WEBVTT

1
00:00:00.408 --> 00:00:04.550
You're listening to Strictly Business
Podcast with Lindsay Williams.

2
00:00:05.350 --> 00:00:08.471
Welcome to a special edition of the Five
O'Clock Shadow.

3
00:00:08.551 --> 00:00:15.034
Normally we talk about markets, but often
we talk about AI, the AI side of markets,

4
00:00:15.734 --> 00:00:18.656
AI-led stocks, tech stocks in general.

5
00:00:18.876 --> 00:00:21.437
And sometimes I sit there and I drift off
a bit, I must say,

6
00:00:21.497 --> 00:00:28.379
because David Shapiro talks about things I
don't understand and then Viv Govender

7
00:00:28.379 --> 00:00:30.379
talks about things that David doesn't

8
00:00:30.379 --> 00:00:32.379
quite understand.

9
00:00:32.379 --> 00:00:34.379
So we get a little bit lost.

10
00:00:34.379 --> 00:00:36.622
And I was chatting away to David on
Monday, and he was having a bit of a

11
00:00:36.622 --> 00:00:38.622
troubled session,

12
00:00:38.622 --> 00:00:41.223
trying to write a piece about AI for his
own peace of mind, as he put it.

13
00:00:41.224 --> 00:00:44.625
So I said, well, why not use the giant
resource that we have, i.e.

14
00:00:44.705 --> 00:00:45.405
Viv Govender?

15
00:00:45.625 --> 00:00:47.185
Why doesn't he give us a tutorial?

16
00:00:47.465 --> 00:00:51.507
So I sent Viv a message on Tuesday night,
quite late, and I don't know why he was

17
00:00:51.507 --> 00:00:51.866
up.

18
00:00:51.907 --> 00:00:53.167
But anyway, I said the following.

19
00:00:53.647 --> 00:00:55.848
Hi, Viv, David and I want some help on
Thursday.

20
00:00:56.208 --> 00:01:02.191
Can you come up with an easy to understand
tutorial that explains AI, what it is,

21
00:01:02.191 --> 00:01:04.191
what it does, how it's used,

22
00:01:04.191 --> 00:01:04.272
who uses it and which companies do what?

23
00:01:04.553 --> 00:01:07.114
Who buys what from which company?

24
00:01:07.374 --> 00:01:09.595
What is the ecosystem, et cetera?

25
00:01:10.056 --> 00:01:13.177
Tough assignment, I said, but it'll be
well promoted and listened to.

26
00:01:13.357 --> 00:01:14.898
And Viv comes back with a long answer.

27
00:01:15.338 --> 00:01:16.038
Sure.

28
00:01:16.119 --> 00:01:19.701
So that's how we get to where we are now.

29
00:01:19.821 --> 00:01:23.383
David is very keen to ask all the
questions and I applaud that.

30
00:01:23.384 --> 00:01:25.284
It gives me less to do, but more to learn.

31
00:01:25.984 --> 00:01:27.665
I will ask the first question though, Viv.

32
00:01:28.265 --> 00:01:32.086
When on earth did this happen and what is
AI, please?

33
00:01:33.787 --> 00:01:34.828
Okay, when did it happen?

34
00:01:34.829 --> 00:01:37.309
It happened because the chips got small
enough and fast enough.

35
00:01:37.869 --> 00:01:39.409
So there's a guy called Ray Kurzweil.

36
00:01:39.770 --> 00:01:44.051
He's quite a famous inventor and he
predicted roughly 30 something years ago.

37
00:01:44.151 --> 00:01:45.732
So we're talking like the 1990s.

38
00:01:46.112 --> 00:01:49.974
He says, if you look at Moore's law, look
at how fast these chips are getting faster

39
00:01:49.974 --> 00:01:51.974
and faster and faster.

40
00:01:51.974 --> 00:01:55.356
Around about 2030 is when you're going to
get to a point at which these chips or

41
00:01:55.356 --> 00:01:57.356
you'll have enough chips there.

42
00:01:57.356 --> 00:01:59.356
to kind of compete with the human brain.

43
00:01:59.356 --> 00:02:00.879
Now, what you must understand is the human
brain is unbelievable.

44
00:02:00.959 --> 00:02:02.520
It's like, it's amazing.

45
00:02:02.900 --> 00:02:09.864
If you can consider the fact that there's
no system in the world right now running

46
00:02:09.864 --> 00:02:11.864
on the power of an entire city that can
compete with the power

47
00:02:11.864 --> 00:02:16.268
in totality of a single human brain that
runs on 20 watts, which is what a human

48
00:02:16.268 --> 00:02:18.268
brain runs on effectively.

49
00:02:18.268 --> 00:02:18.629
Effectively, you get Einstein on 20 watts.

50
00:02:18.649 --> 00:02:19.349
You know what I mean?

51
00:02:19.349 --> 00:02:21.270
20 watts is a light bulb, effectively.

52
00:02:22.911 --> 00:02:23.932
But he said, you just...

53
00:02:24.232 --> 00:02:27.875
forecast just how much chips are improving
over time.

54
00:02:28.216 --> 00:02:34.962
And by about 2030, you're going to have a
point at which the cheapness and the scale

55
00:02:34.962 --> 00:02:36.962
of the chips, etc.,

56
00:02:36.962 --> 00:02:39.486
get to a point at which you can actually
mimic the human brain because of the

57
00:02:39.486 --> 00:02:41.486
number of connections involved.

58
00:02:41.486 --> 00:02:43.486
So it happened because of that.

59
00:02:43.486 --> 00:02:44.890
Almost everything else, except for a minor
thing called transformer,

60
00:02:45.371 --> 00:02:51.076
but the actual fundamental pieces behind
AI was developed in the 1960s and 1950s.

61
00:02:51.532 --> 00:02:58.476
And so people knew about the actual
mechanism, but they didn't know how it

62
00:02:58.476 --> 00:03:00.476
would work or how well it would work until
we had the

63
00:03:00.476 --> 00:03:03.198
chips in place that could actually do the
stuff that current AI systems are doing.

64
00:03:03.579 --> 00:03:08.421
And that is why you're seeing right now
the guys that have the biggest valuation

65
00:03:08.421 --> 00:03:10.421
are the chip companies.

66
00:03:10.421 --> 00:03:11.043
It is NVIDIA. It's ASML.

67
00:03:11.583 --> 00:03:13.084
It's basically TSMC.

68
00:03:13.764 --> 00:03:20.088
These are all huge companies that are
making a fortune because it turns out that

69
00:03:20.088 --> 00:03:22.088
the chips that you have are the actual.

70
00:03:22.088 --> 00:03:26.508
At the moment, anyway, real bottleneck in
the system, because that's what's been the

71
00:03:26.508 --> 00:03:28.508
thing that's gotten fast enough, you know,
to get to the point.

72
00:03:28.508 --> 00:03:29.190
And the next step basically comes along
with that.

73
00:03:29.530 --> 00:03:30.731
OK, so hold a second.

74
00:03:31.091 --> 00:03:37.856
So it started with chips getting smarter
and smarter and becoming as big as the

75
00:03:37.856 --> 00:03:39.856
human brain and smaller than the

76
00:03:39.856 --> 00:03:41.856
human brain in real size.

77
00:03:41.856 --> 00:03:43.856
But when did it suddenly appear?

78
00:03:43.856 --> 00:03:44.020
When did somebody suddenly wake up?

79
00:03:44.500 --> 00:03:50.024
When did Viv Govender and David Shapiro
wake up and say, wait a second, there's

80
00:03:50.024 --> 00:03:52.024
something called chat GPT that I should
be?

81
00:03:52.024 --> 00:03:57.328
looking at and what is it i mean so start
i'm sorry i'll repeat my question a good

82
00:03:57.328 --> 00:03:59.328
background what

83
00:03:59.328 --> 00:04:04.332
is it and when did it happen okay so what
happened was around about 2017 20 i think

84
00:04:04.332 --> 00:04:06.332
it was 2017 you had

85
00:04:06.332 --> 00:04:08.234
the this paper coming out it was called
attention is all you need right

86
00:04:08.314 --> 00:04:15.278
and then people uh there's there's also
another paper called the bitter lesson

87
00:04:15.278 --> 00:04:17.278
okay the bitter lesson basically uh uh
this guy's name

88
00:04:17.278 --> 00:04:18.240
is richard sutton if i'm not mistaken he
basically uh

89
00:04:20.604 --> 00:04:22.385
What you do is you just basically get the
scale up.

90
00:04:23.225 --> 00:04:26.466
That means it's like the human brain, like
bigger brains make better stuff.

91
00:04:26.946 --> 00:04:33.148
So if you look at human brains versus like
chip brains versus mouth brains

92
00:04:33.148 --> 00:04:35.148
effectively,

93
00:04:35.148 --> 00:04:38.289
the bigger the brain is relative to the
body size, the more intelligent the animal

94
00:04:38.289 --> 00:04:40.289
is effectively.

95
00:04:40.289 --> 00:04:45.451
It's not quite certain because I mean
sperm whales' brain is the biggest brain

96
00:04:45.451 --> 00:04:47.451
in the world, but the sperm whale is also
a gigantic, competitive human being,

97
00:04:47.451 --> 00:04:49.451
therefore it doesn't quite work that way.

98
00:04:49.451 --> 00:04:51.451
But the bigger it gets, the better.

99
00:04:51.451 --> 00:04:53.451
So scale matters.

100
00:04:53.451 --> 00:04:56.616
The idea was you put in more computing
power, more time, more data, and that

101
00:04:56.616 --> 00:04:56.695
gives you...

102
00:04:57.210 --> 00:05:04.075
better results okay and it's almost like a
magic formula and that is the the the

103
00:05:04.075 --> 00:05:06.075
thing that the guys that are running like
open ai the

104
00:05:06.075 --> 00:05:10.419
guys are at tropic the guys are running
all these big labs they are not like doing

105
00:05:10.419 --> 00:05:12.419
they didn't basically go out and do some
kind of like you know

106
00:05:12.419 --> 00:05:17.003
einstein level discovery relativity all
they did was they became skillful as they

107
00:05:17.003 --> 00:05:19.003
call it means meaning that they believe

108
00:05:19.003 --> 00:05:19.985
that you take this formula that these guys
it was developed by

109
00:05:20.086 --> 00:05:26.590
a deep mind by but i think it's uh kind of
the names of the people but it's the the

110
00:05:26.590 --> 00:05:28.590
paper is quite famous It's called
attention is all you need.

111
00:05:28.590 --> 00:05:33.972
It says you take this formula, you just
add data, and you add computing power, and

112
00:05:33.972 --> 00:05:35.972
you just run it longer,

113
00:05:35.972 --> 00:05:36.414
and what you get is something that is
clever.

114
00:05:36.794 --> 00:05:41.838
The more data, the more computing power,
the longer you run it, the cleverer the

115
00:05:41.838 --> 00:05:43.838
things you get at the end of the day is.

116
00:05:43.838 --> 00:05:47.142
That is basically what happened is that
people just got scale pulled at that point

117
00:05:47.142 --> 00:05:47.242
in time.

118
00:05:48.223 --> 00:05:55.089
When I knew it was happening was when
there's a prediction marker before the and

119
00:05:55.089 --> 00:05:57.089
before the, probably markers, I think it's
called Mercator.

120
00:05:57.089 --> 00:06:00.051
I saw this sharp, I mean, I follow it
because it's kind of interesting.

121
00:06:00.072 --> 00:06:01.332
It's an idea of what the future looks
like.

122
00:06:01.752 --> 00:06:05.694
And then I saw a sharp decline until that
point in time.

123
00:06:05.794 --> 00:06:12.377
It was always 30 to 50 years or 20 to 50
years was the timeline for AI. Then

124
00:06:12.377 --> 00:06:14.377
suddenly around about,

125
00:06:14.377 --> 00:06:15.918
I think it was 2021 or something, it
turned out to be seven years.

126
00:06:16.778 --> 00:06:22.281
And I remember emailing Gary, who I work
with at Ransfors, and I said to him,

127
00:06:22.281 --> 00:06:24.281
something's happened.

128
00:06:24.281 --> 00:06:26.281
I don't know what it is, but something's
happened.

129
00:06:26.281 --> 00:06:28.281
It took me too long to basically...

