#107 Unraveling the Future of AI: The Developer's Perspective with Steven Schrembeck

1
00:00:10,000 --> 00:00:12,900
Hello, and welcome to a new
episode of the city of show with

2
00:00:12,900 --> 00:00:14,300
mammoths.
My name is Muhammad.

3
00:00:14,300 --> 00:00:16,700
And as you know, in each
episode, I discussed different

4
00:00:16,700 --> 00:00:18,600
topics about emerging
Technologies.

5
00:00:19,300 --> 00:00:23,400
A I just, in fact formation
cyber security and also, I

6
00:00:23,400 --> 00:00:25,800
sometimes have guests with me
who are subject.

7
00:00:25,800 --> 00:00:29,600
Matter experts in one of the
domains I usually talk about and

8
00:00:29,600 --> 00:00:31,200
today I'm very pleased to have
with me.

9
00:00:31,500 --> 00:00:35,400
Steven shainberg, who's joining
me from the United States from

10
00:00:35,400 --> 00:00:36,900
Georgia?
Steven.

11
00:00:36,900 --> 00:00:39,100
Thank you very much for being on
the show today.

12
00:00:39,100 --> 00:00:40,900
I will keep it.
It to you, to introduce

13
00:00:40,900 --> 00:00:43,800
yourself, what you do, and what
you are up to.

14
00:00:45,400 --> 00:00:50,300
Mmm, some thanks for having me
on so I do three things that are

15
00:00:50,300 --> 00:00:54,000
relevant once I'm the founder of
a start-up called impossible

16
00:00:54,000 --> 00:00:58,500
Labs so it builds a i stuff, it
sounds very fancy but I'm

17
00:00:58,500 --> 00:01:04,300
basically a solopreneur 0 plus
AI plus contractors and I really

18
00:01:04,300 --> 00:01:08,200
liked it that way beyond that I
make videos on a channel called

19
00:01:08,200 --> 00:01:10,900
seeking Minima.
So that's what brought us here

20
00:01:10,900 --> 00:01:13,200
today.
That channel has one purpose and

21
00:01:13,200 --> 00:01:17,100
that is to help people.
Use technology instead of being

22
00:01:17,100 --> 00:01:19,900
used by it and that's sort of my
goal.

23
00:01:20,400 --> 00:01:22,400
And then the last part is I
write stories.

24
00:01:22,800 --> 00:01:25,400
Science Fiction and Fantasy.
Oh wow.

25
00:01:25,800 --> 00:01:28,900
I love this combination really
now.

26
00:01:29,500 --> 00:01:31,500
I know.
Stephen also widened like you

27
00:01:31,500 --> 00:01:34,600
you do software development as
well, right?

28
00:01:35,600 --> 00:01:37,300
That's right.
Yeah, I didn't mention the one

29
00:01:37,300 --> 00:01:39,200
thing that like pays the bills,
right?

30
00:01:39,800 --> 00:01:43,500
So I guess AI does pay the bills
but yeah.

31
00:01:43,500 --> 00:01:47,100
So I've been a software
developer Her for 12 years,

32
00:01:47,100 --> 00:01:50,200
something like that big
corporate software developer.

33
00:01:50,600 --> 00:01:52,700
I made the rounds.
I worked at Amazon for a while

34
00:01:53,400 --> 00:01:58,600
working together companies.
So it's been good to me, but

35
00:01:58,700 --> 00:02:02,600
it's not, it's not as exciting
as cutting-edge Tech which is

36
00:02:02,600 --> 00:02:06,800
always where my heart is.
That That's great.

37
00:02:06,800 --> 00:02:11,500
So, stupid best wore out of
curiosity, like you have a mix

38
00:02:11,500 --> 00:02:14,500
of, you know, multiple things at
the same time.

39
00:02:14,900 --> 00:02:21,800
So what I would say, drove you
to choose this path for your

40
00:02:21,800 --> 00:02:28,600
career.
So, I've moved to different

41
00:02:28,600 --> 00:02:32,100
kinds of software development.
And first of all, I can't help

42
00:02:32,100 --> 00:02:33,900
myself.
That's the easy answer, right?

43
00:02:34,300 --> 00:02:37,400
It's, I can't help it, be
interested in things writing is

44
00:02:37,400 --> 00:02:39,700
something I haven't been able to
put down for a long time.

45
00:02:39,700 --> 00:02:41,800
So that's just like, something
they won't leave me alone.

46
00:02:43,200 --> 00:02:45,000
And same with cutting-edge tech,
right?

47
00:02:45,200 --> 00:02:49,200
I moved into software
development because I liked

48
00:02:49,400 --> 00:02:52,400
building things, I'm not one of
those people that loves code for

49
00:02:52,400 --> 00:02:55,700
code sake.
I like it as a tool.

50
00:02:56,400 --> 00:02:58,700
And while I am fascinated with
building things and

51
00:02:58,700 --> 00:03:02,500
understanding how they work, I'm
more interested in the

52
00:03:02,500 --> 00:03:06,000
opportunity to solve cool
problems and help people.

53
00:03:06,300 --> 00:03:08,800
And software was just the
easiest way to do that because,

54
00:03:08,800 --> 00:03:11,400
you know, it's easy to set up
tear down like there's no

55
00:03:11,400 --> 00:03:14,400
sandbox like software.
So I think that's really what

56
00:03:14,400 --> 00:03:18,200
drew me in and then AI is just
honestly, it felt like

57
00:03:18,200 --> 00:03:21,200
modern-day magic.
I started building models and

58
00:03:21,200 --> 00:03:25,300
2016 when deep learning was
having a first or second

59
00:03:25,300 --> 00:03:27,000
Renaissance to bring it.
How you look at it.

60
00:03:27,300 --> 00:03:31,100
And I knew that this was
computer magic and I wanted to

61
00:03:31,100 --> 00:03:34,800
know how to do it, so that's
what drew me into AI.

