Latest / UK Column Radio / The AI Race Just Changed Forever | Silk and Steel

Transcript

The AI Race Just Changed Forever | Silk and Steel · UK Column Radio

• 0:01
  Speaker 1: Hello, and welcome to another episode of the Silk and
• 0:05
  Speaker 1: Steal Podcast live stream. I'm your host Carl zab and
• 0:09
  Speaker 1: today we have back the crowd's favorite expert on Chinese
• 0:14
  Speaker 1: tech and a I T. P. Whue. Welcome back to
• 0:17
  Speaker 1: the show. TP. You know, we just I feel like
• 0:23
  Speaker 1: we just talked about this. You know, we actually talked
• 0:26
  Speaker 1: about we had a show on the Chinese space development
• 0:29
  Speaker 1: last week, but the week before that, we just talked
• 0:32
  Speaker 1: about the breakthrough by Chinese AI company like Zuppou, like
• 0:37
  Speaker 1: their z Ai And now we have another. Uh. Some
• 0:42
  Speaker 1: analysts are calling the second Deep Deep Seek moment, which
• 0:47
  Speaker 1: is a release of Kimmy Kimmy K three model. Uh.
• 0:51
  Speaker 1: Can you tell us why it's such a big deal?
• 0:55
  Speaker 2: Yeah, So I think this is I think let's step
• 0:59
  Speaker 2: back a little bit now.
• 1:01
  Speaker 3: Personally, I've been a user of Kidney for you know,
• 1:06
  Speaker 3: not be paying money for them for several months now,
• 1:08
  Speaker 3: even before this model came out, because evers ever since
• 1:13
  Speaker 3: the start of this year, they have you know, they
• 1:16
  Speaker 3: it's it's been between them and the AI which one
• 1:19
  Speaker 3: is like further along in the model development, and.
• 1:23
  Speaker 2: You know, they're kind of the standard bearer of the China's.
• 1:27
  Speaker 3: Industry at the moment, basically, So if we step back
• 1:31
  Speaker 3: to basically twenty twenty five, when deep sekr one came out,
• 1:37
  Speaker 3: that was a big deal all the time because that
• 1:40
  Speaker 3: was like the first time that the Chinese models had
• 1:43
  Speaker 3: started out the reasoning stuff, and we thought it was
• 1:46
  Speaker 3: the shorter gap between R one and what the open
• 1:50
  Speaker 3: a open AI had and then but you know, in
• 1:54
  Speaker 3: terms of actual proficiency of the tech stack, you know,
• 1:59
  Speaker 3: the gap between the Chinas and the USAI industry.
• 2:02
  Speaker 2: Was probably larger back then than what it is now, you.
• 2:07
  Speaker 4: Know, But because d C figured out the reasoning stuff first,
• 2:12
  Speaker 4: it felt like the gap was smaller, right, But what
• 2:17
  Speaker 4: we're seeing right now is what we saw.
• 2:20
  Speaker 3: Was the US LAP a few months ago, which is
• 2:26
  Speaker 3: the sudden acceleration of model development.
• 2:29
  Speaker 2: Right.
• 2:31
  Speaker 3: I'm gonnappear on your show more often now because things
• 2:34
  Speaker 3: are happening more frequently.
• 2:36
  Speaker 2: So for example, back then, maybe last year.
• 2:41
  Speaker 3: D C would release on a model every three months
• 2:43
  Speaker 3: or releasing updated every three months, and things will get
• 2:47
  Speaker 3: a little better, right, the gap.
• 2:49
  Speaker 2: Will close a little more. And but then during those times.
• 2:55
  Speaker 3: You know, opening app will come up with something in
• 2:57
  Speaker 3: the tropic or Google will come up with something, right,
• 3:01
  Speaker 3: but right now it feels.
• 3:03
  Speaker 2: Like this is happening every whole week.
• 3:06
  Speaker 5: Yeah, yeah, right, So the reason it feels like it's
• 3:12
  Speaker 5: happening every whole.
• 3:13
  Speaker 3: Week is because it is happening every coal.
• 3:15
  Speaker 6: Week and and and you're seeing this because of what
• 3:20
  Speaker 6: we call there's some you know, I don't want to
• 3:23
  Speaker 6: use the phrase fat take off, but there is some
• 3:26
  Speaker 6: level of acceleration.
• 3:27
  Speaker 3: Going on with SMADEL development because of the fact that
• 3:31
  Speaker 3: you're no longer using just humans to do all the
• 3:35
  Speaker 3: research and uh, you know, experimentation. So when it comes
• 3:41
  Speaker 3: to a model development, right, you have to get the
• 3:45
  Speaker 3: data which takes labeling and takes identifications and things like that,
• 3:53
  Speaker 3: and then you do the you know, the model of
• 3:56
  Speaker 3: general the training runs. But then after the training run
• 4:00
  Speaker 3: you have to do validations. You know, you have to
• 4:02
  Speaker 3: know if your results is good or not right, and
• 4:04
  Speaker 3: then there's also you know, you have to do this continuously.
• 4:09
  Speaker 3: In this pre training process is also like reinforcement training
• 4:13
  Speaker 3: and where you feed it, you know, some kind of input,
• 4:18
  Speaker 3: and then you can by verifying that the output is
• 4:21
  Speaker 3: good or not you do like these post training things.
• 4:24
  Speaker 3: But a lot of this stuff is self reinforcement. So
• 4:32
  Speaker 3: once you have a process that can automate this, you
• 4:36
  Speaker 3: know what, use as many engineers to do this process.
• 4:40
  Speaker 3: You can just use the AI to validate itself, to
• 4:44
  Speaker 3: do its own like you know, AI data labeling and
• 4:47
  Speaker 3: things like that.
• 4:49
  Speaker 2: That speeds up to the higher process.
• 4:50
  Speaker 3: And if you're and if you're looking at the design
• 4:53
  Speaker 3: like a better kernel, for example, you.
• 4:56
  Speaker 2: Just ask AI to do it.
• 4:58
  Speaker 3: And if AI can do all a better kernel for you,
• 5:02
  Speaker 3: then you're basically able to have more efficient chips that
• 5:06
  Speaker 3: run your run your processes and your inference, which you
• 5:13
  Speaker 3: know allows you to do more research also, and then
• 5:17
  Speaker 3: if you have more more at these inference chips, you
• 5:20
  Speaker 3: can then tell it to do more research work like, oh,
• 5:25
  Speaker 3: figure out how you can do some analysis and figure
• 5:28
  Speaker 3: out how you can run this better.
• 5:30
  Speaker 2: Right, So what we're seeing right now.
• 5:33
  Speaker 3: Is with g IM and Kimmy, starting from five point
• 5:38
  Speaker 3: one and two point six, they got to the phase
• 5:41
  Speaker 3: where I thought they were doing some form of self
• 5:44
  Speaker 3: improvement and then that was validated with the five point
• 5:47
  Speaker 3: five point two and now kimny K three where we're
• 5:50
  Speaker 3: seeing the fruit of that. Basically, the g IM five
• 5:55
  Speaker 3: point one to five point two was about a two months.
• 5:58
  Speaker 2: Gap between the.
• 6:02
  Speaker 3: Kimmy two point six and KNNY three was about three months.