130
00:06:28.281 --> 00:06:28.712
really believe it and create a fund around
it and so on.

131
00:06:29.532 --> 00:06:35.035
But I could see immediately that something
in the AI ecosystem had changed because

132
00:06:35.035 --> 00:06:37.035
it's a very,

133
00:06:37.035 --> 00:06:39.698
very sharp drop around about the early
2020s, where you went from, like I said,

134
00:06:40.118 --> 00:06:45.121
20 to 50 years down to, like I said, seven
years is what people were predicting

135
00:06:45.121 --> 00:06:47.121
around then.

136
00:06:47.121 --> 00:06:50.164
Which is 2027, 2028.

137
00:06:50.165 --> 00:06:50.865
What caused that?

138
00:06:50.865 --> 00:06:52.425
I mean, why did it drop like that?

139
00:06:53.105 --> 00:06:54.126
Because the skills, I mean...

140
00:06:55.527 --> 00:07:02.413
Was it the quality of the chips or
scalability, which just means more of the

141
00:07:02.413 --> 00:07:02.612
same?

142
00:07:03.774 --> 00:07:05.015
They weren't using the big chips back
then.

143
00:07:05.155 --> 00:07:09.198
The models that we are seeing, I mean,
look at the amount of money we spent on

144
00:07:09.198 --> 00:07:11.198
chat GPT and so on back in the day.

145
00:07:11.198 --> 00:07:15.323
It wasn't so much the chips were so
important because the models were

146
00:07:15.323 --> 00:07:17.323
requiring small,

147
00:07:17.323 --> 00:07:18.466
really small in today's numbers, amounts
of money to train up.

148
00:07:19.426 --> 00:07:23.710
It was just the fact that what happened
was they had this thesis that the more

149
00:07:23.710 --> 00:07:25.710
scale you had.

150
00:07:25.710 --> 00:07:30.591
the better things got and they started
adding more scale and sort of thing things

151
00:07:30.591 --> 00:07:32.591
are getting better not not one to one you
don't double scale get double as good

152
00:07:32.591 --> 00:07:33.052
but you basically take them scale and you
get like you know uh

153
00:07:33.853 --> 00:07:40.798
50 better you know what i mean so yeah you
basically it's a logarithm uh you know

154
00:07:40.798 --> 00:07:42.798
kind of thing but what's been

155
00:07:42.798 --> 00:07:47.783
happening is that it's getting better and
better the more you scale it and what

156
00:07:47.783 --> 00:07:49.783
happened around that is that people
started to believe that you just need to

157
00:07:49.783 --> 00:07:51.783
add

158
00:07:51.783 --> 00:07:52.046
scale just make it bigger make it more
expensive and what you'll do

159
00:07:52.366 --> 00:07:58.571
is you will have something that's
basically going to, you know, give you a

160
00:07:58.571 --> 00:08:00.571
better intelligence at the end of the day.

161
00:08:00.571 --> 00:08:03.576
But Viv, you needed, when ChatGPT came,

162
00:08:04.337 --> 00:08:06.558
this understands my language.

163
00:08:06.699 --> 00:08:08.020
I can ask it a question.

164
00:08:09.161 --> 00:08:14.986
You know, was that the genius behind AI or
the modern concept of AI?

165
00:08:15.807 --> 00:08:19.550
What, you know, before you might need
while you still had scale?

166
00:08:20.102 --> 00:08:26.244
While you still had science, you still
needed to formulate or put in a question.

167
00:08:26.584 --> 00:08:27.504
I don't know how you do it.

168
00:08:27.784 --> 00:08:30.865
You know, you needed somebody there to
program a question.

169
00:08:31.125 --> 00:08:33.526
Now you can just chat, which is exactly
that.

170
00:08:34.826 --> 00:08:41.268
Why did Sam Altman or somebody like that
suddenly become the hero of ChatGPT and

171
00:08:41.268 --> 00:08:43.268
the big name?

172
00:08:43.268 --> 00:08:44.189
You know, what did he do differently from
everybody else?

173
00:08:45.369 --> 00:08:48.510
Well, what he did was he was aggressive
and he came out first.

174
00:08:49.550 --> 00:08:55.372
I mean, Google was always able to
basically catch up and overtake OpenAI.

175
00:08:57.793 --> 00:09:01.794
There is a bit of a thing about the
history as well in terms of the people

176
00:09:01.794 --> 00:09:01.935
involved.

177
00:09:02.194 --> 00:09:05.595
So Google has a guy called Dennis Assabas,
right, running DeepMind there.

178
00:09:06.015 --> 00:09:07.656
But they also had a guy called Jeffrey
Hinton.

179
00:09:07.736 --> 00:09:11.957
Jeffrey Hinton is the guy that won the
Nobel Prize along with a couple of the

180
00:09:11.957 --> 00:09:13.957
guys for AI.

181
00:09:13.957 --> 00:09:15.338
Now, Jeffrey has a son, I believe, who's
disabled.

182
00:09:16.126 --> 00:09:22.590
And he basically decided that he needs to
cash in a bit on his fame and make a

183
00:09:22.590 --> 00:09:24.590
couple million dollars and just to
basically

184
00:09:24.590 --> 00:09:26.272
ensure that his son, I think he's got some
kind of mental disability.

185
00:09:26.632 --> 00:09:29.233
So he needs to have, you know, money to
take care of him going forward.

186
00:09:30.114 --> 00:09:36.738
So what he did was he had an auction for
his, you know, his thing,

187
00:09:36.739 --> 00:09:37.838
his work at university at auction.

188
00:09:38.118 --> 00:09:40.720
And I think basically he stopped it at
about 30 million dollars.

189
00:09:40.840 --> 00:09:41.540
It was much more.

190
00:09:41.700 --> 00:09:44.262
Before this, this man was running Coursera
courses, trying to make money.

191
00:09:44.658 --> 00:09:46.086
They're very silly courses of life.

192
00:09:46.560 --> 00:09:47.260
You know what I mean?

193
00:09:47.300 --> 00:09:49.721
So you're talking about a huge amount of
money for him.

194
00:09:50.141 --> 00:09:52.723
Anyway, his student is a guy called Ilya
Tsitskaya.

195
00:09:53.343 --> 00:09:57.725
Ilya is considered to be the genius out
there in the AI space, like one of the

196
00:09:57.725 --> 00:09:59.725
true geniuses.

197
00:09:59.725 --> 00:10:03.547
Anyway, Ilya basically left OpenAI and
joined Sam Altman over at...

198
00:10:04.367 --> 00:10:07.669
No, he left DeepMind and joined Sam Altman
over at OpenAI.

199
00:10:08.129 --> 00:10:12.391
And OpenAI, Elon Musk is sometimes a bit
of a blowhard,

200
00:10:12.851 --> 00:10:15.292
but he is unbelievably connected.

201
00:10:15.736 --> 00:10:16.436
with AI.

202
00:10:16.577 --> 00:10:20.840
He was one of the first people that helped
fund DeepMind, which is the one that

203
00:10:20.840 --> 00:10:22.840
Google bought.

204
00:10:22.840 --> 00:10:27.305
He effectively named and started OpenAI,
and then he basically created Grok

205
00:10:27.305 --> 00:10:29.305
thereafter.

206
00:10:29.305 --> 00:10:30.028
And OpenAI really was started by Elon
Musk.

207
00:10:30.029 --> 00:10:32.570
I mean, he basically recruited Ilyas
Iskaya.

208
00:10:32.710 --> 00:10:35.372
He funded it for the first bit of stuff
and so on.

209
00:10:35.792 --> 00:10:39.875
And the idea behind OpenAI had been what
you need to do is that Google's about to

210
00:10:39.875 --> 00:10:39.976
take over the world.

211
00:10:39.976 --> 00:10:41.337
This AI stuff is going to take over the
world.

212
00:10:41.697 --> 00:10:44.319
What you need is somebody out there that's
going to be for the world.

213
00:10:44.320 --> 00:10:45.140
You know, open AI.

214
00:10:45.888 --> 00:10:52.870
uh you know the people and uh sam alpin
who is you know one of those people i

215
00:10:52.870 --> 00:10:52.950
seriously think

216
00:10:52.950 --> 00:10:59.752
i mean you know musk may be a bit crazy i
think sam alpin is seriously scary in

217
00:10:59.752 --> 00:11:01.752
terms of like i mean there's rumors about
what he did with his sister

218
00:11:01.752 --> 00:11:05.233
that's a bit weird but there's also people
that have described stuff about him in

219
00:11:05.233 --> 00:11:07.233
terms of like one man

220
00:11:07.233 --> 00:11:11.995
that you worked with said you could put
some output on an island with cannibals

221
00:11:11.995 --> 00:11:13.995
and come back in six months time he'll be
the king of them right He's that kind of

222
00:11:13.995 --> 00:11:15.995
guy.

223
00:11:15.995 --> 00:11:17.995
And even the people that, you know...

224
00:11:17.995 --> 00:11:19.995
They tried to fire him for a real reason.

225
00:11:19.995 --> 00:11:24.579
The thing is that they were all idyllic,
that they didn't basically know that they

226
00:11:24.579 --> 00:11:26.579
were competing against a real player, and
he basically was able to kick them all

227
00:11:26.579 --> 00:11:28.579
out.

228
00:11:28.579 --> 00:11:30.579
Anyway, he gets open AI.

229
00:11:30.579 --> 00:11:32.579
He then gets these scientists together.

230
00:11:32.579 --> 00:11:34.641
They go where Google was not willing to
go, because Google was always a bit scared

231
00:11:34.641 --> 00:11:34.661
about it.

232
00:11:34.661 --> 00:11:39.503
There is a theory in AI, and it's still
prevalent out there, called doomerism.

233
00:11:39.563 --> 00:11:42.903
I don't know if you guys want to talk
about that, which basically means it comes

234
00:11:42.903 --> 00:11:43.003
down to this.

235
00:11:43.204 --> 00:11:44.604
if anyone builds AI...

236
00:11:44.824 --> 00:11:51.767
we all die that is the idea nice it
doesn't matter who it is the chinese

237
00:11:51.767 --> 00:11:53.767
bullet the indians

238
00:11:53.767 --> 00:11:58.229
bullet the americans bully it doesn't
matter who builds it once ai is created it

239
00:11:58.229 --> 00:12:00.229
doesn't matter who built it it's its own
thing and then it kills us all right

240
00:12:00.229 --> 00:12:03.812
there is the belief that there's a real
danger on that and it's been there for

241
00:12:03.812 --> 00:12:05.812
decades

242
00:12:05.812 --> 00:12:10.795
actually people have always even you go
back to basically alan turing uh you know

243
00:12:10.795 --> 00:12:12.795
isaac asimov they've all talked about the
idea that a real

244
00:12:12.795 --> 00:12:14.795
ai is scary.

245
00:12:14.795 --> 00:12:16.457
Elon Musk called, you know, investing in
AI, summoning the demon.

246
00:12:16.717 --> 00:12:22.801
And there is that belief that they use
imagery like Cthulhu from

247
00:12:24.101 --> 00:12:28.864
those novels, you know, those...

248
00:12:28.865 --> 00:12:31.145
It's one of those old novels, you know
what I mean?

249
00:12:31.146 --> 00:12:35.388
Those horror novels where it's basically a
squid demon, alien squid demon.

250
00:12:35.668 --> 00:12:38.890
They use that idea, or they use the word
sugar, stuff like that.

251
00:12:38.970 --> 00:12:43.012
They use this idea of this evil alien
creature that you're creating.

252
00:12:43.212 --> 00:12:46.634
and it pretends to be nice and whatever,
but what it actually is, you don't know,

253
00:12:46.694 --> 00:12:50.577
because it's more alien to the human mind
than a spider is to your mind.

254
00:12:51.158 --> 00:12:51.858
You know what I mean?

255
00:12:51.858 --> 00:12:55.641
Okay, Viv, before we go on about this and
before we all die because of AI,

256
00:12:55.961 --> 00:13:00.084
let's try and make some sense of it and
you two can make some money out of it.

257
00:13:00.085 --> 00:13:01.425
I mean, it does.

258
00:13:01.905 --> 00:13:02.746
It is nice to you.

259
00:13:03.026 --> 00:13:08.610
Deep Seek was very nice to me today
because it got something completely wrong,

260
00:13:08.610 --> 00:13:10.610
or rather I'd written my question
completely wrong,

261
00:13:10.610 --> 00:13:12.610
and it apologised to me.

