Latest / Trumponomics / The Tough AI Questions Looming Over the Xi-Trump Summit

Transcript

The Tough AI Questions Looming Over the Xi-Trump Summit · Trumponomics

0:02
Speaker 1: Bloomberg Audio Studios. Podcasts. Radio. News.
0:08
Speaker 2: President Xi and I have a good relationship. We're going
0:10
Speaker 2: to have a lot of different deals here.
0:12
Speaker 3: We intend, we're not just doing it for our health.
0:14
Speaker 1: I would honestly say if you considered how dependent we
0:17
Speaker 1: are in China now compared to five years ago, roughly
0:19
Speaker 1: around the time of the passage of the CHIPS Act,
0:21
Speaker 1: we're probably more dependent now because we've improved in one
0:24
Speaker 1: category and become much more dependent on others. And so
0:27
Speaker 1: I don't think this relationship or the U.S. could really
0:30
Speaker 1: bear a true impact deglobalization or sort of this kind
0:34
Speaker 1: of two technological spheres that's been discussed and so this
0:37
Speaker 1: relationship will just have to be managed maybe six months
0:40
Speaker 1: at a time three months at a time which now
0:42
Speaker 1: seems to be the pattern going forward so.
0:53
Speaker 3: I'm Stephanie Flanders, Head of Government and Economics at Bloomberg,
0:56
Speaker 3: and this is Trumponomics, the podcast that looks at everything
1:00
Speaker 3: in the economic world of Donald Trump. Now, we're recording
1:03
Speaker 3: this Wednesday morning, the 23rd of September. We're several hours
1:09
Speaker 3: ahead of President Donald Trump meeting Chinese President Xi Jinping's
1:14
Speaker 3: plane in person at Andrews Air Force Base, just outside Washington. Now,
1:19
Speaker 3: there'll be two notable features of this tarmac encounter. The
1:24
Speaker 3: first is that no one can remember any previous president
1:26
Speaker 3: taking the trouble to go. I guess he probably won't
1:29
Speaker 3: need to hold up a sign with the Chinese leader's
1:32
Speaker 3: name on it. The second historical quirk is that the
1:36
Speaker 3: event is likely to be recorded only by a Chinese
1:40
Speaker 3: state TV camera, the regular White House pool cameras being
1:43
Speaker 3: on strike in response to CNN and two other outlets
1:46
Speaker 3: being banned from the White House. Now, the value of
1:50
Speaker 3: these lofty summits of powerful men is supposed to be
1:53
Speaker 3: they can talk directly on the major issues of the day,
1:55
Speaker 3: cut through bureaucratic hurdles, get to the meat of the problem,
1:59
Speaker 3: maybe even get closer to an answer. But that's not
2:03
Speaker 3: how Donald Trump rolls. And that's not been the mood
2:06
Speaker 3: music around this summit. In fact, the risk is the
2:09
Speaker 3: mood music and pageantry might be pretty much all there is,
2:13
Speaker 3: when in fact, there really are some big issues we
2:15
Speaker 3: might want these two guys to get to grips with.
2:18
Speaker 3: Not least, the prospect that in America and China's great
2:22
Speaker 3: race for AI supremacy, humanity might be accidentally wiped out. Now,
2:27
Speaker 3: President Trump has dismissed most of those kind of concerns
2:30
Speaker 3: as a hoax. Beijing, for its part, has signalled it
2:33
Speaker 3: is willing to discuss a narrow category of genuinely dangerous
2:38
Speaker 3: AI capabilities. But China also wants Washington to ease restrictions
2:42
Speaker 3: on advanced chips it needs to power its AI models.
2:47
Speaker 3: And trust between the two is low. So what can
2:50
Speaker 3: we realistically expect to come out of this summit? And
2:53
Speaker 3: what would be needed to reduce the risk that AI
2:57
Speaker 3: will kill us all? Well, joining me from Hong Kong
2:59
Speaker 3: again is Dan Tenkate, Bloomberg's Executive Editor for Economics and
3:04
Speaker 3: Politics in Asia. Dan, thanks for joining us so late
3:07
Speaker 3: in your day.
3:08
Speaker 2: Great to be here.
3:10
Speaker 3: And joining us from Washington, our guru on all things AI,
3:13
Speaker 3: Michael Deng, Geoeconomics Technology Analyst at Bloomberg Economics. And before
3:18
Speaker 3: coming to us, he served as a Policy Advisor and
3:21
Speaker 3: Presidential Management Fellow in the CHIPS Programme Office at the
3:24
Speaker 3: US Department of Commerce. Michael, thanks for coming back.