62
00:03:36,400 --> 00:03:40,900
That's very, very cool.
I would say, now, do and I know

63
00:03:40,900 --> 00:03:44,100
this is the Hot Topic.
It's on everyone's mind, right?

64
00:03:44,100 --> 00:03:49,600
So it's a, i how do you perceive
it from?

65
00:03:50,200 --> 00:03:53,200
Not only a developer
perspective, a solopreneur

66
00:03:53,200 --> 00:03:55,800
perspective.
How do you do to a?

67
00:03:55,800 --> 00:03:59,900
I do you, do you see it as you
know, the technology that were,

68
00:04:01,000 --> 00:04:05,100
you know, make really people's
life easy or do you see it?

69
00:04:05,200 --> 00:04:10,700
It more as a just a cool
technology that we can do cool

70
00:04:10,700 --> 00:04:14,100
stuff with it.
How do you perceive really the

71
00:04:14,100 --> 00:04:16,800
technology because there are not
having that is now asking you

72
00:04:16,800 --> 00:04:21,000
step at this question, there are
a lot of debates recently with

73
00:04:21,100 --> 00:04:24,800
all the hype that happened after
chat GPT, and all these

74
00:04:24,800 --> 00:04:27,800
Technologies what's next.
What's going to happen in the

75
00:04:27,800 --> 00:04:30,800
world.
So from a developer perspective,

76
00:04:30,800 --> 00:04:38,500
a solopreneur perspective, what
you can tell us about this So, a

77
00:04:38,508 --> 00:04:40,200
lot of the time the hype isn't
real.

78
00:04:40,600 --> 00:04:44,300
So I also, you know, I'm very
interested in web three mostly

79
00:04:44,300 --> 00:04:46,200
can for like coordination
Technologies.

80
00:04:46,400 --> 00:04:50,000
I'm terrible at a terrible
writing the height wave.

81
00:04:50,000 --> 00:04:51,900
I'm always interested in the
tech in the values.

82
00:04:52,900 --> 00:04:55,900
So for AI this time the hype is
kind of real.

83
00:04:56,200 --> 00:04:59,800
And it was last time too.
If you'll recall, I don't know

84
00:04:59,800 --> 00:05:01,900
what 45 years ago.
The last time deep learning

85
00:05:01,900 --> 00:05:03,900
really took off, it was for a
vision networks.

86
00:05:04,100 --> 00:05:08,000
And this is really where the
killer use case was and The hype

87
00:05:08,000 --> 00:05:12,000
was real, it just wasn't evenly
distributed now, as you know,

88
00:05:12,000 --> 00:05:14,500
diffusion networks for a
generative Ai and large

89
00:05:14,500 --> 00:05:17,400
language, models are the latest
hotness in.

90
00:05:17,400 --> 00:05:22,500
The hyperbole is real again.
And I think the reason that we

91
00:05:22,500 --> 00:05:28,000
might not see another AI winter
between here, and I really doing

92
00:05:28,500 --> 00:05:33,900
very impactful things to society
is because we have a lot of Lego

93
00:05:33,900 --> 00:05:37,500
blocks that we built up.
If you followed AI development,

94
00:05:37,600 --> 00:05:40,900
The past decade or so, it keeps
increasing.

95
00:05:41,400 --> 00:05:44,400
But what has happened is that
everyone was increasing in their

96
00:05:44,400 --> 00:05:46,900
own silos.
So you know, latch natural

97
00:05:46,900 --> 00:05:51,100
language processing was getting
better time, series analysis was

98
00:05:51,100 --> 00:05:53,600
getting better, just Transformer
networks.

99
00:05:53,600 --> 00:05:56,100
All all the vision networks.
All of these things were

100
00:05:56,100 --> 00:05:57,800
improving.
Diffusion, autoencoders.

101
00:05:57,800 --> 00:05:59,900
All these things were improving
separately.

102
00:06:00,500 --> 00:06:05,200
And what happened is people from
One domain connected at these

103
00:06:05,500 --> 00:06:08,000
two, another domains.
And this case it was Enforcement

104
00:06:08,000 --> 00:06:12,700
learning plus natural language
processing and plus, and that

105
00:06:12,700 --> 00:06:15,600
was basically you connect two
pieces from two verticals that

106
00:06:15,600 --> 00:06:18,600
have been advancing for a long
time and the result was

107
00:06:18,600 --> 00:06:21,000
exponential increase.
It could do things.

108
00:06:21,000 --> 00:06:24,400
The difference was, we have
generative networks, we've had

109
00:06:24,400 --> 00:06:27,300
Gans for a long time that there
are sort of obtuse and really

110
00:06:27,300 --> 00:06:31,700
hard to train.
The difference is we needed we

111
00:06:31,700 --> 00:06:34,900
needed an interface that
understood what we meant.

112
00:06:35,200 --> 00:06:37,000
That was the big, like unlock
here.

113
00:06:37,600 --> 00:06:41,400
Is that we could always sort of
get networks to do really good

114
00:06:41,400 --> 00:06:44,800
things narrowly, but we couldn't
interface with them in a very

115
00:06:44,800 --> 00:06:48,500
good way.
If they know something like like

116
00:06:48,500 --> 00:06:51,500
a large language model does how
do you get the information out

117
00:06:51,500 --> 00:06:53,900
of it?
And I think that natural

118
00:06:53,900 --> 00:06:56,200
language processing was an
interface moment.

119
00:06:56,500 --> 00:07:00,300
So the combination of all these
individual and advancements.

120
00:07:00,600 --> 00:07:05,700
Is really why you're seeing huge
changes now and the amount of

121
00:07:05,700 --> 00:07:07,600
individual advancement has not
stopped.

122
00:07:07,600 --> 00:07:10,900
So I would be very surprised if
there were not more verticals to

123
00:07:10,900 --> 00:07:15,400
combine and again, create
another ridiculous pivotal

124
00:07:15,400 --> 00:07:18,900
moment every six months or so,
but you might see lows in the

125
00:07:18,900 --> 00:07:20,900
meantime so I'd say, the hype is
real.