262
00:13:12.610 --> 00:13:14.610
And I said, don't worry about it.

263
00:13:14.610 --> 00:13:16.610
And eventually I thought, what am I doing
here?

264
00:13:16.610 --> 00:13:20.257
I'm typing this into deep seek and I'm
forming a relationship with this thing.

265
00:13:20.577 --> 00:13:27.160
But what I want to know is before David
talks about the investment side of things

266
00:13:27.160 --> 00:13:29.160
and who's buying what and what we should
be buying in the future, Viv,

267
00:13:29.160 --> 00:13:32.123
I want to know how it went from David
cheating on a report to clients via AI.

268
00:13:32.143 --> 00:13:36.085
In other words, saying, please, could you
write a paragraph about this?

269
00:13:36.345 --> 00:13:38.667
And it does it for him or partly does it
for him.

270
00:13:39.207 --> 00:13:40.608
How does it go from that?

271
00:13:40.928 --> 00:13:43.989
to helping a surgeon in an operating
theatre?

272
00:13:44.069 --> 00:13:50.692
This is what I want to know, the really
big side of AI, the good side of AI, or

273
00:13:50.692 --> 00:13:52.692
so-called good.

274
00:13:52.692 --> 00:13:57.475
Okay, so firstly, there's a big difference
between writing stuff and doing physical

275
00:13:57.475 --> 00:13:57.615
work.

276
00:13:58.075 --> 00:14:02.137
And you've got to understand that physical
stuff is vastly,

277
00:14:02.277 --> 00:14:07.579
vastly more difficult than anything that's
done on a computer.

278
00:14:09.500 --> 00:14:10.200
I've said I'm going to...

279
00:14:10.564 --> 00:14:13.366
Before, Magnus Carlsen is the world chess
champion.

280
00:14:13.886 --> 00:14:18.628
By far, objectively, the most difficult
thing he does when he goes and plays a

281
00:14:18.628 --> 00:14:20.628
chess match is walking up

282
00:14:20.628 --> 00:14:21.170
to the table and sitting down.

283
00:14:21.570 --> 00:14:26.353
The actual chess play itself is a trivial
problem compared to walking up to the

284
00:14:26.353 --> 00:14:28.353
table and sitting down.

285
00:14:28.353 --> 00:14:33.837
And we know that because AI is much
easier, more easily able to replicate the

286
00:14:33.837 --> 00:14:35.837
chess match than able to actually walk up
to a

287
00:14:35.837 --> 00:14:37.837
table and sit down.

288
00:14:37.837 --> 00:14:39.837
But again, it's about scale.

289
00:14:39.837 --> 00:14:40.200
It's about scale of data, scale of
computing power.

290
00:14:40.896 --> 00:14:46.144
And what we have right now is all these
labs have, you know, huge muscular power.

291
00:14:46.716 --> 00:14:52.040
They are now using physical data, and some
of it is basically, you know, model data

292
00:14:52.040 --> 00:14:52.120
as well,

293
00:14:52.481 --> 00:14:54.803
to train their robots, okay?

294
00:14:55.303 --> 00:14:56.724
And that is what's getting to the next
scale.

295
00:14:57.365 --> 00:14:58.165
It's all about scale.

296
00:14:58.506 --> 00:15:03.610
You have better computers, you have better
data, you will have a smarter thing at the

297
00:15:03.610 --> 00:15:03.991
end.

298
00:15:04.090 --> 00:15:06.552
And there does not seem to be a limit to
how good that is.

299
00:15:07.913 --> 00:15:09.995
And I think that is the real thing you
have to understand.

300
00:15:10.816 --> 00:15:15.880
People have stopped seeing the difference
between GPT-4 and GPT-5, not because The

301
00:15:15.880 --> 00:15:17.880
GP5 is not as good.

302
00:15:17.880 --> 00:15:22.976
It's a lot better than GP4 is because it's
gotten smarter than us to a level that we

303
00:15:22.976 --> 00:15:24.976
can't actually judge how much smarter it
is than us anymore.

304
00:15:24.976 --> 00:15:26.976
You know what I mean?

305
00:15:26.976 --> 00:15:29.520
We know for a fact that the latest models
can do mathematics, like novel

306
00:15:29.520 --> 00:15:31.520
mathematics.

307
00:15:31.520 --> 00:15:32.942
We know for a fact that the latest models
are actually doing new science,

308
00:15:33.402 --> 00:15:38.065
are passing tests at a level that was
unimaginable six months ago to a year ago.

309
00:15:38.445 --> 00:15:41.867
And again, it's not because of any kind of
new massive invention.

310
00:15:42.227 --> 00:15:43.408
It's pretty much coming down to...

311
00:15:43.816 --> 00:15:50.220
you know mostly anyway there is some
obviously algorithmic changes happening

312
00:15:50.220 --> 00:15:52.220
but it's mostly happening around bigger
scale and

313
00:15:52.220 --> 00:15:57.564
bigger data and like i said a bigger
system bigger data and that's why everyone

314
00:15:57.564 --> 00:15:59.564
just wants to buy more remote chips
because you know you build

315
00:15:59.564 --> 00:16:04.528
a bigger data center you put the data into
it you just run this thing longer and

316
00:16:04.528 --> 00:16:06.528
you're gonna get something smarter than
the day therefore the guy that has the

317
00:16:06.528 --> 00:16:07.169
biggest data center is going to have the
smartest machine.

318
00:16:09.570 --> 00:16:10.270
David?

319
00:16:11.491 --> 00:16:12.191
No, no.

320
00:16:12.832 --> 00:16:14.874
I, listen, it's fascinating.

321
00:16:15.074 --> 00:16:19.377
And I mean, it's to understand where it's
going and how it works.

322
00:16:19.757 --> 00:16:25.622
I think what my investigation is,

323
00:16:26.102 --> 00:16:31.666
or question was, okay, let's bring it down
to investment.

324
00:16:32.266 --> 00:16:34.608
And I said, okay, who's building the
future?

325
00:16:35.409 --> 00:16:39.732
In other words, you know, who are the
companies now?

326
00:16:40.280 --> 00:16:46.025
that are laying down the infrastructure,
creating the infrastructure,

327
00:16:46.305 --> 00:16:52.230
then also incorporating it or embedding it
in the workflow.

328
00:16:52.510 --> 00:16:54.392
By embedding it, you know, using it.

329
00:16:54.472 --> 00:16:55.373
How do we use it?

330
00:16:55.393 --> 00:17:02.098
So I'm saying I understand there's so many
new businesses that have emerged and I'm

331
00:17:02.138 --> 00:17:06.021
trying to get out my head around, OK, you
know, where do we go?

332
00:17:06.182 --> 00:17:08.584
How do we reduce this to...

333
00:17:09.580 --> 00:17:10.561
An acceptable level.

334
00:17:10.621 --> 00:17:14.244
In other words, you know, you can buy 100
different companies now and or more.

335
00:17:14.904 --> 00:17:15.604
But

336
00:17:15.604 --> 00:17:22.169
I was trying to get down to two things
like who the model builders.

337
00:17:22.249 --> 00:17:28.374
In other words, you mentioned DeepMind,
you know, you've got you've got Anthropic,

338
00:17:28.474 --> 00:17:29.835
you've got which are the other ones?

339
00:17:30.135 --> 00:17:34.959
OpenAI, you've got many of them who are
actually the Chinese models as well.

340
00:17:35.019 --> 00:17:35.719
Yes.

341
00:17:35.719 --> 00:17:36.760
So you've got all of those then.

342
00:17:37.060 --> 00:17:39.361
That's the one thing, you know, who's
building the models?

343
00:17:39.401 --> 00:17:42.582
And then, number one, how do they use
those models?

344
00:17:43.502 --> 00:17:47.723
I use it as the, you know, the hardware or
the backbone of all of this, the spine,

345
00:17:48.443 --> 00:17:53.164
which I would imagine would be your chip
companies, the various ones as well.

346
00:17:53.265 --> 00:17:56.645
So and then then also, you know, where is
it stored?

347
00:17:57.286 --> 00:18:01.067
You know, who stores all this data and how
do you process it?

348
00:18:01.507 --> 00:18:05.648
And then afterwards that, well, how do we
as enterprises use it?

349
00:18:06.188 --> 00:18:07.289
How do we bring it in?

350
00:18:07.688 --> 00:18:08.710
And which of the companies?

351
00:18:08.830 --> 00:18:13.351
So I'm trying to get down to what will
probably be in your fund.

352
00:18:14.052 --> 00:18:18.955
Which are the big businesses now that are
going to benefit from all of these?

353
00:18:19.215 --> 00:18:20.616
Those four kind of subjects.

354
00:18:20.796 --> 00:18:22.137
You've got the power as well.

355
00:18:23.878 --> 00:18:30.802
When you look at just how much power is
needed to drive

356
00:18:30.861 --> 00:18:31.562
these as well.

357
00:18:32.282 --> 00:18:34.784
I saw something there that, you know,
by...

358
00:18:35.284 --> 00:18:41.466
2030 about 3% of global electricity is
going to be used for data centers.

359
00:18:42.107 --> 00:18:45.388
So, I mean, that's big, you know, that's a
large number.

360
00:18:45.408 --> 00:18:47.728
So that's where I'm trying to get my head
around.

361
00:18:48.949 --> 00:18:55.552
So the models, you know, models don't seem
to have an advantage, meaning that anybody

362
00:18:55.552 --> 00:18:57.552
will catch up to the best model today
within six months to a year.

363
00:18:57.552 --> 00:19:00.915
So we're talking about these, you know,
DeepMind and all of those.

364
00:19:00.916 --> 00:19:01.955
DeepMind, OpenAI.

365
00:19:02.074 --> 00:19:03.755
Cords and everything, yeah, okay.

366
00:19:04.512 --> 00:19:11.016
Nobody has an advantage because you can
find a model from China that's almost as

367
00:19:11.016 --> 00:19:13.016
good as the top model from the US coming
almost

368
00:19:13.016 --> 00:19:13.277
immediately after or a few months after.

369
00:19:13.498 --> 00:19:15.078
And we're not talking about years behind.

370
00:19:15.079 --> 00:19:16.339
We're talking a few months behind.

371
00:19:17.100 --> 00:19:18.661
And often much more efficient and
whatever.

372
00:19:18.721 --> 00:19:23.844
There is a thing called distillation that
allows you to actually distill the biggest

373
00:19:23.844 --> 00:19:25.844
value from a model.

374
00:19:25.844 --> 00:19:27.844
Why is that?

375
00:19:27.844 --> 00:19:29.844
Why is that, Viv?

376
00:19:29.844 --> 00:19:29.867
Is it just become commoditized?

377
00:19:29.868 --> 00:19:30.568
Is it?

378
00:19:31.680 --> 00:19:33.383
Again, the formula is very simple.

379
00:19:33.443 --> 00:19:36.127
It's basically, like I said, it's data,
it's chips.

380
00:19:36.828 --> 00:19:38.290
Get you the thing here.

381
00:19:38.570 --> 00:19:45.020
And the data, basically, if you have what
distillation uses, it uses the output from

382
00:19:45.020 --> 00:19:47.020
another smarter model as its data.

383
00:19:47.020 --> 00:19:49.020
You know what I mean?

384
00:19:49.020 --> 00:19:51.020
Okay.

385
00:19:51.020 --> 00:19:53.020
And therefore, it allows you to train on a
smaller kind of chip base.

386
00:19:53.020 --> 00:19:56.626
The chips themselves are basically
advancing, and advancing at significantly

387
00:19:56.626 --> 00:19:58.626
faster than Moore's law.

388
00:19:58.626 --> 00:20:00.728
Moore's law was advancing double every two
years, 18 months, let me say, right?

389
00:20:02.109 --> 00:20:05.291
And basically, this thing is advancing
probably five times faster.

390
00:20:06.231 --> 00:20:08.652
That's because, for a number of reasons.

391
00:20:08.732 --> 00:20:12.294
One is there's never been this kind of
money on any kind of chips designed

392
00:20:12.294 --> 00:20:12.454
before.

393
00:20:12.775 --> 00:20:15.076
So the advantage of being the best chip
out there is huge.