3:27
Speaker 1: Yeah, of course. Happy to be here.
3:32
Speaker 3: Now, Dan, you've recently returned from Beijing, and I know
3:34
Speaker 3: you're in close contact with our team there. What is
3:37
Speaker 3: the mood around this summit today? sitting where they sit.
3:41
Speaker 2: Yeah, I was in Beijing a couple of weeks ago
3:43
Speaker 2: talking with people there as well. Definitely China wants stability
3:48
Speaker 2: out of this summit. The expectations are pretty low for
3:52
Speaker 2: any deliverables. I don't think they have too much ambition
3:57
Speaker 2: for what could actually be accomplished. They would like to
3:59
Speaker 2: see an extension of the trade truce for two years
4:03
Speaker 2: through Trump's term. That's not going to happen. Jameson Greer
4:07
Speaker 2: has already made that pretty clear. But nonetheless, I think
4:11
Speaker 2: China views the leader to leader diplomacy, as I like
4:15
Speaker 2: to say, between Trump and Xi as something that will
4:19
Speaker 2: stabilize this relationship. And they also think they could probably
4:24
Speaker 2: keep Taiwan in check, they could keep exports going. So
4:28
Speaker 2: they have a lot of interest in just meeting and
4:30
Speaker 2: keeping the relationship from blowing up.
4:33
Speaker 3: One of the phrases I've heard a lot about this
4:34
Speaker 3: summit has been calmer, but not closer. There's a sort
4:37
Speaker 3: of Yeah, I think that's a great point. around AI
4:58
Speaker 3: safety and the risks to future development of AI. Is
5:01
Speaker 3: that discussion as live in Beijing as it is in
5:06
Speaker 3: the US and other parts of the Western world?
5:08
Speaker 2: Not in the same way. I think they're basically technology optimists.
5:12
Speaker 2: They think that AI will create more jobs than it
5:16
Speaker 2: will destroy. But the focus there is really employment. They're
5:20
Speaker 2: worried that the transition to AI will lead to a
5:23
Speaker 2: lot of job losses, and that could lead to instability
5:26
Speaker 2: for the Xi government. And so that is where most
5:30
Speaker 2: of the discussion is focused, not so much on the
5:33
Speaker 2: safety risks. I think China views themselves as very much
5:37
Speaker 2: behind the U.S. in this race. They think they can
5:39
Speaker 2: catch up, but they see themselves making progress, and they
5:42
Speaker 2: don't really want to slow down on that front. At
5:46
Speaker 2: the same time, the country's top spy chief, had a
5:49
Speaker 2: paper that came out within the past couple of weeks
5:52
Speaker 2: that identified a lot of risks to AI. And surprise, surprise,
5:56
Speaker 2: the top one he flagged was a risk to China's
6:00
Speaker 2: political security, so the regime itself. And I think that
6:03
Speaker 2: that shows you the thinking. Of course, it's always about
6:07
Speaker 2: the Communist Party. And if they see a threat to
6:10
Speaker 2: their own power, then we've seen them be able to
6:13
Speaker 2: take action very quickly. So they are watching that very closely.
6:16
Speaker 2: But there's not a lot of talk about extinction of
6:19
Speaker 2: humanity or anything like that.
6:21
Speaker 3: Not yet, baby. We've long framed this as a race
6:27
Speaker 3: between the US and China, but safety is now coming
6:30
Speaker 3: much more to the fore. What usefully do we think
6:34
Speaker 3: they can discuss on that topic?
6:36
Speaker 1: So I would say for this dialogue in particular, I
6:39
Speaker 1: would keep expectations pretty modest. I think From everything we've
6:42
Speaker 1: heard so far and that we might hear once the
6:45
Speaker 1: dialogue starts, the expectation should largely be about beginning to
6:49
Speaker 1: set up this process rather than anything seriously concrete or
6:53
Speaker 1: that actually advances any key objectives. We've heard Secretary Besant
6:57
Speaker 1: say that there's some kind of incident notification mechanism. There's
7:00
Speaker 1: other reporting that indicates that may largely just be an
7:03
Speaker 1: open line of communication between him and He Lifeng. So
7:07
Speaker 1: I think the key here is to establish trust and
7:09
Speaker 1: get a dialogue going in the first place. We may
7:11
Speaker 1: hear positive signals about that, but I think the challenge
7:14
Speaker 1: is going to be once they run into having to
7:17
Speaker 1: agree on basic definitions, for example, about what constitutes an
7:21
Speaker 1: advanced capability, what's normal technological competition, what it might require
7:25
Speaker 1: of each side in terms of slowing down, which both
7:28
Speaker 1: sides now have denied wanting to do, how do they
7:31
Speaker 1: get their governments and these companies to actually commit to
7:33
Speaker 1: it and to be serious about these safety commitments? So
7:36
Speaker 1: I think anything we hear out of this dialogue initially
7:38
Speaker 1: is going to be quite easy to And the hard
7:40
Speaker 1: part is going to come later, potentially at the Shenzhen
7:43
Speaker 1: meeting that's mentioned in two months.