126
00:07:20,900 --> 00:07:25,500
Sort of right?
People always tend to overhype,

127
00:07:25,500 --> 00:07:29,200
no matter what you do, he
actually Stephen.

128
00:07:29,200 --> 00:07:34,000
The you said, An important which
I discussed couple of episodes

129
00:07:34,000 --> 00:07:37,000
back.
I go solo sometime and I said

130
00:07:37,000 --> 00:07:40,700
like I'm a no not very well guy
but I'm an old guy enough to

131
00:07:40,700 --> 00:07:45,200
remember all the times in
history and I don't remember him

132
00:07:45,200 --> 00:07:48,000
moment.
No similar to what we are seeing

133
00:07:48,000 --> 00:07:53,000
today because I was like, maybe
10, 11 years old, when the first

134
00:07:53,000 --> 00:07:56,800
time I heard about the internet,
you know, I was 14 years old

135
00:07:56,800 --> 00:08:00,100
when I tried the internet and
then you know, later on the

136
00:08:00,700 --> 00:08:05,400
Smartphone that web to.
Yeah.

137
00:08:05,400 --> 00:08:08,700
Like then was sometimes
exaggeration sometimes, you know

138
00:08:08,700 --> 00:08:11,700
things going.
But you know that the cycle used

139
00:08:11,700 --> 00:08:14,500
to take very long until we see
the next thing.

140
00:08:14,500 --> 00:08:19,200
But with a II mean generative Ai
and the in LPS and you know,

141
00:08:19,200 --> 00:08:21,200
this all models that we are
seeing today.

142
00:08:21,200 --> 00:08:24,200
I think we, as you said, I like
the word.

143
00:08:24,200 --> 00:08:27,400
You set an ER we get we will not
see an AI wind.

144
00:08:27,800 --> 00:08:31,500
I love this expression and I
think we are Are going very

145
00:08:31,500 --> 00:08:33,900
fast.
Now from me I would say

146
00:08:33,900 --> 00:08:38,700
developer perspective, right?
What do you think would be the

147
00:08:38,700 --> 00:08:43,200
best applications?
That let me make it very certain

148
00:08:43,200 --> 00:08:47,800
that everyone can benefit on
other than you know the I would

149
00:08:47,800 --> 00:08:52,400
say the sometimes very simple
things, we see, write me an

150
00:08:52,400 --> 00:08:56,100
email right miss this.
So I believe like from your

151
00:08:56,100 --> 00:08:58,300
perspective, Stephen you see a
bigger picture?

152
00:08:58,300 --> 00:09:00,500
Can you shave that with from
your past?

153
00:09:02,200 --> 00:09:05,600
Yeah.
I think another reason why this

154
00:09:05,600 --> 00:09:10,600
time it's different is we have a
very general model a shockingly

155
00:09:10,600 --> 00:09:13,500
general model.
Meaning you can, it can reason

156
00:09:13,500 --> 00:09:16,300
about things, right?
Whether it doesn't really matter

157
00:09:16,300 --> 00:09:18,400
what's happening under the hood,
all that matters is the input

158
00:09:18,400 --> 00:09:21,200
output results in something that
looks like commodified

159
00:09:21,200 --> 00:09:23,500
intelligence.
It's a brain in a box you can

160
00:09:23,500 --> 00:09:25,800
spin it up.
That's not super smart.

161
00:09:25,800 --> 00:09:29,000
It's like maybe average adult
smart and then in a couple

162
00:09:29,000 --> 00:09:33,100
domains it's exceptionally smart
like law or some solving sat.

163
00:09:33,200 --> 00:09:36,800
Holmes.
But what matters is that you

164
00:09:36,800 --> 00:09:39,200
have common sense reasoning in a
box.

165
00:09:39,300 --> 00:09:42,200
That's the kind of thing that
honestly I expected to take a

166
00:09:42,200 --> 00:09:44,800
lot longer and I think a lot of
people did I did not see this

167
00:09:44,800 --> 00:09:46,700
one coming.
Next there was a number of

168
00:09:46,708 --> 00:09:49,500
hurdles to artificial general
intelligence and I didn't think

169
00:09:49,500 --> 00:09:52,400
this one was going to get
knocked over first or next.

170
00:09:52,400 --> 00:09:55,700
Anyway.
So really when you think about

171
00:09:55,700 --> 00:09:59,800
how can I use this What could
you ask somebody that you could

172
00:09:59,800 --> 00:10:03,100
pay minimum wage or just
somebody with no training of

173
00:10:03,100 --> 00:10:05,500
average intelligence to do for
you on a computer?

174
00:10:06,500 --> 00:10:10,600
Now you can do that for instead
of fifteen dollars an hour

175
00:10:10,600 --> 00:10:15,000
depending where you live, you
can do that for 20 cents and you

176
00:10:15,000 --> 00:10:19,300
can do it a thousand times
faster and you can have 100,000

177
00:10:19,300 --> 00:10:23,000
of them working.
So what can you do with that?

178
00:10:23,000 --> 00:10:28,000
Well, any information processing
is if your job is in Mission

179
00:10:28,000 --> 00:10:32,200
processing.
That's a i job now and you're

180
00:10:32,200 --> 00:10:34,300
going to do something else.
So it's kind of hollowing out

181
00:10:34,300 --> 00:10:37,500
the middle.
It's hollowing out people who

182
00:10:37,500 --> 00:10:40,500
work on if your job is to turn
one piece of information into

183
00:10:40,500 --> 00:10:44,800
another piece of information, if
there is a broad skill set, and

184
00:10:44,800 --> 00:10:47,700
data set for what you do, you're
going to be replaced pretty

185
00:10:47,700 --> 00:10:52,000
quickly depending on the
economic output of what it's

186
00:10:52,000 --> 00:10:56,200
worth to replace you.
Except if you're a specialist

187
00:10:56,400 --> 00:10:59,600
there is is very hard to replace
specialist.