• 6:06
  Speaker 3: Uh so during this time you saw the huge jumping
• 6:11
  Speaker 3: capabilities because they're able to outmate a lot of this
• 6:16
  Speaker 3: stuff now. And in Kenney's case, it not only not
• 6:21
  Speaker 3: only improved itself, it went from like a one point
• 6:23
  Speaker 3: five showing parameter model one point six to two point eight,
• 6:28
  Speaker 3: so that CAIRD a lot of work and all this
• 6:30
  Speaker 3: stuff got sped up by the fact that the AI
• 6:33
  Speaker 3: models come batterfully, can you know, speed up of the process.
• 6:37
  Speaker 1: The wait, wait, wait did I hear you right? Did
• 6:39
  Speaker 1: you say that AI can now do data labeling by itself? So,
• 6:45
  Speaker 1: because I remember, uh, there used to be a you know,
• 6:49
  Speaker 1: Alexander WANs company Scale AI. Its whole premise is to
• 6:54
  Speaker 1: hire these uh cheap labors in served world countries to
• 6:58
  Speaker 1: do like data lay bullying for the big AI companies.
• 7:03
  Speaker 1: Is that job? Is that job basically obsolete? Now?
• 7:07
  Speaker 3: I don't think that job is obsolete per se, But
• 7:09
  Speaker 3: like in my opinion, at least from what I've been hearing,
• 7:14
  Speaker 3: like you can do more of that with well, you
• 7:18
  Speaker 3: can automate through more of more of this stuff now
• 7:20
  Speaker 3: that's my And over time, this is only going to increase, right,
• 7:25
  Speaker 3: like where we're not gonna need as many Filipino laborers
• 7:27
  Speaker 3: over time to do the stuff.
• 7:30
  Speaker 2: And one of.
• 7:31
  Speaker 3: The advantages that you have had over China for a
• 7:33
  Speaker 3: long time was that you had this hard data industry
• 7:38
  Speaker 3: that was just harder doing high quality data, right and
• 7:43
  Speaker 3: you know, trying to developing its own thing too. Now
• 7:46
  Speaker 3: that's another part of this, like you need to have
• 7:50
  Speaker 3: a domestic AI stack to help with everything.
• 7:53
  Speaker 1: Are you telling me, Alexander one scale AI is not
• 7:57
  Speaker 1: worth the seven billion dollars evaluate.
• 8:01
  Speaker 2: Oh, actually was twenty twenty million.
• 8:04
  Speaker 1: Sorry, his own share was seven billion dollars. That's how Yeah, okay.
• 8:12
  Speaker 7: You know the industry, Yeah, it was a it was
• 8:16
  Speaker 7: a it was a sweet yet it was basically digital sweatshop.
• 8:20
  Speaker 1: He just off sourced all these uh uh basic data
• 8:25
  Speaker 1: labeling job to countries like Philippines or like Kenya or
• 8:30
  Speaker 1: like some African countries and pay them below minimum wage
• 8:34
  Speaker 1: even in those countries.
• 8:36
  Speaker 3: Uh.
• 8:36
  Speaker 1: And and you know Wall Street loved him for it.
• 8:39
  Speaker 1: So I guess is this why? Uh but but why
• 8:43
  Speaker 1: is Wall Street panicking right now about uh Kimmy coming out?
• 8:50
  Speaker 2: Yeah?
• 8:50
  Speaker 3: I think the a lot of the AI hype or
• 8:56
  Speaker 3: a lot of that AI money is I actually depending
• 9:00
  Speaker 3: on the success of open Ai. Like a lot of
• 9:03
  Speaker 3: the investments people are making are based on some long
• 9:07
  Speaker 3: term like people think that there's like three laps, but
• 9:11
  Speaker 3: like a lot of early money went into AI with
• 9:15
  Speaker 3: the premise that opening AI was going to have the
• 9:19
  Speaker 3: best model forever, even though that's actually out of the
• 9:21
  Speaker 3: case anymore and Throphic exactly leaving them. But you know
• 9:25
  Speaker 3: Opening I had the early money. You just think about
• 9:27
  Speaker 3: how much money oracles it is dumping into building projects
• 9:32
  Speaker 3: for the support open Ai, and the same with you know,
• 9:36
  Speaker 3: soft Bang for example, and all these guys, right, And
• 9:41
  Speaker 3: I think.
• 9:43
  Speaker 2: If the frontier lab valuation drops, if.
• 9:47
  Speaker 3: They can't keep burning money at the rate that they are,
• 9:51
  Speaker 3: then it's kind.
• 9:51
  Speaker 2: Of crushing for these guys.
• 9:53
  Speaker 3: So essentially what you have right now is in the
• 9:57
  Speaker 3: US AI world, you have like a very rush processes
• 10:01
  Speaker 3: of everyone trying to build as many data centers as
• 10:03
  Speaker 3: they can because they have some concept of if they win,
• 10:07
  Speaker 3: they need this.
• 10:08
  Speaker 2: Is to compute so they don't lose customers.
• 10:10
  Speaker 3: Right, But it's like if everyone's everyone can't win in reality,
• 10:16
  Speaker 3: only right now, only two players having the need for
• 10:21
  Speaker 3: compute right now, Open AI and the Rocks. Right. I
• 10:25
  Speaker 3: just foke with my friend who's in charge of Meta data.
• 10:29
  Speaker 2: Center build out recently. He said that they're the ROI
• 10:32
  Speaker 2: on their data center build out. Horrible because why and
• 10:35
  Speaker 2: the reason is they built these huge data center because.
• 10:38
  Speaker 3: They wanted to have their own meta track et app, right, And.
• 10:44
  Speaker 2: The goal of this map is to because they have
• 10:47
  Speaker 2: such a.
• 10:47
  Speaker 3: Huge members set with meta with Instagram and Facebook right,
• 10:53
  Speaker 3: that they can get these users to use their AI app.
• 10:57
  Speaker 2: And they can charge like ten dollars a month feet
• 11:00
  Speaker 2: or something like that.
• 11:02
  Speaker 3: Right, So once that happens, then the ROI will be
• 11:07
  Speaker 3: worth it and they will have like a front chier lab,
• 11:09
  Speaker 3: uh you know, to on top of that. But as
• 11:14
  Speaker 3: the turnout because of their AI is so bad that
• 11:16
  Speaker 3: nobody wants to use their AI, so they're stranded with
• 11:21
  Speaker 3: this huge data center investment and they're still using a
• 11:25
  Speaker 3: lot of that for research, but there's just no interest
• 11:28
  Speaker 3: in it.
• 11:29
  Speaker 2: So so so now they're saying, Okay, we're already paid.
• 11:32
  Speaker 3: All this money and we're already committed to this much
• 11:36
  Speaker 3: cap act, so we have to use it somehow.
• 11:38
  Speaker 2: Right, So what do you do, Well, you have to
• 11:42
  Speaker 2: then leave out those capacity to somebody.
• 11:45
  Speaker 3: So then you have this case where a bunch of
• 11:47
  Speaker 3: people thought they were gonna win, they build out all
• 11:49
  Speaker 3: these like AI capacity and there's only two consumers basically
• 11:54
  Speaker 3: two customers. In China, we haven't been different. The AI
• 12:00
  Speaker 3: for themselves are not actually involved in a lot of
• 12:04
  Speaker 3: the buildouts, right, very little of it. The the big
• 12:10
  Speaker 3: attacking kin are not spending ridiculous amount of money on
• 12:14
  Speaker 3: attack although so it is very a lot of the
• 12:19
  Speaker 3: build out he's actually carried out by the sees, you know,
• 12:24
  Speaker 3: their local governments and things like that. You know, the
• 12:27
  Speaker 3: things are just a cosmaric loss to the building in China.