7:45
Speaker 3: Scott Besson, the Treasury Secretary, he had this kind of
7:47
Speaker 3: prior discussions with his Chinese counterpart and then there was
7:52
Speaker 3: this discussion around starting or there was a somewhat of
7:55
Speaker 3: announcement saying they've agreed to establish an AI dialogue. I
7:59
Speaker 3: noticed now maybe this is just but the AI dialogue
8:02
Speaker 3: with no capital on D so it's not like an
8:04
Speaker 3: official ongoing it's not like the various commissions and the
8:08
Speaker 3: sort of committees the joint economic committees that we've had
8:10
Speaker 3: in the past or anything like that it is just
8:12
Speaker 3: literally a conversation and they've each got each other's number
8:16
Speaker 3: kind of thing.
8:17
Speaker 2: Yeah and let's be clear, they were going to talk anyway.
8:21
Speaker 2: And so Besant and He Lefong have been talking since
8:25
Speaker 2: Trump's Liberation Day tariffs. They were talking all last year.
8:28
Speaker 2: They struck the trade truce. And this is now the
8:31
Speaker 2: main mechanism of communication between the U.S. and China on
8:35
Speaker 2: the relationship overall. So they just added AI to the
8:39
Speaker 2: list right now. There's nothing else there for the moment.
8:42
Speaker 2: And so that's where it's going to stay unless they
8:45
Speaker 2: put some meat on the bones. But as Michael said,
8:47
Speaker 2: it's really really thin at the moment.
8:49
Speaker 3: Running through a lot of this is a sort of
8:52
Speaker 3: basic level of trust. So although there was this sort
8:56
Speaker 3: of agreement, we're going to carry on talking and everything else,
8:59
Speaker 3: I can't help remembering that Scott Besson was also the
9:01
Speaker 3: person who said recently that there's no day after tomorrow
9:04
Speaker 3: if China wins. I mean, he's talking up the race aspect,
9:09
Speaker 3: even as they pay lip service to some of these
9:12
Speaker 3: security concerns. So where are the points of tension? If
9:16
Speaker 3: they were going to do what we wanted them to do,
9:17
Speaker 3: which is to seriously discuss some shared framework for at
9:22
Speaker 3: least the extreme risks that people like Dario Amadei are
9:27
Speaker 3: worried about, where would the focus be and what are
9:30
Speaker 3: the obstacles to doing that?
9:32
Speaker 2: Yeah.
9:32
Speaker 1: So I think in any discussion, if we talk about
9:34
Speaker 1: the potential risks first that they want to discuss, I
9:37
Speaker 1: think You could largely group the problem with AI safety
9:41
Speaker 1: into three categories of risk. I think the first one
9:44
Speaker 1: is probably malicious use or the idea that you and
9:46
Speaker 1: a model could enable third party rogue actors or criminals
9:50
Speaker 1: to conduct cyber attacks, etc. I think the second one
9:53
Speaker 1: is unexpected autonomous behavior. So an AI agent sort of
9:56
Speaker 1: gaining access or taking actions that the creators didn't expect.