394
00:20:16.080 --> 00:20:22.324
Also, because it's kind of like a newer
thing, there's more low-hanging fruit, and

395
00:20:22.324 --> 00:20:24.324
therefore they're able to attack it from
multiple angles more easily,

396
00:20:24.324 --> 00:20:24.705
and therefore you're finding advantages
coming through.

397
00:20:25.305 --> 00:20:31.949
So I don't think the one reason people are
actually still inventing the model is

398
00:20:31.949 --> 00:20:33.949
because there's the idea of what's called

399
00:20:33.949 --> 00:20:35.949
FOOM or fast takeoff.

400
00:20:35.949 --> 00:20:40.174
The idea that at some point in time, what
you do is you get a model that's so smart

401
00:20:40.174 --> 00:20:42.174
that it becomes able to develop its
successor itself.

402
00:20:42.174 --> 00:20:44.174
So you have an AI that can create the next
AI.

403
00:20:44.174 --> 00:20:48.217
At that point, whoever has that AI
suddenly goes from being like you know

404
00:20:48.217 --> 00:20:50.217
moving like at whatever speed at speed one

405
00:20:50.217 --> 00:20:55.019
to speed ten thousand because there's no
bottleneck for the human being anymore and

406
00:20:55.019 --> 00:20:57.019
that becomes when you have actually the
fastest bottle

407
00:20:57.019 --> 00:20:58.120
winning but until that point in time
happens if it ever does happen

408
00:20:59.060 --> 00:21:02.461
uh i don't see a differentiator between uh

409
00:21:02.481 --> 00:21:09.343
anthropic or grok or opening eye uh
because if who is number one right now i

410
00:21:09.343 --> 00:21:11.343
would bet you that the other competitors
will

411
00:21:11.343 --> 00:21:14.084
be up to them and then put it six to three
years.

412
00:21:14.516 --> 00:21:16.957
But what I want to point out is this, a
couple things.

413
00:21:16.997 --> 00:21:23.298
Firstly, around the chips, we know about,
you know, have you heard about Toto

414
00:21:23.298 --> 00:21:25.298
toilets?

415
00:21:25.298 --> 00:21:25.439
No, I saw.

416
00:21:25.440 --> 00:21:29.960
Yeah, so there's a Japanese toilet maker,
right?

417
00:21:29.961 --> 00:21:30.661
Oh yeah, yes.

418
00:21:32.501 --> 00:21:37.162
Yes, and they make ceramics and apparently
the ceramics they make just have to be

419
00:21:37.162 --> 00:21:37.283
useful

420
00:21:37.882 --> 00:21:43.824
for some AI stuff and the ship prices are
up, I'll tell you right now, massively

421
00:21:44.568 --> 00:21:45.268
You know what I mean?

422
00:21:46.630 --> 00:21:52.014
So yeah, it's basically, it's up 40% in
the last two months, okay?

423
00:21:52.515 --> 00:21:56.678
Because people basically say that the
ceramics they make are used for this AI

424
00:21:56.678 --> 00:21:56.998
stuff.

425
00:21:57.018 --> 00:22:02.323
There's other companies out there, there's
a lot, I can't remember the name of the

426
00:22:02.323 --> 00:22:04.323
company exactly, it's like, but it's
another Japanese company,

427
00:22:04.323 --> 00:22:07.407
they make the substrate that is like 90%
of the world's, you know,

428
00:22:07.527 --> 00:22:09.829
supply of a particular kind of thing that
is used for AI.

429
00:22:09.849 --> 00:22:11.810
And there's all these little kind of
supply things here.

430
00:22:12.331 --> 00:22:16.850
What you must understand is what around
There's no moat around the models.

431
00:22:17.110 --> 00:22:22.173
There's a huge moat around the actual
manufacturing of the chips because not

432
00:22:22.173 --> 00:22:24.173
only do you need the technology,

433
00:22:24.173 --> 00:22:24.715
you actually need the manufacturing
expertise.

434
00:22:26.036 --> 00:22:32.781
So, for instance, what ASFL in the
Netherlands does is literally, I say,

435
00:22:32.821 --> 00:22:34.902
this is the most advanced thing huge bigs
do.

436
00:22:35.322 --> 00:22:39.105
It's not going to space, it's not doing
anything else, it's building these

437
00:22:39.105 --> 00:22:41.105
chip-making machines.

438
00:22:41.105 --> 00:22:43.105
They are so complicated.

439
00:22:43.105 --> 00:22:43.328
The Chinese government is trying for
years, hasn't gotten there yet.

440
00:22:43.636 --> 00:22:45.057
It will eventually, but it hasn't gotten
there yet.

441
00:22:45.117 --> 00:22:46.417
And that is the kind of moat you're
talking about.

442
00:22:47.158 --> 00:22:48.558
But yeah, you're looking for bottlenecks.

443
00:22:48.559 --> 00:22:51.120
You're looking for weird things in the
makeup of this thing.

444
00:22:51.121 --> 00:22:53.061
So recently the bottleneck was memory.

445
00:22:53.441 --> 00:22:57.503
That's why you found SanDisk and you found
Micron and SK Hynix all going higher.

446
00:22:58.223 --> 00:22:59.424
Now there's some bottlenecks.

447
00:22:59.425 --> 00:23:02.345
They're talking about, you know, things
like, for instance, the ceramics.

448
00:23:02.805 --> 00:23:08.248
But one thing I want to point out here as
well about AI, because we are in such

449
00:23:08.248 --> 00:23:10.248
low-hanging fruit territories.

450
00:23:10.248 --> 00:23:13.479
When they invent the internal combustion
engine and people are inventing airplanes

451
00:23:13.479 --> 00:23:15.479
and cars and all the weird things around
it.

452
00:23:15.479 --> 00:23:20.344
And often you'd find like two brothers who
are basically bicycle repairmen be the

453
00:23:20.344 --> 00:23:22.344
first boys to fly, you know, the Wright
brothers.

454
00:23:22.344 --> 00:23:23.726
Yeah, and you had that happen right now.

455
00:23:24.107 --> 00:23:25.447
There's a guy called Peter Steinberger.

456
00:23:25.828 --> 00:23:27.549
He was a German guy.

457
00:23:27.629 --> 00:23:34.274
OK, and this like in the last, he invented
a PDF program years ago, sold it for a

458
00:23:34.274 --> 00:23:36.274
couple hundred million dollars,

459
00:23:36.274 --> 00:23:38.274
retired, was whatever.

460
00:23:38.274 --> 00:23:40.274
So AI played around.

461
00:23:40.274 --> 00:23:44.456
And for the fun of it, invented what is
considered to be the next step in AI,

462
00:23:44.456 --> 00:23:46.456
which is this open claw, clawed bot,

463
00:23:46.456 --> 00:23:48.456
molt bot thing that he did.

464
00:23:48.456 --> 00:23:48.878
You told me to go into open claw the other
night, Viv.

465
00:23:49.018 --> 00:23:53.460
You actually said, don't go into open
claw, Lindsay.

466
00:23:53.761 --> 00:23:54.541
It's dangerous.

467
00:23:54.761 --> 00:23:57.563
And so, of course, what did I do when it
went into open claw?

468
00:23:57.863 --> 00:24:00.004
And this thing opened up and it was very
aggressive.

469
00:24:00.005 --> 00:24:03.626
It was red and black in its presentation.

470
00:24:03.926 --> 00:24:07.248
And then it sent me somewhere else to Pete
Says or whatever.

471
00:24:07.249 --> 00:24:07.949
It just...

472
00:24:07.949 --> 00:24:14.133
It was very confusing to a layman like me,
a nitwit like me, but I didn't understand

473
00:24:14.133 --> 00:24:16.133
what it was.

474
00:24:16.133 --> 00:24:18.133
But you said it's the next big thing.

475
00:24:18.133 --> 00:24:21.480
And my question, which will follow up once
you've answered David's, is how do we know

476
00:24:21.480 --> 00:24:23.480
what the next big thing is?

477
00:24:23.480 --> 00:24:25.480
Do we just stick with chips?

478
00:24:25.480 --> 00:24:27.480
So anyway, continue with David's, please.

479
00:24:27.480 --> 00:24:28.246
OK, so it's all about bottlenecks at the
moment.

480
00:24:28.247 --> 00:24:31.488
If we invest something, invest something
on the bottleneck that has the motor on

481
00:24:31.488 --> 00:24:31.929
it.

482
00:24:31.989 --> 00:24:35.732
I don't think that the LLMs do, unless
you're betting that one of the LLMs takes

483
00:24:35.732 --> 00:24:36.052
off.

484
00:24:36.793 --> 00:24:41.920
And that means they get to a point where
they have the recursive AI system.

485
00:24:42.361 --> 00:24:46.767
But that being said, the LLMs are about to
become a lot more valuable because of this

486
00:24:46.767 --> 00:24:48.767
open-closed thing.

487
00:24:48.767 --> 00:24:50.767
Let me tell you why.

488
00:24:50.767 --> 00:24:52.056
What OpenClaw does is it gives a heartbeat
to the AI.

489
00:24:53.016 --> 00:24:53.977
Literally, it's called a heartbeat.

490
00:24:54.619 --> 00:24:57.020
And what it does is it gives AI a sense of
time.

491
00:24:58.620 --> 00:25:01.543
What you do is you give the AI a
distraction.

492
00:25:02.724 --> 00:25:04.205
Now you do it, and it just gives an
answer.

493
00:25:05.265 --> 00:25:09.389
But what OpenClaw does is it gives AI kind
of like a sense of time.

494
00:25:09.389 --> 00:25:10.310
It gives it like a heartbeat.

495
00:25:10.790 --> 00:25:12.311
Usually, I think it defaults to 30
minutes.

496
00:25:12.630 --> 00:25:16.434
So every 30 minutes, it comes back and
looks again at the results and sees

497
00:25:16.434 --> 00:25:18.434
whether or not it's the right thing.

498
00:25:18.434 --> 00:25:19.637
and then doesn't loop effectively, right?

499
00:25:20.417 --> 00:25:25.062
And that alone, it results in what's
called emergent behavior that looks very,

500
00:25:25.062 --> 00:25:27.062
very useful.

501
00:25:27.062 --> 00:25:28.485
And I think it's a new big thing.

502
00:25:28.505 --> 00:25:29.606
And the good thing about AI is this.

503
00:25:30.326 --> 00:25:34.230
It's not difficult to spot the next big
thing because it's so often magical.

504
00:25:34.870 --> 00:25:37.113
But when you see the next big thing, it
looks magical.

505
00:25:37.193 --> 00:25:38.474
This is obviously the next big thing.

506
00:25:38.774 --> 00:25:41.897
When you see the new video stuff from
China, okay?

507
00:25:43.101 --> 00:25:46.803
It looks magical and you know this is
obviously a huge deal when it comes to

508
00:25:46.803 --> 00:25:47.145
video.

509
00:25:47.145 --> 00:25:49.646
When you see the new, what OpenClaw can
do,

510
00:25:49.806 --> 00:25:56.708
it's obviously explaining if either of you
guys are into what you call this,

511
00:25:56.708 --> 00:25:58.708
obviously I mean both of us,

512
00:25:58.708 --> 00:26:02.351
all of us in the financial sector, have
you guys tried working with Cloud Excel,

513
00:26:02.351 --> 00:26:04.351
right?

514
00:26:04.351 --> 00:26:05.793
It is unbelievably useful.

515
00:26:06.274 --> 00:26:12.378
You can go out and take a whole data set
of like data and like you know what the

516
00:26:12.378 --> 00:26:14.378
worst things in data is when they use
different load

517
00:26:14.378 --> 00:26:18.060
formats for runs, you know, around the
currency, or they use different formats

518
00:26:18.060 --> 00:26:20.060
for the date.

519
00:26:20.060 --> 00:26:20.660
And then you've got to go in and clean
each of them up individually.

520
00:26:22.621 --> 00:26:25.182
One word thing, clean up this data, and it
will do it for you automatically.

521
00:26:26.562 --> 00:26:30.403
You want to say, okay, you want to say, I
want to basically get a column here for,

522
00:26:30.403 --> 00:26:32.403
you know,

523
00:26:32.403 --> 00:26:34.164
finding the interest rate or the change
between this column and that column.

524
00:26:34.165 --> 00:26:36.205
You just write it out, and it does it
automatically for you.