188
00:10:59,600 --> 00:11:04,600
So I know a person whose job is
to audit oil and gas Refinery

189
00:11:05,100 --> 00:11:08,400
machinery for safety and even
though his job is basically, to

190
00:11:08,400 --> 00:11:12,500
analyze data and spit out, okay?
Safe, not safe, do mate.

191
00:11:12,500 --> 00:11:15,900
Most don't do maintenance
replace, he's not had to be

192
00:11:15,900 --> 00:11:19,800
replaced for a very long time
because it's such a specialized

193
00:11:19,800 --> 00:11:25,300
knowledge set that he's fine.
So if your job Is pretty generic

194
00:11:25,700 --> 00:11:29,500
Junior coder with no
specialization or pretty generic

195
00:11:29,500 --> 00:11:32,600
Junior.
Designer who just does websites.

196
00:11:33,900 --> 00:11:35,800
You should consider using other
things.

197
00:11:35,800 --> 00:11:37,600
Now, I think your question was
to ask.

198
00:11:38,200 --> 00:11:41,500
How can people use this?
That was sort of like the - some

199
00:11:41,500 --> 00:11:45,400
dinner, which it's a lot of
attention, but the positive side

200
00:11:45,400 --> 00:11:47,600
is True to.
This is commodified intelligence

201
00:11:47,600 --> 00:11:50,600
in a box.
If something if there's enough

202
00:11:50,600 --> 00:11:54,100
data you can generate it.
That's the very Nothing.

203
00:11:56,100 --> 00:12:01,500
I can tell you that what I'm you
for is to remove is to really

204
00:12:01,500 --> 00:12:05,800
take control of my sort of.
To fight back against the

205
00:12:05,800 --> 00:12:08,700
attention economy I guess, is
the right way of putting it.

206
00:12:08,900 --> 00:12:12,100
So just for instance, yesterday
actually this morning and last

207
00:12:12,100 --> 00:12:16,100
night, I wrote a tool that uses
open my eyes whisper to

208
00:12:16,100 --> 00:12:19,300
transcribe podcasts.
So, takes from audio to text

209
00:12:19,500 --> 00:12:22,600
transcript including speaker
attribution, so who's talking.

210
00:12:23,100 --> 00:12:26,400
And then it takes that pumps it
into chat gbt for through the

211
00:12:26,400 --> 00:12:31,000
API and then I run it through a
prompt where I tell it

212
00:12:31,000 --> 00:12:33,200
basically.
All right, here's the questions.

213
00:12:33,200 --> 00:12:35,800
I'm going to ask hear this.
Things I need to know about this

214
00:12:35,800 --> 00:12:37,300
podcast.
Here are the viewpoints, I want

215
00:12:37,300 --> 00:12:39,900
to see, and then it just curates
this report.

216
00:12:40,100 --> 00:12:42,800
So there are certain podcast
will say, like macro investing,

217
00:12:42,800 --> 00:12:44,300
right?
I really care about this too

218
00:12:44,300 --> 00:12:45,200
much.
I want to know.

219
00:12:45,200 --> 00:12:47,100
Roughly, you know, where the
markets are.

220
00:12:47,400 --> 00:12:48,500
I don't want to pay attention
to.

221
00:12:48,500 --> 00:12:50,900
It news is makes you feel long
enough.

222
00:12:51,300 --> 00:12:54,700
And now with the push of a
button, here's everything I need

223
00:12:54,700 --> 00:12:57,900
to know, distilled down into a
tiny amount and I can curate

224
00:12:57,900 --> 00:13:00,800
that for many podcast feed and
this is just stuff I'm making.

225
00:13:01,100 --> 00:13:06,900
So I am using it to sort of Get
more information and learn more

226
00:13:06,900 --> 00:13:11,200
than I could before, and under
to learn quickly.

227
00:13:12,400 --> 00:13:15,300
And to synthesize information,
I'm using it like intelligence

228
00:13:15,300 --> 00:13:18,800
in a box.
Hey, read this for me, I hate

229
00:13:18,800 --> 00:13:23,400
distill this down but only in
the places where my goal is just

230
00:13:23,400 --> 00:13:25,100
information.
So if you want to be Matrix

231
00:13:25,100 --> 00:13:28,200
mainlined just plug me in.
Like I just want to know how to

232
00:13:28,200 --> 00:13:31,200
do this.
This is a killer use case.

233
00:13:31,200 --> 00:13:34,100
Right now for anybody, you don't
have to know how to program,

234
00:13:34,600 --> 00:13:37,400
just interfacing with the the
chat clients is enough.

235
00:13:38,300 --> 00:13:41,400
You could be learning really,
really quickly, right now.

236
00:13:42,100 --> 00:13:46,000
And so this is the sort of this
is the trade, then we have an

237
00:13:46,000 --> 00:13:49,100
opportunity where AI is really
really useful at helping you

238
00:13:49,100 --> 00:13:52,400
learn just about anything that
makes you very marketable right

239
00:13:52,400 --> 00:13:57,400
now or you can ignore it and sit
around and wait for it to your

240
00:13:57,400 --> 00:14:02,300
job to be slowly eating away.
So it's like you can have you

241
00:14:02,300 --> 00:14:05,300
can really stand out right now.
If you're willing to do the work

242
00:14:05,300 --> 00:14:08,600
and learn and try these things
out or you can, you know, wait

243
00:14:08,600 --> 00:14:11,600
on the sidelines.
Wow.

244
00:14:11,800 --> 00:14:14,100
Do you know that what you are
trying to bring?

245
00:14:14,100 --> 00:14:17,800
This is a something similar
other have been met exactly the

246
00:14:17,800 --> 00:14:22,500
same idea.
But I wanted to I didn't lie or

247
00:14:22,500 --> 00:14:25,200
preparing this podcast on some
news feed.