10:00
Speaker 1: And this is largely actually what happened over the summer
10:02
Speaker 1: with many of these cyber incidents. I think the third
10:04
Speaker 1: is international miscalculation. This is probably the most familiar in
10:08
Speaker 1: terms of government-to-government dialogue. So if you imagine a major
10:11
Speaker 1: AI-enabled incident hitting either government and making sure that it
10:14
Speaker 1: wasn't the other one doing it, I think the notification
10:17
Speaker 1: mechanism we've gotten so far is probably most useful for
10:20
Speaker 1: the third. I think progressing on the first two is
10:22
Speaker 1: going to require greater levels of trust and some kind
10:26
Speaker 1: of actual concrete action or commitment from either side. I
10:29
Speaker 1: think the difficult part that comes from this is then
10:31
Speaker 1: settling on joint definitions. For example, China has signaled they're
10:35
Speaker 1: happy to talk, but they want to draw a very
10:37
Speaker 1: clear line between normal technological competition and this kind of
10:42
Speaker 1: fenced-off area of advanced capability or advanced risk. I think
10:45
Speaker 1: the problem is that line is very fuzzy, and many
10:49
Speaker 1: of the capabilities or sort of training methods that you
10:54
Speaker 1: would use to just simply improve your models within the
10:57
Speaker 1: normal tech competition arena inevitably will bleed into this higher
11:00
Speaker 1: area of risk. And then therefore, what you've seen from
11:03
Speaker 1: Anthropic and OpenAI is they have sort of self-volunteered or
11:06
Speaker 1: done by themselves to sort of weaken their models in
11:08
Speaker 1: these areas. The government hasn't until very recently actually required
11:12
Speaker 1: them to do so. So if you would want China
11:14
Speaker 1: to actually take substantive action on this, how do you
11:16
Speaker 1: convince them that in this narrow area you should specifically
11:19
Speaker 1: slow down your companies when you're in a race with
11:22
Speaker 1: the US as well and worried about being behind. And
11:24
Speaker 1: so that is what really requires trust between these two sides.
11:27
Speaker 3: And that runs pretty deep, Dan. I mean, the perception,
11:30
Speaker 3: which is not wrong, because you hear cabinet ministers, cabinet
11:34
Speaker 3: secretaries saying things along these lines fairly regularly, that the
11:38
Speaker 3: US wants to slow China's development in this area. Are
11:43
Speaker 3: there any areas that you think the leadership would be
11:46
Speaker 3: willing to say, okay, we recognize that this is too important?
11:50
Speaker 3: to be subject to this lack of trust?
11:54
Speaker 2: I think China definitely sees it on that high level.
11:57
Speaker 2: Like we should treat AI like nuclear weapons and we
12:00
Speaker 2: should come and agree not to kill each other. I mean,
12:03
Speaker 2: that is essentially something that is in their interest and
12:07
Speaker 2: that they would want to do. But as Michael said,
12:09
Speaker 2: the line in there is very fuzzy. I mean, we
12:11
Speaker 2: often portray AI as a race, but it's kind of
12:14
Speaker 2: a race that doesn't really have an end per se.
12:19
Speaker 2: So when Scott Besson says, okay, there's no day after
12:22
Speaker 2: tomorrow if China wins, the Chinese researchers I was speaking
12:26
Speaker 2: to in Beijing, they see this going in waves. There's
12:28
Speaker 2: going to be developments where China narrows the gap and
12:32
Speaker 2: then the U.S. will have something and the gap gets
12:34
Speaker 2: a bit wider. So there's not a linear kind of
12:38
Speaker 2: race and then somebody wins and then the world is
12:41
Speaker 2: over type of scenario. China really views this as a
12:45
Speaker 2: long-term thing. They want to be seen as an equal partner,
12:48
Speaker 2: I guess, in any of these talks. So, yeah, anything
12:53
Speaker 2: that appears like it's the U.S. trying to slow China
12:57
Speaker 2: down so the U.S. can get an advantage, China would
12:59
Speaker 2: be very wary of. There's just not a lot of
13:01
Speaker 2: trust between these governments, and they're very skeptical of what's
13:06
Speaker 2: coming out of Washington.
13:07
Speaker 3: So that's why there's so much distrust on the China side.
13:11
Speaker 3: But Michael, as you've pointed out in your piece, on
13:14
Speaker 3: the US side, what has reduced trust, made them less
13:19
Speaker 3: open to this conversation is this issue of distillation. So
13:22
Speaker 3: put basically, it's about, once again, China finding a way
13:26
Speaker 3: to steal a US technology, but talk us through it.
13:30
Speaker 1: So at a high level, distillation is really a normal
13:32
Speaker 1: AI development technique. And I think that's where some of
13:34
Speaker 1: these definitions get muddled. So distillation Essentially, it's about using
13:38
Speaker 1: the outputs of a stronger teacher model to help train
13:41
Speaker 1: a weaker student one. There's various techniques that you can
13:44
Speaker 1: classify as distillation, but that's probably the most general one.