• 12:29
  Speaker 2: So there's just not as.
• 12:31
  Speaker 3: Much of this efficiency as you see in the US system.
• 12:36
  Speaker 2: So this is why people are panicking.
• 12:39
  Speaker 3: Like you know, there's people are a certain certain point
• 12:44
  Speaker 3: like a bubble birth because the economics doesn't supported and
• 12:50
  Speaker 3: you know a lot of people are are feeling it
• 12:52
  Speaker 3: right now in the in the sock market.
• 12:56
  Speaker 1: Is it's give me new model? Is this big deal?
• 13:00
  Speaker 1: What than z Ai which was just released like like
• 13:04
  Speaker 1: less than a month ago.
• 13:06
  Speaker 3: What I actually think that this is very similar, I
• 13:10
  Speaker 3: would say, because to me, like other people will.
• 13:14
  Speaker 2: Say King K three, is it a bigger deal?
• 13:16
  Speaker 3: Right? But for me these are the both cases are
• 13:21
  Speaker 3: exactly the same thing, which is recursion of self improvement.
• 13:26
  Speaker 3: And I actually think for me personally, what we saw
• 13:30
  Speaker 3: this week at the Queen announcement and also with someone
• 13:36
  Speaker 3: of the we saw in the World AI conference, when piece.
• 13:40
  Speaker 2: I've ever seen together, it's actually the bigger news here
• 13:43
  Speaker 2: because what we're seeing is when the.
• 13:47
  Speaker 3: Same it's like thirty I think something like thirty six percent.
• 13:52
  Speaker 2: Owned by Alba, and then when.
• 13:56
  Speaker 3: It released like a kingy K three, it's a okay, well,
• 14:01
  Speaker 3: you know we have a really big model. H It
• 14:04
  Speaker 3: is actually an opt mind super you run from a
• 14:06
  Speaker 3: sixty four AI accelerator superneral like, we recommend that because
• 14:11
  Speaker 3: you know, you have a two point eight parameters and
• 14:15
  Speaker 3: then each parameters is like us FB sixteen and you
• 14:19
  Speaker 3: need five point.
• 14:21
  Speaker 2: Six terrified memory at least, right, so that's a lot.
• 14:27
  Speaker 2: That's a lot of memory. So that's why.
• 14:30
  Speaker 3: They suggested to use at least a sixty four no
• 14:36
  Speaker 3: a supernal will call them. And an interesting part is
• 14:40
  Speaker 3: that Aliba did a presentation recently and it's said basically, okay.
• 14:47
  Speaker 2: We have this new.
• 14:50
  Speaker 3: Supernal that we have that sixty four uh sixty four
• 14:54
  Speaker 3: car like, oh okay, so I guess.
• 14:56
  Speaker 2: That this is pretty obvious that.
• 14:58
  Speaker 3: You know, the kidney machine has worked pretty extensive way
• 15:04
  Speaker 3: because a lot of like the computer they use it
• 15:05
  Speaker 3: is from an alley cloud, so they're obviously using They're saying, okay,
• 15:09
  Speaker 3: we kind of sted down in this alley cloud.
• 15:11
  Speaker 2: Superperan that it works pretty well, right, so obviously you
• 15:15
  Speaker 2: know there.
• 15:15
  Speaker 3: Are some other good computer that they use also, but uh,
• 15:18
  Speaker 3: that's kind of like the minimum they think that they need.
• 15:21
  Speaker 2: To should run to the thing. Practically we would run
• 15:23
  Speaker 2: the same wall and then but the but the other
• 15:26
  Speaker 2: interesting thing is that all of the players in the.
• 15:30
  Speaker 3: Market are continuously building bigger supernanals. And the reason why
• 15:34
  Speaker 3: they need to build bigger the supernanurals to run is
• 15:36
  Speaker 3: they need to run inference on like larger models over time.
• 15:41
  Speaker 3: So up and until like early this year, the Chinese
• 15:46
  Speaker 3: models were stuck like under showing parameters normally because you know,
• 15:52
  Speaker 3: the superpernanurals weren't ready.
• 15:55
  Speaker 2: But now that they have the superpermanals ready.
• 15:58
  Speaker 3: What you're seeing is like larger and larger, the models
• 16:01
  Speaker 3: become unavailable a little because now they're.
• 16:06
  Speaker 2: Thinking we kind of the domestic like computy for this.
• 16:08
  Speaker 2: One of the things that I saw recently a.
• 16:11
  Speaker 3: Kind of interesting thing was they're saying, this year, like
• 16:14
  Speaker 3: the twenty twenty six the models like a comma or
• 16:16
  Speaker 3: two to the three retrouling and parameter ring. But by
• 16:19
  Speaker 3: the next year or twenty and twenty eight, we're going
• 16:21
  Speaker 3: to see these five penny troubling parameter.
• 16:23
  Speaker 1: The model is a lot more Wow.
• 16:26
  Speaker 3: So what we're meaning one, you mean that the data
• 16:29
  Speaker 3: designed for that, right, Like I need a lot more
• 16:31
  Speaker 3: data to train a lot more susie out of data
• 16:33
  Speaker 3: to train these bigger on the models, and that would
• 16:36
  Speaker 3: do you don't really get that.
• 16:37
  Speaker 2: The state or the problem is people will.
• 16:39
  Speaker 3: Keep using these AI over and over leander whatever tasks
• 16:44
  Speaker 3: that they use to generate the quality data. And also
• 16:48
  Speaker 3: you need these like bigger superner run them and once
• 16:51
  Speaker 3: you have like one hundred and twenty eight card supern
• 16:55
  Speaker 3: old and each them have like two hundred and sixteen
• 16:59
  Speaker 3: yo by M for example, that's enough to run like
• 17:05
  Speaker 3: a ten chilling parameter model at a few sixteen So
• 17:11
  Speaker 3: just just by the development of these suffers, which were
• 17:13
  Speaker 3: everywhere in the World I conference, we can see that
• 17:17
  Speaker 3: the the the condition has been set up for the
• 17:20
  Speaker 3: Jachinese AI industry to build larger models and these aren't
• 17:26
  Speaker 3: necessary because of anthropic and open AI.
• 17:30
  Speaker 2: Than themselves had previously been able to get into this
• 17:35
  Speaker 2: twos to chilling parameter range.
• 17:38
  Speaker 3: We we know, we don't know for sure, like we
• 17:41
  Speaker 3: gases around to chilling parameter rings before the Chinese lab
• 17:45
  Speaker 3: we're able to do that, you know. Last part of
• 17:47
  Speaker 3: the reason that they were able to have better models
• 17:50
  Speaker 3: because they have more data.
• 17:51
  Speaker 2: They have more compute and more informantly they had all
• 17:56
  Speaker 2: these like uh, anybody else.
• 17:59
  Speaker 3: Said them to choose that can run proper models, right,
• 18:04
  Speaker 3: and then that will also use the train there you know,
• 18:08
  Speaker 3: minthols or fable whatever, those are probably like five to
• 18:12
  Speaker 3: the tent proper right. So yeah, when once you gets
• 18:18
  Speaker 3: to the when the Chinese models get a little walker,
• 18:20
  Speaker 3: they can do that.