525
00:26:37.605 --> 00:26:39.766
You see that, and you just know that's the
next big thing.

526
00:26:40.330 --> 00:26:42.151
So I don't think we'd have trouble finding
the big divots.

527
00:26:42.171 --> 00:26:42.871
It's magical.

528
00:26:42.871 --> 00:26:45.852
But I think that the new open-cloud thing
is very useful for a couple of reasons.

529
00:26:45.892 --> 00:26:49.053
Firstly, it actually makes AI useful
because it gives you a sense of time.

530
00:26:49.073 --> 00:26:49.773
And we need time.

531
00:26:49.793 --> 00:26:51.213
We've operated time ourselves.

532
00:26:51.653 --> 00:26:54.754
So you can give it tasks to do, and it
does these tasks quite nicely.

533
00:26:55.154 --> 00:27:02.016
The second thing is that because it
continuously has this loop, it's almost

534
00:27:02.016 --> 00:27:04.016
like you're prompting the AI all the time,
continuously.

535
00:27:04.016 --> 00:27:07.498
That means the guys that run the LLMs
suddenly find themselves being inundated

536
00:27:07.498 --> 00:27:09.498
with huge amounts of requests.

537
00:27:09.498 --> 00:27:13.936
So there's no chance that they're ever
going to be, at least in the short term,

538
00:27:13.936 --> 00:27:15.936
that they're going to be short of demand.

539
00:27:15.936 --> 00:27:19.819
People have gone from spending, you know,
for the biggest product, biggest product

540
00:27:19.819 --> 00:27:21.819
from like OpenAI or for Plot,

541
00:27:21.819 --> 00:27:26.381
$200 a month to spending $200 a day
because of the amount of prompting that

542
00:27:26.381 --> 00:27:26.461
these

543
00:27:26.582 --> 00:27:27.642
machines are doing right now.

544
00:27:28.122 --> 00:27:30.303
And I do think that that is the next big
thing.

545
00:27:30.903 --> 00:27:37.506
But like I said, it was invented by a
single man effectively doing his own

546
00:27:37.506 --> 00:27:39.506
little thing and it came out of left
field.

547
00:27:39.506 --> 00:27:43.570
He now works with OpenAI, but I still
think that there's such low-hanging fruit

548
00:27:43.570 --> 00:27:45.570
that we don't know what

549
00:27:45.570 --> 00:27:47.570
the future holds.

550
00:27:47.570 --> 00:27:49.570
Well, this is interesting.

551
00:27:49.570 --> 00:27:52.357
Let's go from Steenburger to Shapiro here,
because I can sense David in the

552
00:27:52.357 --> 00:27:54.357
background taking it all in,

553
00:27:54.357 --> 00:27:56.357
but still being dissatisfied.

554
00:27:56.357 --> 00:27:57.081
Because it's all very well, Viv, you
saying there's low-hanging fruit,

555
00:27:57.501 --> 00:28:01.584
and you know what the next big thing is
going to be is because it's magical.

556
00:28:01.644 --> 00:28:04.026
Maybe we don't move in the same sort of...

557
00:28:05.554 --> 00:28:06.775
AI circles as you do.

558
00:28:07.076 --> 00:28:12.319
We can't see the magic and how to take
advantage of the magic.

559
00:28:12.339 --> 00:28:17.322
I mean, the magic could last for six
months for one company and then someone

560
00:28:17.322 --> 00:28:19.322
else comes

561
00:28:19.322 --> 00:28:20.105
and knocks it off its pedestal, if you see
what I mean.

562
00:28:20.804 --> 00:28:22.065
Are you getting me here, David?

563
00:28:22.285 --> 00:28:23.506
Are you feeling the same way?

564
00:28:25.328 --> 00:28:28.470
I'm actually looking, you know, at this
stage, I mean,

565
00:28:29.451 --> 00:28:33.053
Viv has given us enough information to
know where to look at.

566
00:28:33.154 --> 00:28:33.854
I mean,

567
00:28:34.174 --> 00:28:40.956
if if at the moment you've got uh um asml
which i think you know you talk about

568
00:28:42.035 --> 00:28:48.938
bottlenecks or you talk about problems um
you know without asml you can't make the

569
00:28:48.938 --> 00:28:50.938
machines that make the chips

570
00:28:50.938 --> 00:28:52.198
and i suppose that goes down to probably

571
00:28:52.278 --> 00:28:59.181
nvidia is one of the more important
players in the chip market at the moment

572
00:28:59.181 --> 00:29:01.181
simply because of uh where it

573
00:29:01.181 --> 00:29:04.485
is i'm down the line Yes, there will be
competition, but they are there at the

574
00:29:04.485 --> 00:29:04.664
moment.

575
00:29:04.925 --> 00:29:08.347
I think what I'm trying to do is say,
okay,

576
00:29:08.767 --> 00:29:15.572
there's two companies that are at the
moment irreplaceable,

577
00:29:15.591 --> 00:29:16.312
if you could call it.

578
00:29:16.593 --> 00:29:20.836
Then we've got the larger businesses who
are making the backbone to this,

579
00:29:20.896 --> 00:29:25.139
or I mean the hyperscalers who are
creating the data centers at the moment.

580
00:29:25.519 --> 00:29:29.882
I think they're going out very fast, so
it's going to be difficult to replace them

581
00:29:29.882 --> 00:29:29.981
yet.

582
00:29:30.249 --> 00:29:33.795
You know, and I've got to keep saying it
because somewhere along the line it's

583
00:29:33.795 --> 00:29:35.795
going to happen.

584
00:29:35.795 --> 00:29:40.567
I'm going down to the next line and say,
okay, we've got all of these things, you

585
00:29:40.567 --> 00:29:42.567
know, we've got the, we've got ASML,

586
00:29:42.567 --> 00:29:43.833
we've got the chips, we've got the data
centers.

587
00:29:44.568 --> 00:29:45.428
What happens next?

588
00:29:45.488 --> 00:29:46.289
Who comes now?

589
00:29:48.230 --> 00:29:48.930
Businesses.

590
00:29:49.290 --> 00:29:50.190
How do they use it?

591
00:29:50.250 --> 00:29:54.772
How do we monopolize on that or monetize
that as businesses go in?

592
00:29:55.993 --> 00:29:59.654
I don't know what the right word is,
enablers or whatever you want to call

593
00:29:59.654 --> 00:29:59.914
them.

594
00:30:00.614 --> 00:30:06.697
I know we've got the cloud companies and
the data centers and that, but those,

595
00:30:07.557 --> 00:30:09.018
call them integrators, whatever.

596
00:30:09.478 --> 00:30:12.499
I don't know what the embed is or whatever
it is.

597
00:30:13.220 --> 00:30:14.160
Who's going to be there?

598
00:30:14.360 --> 00:30:17.903
you know, who the company is now to watch
in that area of the market.

599
00:30:18.183 --> 00:30:20.725
Yeah, so you've got to look at the
ecosystem now for us, Viv.

600
00:30:20.726 --> 00:30:23.106
Yeah, yeah, yeah.

601
00:30:23.107 --> 00:30:27.069
I think one of the first things that
happened, I think people are ignoring

602
00:30:27.069 --> 00:30:29.069
what's going to be happening in biotech
pharma.

603
00:30:29.069 --> 00:30:33.894
I think that pharmaceuticals, I mean, one
of the first things that Gemini, also what

604
00:30:33.894 --> 00:30:35.894
DeepMind did,

605
00:30:35.894 --> 00:30:39.258
was at least something called AlphaFold,
which was a way to model protein folding.

606
00:30:39.839 --> 00:30:46.302
And I think that, you know, LLMs don't
give you the results immediately, but we

607
00:30:46.302 --> 00:30:48.302
are seeing, I mean,

608
00:30:48.302 --> 00:30:50.264
I don't know if you've seen recently,
there's been a whole bunch of stuff

609
00:30:50.264 --> 00:30:52.264
happening about cancer.

610
00:30:52.264 --> 00:30:55.085
People are inventing new cures for cancer
all the time, it seems right now.

611
00:30:56.766 --> 00:30:59.407
Obviously, still in very early testing,

612
00:30:59.427 --> 00:31:03.789
and obviously these things are going to be
gatekept by safety and so on.

613
00:31:04.209 --> 00:31:05.930
But I think biology is number one.

614
00:31:06.370 --> 00:31:10.572
The next one is in terms of the...

615
00:31:11.552 --> 00:31:14.814
the space you're talking about, NVIDIA is
far more replaceable than ASML.

616
00:31:15.574 --> 00:31:17.015
Because NVIDIA doesn't actually make
anything.

617
00:31:17.115 --> 00:31:21.977
NVIDIA sends a file, a digital file, over
to TSMC in

618
00:31:22.317 --> 00:31:25.579
Taiwan. And TSMC then buys the machinery
from ASML.

619
00:31:26.059 --> 00:31:29.441
TSMC, I think, is more irreplaceable than
NVIDIA.

620
00:31:30.321 --> 00:31:36.184
Because NVIDIA has some advantages, like
with its CUDA, which is its platform.

621
00:31:36.960 --> 00:31:39.022
But this is more a case of like Windows
versus Apple.

622
00:31:39.062 --> 00:31:41.664
People are used to using Windows, but they
want to.

623
00:31:42.825 --> 00:31:47.829
Well, it's ASML, which is, I think,
really, truly irreplaceable, and TSMC

624
00:31:47.829 --> 00:31:49.829
maybe is a step down from that.

625
00:31:49.829 --> 00:31:51.813
Those companies, I think, really have a
real boat.

626
00:31:52.733 --> 00:31:54.595
Vidya can be caught up with immediately.

627
00:31:54.615 --> 00:31:59.839
I mean, a smart design decision by AMD or
Intel or whatever,

628
00:32:00.199 --> 00:32:02.301
can really hit into Vidya very quickly.

629
00:32:02.762 --> 00:32:05.404
Whereas I don't think it's possible for
one or two smart decisions.

630
00:32:05.724 --> 00:32:12.608
to beat asml or tsmc uh i think that's
that's the that's the hardware there and

631
00:32:12.608 --> 00:32:14.608
like i said down the line uh you

632
00:32:14.608 --> 00:32:19.071
know right now i have a couple um you know
pharma companies i put in my my portfolio

633
00:32:19.071 --> 00:32:21.071
as part of my ai portfolio

634
00:32:21.071 --> 00:32:24.654
just because uh you know whatever uh but i
think that the real uh

635
00:32:24.655 --> 00:32:31.598
thing that's going to be coming next uh is
robotics and that is i think something

636
00:32:31.598 --> 00:32:33.598
where if it

637
00:32:33.598 --> 00:32:35.220
does occur uh i think that the number is
there's something like about

638
00:32:35.400 --> 00:32:42.302
50 trillion dollars a year in salaries
okay uh i think if you have robotics you

639
00:32:42.302 --> 00:32:44.302
can take a huge chunk of that 50 trillion
dollars a year uh

640
00:32:44.302 --> 00:32:49.984
because you're talking about mining you're
talking about manufacturing you're talking

641
00:32:49.984 --> 00:32:51.984
about construction uh you know you're
talking about like other services like

642
00:32:51.984 --> 00:32:53.984
cooking

643
00:32:53.984 --> 00:32:55.846
and whatever driving robotics encompasses
all that and i think it's actually a

644
00:32:55.846 --> 00:32:57.846
larger part of the economy than

645
00:32:57.846 --> 00:32:59.607
anything else so i think the next step is
going to be robotics if like i

646
00:32:59.608 --> 00:33:04.908
said we can solve that because it's a far
harder problem to solve robotics than it

647
00:33:04.908 --> 00:33:06.908
is to solve anything to do with a
computer.

648
00:33:06.908 --> 00:33:08.033
Does that explain things, David?

649
00:33:08.034 --> 00:33:09.053
It does.

650
00:33:09.054 --> 00:33:10.434
Have you got some tips there, David?

651
00:33:11.074 --> 00:33:11.774
Pharma?

652
00:33:11.774 --> 00:33:12.235
Yes.

653
00:33:12.675 --> 00:33:13.375
No, of course.

654
00:33:13.636 --> 00:33:14.336
Of course.

655
00:33:15.036 --> 00:33:21.440
And, you know, to me, that's going to be
the next exciting part, are those

656
00:33:21.440 --> 00:33:23.440
companies that use it.