13:47
Speaker 1: What's happened on the U.S. side is they specifically said
13:50
Speaker 1: that China is conducting these exploitative distillation campaigns against U.S.
13:56
Speaker 1: models where these U.S. models' terms of service specifically prohibit
14:01
Speaker 1: them to do so, and they are using all kinds
14:04
Speaker 1: of methods to basically setting up false accounts, setting up
14:06
Speaker 1: routing methods, essentially to, and even in, you know, Anthropocene
14:10
Speaker 1: accused them of essentially rerouting traffic in terms of prompts
14:14
Speaker 1: being sent to a Chinese model. They are then sending
14:16
Speaker 1: it to Claude, collecting the answers and training on the
14:19
Speaker 1: answers and sending it back to the original user. The
14:21
Speaker 1: Chinese response has been, of course, that this is normal
14:23
Speaker 1: technological competition, that this is a widespread technique in the industry,
14:27
Speaker 1: and it is. But I think the difference is, It's
14:29
Speaker 1: a widespread technique in the industry when you use your
14:32
Speaker 1: own models or when you use it on models where
14:34
Speaker 1: they have allowed you to distill from. What China is
14:37
Speaker 1: doing is a little bit beyond what the quote unquote
14:41
Speaker 1: normal development technique is in terms of they're conducting these
14:44
Speaker 1: campaigns specifically against closed source models that don't want them
14:47
Speaker 1: to do so. And so that's the tricky part. And
14:50
Speaker 1: the difficult thing is it's also quite hard to stop.
14:53
Speaker 1: It's not really like many of the hardware export controls
14:56
Speaker 1: that the U.S. has implemented over the past five years.
14:58
Speaker 1: where there's a physical device and a shipment that you
15:00
Speaker 1: can track. This is really about, you know, at the
15:04
Speaker 1: core of API access. Again, when we talk about fuzzy
15:07
Speaker 1: or clear lines, it's very hard to distinguish between what's
15:11
Speaker 1: an illegitimate user and what is a Chinese AI firm
15:13
Speaker 1: trying to distill, because both are just sending prompts to
15:16
Speaker 1: your model that it's trying to respond to. And so
15:18
Speaker 1: this is one of the biggest areas of distrust from
15:21
Speaker 1: the US. The AI labs are upset about it. Many
15:24
Speaker 1: security agencies are upset about it. Congress is upset about it.
15:27
Speaker 1: And if there was one thing that could really destabilize
15:29
Speaker 1: this conversation moving forward, when we talk about what is
15:32
Speaker 1: enabling China to have these capabilities in the first place,
15:34
Speaker 1: I think it would be this issue.
15:36
Speaker 3: Hearing what you say about distillation, I mean, it would
15:38
Speaker 3: be a helpful thing for the president and President Xi
15:42
Speaker 3: to discuss. Let me put it this way. I mean,
15:45
Speaker 3: President Trump doesn't tend to embrace that kind of technical detail.
15:49
Speaker 3: So it just doesn't feel like it's the kind of
15:50
Speaker 3: thing that's going to be on the talking points.
15:54
Speaker 1: No, I think President Xi Trump's understanding of artificial or
15:59
Speaker 1: super intelligence as he now brands it is somewhat limited.
16:03
Speaker 1: And generally, he seems to be representing points of officials
16:07
Speaker 1: or more technical officials and issues that they bring up
16:09
Speaker 1: in his agenda. I think he would bring this up
16:11
Speaker 1: as advocating, again, how the U.S. is number one, how
16:14
Speaker 1: we're in the race ahead of China. But this kind
16:17
Speaker 1: of distillation issue seems a layer or two below what
16:20
Speaker 1: he would actually be willing to engage in. And it
16:21
Speaker 1: would actually fall in the realm of something like this
16:23
Speaker 1: dialogue which makes it even trickier in terms of allowing
16:26
Speaker 1: it to succeed.
16:40
Speaker 3: Dan, one thing I've noticed that sort of has crept
16:42
Speaker 3: into these conversations, although we talk about the race and
16:44
Speaker 3: we've tended to focus on the frontier, is the biggest
16:47
Speaker 3: story of the last six months that Chinese models, despite
16:51
Speaker 3: being less sophisticated, have actually vastly extended their reach in
16:56
Speaker 3: terms of usage because they're so much cheaper and because
16:59
Speaker 3: they have that open source approach. I mean, in that sense,
17:02
Speaker 3: the Chinese feel like they're coming out ahead, even if
17:05
Speaker 3: they're still behind on this kind of frontier sense.