248
00:14:25,200 --> 00:14:31,200
So and one of the use cases like
thought is Led GPT or whatever

249
00:14:31,200 --> 00:14:34,700
the API read these feeds and
just select which ones are

250
00:14:34,700 --> 00:14:38,000
important.
So I can choose a topic for

251
00:14:38,000 --> 00:14:41,700
example.
Now you mentioned the word from

252
00:14:41,900 --> 00:14:45,700
several times and there was a
hype, if we can call it apart

253
00:14:46,200 --> 00:14:49,600
from the engineers, right?
So prompt engineers and you

254
00:14:49,600 --> 00:14:53,400
know, A New Perspective and
let's discuss it both you know

255
00:14:53,400 --> 00:14:57,500
from a real software development
perspective and plus you know

256
00:14:57,700 --> 00:15:02,500
realistic use case perspective.
How important is the prom.

257
00:15:02,800 --> 00:15:09,500
When interacting with models
like chatty Beauty Ideally over

258
00:15:09,500 --> 00:15:12,000
time, the prompt becomes less
important.

259
00:15:12,400 --> 00:15:15,400
If you have a good natural
language model, the goal is for

260
00:15:15,400 --> 00:15:18,700
to be a better interface.
So I'll just take a step back

261
00:15:18,700 --> 00:15:21,200
and do a little technical
explanation that I guess

262
00:15:21,200 --> 00:15:23,700
beautifully doesn't put people
to sleep, but I think is very

263
00:15:23,700 --> 00:15:27,200
useful for understanding how
these things work.

264
00:15:27,700 --> 00:15:33,000
So at the center of these large
language models and even stuff,

265
00:15:33,000 --> 00:15:37,200
like diffusion models is
something that researchers call

266
00:15:37,200 --> 00:15:40,300
a latent space.
Which in my opinion, they just

267
00:15:40,300 --> 00:15:42,900
use a technical sounding term to
describe something is very

268
00:15:42,900 --> 00:15:45,300
complicated.
I like to think of it as just

269
00:15:45,300 --> 00:15:49,200
this big information soup, it's
encoded in a sort of machine

270
00:15:49,200 --> 00:15:51,100
language.
It's just a representation that

271
00:15:51,100 --> 00:15:54,800
the model has learned about the
word the world in the most dense

272
00:15:54,800 --> 00:15:56,300
format.
This is the center of

273
00:15:56,300 --> 00:15:58,600
everything.
It knows, this is where it's

274
00:15:58,600 --> 00:16:02,100
recorded, all the information,
it knows all the patterns it

275
00:16:02,100 --> 00:16:04,500
needs to find.
You can think of it.

276
00:16:04,600 --> 00:16:08,300
It's like it's repository of
knowledge but it's not in words.

277
00:16:08,400 --> 00:16:12,900
It's in technically in
embeddings, it's just hyper

278
00:16:12,900 --> 00:16:17,500
compressed information so it
knows a lot of stuff about a lot

279
00:16:17,500 --> 00:16:18,800
of stuff.
You know.

280
00:16:18,800 --> 00:16:22,800
The, the open AI models have
read a bulk of the public

281
00:16:22,800 --> 00:16:26,900
internet.
So they know a lot of things.

282
00:16:26,900 --> 00:16:29,400
In fact, I know a lot more than
we can pull out of them.

283
00:16:29,800 --> 00:16:33,300
And so a lot of the trick is,
how do we get out?

284
00:16:33,300 --> 00:16:36,500
I know you can do this.
How do I get you to know what I

285
00:16:36,508 --> 00:16:38,300
mean?
It's like an sent Tha size.

286
00:16:38,800 --> 00:16:41,100
Either do these operations,
write this code.

287
00:16:41,100 --> 00:16:43,000
Perform these tasks.
Help me again.

288
00:16:43,000 --> 00:16:46,000
Read this thing, understand what
I'm trying to get.

289
00:16:46,400 --> 00:16:50,600
So that interface gap between, I
know you have this information

290
00:16:50,600 --> 00:16:53,200
in your latent space and this
ocean of information.

291
00:16:53,300 --> 00:16:56,500
I know you can do this.
I'm trying to get you to focus,

292
00:16:56,500 --> 00:17:00,000
only on what I want you to do
and there's some misalignment

293
00:17:00,000 --> 00:17:02,100
there.
So there's a gap in the

294
00:17:02,100 --> 00:17:05,000
interface which is do, you know
what I mean?

295
00:17:05,200 --> 00:17:07,400
And can use it in.
Pull it out of what you actually

296
00:17:07,400 --> 00:17:10,200
know how to do.
So that's what prompted

297
00:17:10,200 --> 00:17:13,400
engineering right now.
Solves for is, can you shrink

298
00:17:13,400 --> 00:17:16,500
that Gap to be better at
synthesizing?

299
00:17:16,500 --> 00:17:18,700
What you want from these large
language models.

300
00:17:19,000 --> 00:17:24,200
So right now it is useful.
I will probably be useful for a

301
00:17:24,200 --> 00:17:28,900
while but in theory as things
get better, that interface

302
00:17:28,900 --> 00:17:31,900
should get better and the models
will get better and better at

303
00:17:31,900 --> 00:17:35,400
understanding what you want so
that you don't have to use all

304
00:17:35,400 --> 00:17:37,700
these tricks of the trade.
So I wouldn't say there's a

305
00:17:37,700 --> 00:17:41,600
super long Long shelf life on
this skill but I could be wrong.

306
00:17:43,400 --> 00:17:47,700
At that cool fish me, you know
like because yesterday when

307
00:17:47,900 --> 00:17:51,000
actually in the during the week
at one of the things I was

308
00:17:51,000 --> 00:17:58,300
trying to do is to let the model
tells me how best it prefers to

309
00:17:58,300 --> 00:18:04,300
be interacted with.
And yeah I spend some time until

310
00:18:04,300 --> 00:18:10,100
at actively you people might
love but I told Chad GPT after I

311
00:18:10,100 --> 00:18:13,100
get the answer I wanted.
I said okay Okay, can we write

312
00:18:13,100 --> 00:18:17,100
any book about the and put some
examples and they did?