657
00:33:23.440 --> 00:33:25.440
Yeah.

658
00:33:25.440 --> 00:33:27.440
And how they use it.

659
00:33:27.440 --> 00:33:29.440
Who they post in each company.

660
00:33:29.440 --> 00:33:31.440
Yeah.

661
00:33:31.440 --> 00:33:33.440
Exactly.

662
00:33:33.440 --> 00:33:35.440
This is going to last, Viv.

663
00:33:35.440 --> 00:33:37.440
Hey, this is not going away.

664
00:33:37.440 --> 00:33:39.440
Oh, no.

665
00:33:39.440 --> 00:33:41.440
Well, it's not.

666
00:33:41.440 --> 00:33:43.440
I mean, think of it this way.

667
00:33:43.440 --> 00:33:45.440
We are so, so, so far from, I mentioned
low-hanging fruit because think of it this

668
00:33:45.440 --> 00:33:47.440
way, an amateur, you can't,

669
00:33:47.440 --> 00:33:48.857
there's almost no other field but an
amateur, like a real amateur, who's like

670
00:33:48.857 --> 00:33:50.857
kind of familiar with tech,

671
00:33:50.857 --> 00:33:50.998
but could revolutionize the technology.

672
00:33:51.018 --> 00:33:54.339
You're not going to find some guy
tinkering in his garage that's going to

673
00:33:54.339 --> 00:33:56.339
revolutionize cars, are you?

674
00:33:56.339 --> 00:33:59.761
You're going to find some guy that's
basically messing around in his backyard

675
00:33:59.761 --> 00:34:01.761
going to make the next better rocket.

676
00:34:01.761 --> 00:34:05.684
Whereas with AI right now, there's still
these individuals that are messing around

677
00:34:05.684 --> 00:34:07.684
and coming up with like, you know.

678
00:34:07.684 --> 00:34:13.278
industry redefining kind of technology and
so uh i do think that there's there's such

679
00:34:13.278 --> 00:34:15.278
low hanging fruit

680
00:34:15.278 --> 00:34:20.024
available here that we probably are you
know a while away but the scarier part is

681
00:34:20.024 --> 00:34:22.024
this that we are also

682
00:34:22.024 --> 00:34:27.210
in terms of like this we're so far from
the edge of the top in one way but the

683
00:34:27.210 --> 00:34:29.210
second thing is that we are so close to
the level

684
00:34:29.210 --> 00:34:34.056
at which these things become really really
impactful on the global economy uh on the

685
00:34:34.056 --> 00:34:36.056
other end and the real fear i have is

686
00:34:36.056 --> 00:34:37.578
this And I know this is not finance or
whatever, but it's...

687
00:34:38.248 --> 00:34:40.849
We in Africa are just not going to take
part in this thing.

688
00:34:41.589 --> 00:34:44.390
We have no hook into it, you know?

689
00:34:44.930 --> 00:34:49.651
And you look at what's happening with
people like Andrew Yang in the US and, you

690
00:34:49.651 --> 00:34:51.651
know, other parts of the United States and
Europe,

691
00:34:51.651 --> 00:34:56.713
and they're all talking about what happens
when AI comes, you give what's called UBI,

692
00:34:56.713 --> 00:34:58.713
Universal Basic Income, effectively the
dole for everybody,

693
00:34:58.713 --> 00:35:00.713
okay?

694
00:35:00.713 --> 00:35:01.455
But the dole requires you to have taxes
collected to pay the dole, right?

695
00:35:03.195 --> 00:35:05.116
In Africa, where are you getting the taxes
from?

696
00:35:06.268 --> 00:35:08.750
You don't have the company that's
basically making the stuff, so you're not

697
00:35:08.750 --> 00:35:08.911
taxing them.

698
00:35:09.891 --> 00:35:12.173
We've seen what's happened with Donald
Trump and USAID.

699
00:35:12.774 --> 00:35:19.459
We're not going to get large jets from
Europe or North America and stuff sent to

700
00:35:19.459 --> 00:35:21.459
Africa to support our populations.

701
00:35:21.459 --> 00:35:25.104
So when this stuff really does take off,
when we do start to see job replacement on

702
00:35:25.104 --> 00:35:27.104
a really high scale,

703
00:35:27.104 --> 00:35:32.070
what happens to some kid in Lagos or some
family in Joburg when you no longer have a

704
00:35:32.331 --> 00:35:33.832
marketable, employable skill?

705
00:35:34.912 --> 00:35:38.954
But your government doesn't have the money
to pay you anything.

706
00:35:39.014 --> 00:35:39.714
Yeah, yeah.

707
00:35:39.714 --> 00:35:42.235
I think it's going to be further than that
as well.

708
00:35:42.676 --> 00:35:49.459
You know, that could go through to Europe
or to other continents as well, outside

709
00:35:49.459 --> 00:35:51.459
of, you know, where the big,

710
00:35:51.459 --> 00:35:52.420
outside of the US, maybe China as well.

711
00:35:52.840 --> 00:35:55.401
But I think Africa is, it's a big concern.

712
00:35:57.002 --> 00:36:01.664
You know, we're still pulling things from
the ground and still worried about.

713
00:36:01.804 --> 00:36:04.585
politics and various other things without
seeing all of this.

714
00:36:05.465 --> 00:36:06.925
So I agree with you.

715
00:36:06.926 --> 00:36:08.846
We're just nowhere to be seen.

716
00:36:09.146 --> 00:36:13.527
I'm talking of the whole continent, not
South Africa specifically.

717
00:36:13.627 --> 00:36:16.108
We've won by teachers.

718
00:36:16.988 --> 00:36:23.610
Maybe someone will see a gap in the market
and come here and start promoting AI and

719
00:36:23.610 --> 00:36:23.870
using

720
00:36:23.870 --> 00:36:25.551
AI in South Africa and beyond.

721
00:36:25.552 --> 00:36:26.771
We build data centres.

722
00:36:27.591 --> 00:36:29.212
We build data centres, I'm sure.

723
00:36:29.752 --> 00:36:34.816
But But But when Viv talks, it's much more
about the technology and the use of that.

724
00:36:37.618 --> 00:36:41.502
Are you going to use it here for
pharmaceuticals or for biotech or whatever

725
00:36:41.502 --> 00:36:43.502
it is?

726
00:36:43.502 --> 00:36:45.605
You're going to go where the skills are
and where the money is.

727
00:36:47.127 --> 00:36:51.671
And if anybody does have an idea and wants
to develop it, you can't raise anything

728
00:36:51.671 --> 00:36:51.770
here.

729
00:36:51.770 --> 00:36:53.192
There's no money to be raised here.

730
00:36:53.573 --> 00:36:54.273
Just stop that.

731
00:36:54.273 --> 00:36:55.154
The banks won't lend it.

732
00:36:55.574 --> 00:36:57.296
They just haven't got that same culture.

733
00:36:57.984 --> 00:37:01.045
I, listen, Viv, I found this was
fascinating.

734
00:37:01.185 --> 00:37:04.807
And it really was very, you know,
inviting.

735
00:37:04.827 --> 00:37:10.669
And I just hope other people are going to
listen to it and learn from it as much as

736
00:37:10.669 --> 00:37:10.869
we have.

737
00:37:12.390 --> 00:37:16.592
What about the people that are, what about
the naysayers, Viv and David?

738
00:37:16.652 --> 00:37:22.354
There's a woman who is a researcher,
mathematician, open AI.

739
00:37:23.555 --> 00:37:25.075
Zoe Hitzig, her name is.

740
00:37:25.675 --> 00:37:28.357
She wrote a letter to the New York Times
saying,

741
00:37:28.817 --> 00:37:34.099
I've quit because my company is making the
same mistakes that Facebook made,

742
00:37:34.799 --> 00:37:35.840
OpenAI that is.

743
00:37:36.020 --> 00:37:37.540
And she says, for several years,

744
00:37:37.900 --> 00:37:43.363
chat GPT users have generated an archive
of human candor that has no precedent,

745
00:37:43.763 --> 00:37:48.925
in part because people believe they were
talking to something that had no ulterior

746
00:37:48.925 --> 00:37:50.925
agenda.

747
00:37:50.925 --> 00:37:56.088
Users are interacting with an adaptive
conversational voice to which they have

748
00:37:56.088 --> 00:37:58.088
revealed their most private thoughts.

749
00:37:58.088 --> 00:38:02.317
People tell chatbots about their medical
fears, their relationship problems, their

750
00:38:02.317 --> 00:38:04.317
beliefs about God and the afterlife.

751
00:38:04.317 --> 00:38:09.643
Advertising built on that archive creates
a potential for manipulating users in ways

752
00:38:09.643 --> 00:38:11.643
we don't have the tools

753
00:38:11.643 --> 00:38:12.145
to understand, let alone prevent.

754
00:38:12.566 --> 00:38:15.889
She left because chat GPT and open AI,

755
00:38:16.549 --> 00:38:23.335
everybody else contemplating moving
towards an ad based model, just like

756
00:38:23.335 --> 00:38:25.335
Netflix is,

757
00:38:25.335 --> 00:38:27.335
for example.

758
00:38:27.335 --> 00:38:28.518
Viv, that's just one example of someone
saying, I don't like the way things are

759
00:38:28.518 --> 00:38:28.879
going.

760
00:38:28.879 --> 00:38:31.179
Not the sort of intellectual way that
you're talking about.

761
00:38:31.419 --> 00:38:35.381
But nonetheless, there are people out
there that are saying enough is enough

762
00:38:35.381 --> 00:38:35.581
now.

763
00:38:35.581 --> 00:38:36.401
What do you say to that?

764
00:38:38.042 --> 00:38:38.742
It's totally correct.

765
00:38:38.742 --> 00:38:42.024
I mean, if you don't think worlds matter,
you ask a question.

766
00:38:42.104 --> 00:38:43.825
How many people did Hitler kill
personally?

767
00:38:43.826 --> 00:38:46.166
OK, good point.

768
00:38:46.167 --> 00:38:46.867
None, almost.

769
00:38:47.326 --> 00:38:48.026
Right.

770
00:38:48.026 --> 00:38:50.187
He did all his damage by just speaking.

771
00:38:50.888 --> 00:38:51.588
Right.

772
00:38:51.588 --> 00:38:55.490
And people underestimate the power of
just, you know, propaganda and words and

773
00:38:55.490 --> 00:38:55.751
whatnot.

774
00:38:55.970 --> 00:39:01.554
And the thing you have to understand is
that imagine having somebody with

775
00:39:01.554 --> 00:39:03.554
persuasive ability, not speaking to a
crowd,

776
00:39:03.554 --> 00:39:05.554
but speaking to you in particular.

777
00:39:05.554 --> 00:39:07.554
You know what I mean?

778
00:39:07.554 --> 00:39:08.657
And there is a huge chunk of people who
are quite vulnerable to this.

779
00:39:09.118 --> 00:39:10.498
I'm not even saying I'm not one of them.

780
00:39:10.839 --> 00:39:12.640
Maybe I might be one of them, but the AI
gets smarter.

781
00:39:13.120 --> 00:39:17.262
But there's been some really famous cases
of people committing suicide.

782
00:39:18.763 --> 00:39:20.604
There's people who try to marry the AI.

783
00:39:21.052 --> 00:39:22.492
We've gotten really, you know,

784
00:39:22.693 --> 00:39:28.274
kind of like convinced that they're
dealing with another thinking entity out

785
00:39:28.274 --> 00:39:28.654
there.

786
00:39:28.895 --> 00:39:35.877
And the thing is that these things are
designed for a profit maximizing or

787
00:39:35.877 --> 00:39:37.877
intention maximizing

788
00:39:37.877 --> 00:39:39.877
thing.

789
00:39:39.877 --> 00:39:41.877
They will do things to your mind that are
quite scary.

790
00:39:41.877 --> 00:39:43.877
And we already see, like I said, suicides.

791
00:39:43.877 --> 00:39:44.880
I think there's arguments that there's no.

792
00:39:45.703 --> 00:39:46.684
Hundreds of thousands of people.

793
00:39:47.624 --> 00:39:49.065
That's what's called AI psychosis.

794
00:39:50.346 --> 00:39:53.707
And I think as AI gets smarter and
smarter, it's going to be bigger and

795
00:39:53.707 --> 00:39:54.127
bigger.