17:09
Speaker 2: Yeah, exactly. I think China views this as a public good.
17:12
Speaker 2: Of course, there are behind in the race, so they
17:15
Speaker 2: have an advantage using open source models. But they're also
17:20
Speaker 2: undercutting the potential valuations of these big companies like Anthropic
17:25
Speaker 2: and OpenAI in doing so. They're also winning a lot
17:28
Speaker 2: of support throughout the global south because a lot of
17:31
Speaker 2: people can use Chinese AI very cheaply. There is a
17:35
Speaker 2: question about how some of these companies can monetize that
17:40
Speaker 2: at the end of the day. But for the moment,
17:42
Speaker 2: China is getting people hooked on its models, and I
17:47
Speaker 2: guess it depends how sticky that is going forward. You know,
17:50
Speaker 2: if you settle into a certain routine using whatever AI
17:55
Speaker 2: you're using, then maybe you just stick with that, and
17:58
Speaker 2: then maybe you buy add-on products, whatever, as long as
18:02
Speaker 2: that infrastructure is laid down. And so China, at least
18:06
Speaker 2: in my conversations, they plan to stick with this open-source
18:10
Speaker 2: approach model. They view it as a good thing. I
18:13
Speaker 2: think they are worried about data being routed through cloud.
18:17
Speaker 2: For instance, we had the information reporting that there is
18:19
Speaker 2: a probe now into deep seek and moonshot where China
18:25
Speaker 2: is very concerned that sensitive military data, police data and
18:29
Speaker 2: different stuff that's secret is somehow being routed through these companies.
18:34
Speaker 2: And they're also scared about mythos, for example, like, you know,
18:39
Speaker 2: Scott Besson says oh my God, if China wins, it's
18:42
Speaker 2: the day after tomorrow. But from the Chinese perspective, the
18:45
Speaker 2: US already has these tools that they don't have. And
18:48
Speaker 2: so they're very worried that they're going to be used
18:50
Speaker 2: against them as well.
18:51
Speaker 3: Michael, this reliance on open source models, I mean, at
18:55
Speaker 3: least a few months ago, we felt that was one
18:59
Speaker 3: of the biggest risks to particularly the kind of malign
19:03
Speaker 3: actor risk that you were talking about earlier comes from
19:09
Speaker 3: quite a lot of power being in the hands of
19:11
Speaker 3: being embodied in open source models, which then have no
19:16
Speaker 3: kill switch. Presumably, that's still something we should be worried about.
19:18
Speaker 1: I do think it's still a risk. And I will
19:20
Speaker 1: say China's stance is interesting because there is a split
19:25
Speaker 1: between what it does internally and what it does externally. Internally,
19:29
Speaker 1: in terms of regulations and security review processes on Chinese
19:33
Speaker 1: citizens using AI services, they're actually quite stringent. At least
19:37
Speaker 1: definitely there's more regulation than the US, for example.
19:39
Speaker 3: And that's on things like chatbots and children's AI companions
19:43
Speaker 3: and things like that. Exactly.
19:44
Speaker 1: And I would say as a general point, China has
19:46
Speaker 1: far better control over its internal internet than any other country.
19:50
Speaker 1: does over the global internet, right? And so internally, they
19:52
Speaker 1: feel more secure and locked down about potentially open weight risks.
19:55
Speaker 1: And you also have the idea that open weight models,
19:58
Speaker 1: as of now, are still a good few months, six,
20:02
Speaker 1: seven months behind the sort of cyber capabilities of a
20:04
Speaker 1: mythos level model. And the thing with these models in
20:08
Speaker 1: terms of when you talk about gaps is that it's
20:10
Speaker 1: not necessarily a linear progression. It's once these models get
20:13
Speaker 1: to a point of reliability, all of a sudden they
20:15
Speaker 1: have this capability turned on. So that's what happened. That
20:18
Speaker 1: inflection point happened in April for Mythos. It appears that
20:21
Speaker 1: inflection point has not yet happened with these Chinese openweight models.
20:24
Speaker 1: So externally, I believe up till this point, they're still
20:27
Speaker 1: comfortable releasing these openweight models, degrading the US advantage, depressing
20:31
Speaker 1: anthropic and open AI's valuations. When it gets to a
20:34
Speaker 1: point one year or two years down the line where
20:36
Speaker 1: these openweight models are capable of conducting full cyber campaigns
20:39
Speaker 1: all the way to designing and executing an exploit, I
20:42
Speaker 1: wonder if at that point China might reassess its position,
20:45
Speaker 1: because then they will be under attack externally as well.