313
00:18:17,300 --> 00:18:23,000
So as per chair GPT it prefers
to have the following a context

314
00:18:23,400 --> 00:18:28,900
specificity instruction purpose
or goal and limitation.

315
00:18:28,900 --> 00:18:33,300
So this is what I was told by
GPT and it gave me an example,

316
00:18:33,300 --> 00:18:35,000
by the way, if you ask me this
way.

317
00:18:35,200 --> 00:18:36,900
Exactly.
As you said, Stephen, I know

318
00:18:36,900 --> 00:18:40,800
that I should search exactly in
this, you know, I don't have to

319
00:18:40,800 --> 00:18:43,500
go search the whole language.
There's now more than because I

320
00:18:43,500 --> 00:18:46,400
know exactly what I have to go.
So this is maybe where the

321
00:18:46,400 --> 00:18:49,000
prompt engineering, but I
believe, you know, I have a

322
00:18:49,008 --> 00:18:51,400
little bit of technical
background and I know a little

323
00:18:51,400 --> 00:18:54,100
bit how these models work.
I think the more they are

324
00:18:54,100 --> 00:18:57,400
trained, the better they get.
So even you will not need even

325
00:18:58,000 --> 00:19:01,400
know an exact from to get the
information out of it, right?

326
00:19:01,600 --> 00:19:04,800
So that's that's very, very
cool.

327
00:19:04,800 --> 00:19:08,000
I would say.
Now here, I want to ask you.

328
00:19:08,000 --> 00:19:12,400
Now, we have this moment, I
would say, What do you think

329
00:19:12,500 --> 00:19:16,000
other technology is really get
combined or let's say, you know,

330
00:19:16,100 --> 00:19:21,900
used together with the, you
know, all this AI Technologies.

331
00:19:22,100 --> 00:19:25,800
Like for example personally, I
think automation with Arie play

332
00:19:25,800 --> 00:19:28,200
a huge role.
I started to see, you know,

333
00:19:28,200 --> 00:19:32,800
someone talked about autonomous
boats that they can interact,

334
00:19:32,800 --> 00:19:36,200
you know, with each other.
What's your view on this?

335
00:19:37,400 --> 00:19:39,500
All right, so use this.
Let's like three.

336
00:19:39,500 --> 00:19:47,300
Good questions. so, I think that
there's a lot of it will be

337
00:19:47,300 --> 00:19:51,400
combined with everything it's
intelligence, so it gets

338
00:19:51,400 --> 00:19:53,700
combined with everything but I
think the ones that are sort of

339
00:19:53,700 --> 00:19:57,100
Juicy and next, I actually don't
think it's automation.

340
00:19:57,100 --> 00:20:00,100
Now I could be wrong.
There is a lot of value there.

341
00:20:00,900 --> 00:20:03,300
So I think it's more like
narrowly.

342
00:20:03,300 --> 00:20:05,700
There are new kinds of
automation that we just unlocked

343
00:20:05,700 --> 00:20:08,700
and those yeah that's Greenfield
that'll be anything that

344
00:20:08,700 --> 00:20:10,300
required like Common Sense
reasoning.

345
00:20:10,800 --> 00:20:14,200
I think those things there is.
Okay, automation will move

346
00:20:14,200 --> 00:20:16,400
forward.
The reason I don't think that

347
00:20:16,400 --> 00:20:20,300
this is the automation moment.
I mean, physical movement and

348
00:20:20,300 --> 00:20:24,500
Manufacturing stuff in the real
world is because physical stuff

349
00:20:24,500 --> 00:20:28,400
is hard.
It's hard and expensive, and

350
00:20:28,400 --> 00:20:32,800
physics is finicky, and moving
stuff around in the real world

351
00:20:32,800 --> 00:20:36,700
requires precision, and rules
and money.

352
00:20:37,200 --> 00:20:40,000
And a lot of testing that is a
lot slower than doing stuff in

353
00:20:40,000 --> 00:20:41,800
the digital world.
And this is something that

354
00:20:41,800 --> 00:20:46,500
surprised me, but as soon as I
saw, Asian is that year ago,

355
00:20:46,500 --> 00:20:48,800
something like that?
When stable diffusion first came

356
00:20:48,800 --> 00:20:52,500
out I knew immediately that oh I
was wrong.

357
00:20:53,100 --> 00:20:57,200
It's not automation, it's coming
for knowledge workers.

358
00:20:57,200 --> 00:21:02,000
First it coming for everything
so that sounds very Sinister but

359
00:21:02,000 --> 00:21:04,400
like because that's where the
data is not only is that with

360
00:21:04,400 --> 00:21:09,000
it, the data is, but the people
building these things, this is

361
00:21:09,000 --> 00:21:12,100
what they understand.
All right, they understand how

362
00:21:12,100 --> 00:21:15,400
to use computers, how to do
information jobs how How to

363
00:21:15,400 --> 00:21:19,400
program how to do design.
So not only they have the data

364
00:21:19,400 --> 00:21:22,100
sets but this is a set of
domains, they understand.

365
00:21:22,100 --> 00:21:24,800
So knowledge work will be
automated actually before

366
00:21:24,800 --> 00:21:28,100
robots.
So hilariously it, the plumbers

367
00:21:28,200 --> 00:21:31,600
and the mechanics and even the
factory workers of the world

368
00:21:31,600 --> 00:21:34,400
that will probably be gainfully
employed.

369
00:21:34,500 --> 00:21:38,200
You no longer than the you sort
of Silicon Valley types which

370
00:21:38,200 --> 00:21:43,600
isn't highly amusing.
But again, if that sounds sort

371
00:21:43,600 --> 00:21:46,800
of dark, I think the other ways
I would combine this.

372
00:21:46,800 --> 00:21:48,600
I can just tell you what I'm
doing.

373
00:21:49,200 --> 00:21:52,300
One of the products I'm most
excited about is called intent.