• 18:23
  Speaker 1: Now do you think the Kidney's model release has anything
• 18:29
  Speaker 1: to do with the fact that Ali Baba said the
• 18:31
  Speaker 1: Quinn is now going to be open weight.
• 18:36
  Speaker 3: I think there is just a lot of immolusion and
• 18:40
  Speaker 3: competition going on at the moment. I think actually the
• 18:44
  Speaker 3: competition between Kimmy and that aie a little bigger because
• 18:48
  Speaker 3: they're both kind of like Beijing.
• 18:51
  Speaker 2: Grab kind of a formed or professor formed a firm, right, so.
• 19:00
  Speaker 3: They both saw they're pretty close to each other, and
• 19:03
  Speaker 3: it's just I feel like they.
• 19:07
  Speaker 2: Are trying to compete with each other.
• 19:09
  Speaker 3: And it was totally obvious for the last couple of
• 19:13
  Speaker 3: months that them people to the day people have been
• 19:16
  Speaker 3: saying that, oh, we're gonna have like this John five
• 19:19
  Speaker 3: point five by August, and it's gonna be a long
• 19:22
  Speaker 3: larger in size, and like moms are still a little
• 19:27
  Speaker 3: smaller in general, like they had they had set one
• 19:30
  Speaker 3: hundred and fifty yis parameters, so they're thinking that this
• 19:33
  Speaker 3: is gonna be over shown it probably gonna want to
• 19:35
  Speaker 3: keep going parameters now. And with the way that that
• 19:39
  Speaker 3: guy is going, they're they're just getting like new data,
• 19:42
  Speaker 3: they're just coming on line.
• 19:44
  Speaker 2: It's very possible that they're going to be able to train.
• 19:48
  Speaker 3: They're like two three or five children parameter models two
• 19:51
  Speaker 3: because that's just what the addital resources that's coming along
• 19:55
  Speaker 3: wat of them.
• 19:58
  Speaker 2: Especially as you know at become the bill.
• 20:03
  Speaker 3: Uh.
• 20:03
  Speaker 2: Some these Chinese labs now finally use a better domestic
• 20:11
  Speaker 2: cystem containing their models, so they're not stuck to their work.
• 20:17
  Speaker 3: So it's kind of be interesting how are going because
• 20:20
  Speaker 3: you know, has been relying more on black walls do
• 20:24
  Speaker 3: their thing because they they have.
• 20:29
  Speaker 2: So there's a lot of data that has black walls.
• 20:37
  Speaker 1: But that's so much for the U S sanction on
• 20:41
  Speaker 1: blackwell chips going to China. Yeah, there's so many workarounds.
• 20:45
  Speaker 1: Did you think that you know that the moon Shots
• 20:48
  Speaker 1: Kimmy model released is the reason why Ali Baba stock
• 20:52
  Speaker 1: has been shooting up recently, partly.
• 20:58
  Speaker 3: I think. I think also the quint A model of
• 21:01
  Speaker 3: solve being petitive happening. You know there they say that
• 21:06
  Speaker 3: a certain point is that that's good, is as good
• 21:09
  Speaker 3: as a case three.
• 21:09
  Speaker 2: We'll not have to see because they don't really rely
• 21:14
  Speaker 2: yet they're still saying we're still making it, just improve it.
• 21:18
  Speaker 5: And the thing that everyone wants to improving so fast
• 21:21
  Speaker 5: now like it's you can't really release the model unless
• 21:24
  Speaker 5: it's like better than the previous want to do it right,
• 21:27
  Speaker 5: other just looks fast.
• 21:29
  Speaker 2: So uh so that's why like Hay three has to
• 21:32
  Speaker 2: release this months because.
• 21:34
  Speaker 3: Once comes out h they're an all them, all the
• 21:38
  Speaker 3: all the users to go to their platforms like well,
• 21:42
  Speaker 3: the AI he said yesterday, it is time.
• 21:46
  Speaker 2: This is as good as times to subscribe now.
• 21:52
  Speaker 1: Yeah, it's fascinating like now now if you actually look
• 21:57
  Speaker 1: at usage, it's the biggest adoption is actually the Chinese
• 22:03
  Speaker 1: AI model worldwide.
• 22:04
  Speaker 2: Including US.
• 22:06
  Speaker 1: Inside the US, a lot of US companies are switching
• 22:08
  Speaker 1: to Chinese models because they're far, far cheaper. And this
• 22:12
  Speaker 1: is why recently there's noises coming out of the White House.
• 22:16
  Speaker 1: They're saying they're considering a ban or some kind of
• 22:20
  Speaker 1: regulation on Chinese usage of Chinese AI models. They're they're
• 22:25
  Speaker 1: they're serious.
• 22:26
  Speaker 2: Bannon, because it's kind of hard to be open source.
• 22:29
  Speaker 8: I think it's more likely they'll say, like these firms
• 22:32
  Speaker 8: are you know, security risks, and we warmed firms to
• 22:36
  Speaker 8: stay away from them. And then maybe US cloud can
• 22:40
  Speaker 8: host Chinese models because they're security risks, so things like that,
• 22:49
  Speaker 8: and then that will hopefully drive down the desirability of
• 22:54
  Speaker 8: the Chinese models.
• 22:57
  Speaker 3: Yeah.
• 22:57
  Speaker 1: Actually, what one reason the Ali Baba stock had what
• 23:02
  Speaker 1: was going on earlier was because he was listed by
• 23:05
  Speaker 1: United States as a Chinese military Associate company and that
• 23:10
  Speaker 1: Alibaba took them to court and apparently, you know, Ali
• 23:14
  Speaker 1: Baba won surprisingly and not just to to my own
• 23:20
  Speaker 1: horns here, I did. I did say on our last
• 23:24
  Speaker 1: show that Ali Baba back then was looking like it
• 23:29
  Speaker 1: has value. Ali Baba stock looks like it was because
• 23:32
  Speaker 1: he was in the sub hundred dollar territory at that time.
• 23:37
  Speaker 1: And again, this show do not promote stocks. Disclaimer, it's
• 23:43
  Speaker 1: not stock advice, but I did. I'm just saying, hey,
• 23:46
  Speaker 1: I did say back then a couple of weeks ago
• 23:49
  Speaker 1: that Ali Baba looks like it was in the good
• 23:52
  Speaker 1: value territory and then it has jumped twenty percent since then,
• 23:56
  Speaker 1: and I do think, you know, that's a new AI
• 23:59
  Speaker 1: advanced then has something to do with that. So where
• 24:04
  Speaker 1: do we go from here?
• 24:11
  Speaker 3: I think or we're going to see pretty much speeding
• 24:15
  Speaker 3: up on the improvement of the Chinese.
• 24:19
  Speaker 2: You know, it's the move that the progress are not
• 24:24
  Speaker 2: going to be even so in some cases you'll see
• 24:29
  Speaker 2: tiny model a little further behind, like they're.
• 24:31
  Speaker 3: Saying right now out the frontier mare something. It is
• 24:35
  Speaker 3: a little further behind. Another area like front end coding
• 24:40
  Speaker 3: is right at the top. Uh So the reas things
• 24:44
  Speaker 3: when when people say like this model is behind, this
• 24:47
  Speaker 3: models the head, it's it's kind of hard to say
• 24:50
  Speaker 3: just be because like where are you behind or you're heading,
• 24:53
  Speaker 3: You're not.
• 24:53
  Speaker 2: Going to be there.