20:48
Speaker 3: I guess you kind of wonder that they will reassess
20:50
Speaker 3: their position just before they do that, rather than once
20:54
Speaker 3: they realize, you know, with all these things, it seems
20:56
Speaker 3: to also come down to quite a fine judgment about
20:59
Speaker 3: what constitutes a the boundary. And that, I can't help noticing,
21:03
Speaker 3: seems to be shifting forward all the time. You know,
21:05
Speaker 3: it's always six months ahead. But six months ago, we
21:07
Speaker 3: were supposed to be six months away from the really
21:09
Speaker 3: scary stuff. But actually, there's so much forward momentum from
21:14
Speaker 3: both sides that no one really wants to call time,
21:17
Speaker 3: except apparently some of the leaders of these companies themselves.
21:21
Speaker 3: I mean, I guess we should do a full circle
21:25
Speaker 3: on the summit. Dan, As well as this sort of
21:28
Speaker 3: advantage it's built up recently on the use of its AI,
21:32
Speaker 3: if we kind of look at the sort of nuts
21:34
Speaker 3: and bolts of the relationship that we focused on in
21:37
Speaker 3: the past, the trade war truce and the stranglehold that
21:43
Speaker 3: China has on the rare earth supply, which is the
21:47
Speaker 3: reason for the truce, fundamentally, that does seem like quite
21:50
Speaker 3: a stable situation where both sides have sort of got
21:54
Speaker 3: their stick behind their back.
21:56
Speaker 2: Yeah, it's sort of stable, but not stable. I'd say
22:00
Speaker 2: it's below the surface. It's not very stable. You have
22:03
Speaker 2: a lot of measures being taken by the FCC to
22:06
Speaker 2: kind of put restrictions on Chinese hardware. It's kind of
22:10
Speaker 2: a stop and start type of thing. China wants a
22:14
Speaker 2: truce to continue, but the U.S. isn't happy with the
22:17
Speaker 2: amount of rare earths they're getting from China. They don't
22:20
Speaker 2: think China is being transparent and they're not being transparent
22:23
Speaker 2: really in how these things come out. From the U.S. perspective,
22:26
Speaker 2: they think if we give China a one-year or a
22:30
Speaker 2: two-year extension, then China is just not going to give
22:33
Speaker 2: any rarest at all. So the U.S. is still seeking
22:36
Speaker 2: some sort of leverage. And Jameson Greer, I believe, said
22:40
Speaker 2: three months or six months. The truce doesn't stop or
22:44
Speaker 2: expire until November 10. So Trump is supposed to head
22:48
Speaker 2: to Shenzhen right across from Hong Kong in mid-November. And
22:55
Speaker 2: from the U.S. perspective, that's the real tell. Can they
22:59
Speaker 2: afford to blow up this relationship or do they still
23:01
Speaker 2: need the rarest? And this alternative supply chain is not there.
23:05
Speaker 2: So China is very confident that they can withstand any U.S.
23:09
Speaker 2: pressure by using this card, at least for the foreseeable future.
23:15
Speaker 2: And then the other big issue is Taiwan, of course.
23:18
Speaker 2: And China is going to try and make the cost
23:21
Speaker 2: very high for Trump to go ahead with an arms
23:25
Speaker 2: sale to Taiwan. We think eventually it'll be broken up
23:29
Speaker 2: into smaller trenches so it doesn't embarrass the Chinese. But
23:33
Speaker 2: at the same time, China will make this a big
23:36
Speaker 2: issue and will kind of threaten to cut off talks
23:40
Speaker 2: or whatever, probably a lot of unspecified vague threats if
23:45
Speaker 2: this goes forward. But that's kind of top of the agenda.
23:50
Speaker 2: The other interesting thing I'll just add is Trump mentioned
23:53
Speaker 2: that he would be open to a Chinese EV factory
23:58
Speaker 2: in the U.S. if it creates American jobs. And I
24:01
Speaker 2: think if we saw anything like that, that would actually
24:03
Speaker 2: be kind of a big breakthrough given the opposition on
24:08
Speaker 2: Capitol Hill to all of this. It's, I think from
24:11
Speaker 2: the Chinese perspective, very unlikely given the consensus in Washington
24:16
Speaker 2: is very much against this right now. But If you're
24:19
Speaker 2: looking for ways that you could pull back these national
24:23
Speaker 2: security restrictions that are really leaving this relationship in a straitjacket,
24:28
Speaker 2: that would be one sort of tangible thing if it
24:31
Speaker 2: were to happen at all.