374
00:21:53,100 --> 00:21:56,700
So I'm trying to build a system
that literally watches your

375
00:21:56,700 --> 00:21:59,500
screen.
So this is a combination with

376
00:21:59,500 --> 00:22:02,700
other data streams.
So it's, it's using Vision

377
00:22:02,700 --> 00:22:06,600
that's to watch your screen and
it knows what you're doing all

378
00:22:06,600 --> 00:22:09,100
the time private.
So it runs on just your

379
00:22:09,100 --> 00:22:12,000
computer.
I just It's a stream of it

380
00:22:12,000 --> 00:22:12,900
answers.
The question.

381
00:22:13,100 --> 00:22:16,300
What did I do all day?
It literally just keeps a log

382
00:22:16,600 --> 00:22:20,100
because it can understand what
you're up to you right now.

383
00:22:20,100 --> 00:22:22,800
You're doing an interview and it
can log Moment by moment.

384
00:22:22,800 --> 00:22:26,000
So at face value it's just
analytics but the real goal is

385
00:22:26,000 --> 00:22:28,200
to do executive Control
software.

386
00:22:28,400 --> 00:22:32,800
So sort of you tell it what you
want to do and it sort of keeps

387
00:22:32,800 --> 00:22:35,200
you in line.
It's doing that.

388
00:22:35,200 --> 00:22:36,400
Why are you up at 2:00 a.m.
watching?

389
00:22:36,400 --> 00:22:39,000
Anime shouldn't be doing that.
You said you didn't want to.

390
00:22:39,400 --> 00:22:44,000
So it's sort of like out Forcing
your higher mind to keep your

391
00:22:44,000 --> 00:22:48,800
lower mines in check and but
ultimately, I want to combine

392
00:22:49,600 --> 00:22:53,400
information from other things.
So wearables is a big one.

393
00:22:54,000 --> 00:22:56,800
Anything that a I just wants
data.

394
00:22:56,800 --> 00:22:59,300
So anytime it's I have an aura
ring, right?

395
00:22:59,300 --> 00:23:02,600
It has all my sleep data as well
as continuous like Biometrics,

396
00:23:03,000 --> 00:23:04,000
wow.
That would be great to

397
00:23:04,000 --> 00:23:07,400
incorporate into the thing that
knows everything about my goals.

398
00:23:07,400 --> 00:23:10,500
And what I'm doing all the time
and these things just stack on

399
00:23:10,500 --> 00:23:13,500
top of Of each other.
So I would say anything that has

400
00:23:13,500 --> 00:23:17,500
a consistent high quality amount
of data or data and large

401
00:23:17,500 --> 00:23:21,900
volumes is going to be very
useful.

402
00:23:22,800 --> 00:23:26,400
If I had to bet on a dark horse,
interaction with AI that most

403
00:23:26,400 --> 00:23:28,300
people don't see coming.
I'd say it's probably

404
00:23:28,300 --> 00:23:35,000
brain-computer interfaces most
likely in terms of consumer EEG.

405
00:23:35,800 --> 00:23:39,100
Yes, I think meta and apple both
have EG coming out, incorporated

406
00:23:39,100 --> 00:23:41,800
into wearables.
Can do more than you would

407
00:23:41,800 --> 00:23:47,100
expect.
So that's probably plus

408
00:23:47,100 --> 00:23:49,800
information, brain, computer
interface information in.

409
00:23:50,800 --> 00:23:53,700
You can do that through sensory
augmentation will go into it,

410
00:23:54,200 --> 00:23:56,200
but turns out the brain is
pretty good at getting

411
00:23:56,200 --> 00:24:00,100
information into it.
If you train it, that one is

412
00:24:00,100 --> 00:24:02,900
also pretty interesting.
So I think people are going to

413
00:24:02,900 --> 00:24:06,300
start to chafe at how slow
information goes out and into

414
00:24:06,300 --> 00:24:10,700
their brains and they're going
to be interested in more or

415
00:24:10,700 --> 00:24:14,600
less. he's hooking up to these
systems all the time and these

416
00:24:14,600 --> 00:24:18,000
are the tools that enable that
You may have asked another

417
00:24:18,000 --> 00:24:20,000
question but those answers I got
for you.

418
00:24:20,500 --> 00:24:24,600
No, no, that's that's completely
fine and at it made sense

419
00:24:24,600 --> 00:24:27,900
because you know anything which
is as you said today, I think

420
00:24:27,900 --> 00:24:31,700
data actually these language
models lives on data, right?

421
00:24:31,700 --> 00:24:36,000
So they need to select data to
give us back what we want and I

422
00:24:36,000 --> 00:24:42,300
think maybe something later 22.
If we think consumer perspective

423
00:24:42,300 --> 00:24:46,800
sensors, you know anything that
comes from cameras and beasts.

424
00:24:47,500 --> 00:24:52,100
In industrial vertical I can
think about so about sensors and

425
00:24:52,100 --> 00:24:56,100
this kind of thing.
So yeah, now question that came

426
00:24:56,100 --> 00:24:58,100
to my mind, just out of
curiosity.

427
00:24:58,100 --> 00:25:00,600
Honestly.
Like, of course, I'm too much

428
00:25:00,600 --> 00:25:06,000
into the anything similar to
open a i and there are a couple

429
00:25:06,000 --> 00:25:09,800
of others, but when it comes to,
you know, something like

430
00:25:09,800 --> 00:25:16,800
statement of huge and like mine
Journey, like very Delhi to I

431
00:25:16,800 --> 00:25:22,100
want to understand person is so.
So consider me a guy that he's

432
00:25:22,100 --> 00:25:24,100
not from a technical background
now.

433
00:25:24,200 --> 00:25:29,200
So for me, what I see is that
this algorithm is or whatever

434
00:25:29,200 --> 00:25:31,800
you want to call them to
generate for you meet.

435
00:25:31,800 --> 00:25:32,400
Yes.
Right.

436
00:25:32,400 --> 00:25:40,000
So whether it's a photos or kind
of paintings but what are really

437
00:25:40,000 --> 00:25:45,500
the real uses or really word use
cases that we can get out of

438
00:25:45,800 --> 00:25:50,400
this.
Like Steve diffusion or minder.