• 24:54
  Speaker 3: It's not ever going to be consistently like this parade behind. Right.
• 25:00
  Speaker 2: But what apparently this is why probably not to be
• 25:04
  Speaker 2: true about that, was how many months some of the
• 25:07
  Speaker 2: models ahead or behind?
• 25:08
  Speaker 3: Like what is the most important is that they have
• 25:11
  Speaker 3: reaches of this, this, this, this, this, you know the
• 25:13
  Speaker 3: sast iterations process, and it is was how you know
• 25:18
  Speaker 3: competitive Chinese firms are and you know.
• 25:21
  Speaker 2: You have collusion, right, you will have a case where
• 25:24
  Speaker 2: the Chinese as moves significant only.
• 25:28
  Speaker 3: Faster, and it is very possible that you you toue,
• 25:33
  Speaker 3: you have government's own craziness that for the rest of
• 25:36
  Speaker 3: the world that the Chinese models are the best stream
• 25:39
  Speaker 3: you use for the development cybersecurity, figure generations and all
• 25:48
  Speaker 3: these things. Right, even if you have model commun the
• 25:53
  Speaker 3: beast in some areas, if you have government it says
• 25:55
  Speaker 3: that you can't use this because this is an IP
• 25:58
  Speaker 3: and then the Chinese yeah, like oh yeah, you know
• 26:03
  Speaker 3: this is it is much more desirable when you can
• 26:06
  Speaker 3: take open.
• 26:07
  Speaker 2: Source model run on your computer, your own service versus
• 26:10
  Speaker 2: like being troubled by the government. So I think there's
• 26:13
  Speaker 2: a lot of hand to that, and we're going to
• 26:18
  Speaker 2: get to a phase where the.
• 26:22
  Speaker 3: The the compute is less than an issue, where the
• 26:26
  Speaker 3: data gap is less than an issue, and where I think, uh, the.
• 26:36
  Speaker 2: The gap will just slowly way.
• 26:40
  Speaker 3: Over time, and whichever leads will be based on who
• 26:45
  Speaker 3: releases related one one.
• 26:47
  Speaker 1: Is Chinese AI company still working from a handicap versus
• 26:53
  Speaker 1: the U s AI companies.
• 26:57
  Speaker 2: I would say the data gap is still real. US has.
• 27:04
  Speaker 3: So it has a more robust industry, and the US
• 27:10
  Speaker 3: AI industry also have lot and probably the biggest one
• 27:18
  Speaker 3: funding like the US not so much more money that
• 27:21
  Speaker 3: are that they got.
• 27:22
  Speaker 2: Access to the flip.
• 27:27
  Speaker 3: You know, there's the money here from the from there
• 27:30
  Speaker 3: money them. I don't know.
• 27:34
  Speaker 1: I don't know, Uh well open a I was kinding
• 27:39
  Speaker 1: out that era of money, but I don't. I think
• 27:42
  Speaker 1: a lot less it will be force coming now because
• 27:45
  Speaker 1: they're they're they're being bombarded by Irani and Jones and
• 27:48
  Speaker 1: themselves each you know, speaking of data. You know, I
• 27:54
  Speaker 1: understand US have a significant lead in terms of funding
• 27:58
  Speaker 1: and in terms of compete, But why why why does
• 28:02
  Speaker 1: US have an edge in data? You know, I would
• 28:05
  Speaker 1: think China was its larger population, wouldn't they have more
• 28:10
  Speaker 1: data to feed the AI models.
• 28:14
  Speaker 2: Well, this is where people in the Chinese like they
• 28:18
  Speaker 2: were all using.
• 28:24
  Speaker 3: UH even that I were.
• 28:28
  Speaker 2: Originally use the run, so you can't really Yeah, you
• 28:38
  Speaker 2: can't really have data whenever one tatters like UH program.
• 28:46
  Speaker 3: Right, So that was that was the main issue in industries.
• 28:54
  Speaker 2: That they had to leave that for a while.
• 28:58
  Speaker 1: But but like why either adoption of the Chinese model world,
• 29:03
  Speaker 1: Why wouldn't you know, US and China be able to
• 29:08
  Speaker 1: leap ahead in time time in terms of data?
• 29:13
  Speaker 3: Yeah, that whole way, right, that's that's that's probably my
• 29:15
  Speaker 3: own way that can happen because althotherwise, if if you
• 29:21
  Speaker 3: keep using block hold like the there's just no over
• 29:24
  Speaker 3: for them for China to have the data lead here.
• 29:27
  Speaker 3: So I think it's very personally when when I use
• 29:31
  Speaker 3: the right, I did not use poles. That is why
• 29:35
  Speaker 3: I personally and again using alot hold because I don't
• 29:39
  Speaker 3: want to give data workers even though.
• 29:44
  Speaker 2: Like myself, I was probably not not that meaning.
• 29:46
  Speaker 3: But I think he has had so much intocropic UH
• 29:54
  Speaker 3: data engine because because all the free people.
• 29:59
  Speaker 2: Be giving them.
• 30:01
  Speaker 1: Yeah, okay, so so so we are basically the AI
• 30:09
  Speaker 1: race is still wide open. It sounds like, uh, you know,
• 30:13
  Speaker 1: between various Chinese AI companies, but also the the frontier
• 30:19
  Speaker 1: model of open AI and and tropics.
• 30:23
  Speaker 3: Is that the case, Yeah, I would say that's like, okay,
• 30:26
  Speaker 3: I think that even the Chinese AI labs are like
• 30:30
  Speaker 3: a couple of months behind.
• 30:32
  Speaker 2: It's not like the end of the world, right we we.
• 30:35
  Speaker 3: We saw some regionally, like the biggest a long time
• 30:38
  Speaker 3: was the Chinese labs behind cybersecurity. So then when you
• 30:43
  Speaker 3: have launched China Defend.
• 30:46
  Speaker 2: That was like the biggest guitar, that was the thing
• 30:49
  Speaker 2: that andff like that. But uh, if the biggest lab
• 30:55
  Speaker 2: for most of them, for the models are like all.
• 30:58
  Speaker 3: Some Chinese leading colding models don't do recognition as well,
• 31:05
  Speaker 3: or that it's.
• 31:07
  Speaker 2: Front matters as well, It's just not like that you
• 31:11
  Speaker 2: can have that. Like the most part.
• 31:13
  Speaker 3: Are can do come like long tasks, long horizon past
• 31:20
  Speaker 3: and if you look at a ranking recently for the
• 31:25
  Speaker 3: day tap, which is one of the best measures for that.
• 31:30
  Speaker 2: Basically kam K three is.
• 31:34
  Speaker 3: There's the top level which is stable that I want
• 31:37
  Speaker 3: to everyone.
• 31:38
  Speaker 2: And then then the next level, which is over.
• 31:40
  Speaker 3: The five point six five and the king came near
• 31:45
  Speaker 3: like abobout saying level. So it's like, so it is,
• 31:50
  Speaker 3: you know, pretty high up for that. So that's why
• 31:54
  Speaker 3: recently when people were coming there's quite a few like
• 31:57
  Speaker 3: us uh.
• 31:58
  Speaker 2: Tech exactly what we're saying.