24:33
Speaker 3: And there's been some criticism on the Hill of even
24:35
Speaker 3: this decision of the president to go to the airport
24:38
Speaker 3: and all of the suggestion that we are somewhat rolling
24:43
Speaker 3: out the red carpet for China at a time when
24:46
Speaker 3: there are all these concerns. Michael, I guess, you know,
24:49
Speaker 3: as a final word for you, it just seems there's
24:50
Speaker 3: a sort of irony here that's been brought out by
24:53
Speaker 3: the discussion on rare earths. We had a story on
24:57
Speaker 3: Bloomberg today about how well a lot of Chinese basic
25:00
Speaker 3: sort of equipment manufacturers are doing out of the AI
25:03
Speaker 3: boom in the US because the sort of basic things
25:05
Speaker 3: you need to build the data centres and to do
25:08
Speaker 3: all of the production work are actually things that the
25:12
Speaker 3: US industry is still very reliant on Chinese part for.
25:16
Speaker 3: I mean, despite this being the president who has encouraged
25:19
Speaker 3: us all to de-globalize and bring production home, AI more
25:23
Speaker 3: than anything else is making that impossible and highlighting how
25:28
Speaker 3: integrated we still are.
25:29
Speaker 1: Yeah, absolutely. I think we remain incredibly interdependent, especially through AI.
25:36
Speaker 1: I mean, some of the resilience measures that have been
25:39
Speaker 1: talked about, like the CHIPS Act, et cetera, at most
25:41
Speaker 1: have moved part of a single node of the supply
25:44
Speaker 1: chain to the U.S. when there's many, many other dependencies
25:48
Speaker 1: between the U.S. and China. If not directly, then with
25:52
Speaker 1: one or two countries as intermediaries or just separated by
25:54
Speaker 1: a few stages in the supply chain. Ultimately, China is
25:57
Speaker 1: a key part of many of these component supply chains,
26:01
Speaker 1: material supply chains, even assembly supply chains that make up A.I.,
26:05
Speaker 1: they're becoming increasingly prominent in many of these other older
26:08
Speaker 1: chip categories that extend into U.S. industry or other sectors
26:12
Speaker 1: that are in technology as well. And so I would
26:14
Speaker 1: honestly say if you considered how dependent we are in
26:17
Speaker 1: China now compared to five years ago, roughly around the
26:19
Speaker 1: time of the passage of the CHIPS Act, we're probably
26:21
Speaker 1: more dependent now because we've improved in one category and
26:24
Speaker 1: become much more dependent on others. And so, yes, I
26:28
Speaker 1: don't think this relationship or the U.S. could really bear
26:33
Speaker 1: a true deglobalization or sort of this kind of two
26:36
Speaker 1: technological spheres that's been discussed. And so this relationship will
26:39
Speaker 1: just have to be managed maybe six months at a time,
26:42
Speaker 1: three months at a time, which now seems to be
26:44
Speaker 1: the pattern going forward.
26:46
Speaker 3: And there could be no better demonstration of that than
26:48
Speaker 3: the fact that along with everything else, the U.S. president
26:51
Speaker 3: is also, at least today, dependent on China for his
26:54
Speaker 3: TV footage. Michael Deng, Dante and Kate, thank you very much.
26:58
Speaker 2: Thank you.
26:59
Speaker 1: Thank you.
27:08
Speaker 3: Thanks for listening to Trumponomics from Bloomberg. It was hosted
27:11
Speaker 3: by me, Stephanie Flanders. I was joined by Dante and
27:14
Speaker 3: Kate and Michael Deng. Trumponomics was produced by Sama Sadi
27:17
Speaker 3: and Moses Andam with editorial guidance from Megan O'Neill from
27:21
Speaker 3: Bloomberg Economics. Sound design was by Blake Maples and Kelly Gary.
27:26
Speaker 3: Amy Keene is executive producer of Talk podcast for Bloomberg
27:30
Speaker 3: and Cheryl Brumley is the head of podcast. Now, to
27:34
Speaker 3: help others find us and enjoy Trumponomics, please rate and
27:39
Speaker 3: review us highly wherever you listen.