with a probability and also the lifespan, the time taken to get there.
15:50
There is always a trade-off between the cost, the missed opportunity, and also the risk that you actually substantially experience.
15:58
In this particular case, I'm with Jensen and with Mark because I think AI is not developing fast enough.
18:07
How much do you feel as though that open model is going to act as an effective self-regulatory mechanism for the development of AI going forward?
26:48
And I'm just wondering, as you deal with new cohorts, are you discovering skepticism from international students that are there about AI? Is anything changing? Do you feel at all what I'm talking about here?
27:04
I, I kinda appreciate, uh, you know, uh, many of your points in terms of the AI skepticism.
27:09
In fact, you know, every new disruptive technology, you know, uh, come to the world, uh, you have the same skepticism.
27:16
In fact, uh, you know, when printing press was brought to Europe, you know, you have this, uh, huge skepticism of losing control, that, uh, people, you know, uh, could have their own books, could, people could read their, their own way of their Bibles and therefore, uh, of course, also job displacement because the calligraphers could not, uh, find the jobs and so forth, right? But, um, at the end of the day, you know, it may take longer, but eventually people realize that technological involvement, uh, uh, advancement, you know, uh, you know, uh, have the ultimate benefit of moving civilization forward, as long as you treat it as a technology rather than as a religion or as a doctrine.
27:57
In here, I think AI, um, is, after all, a tool.
28:02
If you view that as a tool without submitting it, you know, to be a slave to a tool, you actually could, uh, ride on top of it and make use of it, uh, the way you want, or maybe you just put it off and not using it.
28:16
What's happening or what is really, you know, underpinning, in my opinion, the AI skepticism, is the fear that people may submit to AI and then letting go AI to be our master and also to come back, you know, uh, to harm us, you know, uh, to the point that we cannot even control it, without realizing that AI is not basically, you know, a agentive creature that is, uh, you know, not driven by a human stewardship before, uh, be behind that.

6 MINS LATER

35:01
And, and I was hoping that before we wrapped up here, you could talk a little bit about that.
26:47
And I'm just wondering, as you deal with new cohorts, are you discovering skepticism from international students that are there about AI? Is anything changing? Do you feel at all what I'm talking about here?
27:02
I, I kinda appreciate, uh, you know, uh, many of your points in terms of the AI skepticism.
27:08
In fact, you know, every new disruptive technology, you know, uh, come to the world, uh, you have the same skepticism.
27:14
In fact, uh, you know, when printing press was brought to Europe, you know, you have this, uh, huge skepticism of losing control, that, uh, people, you know, uh, could have their own books, could, people could read their, their own way of their Bibles and therefore, uh, of course, also job displacement because the calligraphers could not, uh, find the jobs and so forth, right? But, um, at the end of the day, you know, it may take longer, but eventually people realize that technological involvement, uh, uh, advancement, you know, uh, you know, uh, have the ultimate benefit of moving civilization forward, as long as you treat it as a technology rather than as a religion or as a doctrine.
27:55
In here, I think AI, um, is, after all, a tool.
28:01
If you view that as a tool without submitting it, you know, to be a slave to a tool, you actually could, uh, ride on top of it and make use of it, uh, the way you want, or maybe you just put it off and not using it.
28:15
What's happening or what is really, you know, underpinning, in my opinion, the AI skepticism, is the fear that people may submit to AI and then letting go AI to be our master and also to come back, you know, uh, to harm us, you know, uh, to the point that we cannot even control it, without realizing that AI is not basically, you know, a agentive creature that is, uh, you know, not driven by a human stewardship before, uh, behind that.

6 MINS LATER

34:59
And, and I was hoping that before we wrapped up here, you could talk a little bit about that.
Nicholas ThompsonMODERATOR
10:50
In case you think you've had a whole series of innovations in it, explain what you've done that's novel and what's different from the amazing stuff that Yijing and Yashua are working on.
11:00
Well, there's my boot.
11:02
So yeah, as Nick just mentioned, we at MPCI are among the few, maybe the only university that is actually building those foundation models from scratch.
11:13
From scratch meaning that you gather your own data, you increment your own algorithm, you build your own machine, and then you train from top to bottom, and then you release and serve the whole process.
11:25
I thought that it's important for academic to be a player like this so that we can share the knowledge to the public so that people can study many of the nuances in building this and also understand the safety and the risk issue.
11:38
In fact, I want to say that it is by no means easy.
11:41
It's very, very difficult.

29 MINS LATER

Nicholas ThompsonMODERATOR
40:38
Either

We value your privacy

We use cookies to understand how you use our platform and to improve your experience. Click “Accept All” to consent, or “Decline non-essential” to opt out of non-essential cookies. Read our Privacy Policy.