• 32:00
  Speaker 3: You know, we can use clot Fable or five point
• 32:04
  Speaker 3: six all because the handicapped put out the government, but
• 32:11
  Speaker 3: case resup running these things because there's just just like okay,
• 32:17
  Speaker 3: we don't have any governments on this, so uh, all
• 32:21
  Speaker 3: of this stuff, all the fiber security path that they
• 32:24
  Speaker 3: being asked to actually collect data like this is really
• 32:28
  Speaker 3: really good that the people are using munittree globally to
• 32:33
  Speaker 3: do that every day have because you can only get
• 32:36
  Speaker 3: better from more data or over time. Ye still haven't
• 32:42
  Speaker 3: moved away from the transformer uh you know transformer uh
• 32:48
  Speaker 3: type about yet, so that's why it's still at data
• 32:51
  Speaker 3: you have.
• 32:53
  Speaker 1: So so you would say, uh, you know, in terms
• 32:56
  Speaker 1: of cybersecurity, China has each some sort of parody with say,
• 33:03
  Speaker 1: you know Fable or a mythel evanthrophics.
• 33:09
  Speaker 2: Yeah, we'll get there in the next few months. So
• 33:16
  Speaker 2: so so the current level.
• 33:19
  Speaker 3: Where you want to is actually not get who we're
• 33:22
  Speaker 3: favorable is now, but rather we're favorable will be in
• 33:26
  Speaker 3: a few months, right and h to get I think
• 33:33
  Speaker 3: we will have to get to the five crowt like
• 33:36
  Speaker 3: next year to get get I think that, uh, I
• 33:41
  Speaker 3: think they can get where we're favorable five is now
• 33:45
  Speaker 3: pretty still, but where if we.
• 33:48
  Speaker 2: Want to catch up to the you know, the thropic
• 33:51
  Speaker 2: or g.
• 33:54
  Speaker 3: Which is expected to come out, so they they didn't
• 33:57
  Speaker 3: need yellow order and then that's like that's your thing.
• 34:02
  Speaker 3: But you know, I think they're clearly effort right now
• 34:08
  Speaker 3: and it's probably not as far away as people.
• 34:13
  Speaker 1: Did you did you watch that interview of the Moonshot CEO, uh,
• 34:18
  Speaker 1: the recent most recent interview where he was asked about
• 34:22
  Speaker 1: a g I and he said, uh, you know, maybe
• 34:25
  Speaker 1: AGI is not an endpoint. It's more of a direction
• 34:28
  Speaker 1: because because it's not like moon landing. Once you landed
• 34:32
  Speaker 1: on the moon, you are you already reached your destination.
• 34:35
  Speaker 1: But with AI, you are were just exploring the boundary
• 34:40
  Speaker 1: of the knowledge. But it might turns out that there's
• 34:44
  Speaker 1: really no limit, you know, because like it's using the
• 34:48
  Speaker 1: analogy of climbing a mountain and maybe there's really no peak.
• 34:53
  Speaker 1: You know, once you climb what you thought was a peak,
• 34:55
  Speaker 1: you gain a new perspective. But there's more. You just
• 34:59
  Speaker 1: discover more layers to climb. Uh, what did you think
• 35:04
  Speaker 1: of that?
• 35:06
  Speaker 3: I made a lot a lot of sense to me,
• 35:08
  Speaker 3: like I was already better name and a lot of
• 35:11
  Speaker 3: tasks right right, So that never really named a lot
• 35:16
  Speaker 3: a lot of sense what we were talking about and
• 35:19
  Speaker 3: something like like that superlogence and stuff like that.
• 35:23
  Speaker 2: I think it was like mardating and over time, I
• 35:27
  Speaker 2: think the only way way to get to uh superherd
• 35:32
  Speaker 2: tellogence is through you know know, just more just continue.
• 35:42
  Speaker 3: Self improvement, like like what we would have seen. But
• 35:45
  Speaker 3: but I do think at some point this is what
• 35:49
  Speaker 3: just because there's only so so much stuff we can
• 35:53
  Speaker 3: verify through this means.
• 35:58
  Speaker 2: Like if you're uh drug, for example, you cannot you
• 36:03
  Speaker 2: cannot verify drug computer computer you still have to still
• 36:08
  Speaker 2: have for example, right, so right, So uh that I
• 36:14
  Speaker 2: think that I think, uh that I think that limit.
• 36:18
  Speaker 3: The limits page world real world application validation that we're
• 36:24
  Speaker 3: just not we're just not there yet yet. Yeah.
• 36:31
  Speaker 1: I want to ask some question about the recently how
• 36:35
  Speaker 1: World AI Conference that was held in China. What is
• 36:39
  Speaker 1: the significance of that?
• 36:41
  Speaker 3: Yeah, I think the general intelligence is a few general
• 36:45
  Speaker 3: importance is a few things. One, there are a lot
• 36:47
  Speaker 3: of Chinese players in the AI world in terms of
• 36:52
  Speaker 3: making like advanced servers. The superno was like a really
• 36:56
  Speaker 3: important one when that I was looking at. Another major
• 36:59
  Speaker 3: one are the AI robots. They're just like everywhere in
• 37:02
  Speaker 3: China right now. And then there is also the talk
• 37:07
  Speaker 3: of like AI large language models versus world models.
• 37:12
  Speaker 2: So world models are what the the robots use.
• 37:17
  Speaker 3: So if you talk to like a Chinese someone involved
• 37:20
  Speaker 3: in the Chinese AI industry, they talk a lot about
• 37:22
  Speaker 3: the world models, which is what the all the humanoid
• 37:27
  Speaker 3: robots and what all the the robot dogs use, and
• 37:31
  Speaker 3: those are typically a little smaller now, but there's a
• 37:34
  Speaker 3: lot of potential in the the robot model world because
• 37:40
  Speaker 3: that one you can actually have a real world reinforcement learning, like,
• 37:47
  Speaker 3: for example, you should have humanoid robot learning to pick
• 37:54
  Speaker 3: up clothing right, and it's pretty easy to identify how
• 37:59
  Speaker 3: you move, how the how robot moves its hand, whether
• 38:03
  Speaker 3: it picked up the piece of clothing it wants to
• 38:07
  Speaker 3: fold or not. Like in real world you can run
• 38:09
  Speaker 3: real world reinforcement learning, like that's something that uh you
• 38:15
  Speaker 3: can't do uh with this digital.
• 38:19
  Speaker 2: Large language model.
• 38:20
  Speaker 3: So uh so so in America, uh, here we're very
• 38:26
  Speaker 3: much stuck on just large language models. But in China
• 38:30
  Speaker 3: it has both the large language models that are that's
• 38:33
  Speaker 3: digital and also the physical world humanoid robots and other
• 38:38
  Speaker 3: kind of AI robots. Yeah. So I think that's like
• 38:41
  Speaker 3: the the big takeaway, and also the big takeaway is
• 38:45
  Speaker 3: that the Chinese government is very supportive of this.
• 38:48
  Speaker 1: This is all very exciting stuff. Is there any other
• 38:54
  Speaker 1: area of.
• 38:55
  Speaker 3: I just think that what we saw recently is very interesting,
• 39:00
  Speaker 3: and that the models are just getting so much larger now,
• 39:03
  Speaker 3: like they're they're also projecting the models to get really large.
• 39:07
  Speaker 1: And because oh oh, TP says she has to step away,
• 39:12
  Speaker 1: had some last minute tasks that my wife just dragged
• 39:16
  Speaker 1: me into. Okay, no problem, no problem, TP, thank you,
• 39:22
  Speaker 1: thank you for making the time to speak to us.