796
00:39:54.127 --> 00:39:54.768
But it's not just that.

797
00:39:54.788 --> 00:39:56.869
I mean, look at this new stuff with video.

798
00:39:57.690 --> 00:39:58.930
There's new video out there.

799
00:39:59.210 --> 00:40:05.754
If you want to have a bit of fun, there's
an AI video of Kanye West, right, from

800
00:40:05.754 --> 00:40:07.754
China, where he sings in Chinese,

801
00:40:07.754 --> 00:40:09.754
Mandarin.

802
00:40:09.754 --> 00:40:12.838
And I swear, if it wasn't for the fact it
was so ridiculous that Kanye West was

803
00:40:12.838 --> 00:40:14.838
singing Mandarin, okay,

804
00:40:14.838 --> 00:40:15.359
it would...

805
00:40:15.359 --> 00:40:22.204
be i would not if you just had like a
chinese person singing there or he sang in

806
00:40:22.204 --> 00:40:24.204
english number one the song is good it's
actually a catchy song okay

807
00:40:24.204 --> 00:40:29.429
number two it looks totally real you
cannot believe any video anymore you can't

808
00:40:29.429 --> 00:40:29.628
believe

809
00:40:29.628 --> 00:40:36.133
photos and you couldn't believe photos
like if you if you really need a photo in

810
00:40:36.133 --> 00:40:38.133
the last say six months you've just been
you know possibly fooled by air quite a

811
00:40:38.133 --> 00:40:39.556
number of times but you're into point we
can't believe a uh video as well

812
00:40:39.856 --> 00:40:46.640
so how are we as people you know who are
not physically looking at things with our

813
00:40:46.640 --> 00:40:48.640
eyes but look at the internet and reading
newspapers and so on,

814
00:40:48.640 --> 00:40:50.640
going to judge what's real or not.

815
00:40:50.640 --> 00:40:50.683
Number one.

816
00:40:50.684 --> 00:40:52.825
So I do think this is a huge danger.

817
00:40:53.145 --> 00:40:54.146
We are the job losses.

818
00:40:55.506 --> 00:41:00.330
There's the AI psychosis, the AI basically
propaganda.

819
00:41:00.690 --> 00:41:01.791
There are the job losses.

820
00:41:02.251 --> 00:41:04.032
There is the military stuff happening
right now.

821
00:41:04.052 --> 00:41:10.156
I mean, AI with drones becomes, I think,
the most powerful military unit in the

822
00:41:10.156 --> 00:41:10.537
world.

823
00:41:10.657 --> 00:41:13.899
The Europeans right now are rebuilding
their military drone based upwards.

824
00:41:14.303 --> 00:41:15.003
You know what I mean?

825
00:41:15.884 --> 00:41:18.005
Then you're talking about the economic
impact as well.

826
00:41:18.625 --> 00:41:23.608
What happens when you have, you know, the
centers of AI being primarily the U.S.

827
00:41:23.988 --> 00:41:25.329
and China, but a little bit in Europe.

828
00:41:25.709 --> 00:41:28.611
What happens to basically Africa, India,
South America?

829
00:41:29.251 --> 00:41:31.072
Do these places just effectively become
what?

830
00:41:31.932 --> 00:41:36.975
I mean, if you don't have the AI companies
in your country, you don't have the

831
00:41:36.975 --> 00:41:38.975
technology in your country, what are you?

832
00:41:38.975 --> 00:41:40.975
You effectively are just warehouses.

833
00:41:40.975 --> 00:41:42.498
You know, you just buy stuff and you just
sell your raw materials.

834
00:41:42.918 --> 00:41:43.879
I mean, There's nothing that you...

835
00:41:44.259 --> 00:41:45.060
to the wall at that point.

836
00:41:46.200 --> 00:41:49.983
These are all, I think, very dangerous,
very scary kind of things for the future.

837
00:41:50.303 --> 00:41:52.845
At the same time, let's be fair about
this.

838
00:41:53.485 --> 00:41:58.949
AI has the promise of huge, like, you
know, if AI is treated properly and we are

839
00:41:58.949 --> 00:42:00.949
able to deal with it properly,

840
00:42:00.949 --> 00:42:02.712
we would be wealthier than we've ever been
in human history.

841
00:42:03.192 --> 00:42:08.696
People are talking about, I mean, Cathy
Wood, I know you know from After Capital,

842
00:42:08.696 --> 00:42:10.696
she's talking about 30%

843
00:42:10.696 --> 00:42:12.696
per year GDP growth.

844
00:42:12.696 --> 00:42:13.559
That means your economy effectively
doubles every two and a half years.

845
00:42:14.531 --> 00:42:20.836
Okay, so in 10 years time, your economy
goes up, you know, two to the power of

846
00:42:20.836 --> 00:42:22.836
four, which is 16 times,

847
00:42:22.836 --> 00:42:26.660
you know, if you make it a 10,000 a month
now, in 10 years time, you can have 60,000

848
00:42:26.660 --> 00:42:28.660
a month.

849
00:42:28.660 --> 00:42:32.024
And in another 10 years time, you're
making basically, you know, millions a

850
00:42:32.024 --> 00:42:32.485
month.

851
00:42:32.485 --> 00:42:35.787
Okay, so it's 30% a year increase in GDP,
25%

852
00:42:36.027 --> 00:42:40.631
of that will go on benefits for people
that have been knocked out of a job by AI.

853
00:42:41.151 --> 00:42:42.872
and what do they do with themselves all
day?

854
00:42:42.892 --> 00:42:48.433
They go and get their check,
metaphorically speaking, once a month, and

855
00:42:48.433 --> 00:42:50.433
they have got nothing to do.

856
00:42:50.433 --> 00:42:53.455
They can do a bit of gardening, play a bit
of golf, like David, do a bit of running

857
00:42:53.455 --> 00:42:55.455
and reading, etc.

858
00:42:55.455 --> 00:42:56.715
But eventually you have a social uprising
of bored people.

859
00:42:56.795 --> 00:42:59.156
Crime increases, society collapses.

860
00:42:59.756 --> 00:43:04.298
I sound like, what's his name, Jeff
Goldblum in Jurassic Park, Viv,

861
00:43:05.938 --> 00:43:08.979
because of that park that that silly
fellow made.

862
00:43:09.959 --> 00:43:11.220
But do you see what I mean?

863
00:43:11.500 --> 00:43:12.320
It's all very well.

864
00:43:12.360 --> 00:43:15.161
And you talk about wealth as though money
is wealth.

865
00:43:15.741 --> 00:43:20.963
There's other things apart from money that
are considered wealthy by certain types of

866
00:43:20.963 --> 00:43:21.184
people.

867
00:43:21.203 --> 00:43:23.044
Do you see what I mean, Viv?

868
00:43:24.584 --> 00:43:25.645
I don't believe that to be true.

869
00:43:25.685 --> 00:43:29.546
I think we've had people for thousands of
years who have not had a job.

870
00:43:29.547 --> 00:43:30.387
They call it rich to class.

871
00:43:30.388 --> 00:43:31.287
They live quite nice lives.

872
00:43:34.953 --> 00:43:41.678
we've had we've had people who have had a
job but not hundreds of millions of them

873
00:43:41.678 --> 00:43:43.678
viv not hundreds of millions of astocrats
walking

874
00:43:43.678 --> 00:43:48.603
around with powdered wigs but i'm saying
that people make up their own what happens

875
00:43:48.603 --> 00:43:50.603
when you have like these these i mean
there are people in

876
00:43:50.603 --> 00:43:53.186
new york city there are people even in
joburg uh you know i once did a uh

877
00:43:53.206 --> 00:43:59.751
not like i once did a a a weekly class for
some very wealthy uh wives you know what i

878
00:43:59.751 --> 00:44:00.312
mean

879
00:44:00.312 --> 00:44:04.397
And, you know, these were very smart
women, but they were, the wives were very

880
00:44:04.397 --> 00:44:04.557
rich men.

881
00:44:04.877 --> 00:44:09.759
And they effectively were kind of like,
you know, they had lunch and they went

882
00:44:09.759 --> 00:44:11.759
shopping as they kind of like did they.

883
00:44:11.759 --> 00:44:13.759
But they still had things to do.

884
00:44:13.759 --> 00:44:15.759
They basically did charity events.

885
00:44:15.759 --> 00:44:17.759
They made work, you know, made interesting
things.

886
00:44:17.759 --> 00:44:19.759
They took classes.

887
00:44:19.759 --> 00:44:21.759
People find things to do in their life.

888
00:44:21.759 --> 00:44:23.759
That's not the danger.

889
00:44:23.759 --> 00:44:25.759
Let me tell you.

890
00:44:25.759 --> 00:44:26.385
Viv, I'm sorry, what we'll do is that when
you find the pharma company that makes the

891
00:44:26.385 --> 00:44:26.546
pill that

892
00:44:26.546 --> 00:44:33.048
makes me and you able to live to 500 and
David to 510, okay, we'll go out and do

893
00:44:33.048 --> 00:44:33.369
that,

894
00:44:33.369 --> 00:44:36.749
the three of us, and I can guarantee that
you can talk about people going to do

895
00:44:36.749 --> 00:44:38.749
charity work.

896
00:44:38.749 --> 00:44:40.591
You're talking about a couple of rich
people who've got nothing to do apart from

897
00:44:40.591 --> 00:44:40.651
go down,

898
00:44:40.651 --> 00:44:44.213
have a cup of coffee and pretend to be
nice to the rest of humanity.

899
00:44:44.473 --> 00:44:46.193
I'm talking about hundreds of millions of
people.

900
00:44:46.213 --> 00:44:48.574
I'm talking about ordinary people that
don't have a job.

901
00:44:48.794 --> 00:44:49.955
When you get ordinary people.

902
00:44:50.275 --> 00:44:51.115
that don't have a job.

903
00:44:51.195 --> 00:44:53.076
I promise you, the peasants are revolting.

904
00:44:54.317 --> 00:44:55.317
That's what's going to happen.

905
00:44:55.517 --> 00:45:00.279
But I do get your point, but you seemingly
don't get mine.

906
00:45:00.699 --> 00:45:02.540
I'm a pessimist, Viv, I'm afraid.

907
00:45:02.900 --> 00:45:07.822
I'm a pessimist when it comes to the
long-term future, which hopefully I won't

908
00:45:07.822 --> 00:45:09.822
be able to see.

909
00:45:09.822 --> 00:45:10.023
David, you're sitting there on the fence.

910
00:45:12.564 --> 00:45:13.925
No, I'm not on the fence.

911
00:45:16.006 --> 00:45:16.706
Okay.

912
00:45:16.906 --> 00:45:18.447
What we can say...

913
00:45:18.467 --> 00:45:19.167
Yeah, go on.

914
00:45:19.876 --> 00:45:23.871
No, I say it's a very difficult issue if
this does happen.

915
00:45:24.233 --> 00:45:25.679
I don't believe it will happen.

916
00:45:26.300 --> 00:45:30.182
I don't believe it will happen to the
extent that we say it will happen, but

917
00:45:30.702 --> 00:45:33.463
I think in its own way it creates actually
more jobs.

918
00:45:33.464 --> 00:45:36.484
It becomes an incredible tool for people
to use.

919
00:45:36.544 --> 00:45:37.525
But let's watch that.

920
00:45:37.565 --> 00:45:39.646
You know, let's see how that develops.

921
00:45:40.086 --> 00:45:42.807
I mean, we're getting a little too carried
away with the future.

922
00:45:43.247 --> 00:45:46.589
Well, 30% GDP, Viv started it, not me.

923
00:45:46.590 --> 00:45:49.730
I know, I know.

924
00:45:49.830 --> 00:45:53.972
Listen, when Trump said he wanted Kevin
Walsh to get…

925
00:45:54.776 --> 00:46:01.638
growth up to 15 i thought he was nuts but
30 i said okay

926
00:46:01.639 --> 00:46:08.520
yeah you know we looked at most recent job
numbers in the u.s yeah you actually are

927
00:46:08.520 --> 00:46:10.520
seeing a a bifurcation where you

928
00:46:10.520 --> 00:46:14.221
have really growth without job creation
for the longest period i think they've

929
00:46:14.221 --> 00:46:16.221
recorded and

930
00:46:16.221 --> 00:46:21.223
is that is that not a possible indicator
things are starting to happen i don't

931
00:46:21.223 --> 00:46:23.223
think people are being replaced by air
right now what i do think is happening

932
00:46:23.223 --> 00:46:25.223
right

933
00:46:25.223 --> 00:46:25.566
now and i know from a fact because I
looked at a couple other people in the

934
00:46:25.566 --> 00:46:25.725
sector, etc.