• 39:25
  Speaker 1: All right, guys, I guess that wraps are yeah TP.
• 39:30
  Speaker 1: TP has to do some house shore. His wife, his
• 39:33
  Speaker 1: boss his wife is is calling. Uh why Indonesia, Well,
• 39:40
  Speaker 1: I'm I mean Bali man. Bali is is a tropical paradise.
• 39:44
  Speaker 1: I can do Bali to surf, and I just never
• 39:47
  Speaker 1: left because this is the best island for surfing. Uh,
• 39:50
  Speaker 1: there's no reason for me to leave.
• 39:52
  Speaker 3: And I love the.
• 39:53
  Speaker 1: Tropical weather here. And now my I'm married, my wife
• 39:58
  Speaker 1: is local, my pit my both of my children are
• 40:01
  Speaker 1: born and grew up in Bali, so I'm not I'm
• 40:04
  Speaker 1: not leaving here. You know, Bali is great, although I
• 40:09
  Speaker 1: would love to take my children uh to China more
• 40:15
  Speaker 1: especially you know, my my my son, my my oldest
• 40:20
  Speaker 1: song already started school. So the only time I have
• 40:23
  Speaker 1: is really summer. But it's been really hot in China
• 40:29
  Speaker 1: this last couple of summers. So TV said sorry about this,
• 40:35
  Speaker 1: did not foresee it happening. No problem to be, no,
• 40:38
  Speaker 1: no problem, no problem at all. He's stole enough money
• 40:43
  Speaker 1: from Chinese and now he's doing big life in Bali.
• 40:47
  Speaker 1: What the heck? Well, okay, whatever, I how how I
• 40:52
  Speaker 1: don't know how I'm as I'm stealing money from Chinese
• 40:55
  Speaker 1: because hal's time. When I was doing my growing up
• 40:58
  Speaker 1: in China, I was, you know, thirteen years old back
• 41:01
  Speaker 1: in nineteen ninety. So I don't know what this person
• 41:04
  Speaker 1: is talking about. But I love theaters. I love people
• 41:07
  Speaker 1: who hate watch my show because you are helping with
• 41:11
  Speaker 1: my YouTube metrics. So thank you, you know, I this
• 41:14
  Speaker 1: is why I don't block people anyway. People people can
• 41:18
  Speaker 1: can can can can come and help help me with
• 41:22
  Speaker 1: a YouTube AI with all your stupid comments. All right, guys,
• 41:28
  Speaker 1: average hundred square meter house in Bali price, Well, it
• 41:33
  Speaker 1: depends it's location, location, location, right, Bali is still a
• 41:37
  Speaker 1: big island. It it really depends where where in Bali's
• 41:43
  Speaker 1: There are places in Bali that's quite expensive and there's
• 41:47
  Speaker 1: places in Bali that's quite cheap. Uh you know, I
• 41:50
  Speaker 1: I have.
• 41:51
  Speaker 2: We we build our.
• 41:52
  Speaker 1: House you know, in the boondogs, so it's it's cheap.
• 41:56
  Speaker 1: And I got the COVID price, you know. So we
• 41:59
  Speaker 1: bought our land during during the pandemic time, so we
• 42:03
  Speaker 1: locked out. We got the COVID price. But in some areas,
• 42:08
  Speaker 1: you know, more touristy areas, especially now, the housing prices
• 42:13
  Speaker 1: can be quite quite expensive because a lot of foreigners
• 42:16
  Speaker 1: are coming here buying our property. Actually, so I'm I
• 42:21
  Speaker 1: live among the locals, you know, I literally live among
• 42:24
  Speaker 1: the farms. We are the among the first couple of
• 42:28
  Speaker 1: families who uh who start building a house there. So
• 42:34
  Speaker 1: when I first moved there, it was all rice fields
• 42:38
  Speaker 1: and jungle so around this, which is great. I loved it.
• 42:42
  Speaker 1: Only set back is I don't have infrastructure, you know,
• 42:45
  Speaker 1: like I'm my area still don't have five G. So
• 42:49
  Speaker 1: now where I'm now, like my where me and my
• 42:52
  Speaker 1: wife were doing, we're renovating our house right now.
• 42:55
  Speaker 3: So now we're.
• 42:56
  Speaker 1: Renting this current apartment which is closer to the urban
• 43:00
  Speaker 1: center Abouli in the pasta and five G is so great.
• 43:06
  Speaker 1: Five G is so great, My god, it makes all
• 43:10
  Speaker 1: the difference, uh the internet speed. So I'm hoping eventually,
• 43:16
  Speaker 1: once my area gets more populated, maybe I will also
• 43:20
  Speaker 1: get on get get five G once my house is complete. Anyway,
• 43:25
  Speaker 1: so uh, okay, guys, thanks for coming to the show.
• 43:33
  Speaker 1: Uh you know, originally I was hoping to do one
• 43:36
  Speaker 1: hour show with uh with with TP. But but TP,
• 43:40
  Speaker 1: how much are you paying for rent? Well, I don't
• 43:42
  Speaker 1: pay rent. I own my house. Or right now I'm
• 43:47
  Speaker 1: paying rent because this current place I'm staying in is
• 43:51
  Speaker 1: temporary because I'm waiting to uh finish renovating my house.
• 43:56
  Speaker 1: The current place I'm paying about two million uh indonation Rupia.
• 44:04
  Speaker 1: How much is twenty two million I DR two USD.
• 44:08
  Speaker 1: Let me let me calculate, that's about twelve thousand dollars
• 44:12
  Speaker 1: and then twelve hundred sorry, not twelve dollars, twelve hundred
• 44:14
  Speaker 1: dollars a month. So I'm paying twelve hundred dollars for
• 44:18
  Speaker 1: two bedroom with a pool. Again, this this, this area
• 44:24
  Speaker 1: is more pricey because it's more central location. My son
• 44:28
  Speaker 1: goes to an international school and my our current apartment
• 44:34
  Speaker 1: is like five minutes away, and and so, and it's
• 44:38
  Speaker 1: closer to all the cities and shopping area and the beach.
• 44:42
  Speaker 1: So so this, this area is a lot more pricey
• 44:44
  Speaker 1: than where I where we live, where are houses houses located.
• 44:52
  Speaker 1: So Rupia is going the same way as yeah, indonation.
• 44:57
  Speaker 1: Rupia now is at the lowest level versus a dollar.
• 45:01
  Speaker 2: UH.
• 45:02
  Speaker 1: It's it went lower than the nineteen ninety eight Asian
• 45:06
  Speaker 1: financial crisis. And part of that is due to the
• 45:11
  Speaker 1: straight or home moose crisis. Because Indonesia imports about one
• 45:15
  Speaker 1: million barrels of oil per day, and and that oil
• 45:21
  Speaker 1: is pricing dollars unfortunately, and so so right now with
• 45:25
  Speaker 1: a with a high oil prices, UH, Indonesia well Indonesian
• 45:29
  Speaker 1: government subsidized all the field, the subsidized electricity, the subsidized
• 45:34
  Speaker 1: feel so right now that's eating into the Indonesian government budget.
• 45:40
  Speaker 3: And this is.
• 45:42
  Speaker 1: And and and and and this is one reason for
• 45:46
  Speaker 1: the sharp devaluation of the Indonesian currency right now. But
• 45:50
  Speaker 1: this is a problem coming throughout Asia. You know, all
• 45:53
  Speaker 1: Asian currencies right now, not just Indonesia or the Philippines,
• 45:58
  Speaker 1: but also developed economies like South Korea and Japan. There
• 46:03
  Speaker 1: there their currency also devalue it in like crazy versus
• 46:06
  Speaker 1: US dollars because all of them share one feature. Guess what,
• 46:10
  Speaker 1: They get most of their oil from Persian golf and
• 46:15
  Speaker 1: right now that that that that that that became a
• 46:21
  Speaker 1: very precious commodity. So so the only major Asian currency
• 46:27
  Speaker 1: that's actually rising against US dollar right now is guess
• 46:32
  Speaker 1: what R and B you know, people's money, Chinese Chinese Ui.
• 46:38
  Speaker 1: That's the only only major Asian currency that's actually holding
• 46:42
  Speaker 1: up its value versus a dollar at the moment. So yeah,
• 46:48
  Speaker 1: and uh yeah, yeah, don't don't don't import ferrari to Indonesia, man,
• 46:52
  Speaker 1: And uh why would you do that?
• 46:54
  Speaker 3: Uh? I?
• 46:55
  Speaker 1: I myself, I got a Chinese ev you know, like
• 46:59
  Speaker 1: a like a budget budget a Chinese EV. I got
• 47:02
  Speaker 1: a small Ruling Bingual because the Bally's roads are small,
• 47:08
  Speaker 1: you know, like, what's the point of getting a large,
• 47:10
  Speaker 1: huge ass s uvs. You probably run into people. So
• 47:13
  Speaker 1: I got a small Chinese CV and I love it
• 47:16
  Speaker 1: so far, and my my fuel is so cheap compared
• 47:20
  Speaker 1: to gas cars and also thanks to Indonesian government subsidy
• 47:25
  Speaker 1: of electricity. But I will probably install solar on my
• 47:31
  Speaker 1: house once my house renovation is done, because you know,
• 47:35
  Speaker 1: some energy is abundant here and pretty soon I will
• 47:39
  Speaker 1: be able to drive my car for free. You know,
• 47:42
  Speaker 1: I just just just just get it for the sample.
• 47:45
  Speaker 1: I'm waiting maybe for mass production of the sodian ion
• 47:51
  Speaker 1: battery from China that they just started uh this year,
• 47:56
  Speaker 1: and they will really they will ramp up their their
• 47:59
  Speaker 1: production by twenty twenty seven next year. So I'm waiting
• 48:02
  Speaker 1: a year for the sodium price to sodium battery price
• 48:07
  Speaker 1: to come down. Currently, the sodium battery price is about
• 48:12
  Speaker 1: on par with a Liftian iron phosiphate battery in China
• 48:19
  Speaker 1: right now, but that's soon going to change because China
• 48:21
  Speaker 1: just decided they are going to start place taxes. They're
• 48:26
  Speaker 1: going to start tax as lithium batteries. You know, before
• 48:31
  Speaker 1: the lifting battery enjoys zero percent tax. Now they're going
• 48:36
  Speaker 1: to put the tax back at two percent eventually four percent,
• 48:40
  Speaker 1: but the sodium battery will continue to enjoy the zero
• 48:44
  Speaker 1: percent tax status. So so you know, by next year,
• 48:47
  Speaker 1: sodium battery in China will be so cheap. They already
• 48:52
  Speaker 1: started to put them in cars. Tangan which is an
• 48:55
  Speaker 1: auto manufacturer in my hometown's Jin, They're already putting a
• 49:00
  Speaker 1: sodium battery in their car. So but they're still just
• 49:02
  Speaker 1: started doing that. I'm expecting the price to come down
• 49:07
  Speaker 1: significantly uh next year because the sodium batteries are costs
• 49:12
  Speaker 1: compared to the lifting batteries, like forty sixty percent cheaper.
• 49:16
  Speaker 1: So I'm just waiting one more year two two to
• 49:22
  Speaker 1: install solar at my house and I will enjoy free
• 49:25
  Speaker 1: electricity uh free free uh electricity for my car, for
• 49:32
  Speaker 1: my house. You'll be great, you know, I'm I But yeah,
• 49:38
  Speaker 1: so guys, thanks for again, thanks for I don't see
• 49:43
  Speaker 1: any specific question I'm going I'm trying to go in
• 49:46
  Speaker 1: through the your comments, but I don't really see any
• 49:49
  Speaker 1: specific questions for me. Uh, so I if there's no
• 49:56
  Speaker 1: more questions, let's uh, let's call it. Let's call it
• 50:00
  Speaker 1: of the day. I'm just going through your comments as
• 50:08
  Speaker 1: quickly as possible. Blah blah blah blah blah blah. Okay,
• 50:12
  Speaker 1: I don't I don't see any questions. Uh, tell us
• 50:17
  Speaker 1: about the gated community in China. Why why would I
• 50:20
  Speaker 1: know about gated community in China? Man, I don't live
• 50:23
  Speaker 1: in China. I haven't lived in China for like freaking
• 50:26
  Speaker 1: thirty years. Man, why am I telling about data? Just
• 50:30
  Speaker 1: travel to China, shit, Man, Just go to China. I'm
• 50:34
  Speaker 1: I'm gonna go to China later this year.
• 50:36
  Speaker 3: Again.
• 50:37
  Speaker 2: It's great.
• 50:38
  Speaker 1: China is a great place to visit.
• 50:40
  Speaker 2: Right now.
• 50:41
  Speaker 1: Everything is so cheap, like the hotel accommodation. A lot
• 50:45
  Speaker 1: of the like the five four five stars level hotels
• 50:49
  Speaker 1: in China that will cost your arm in a lake
• 50:51
  Speaker 1: in Europe and North America, but extremely affordable in China.
• 50:55
  Speaker 1: I highly recommend it. And you know, because right I
• 51:00
  Speaker 1: think I have a feeling. Right now is a limited
• 51:05
  Speaker 1: window for people to enjoy that because the R and
• 51:09
  Speaker 1: B is for sure going to continue to appreciate in value.
• 51:13
  Speaker 1: So right now, is the best time if you want
• 51:16
  Speaker 1: to visit China. Go to China now. Don't wait, you
• 51:20
  Speaker 1: know a couple of years when the R and B
• 51:23
  Speaker 1: would have risen in value, you know, right now, right now,
• 51:26
  Speaker 1: is that like a sweet time window for you when
• 51:29
  Speaker 1: you're whether your dollar or whatever currency you have can
• 51:33
  Speaker 1: still go up very long way in China. And and
• 51:37
  Speaker 1: you know, like like even my wife when she goes
• 51:40
  Speaker 1: to China, she's surprised how affordable everything is, basically because
• 51:44
  Speaker 1: everything is made in China, you know, like in bowling,
• 51:47
  Speaker 1: we have to import them from China. You have to
• 51:51
  Speaker 1: incurage all the extra costs of shipping and tariff.
• 51:54
  Speaker 3: And all that.
• 51:55
  Speaker 1: So so like, you know, man, just go to China.
• 51:59
  Speaker 1: All right, guys, thanks for joining. Uh next time, we'll
• 52:04
  Speaker 1: have a have a longer show for you, guys. It
• 52:07
  Speaker 1: was just due to unforeseen evidence that the TP has
• 52:11
  Speaker 1: to step away to uh, you know, to do some
• 52:16
  Speaker 1: you know, house house household chores. All right, guys, Thanks, Bye,