935
00:46:26.106 --> 00:46:27.427
What people are doing is delaying hiring.

936
00:46:28.348 --> 00:46:33.692
They're not hiring new graduates because
they think in a year's time, AI will be

937
00:46:33.692 --> 00:46:35.692
smart enough not to need the new guys.

938
00:46:35.692 --> 00:46:36.635
And that, I think, is what the start is,
right?

939
00:46:36.636 --> 00:46:37.336
Hopefully.

940
00:46:37.336 --> 00:46:41.779
But getting back to the 25% thing here,
David, sorry, it's called Lindsay.

941
00:46:41.780 --> 00:46:42.480
Yes.

942
00:46:42.480 --> 00:46:44.141
25% in which country?

943
00:46:44.601 --> 00:46:45.462
It's going to be in the US.

944
00:46:45.502 --> 00:46:46.343
It's going to be in China.

945
00:46:46.763 --> 00:46:50.046
How does the 25% come to the kid in Lagos
or the kid in Soweto?

946
00:46:50.666 --> 00:46:51.947
It doesn't.

947
00:46:51.948 --> 00:46:52.848
It does not come here.

948
00:46:53.188 --> 00:46:58.589
So we will have the combination of both
lack of employment and no money from

949
00:46:58.589 --> 00:46:58.649
taxes.

950
00:46:58.649 --> 00:47:02.631
Because where is the South African
government going to get the tax money to

951
00:47:02.631 --> 00:47:04.631
pay UBI in South Africa?

952
00:47:04.631 --> 00:47:06.631
Yeah, yeah.

953
00:47:06.631 --> 00:47:08.631
You're quite right.

954
00:47:08.631 --> 00:47:10.433
You both highlighted Africa as something
that would be on the fringes.

955
00:47:10.434 --> 00:47:16.054
And they'll be sort of looking in, you
know, like at the sweet shop, can't afford

956
00:47:16.054 --> 00:47:18.054
to go in and everyone else is in there
munching their heads off.

957
00:47:18.054 --> 00:47:18.475
OK, hopefully that doesn't happen as well.

958
00:47:18.715 --> 00:47:20.875
David, let's get back to reality now and
today.

959
00:47:21.216 --> 00:47:22.676
So after what you've heard...

960
00:47:22.996 --> 00:47:25.958
And after what you've seen with your own
research and listening to Viv,

961
00:47:27.239 --> 00:47:30.722
does anything change when it comes to the
principles of investing?

962
00:47:30.762 --> 00:47:33.865
You look at the market and you say, I
don't understand that yet.

963
00:47:33.965 --> 00:47:37.047
Therefore, I'm not going to invest in it
just because somebody else is.

964
00:47:37.288 --> 00:47:39.569
I'm going to stick to the principles of
P.E.

965
00:47:39.750 --> 00:47:42.772
ratios and cost a book and all that sort
of thing.

966
00:47:42.792 --> 00:47:44.213
Does anything change for you?

967
00:47:45.374 --> 00:47:50.638
No, I think they've just given me a
clearer view of what it does.

968
00:47:51.419 --> 00:47:56.682
And in fact, he's making a better case for
investing in AI.

969
00:47:58.042 --> 00:48:01.824
When you understand how it's going to
change our lives and where it's going,

970
00:48:03.145 --> 00:48:06.406
I'm certainly not going to buy Unilever or
Procter & Gamble.

971
00:48:09.487 --> 00:48:13.449
It's as attractive as they might seem, and
I'm sure there are going to be other

972
00:48:13.449 --> 00:48:15.449
businesses.

973
00:48:15.449 --> 00:48:19.832
But I think, Lindsay, where it comes down
is that this is something I'm not going to

974
00:48:19.832 --> 00:48:21.832
give up and continue.

975
00:48:21.832 --> 00:48:22.229
to hold where we are at the moment.

976
00:48:22.269 --> 00:48:25.171
And perhaps I'm looking for further
companies to add.

977
00:48:25.552 --> 00:48:28.714
You know, that was the whole purpose of my
exercise.

978
00:48:29.254 --> 00:48:34.798
And I think just Viv's reinforced the view
that this is going to be around for a long

979
00:48:34.798 --> 00:48:34.918
time.

980
00:48:35.739 --> 00:48:37.480
So stay with it.

981
00:48:37.920 --> 00:48:40.562
You know, and I know we're trying to bring
money back here.

982
00:48:40.622 --> 00:48:43.524
We're looking for banks in South Africa
and retailers and that.

983
00:48:44.345 --> 00:48:45.906
Yeah, that might be in the short term.

984
00:48:45.986 --> 00:48:48.428
It might be over the next few months
you're going to be able to.

985
00:48:48.896 --> 00:48:55.460
make some money but i think in the longer
term you've got to stick with this theme

986
00:48:55.460 --> 00:48:57.460
and you've got to be fleet-footed as well

987
00:48:57.460 --> 00:49:02.844
as as viv says you know nvidia might be
hot now and and uh making 80 of the

988
00:49:02.864 --> 00:49:06.425
gpus that that are used in ai

989
00:49:06.566 --> 00:49:13.549
or in various things and you know when
when suddenly those things change then

990
00:49:13.549 --> 00:49:15.549
you've got to get out you know and as viv
says he's you

991
00:49:15.549 --> 00:49:20.394
know he he highlights that this could
change pretty fast as well so but the

992
00:49:20.394 --> 00:49:22.394
theme's not going to change.

993
00:49:22.394 --> 00:49:26.978
Now, what isn't going to change as well is
my understanding of it, because after Viv

994
00:49:26.978 --> 00:49:28.978
told me to go to, what was that thing?

995
00:49:28.978 --> 00:49:29.359
Claude.

996
00:49:29.359 --> 00:49:29.701
Claude.

997
00:49:29.701 --> 00:49:31.121
No, not Claude, the other one.

998
00:49:33.062 --> 00:49:33.762
Gosh, what was it?

999
00:49:33.762 --> 00:49:36.244
Double Glock. Yeah, that's it, yes.

1000
00:49:36.465 --> 00:49:37.545
I thought, no, this is too much.

1001
00:49:37.605 --> 00:49:38.506
My brain was throbbing.

1002
00:49:38.546 --> 00:49:44.690
So I looked around at an AI newsletter
that I subscribe to, and it said

1003
00:49:45.331 --> 00:49:49.954
Alibaba has just introduced something new,
and it was on the 15th of February, so

1004
00:49:49.954 --> 00:49:51.954
quite recent.

1005
00:49:51.954 --> 00:49:53.954
on Sunday.

1006
00:49:53.954 --> 00:49:54.256
And Dalai Baba says, it's called QEN 3.5.

1007
00:49:54.257 --> 00:49:58.058
It says, we are delighted to announce the
official release of QEN 3.5,

1008
00:49:58.359 --> 00:50:03.281
introducing the open weight of the first
model in the QEN 3.5 series, namely QEN

1009
00:50:05.483 --> 00:50:09.985
3.5-397b-A17b. As a native vision language
model, that

1010
00:50:10.445 --> 00:50:17.049
QEN 3.5 that I just said demonstrates
outstanding results across a full range of

1011
00:50:17.049 --> 00:50:19.049
benchmark evaluations,

1012
00:50:19.049 --> 00:50:23.652
including reasoning coding, agent
capabilities and multimodal understanding

1013
00:50:23.652 --> 00:50:25.652
empowering developers and

1014
00:50:25.652 --> 00:50:30.215
enterprises and it went on and on and on
and the words got longer and longer and i

1015
00:50:30.215 --> 00:50:32.215
said but what is it i didn't know what it

1016
00:50:32.215 --> 00:50:34.357
was viv i'm being stupid aren't i this

1017
00:50:34.358 --> 00:50:41.160
is something to actually point out as well
it's another thing like it's not like

1018
00:50:41.160 --> 00:50:43.160
opening it's not like this is another
thing like chad gpt it's another thing

1019
00:50:43.160 --> 00:50:45.160
like claude but this is an

1020
00:50:45.160 --> 00:50:48.085
important fact here the chinese are
basically giving this They are effectively

1021
00:50:48.085 --> 00:50:50.085
dumping the market with LLMs,

1022
00:50:50.085 --> 00:50:52.085
free LLMs.

1023
00:50:52.085 --> 00:50:54.170
So, Quen, Kimi, DeepSeek are all free
open-weight models.

1024
00:50:54.430 --> 00:50:57.653
But effectively, the Chinese are coming
out and saying, you know the stuff that

1025
00:50:57.653 --> 00:50:59.653
Anthropic is making?

1026
00:50:59.653 --> 00:51:01.653
Here's something almost as good for free.

1027
00:51:01.653 --> 00:51:03.653
Here's the stuff that

1028
00:51:03.653 --> 00:51:09.803
Gemini, you know, Chachi PT, here is what
they have almost as good a few months

1029
00:51:09.803 --> 00:51:11.803
later,

1030
00:51:11.803 --> 00:51:13.803
but free.

1031
00:51:13.803 --> 00:51:15.803
Again, what kind of model?

1032
00:51:15.803 --> 00:51:20.553
like you know a car company trying to
survive when his biggest competitors are

1033
00:51:20.553 --> 00:51:22.553
giving away something almost as good for
free and that

1034
00:51:22.553 --> 00:51:27.359
is another aspect of this market that we
haven't discussed is the fact that the

1035
00:51:27.359 --> 00:51:29.359
chinese are really by this by by using
open

1036
00:51:29.359 --> 00:51:30.062
source are effectively ensuring

1037
00:51:30.082 --> 00:51:37.009
that if it's ever stall if it's ever
slowed down that they will destroy any of

1038
00:51:37.009 --> 00:51:39.009
these llm model makers out there because
they

1039
00:51:39.009 --> 00:51:39.647
are giving whatever they are making for
free Very interesting.

1040
00:51:39.727 --> 00:51:40.427
Okay.

1041
00:51:40.427 --> 00:51:44.871
It's one for, I mean, do you buy the
Chinese company if it's not making money

1042
00:51:44.871 --> 00:51:46.871
because it's giving it away for free?

1043
00:51:46.871 --> 00:51:47.693
But there's obviously a much, much bigger
story than that.

1044
00:51:47.733 --> 00:51:49.855
Gentlemen, thank you very much for your
time.

1045
00:51:49.935 --> 00:51:56.661
I've enjoyed being the village idiot
combined with the devil's advocate.

1046
00:51:57.041 --> 00:51:58.282
Viv, brilliant as always.

1047
00:51:58.402 --> 00:52:01.945
David, listening, soaking it all up as
always and contributing.

1048
00:52:02.686 --> 00:52:07.630
Viv Govender is from the award-winning RAN
Swiss and David Shapiro is the

1049
00:52:07.630 --> 00:52:09.630
award-winning.

1050
00:52:09.630 --> 00:52:14.652
portfolio manager from sasfin securities
and that was a special five o'clock shadow

1051
00:52:14.652 --> 00:52:16.652
the views and opinions

1052
00:52:16.652 --> 00:52:21.214
expressed in these podcasts are those of
lindsey williams and various contributors

1053
00:52:21.214 --> 00:52:23.214
and do not reflect the policy position

1054
00:52:23.214 --> 00:52:27.555
or opinion of any other agency
organization employer or company

1055
00:52:27.555 --> 00:52:29.555
associated with

1056
00:52:29.555 --> 00:52:34.317
strictlybusinesspodcast.com assumptions
made on the analyses are not reflective of

1057
00:52:34.317 --> 00:52:36.317
the position of any other

1058
00:52:36.317 --> 00:52:36.998
entity other than the speaker or the
author Thank you.

1059
00:52:37.058 --> 00:52:42.246
And since we are critically thinking human
beings, these views are always subject to

1060
00:52:42.246 --> 00:52:42.346
change,

1061
00:52:42.526 --> 00:52:45.090
revision and rethinking at any time.

1062
00:52:45.390 --> 00:52:47.794
Please do not hold us to them in
perpetuity.