439
00:25:52,200 --> 00:25:55,700
I think people would probably be
shocked about what is currently

440
00:25:55,700 --> 00:25:59,300
in development.
It takes a long time to build

441
00:25:59,300 --> 00:26:04,300
production systems for really
you get this Cambrian explosion

442
00:26:04,300 --> 00:26:08,200
of, like, people eat up the easy
use cases first just slapping

443
00:26:08,200 --> 00:26:10,800
user interface on chat GPT or
stable.

444
00:26:10,800 --> 00:26:13,100
Diffusion cool, there you go.
Now you can edit photos.

445
00:26:13,200 --> 00:26:16,800
That's great cool you make
people disappear in photos.

446
00:26:16,900 --> 00:26:19,100
All right?
Now what you can generate our

447
00:26:19,100 --> 00:26:23,500
who cares But you should really
consider this more.

448
00:26:23,500 --> 00:26:27,100
Like, by the way, there are
other really good generative

449
00:26:27,100 --> 00:26:31,700
networks Transformer Network are
competing against diffusion now

450
00:26:31,800 --> 00:26:33,900
and I think it's only a matter
of time until they end up

451
00:26:33,900 --> 00:26:35,800
working together.
They have different strengths.

452
00:26:35,800 --> 00:26:40,500
So, generative information is
really it's not slowing down

453
00:26:40,500 --> 00:26:42,400
either.
I think you should look at this

454
00:26:42,400 --> 00:26:45,300
not as generating media,
although it does do that.

455
00:26:45,400 --> 00:26:49,400
And that will have a profound
impact particularly in terms of

456
00:26:49,400 --> 00:26:53,800
meeting identity Solutions and
probably the death of the public

457
00:26:53,800 --> 00:26:57,600
Anonymous internet.
In a way, but that's okay.

458
00:26:58,200 --> 00:27:01,800
It will bring other good things
too, but it's more like

459
00:27:01,800 --> 00:27:05,600
generative information.
So, here's an example that I

460
00:27:05,600 --> 00:27:11,700
like to give you should think of
these things as being Really

461
00:27:11,700 --> 00:27:15,400
good at generating things to
about 80 or 90% Fidelity.

462
00:27:16,300 --> 00:27:18,600
They're not so good at
generating it to 100%.

463
00:27:19,200 --> 00:27:21,600
So let's say you're, I don't
know, you're an aircraft

464
00:27:21,600 --> 00:27:25,200
designer at Boeing, or you
design cars, you your job is

465
00:27:25,200 --> 00:27:29,300
going to look more like sitting
back crossing your arms and

466
00:27:29,300 --> 00:27:33,300
saying OK, I want something.
That is Sleek powerful uses this

467
00:27:33,300 --> 00:27:36,600
last year's model, show me forty
different combinations.

468
00:27:36,700 --> 00:27:38,400
Whoop, here they are.
That one.

469
00:27:39,000 --> 00:27:42,700
Okay, now given this One make
four different variations of

470
00:27:42,700 --> 00:27:46,500
this, run it through the, you
know, the stream or the the

471
00:27:46,500 --> 00:27:49,500
tests for wind resistance and
all this stuff, great.

472
00:27:49,600 --> 00:27:51,700
I'll come back after lunch.
Okay, here it is.

473
00:27:51,800 --> 00:27:54,000
Here's your model.
That would have taken you how

474
00:27:54,000 --> 00:27:57,200
long with clay models or even
just like going in with computer

475
00:27:57,200 --> 00:28:00,600
models?
Well unbelievable, the amount of

476
00:28:00,600 --> 00:28:08,200
testing an idea Yet the cost of
testing out ideas is going to

477
00:28:08,200 --> 00:28:12,500
zero and that's going to have a
very big impact on Innovation,

478
00:28:12,800 --> 00:28:15,600
which is not really well
captured right now or understood

479
00:28:16,100 --> 00:28:20,700
when it cost you a hundred or a
thousand or even ten dollars to

480
00:28:20,700 --> 00:28:25,700
do something and it takes you
10, 10 hours, 20 hours to test

481
00:28:25,700 --> 00:28:28,200
it.
You don't test that many things.

482
00:28:28,200 --> 00:28:29,900
You have to go with by
necessity.

483
00:28:29,900 --> 00:28:32,400
The things that look like
they're going to have Roi, but

484
00:28:32,400 --> 00:28:35,200
when the Cost of testing, a
thousand things is basically

485
00:28:35,200 --> 00:28:38,500
zero and you can test it
thousand things and a minute.

486
00:28:39,400 --> 00:28:43,000
Things change dramatically, you,
you can explore horizontally.

487
00:28:43,000 --> 00:28:45,300
A lot more than you could
before, you can test a really

488
00:28:45,300 --> 00:28:48,800
dumb ideas and sometimes those
dumb ideas turned out to be

489
00:28:48,800 --> 00:28:52,100
really interesting useful ideas
and that's going to happen

490
00:28:52,100 --> 00:28:53,800
everywhere.
So it's not just generating

491
00:28:53,800 --> 00:28:58,500
media, it's generating
information of any kind but it

492
00:28:58,500 --> 00:29:00,900
will be.
The reason we see media is

493
00:29:00,900 --> 00:29:04,200
because we have a public open
internet and everyone dumps all

494
00:29:04,200 --> 00:29:07,400
their data online.
So guess what they trained on is

495
00:29:07,400 --> 00:29:10,600
the stuff that there was a lot
of data for but there is data

496
00:29:10,600 --> 00:29:13,800
for other stuff.
It's a little harder to get, and

497
00:29:13,900 --> 00:29:16,700
it's a little less, lucrative up
front and a little harder to

498
00:29:16,700 --> 00:29:22,900
build, but it's coming.
So any information fundamentally

499
00:29:23,400 --> 00:29:26,200
whether it's science
experiments, you know,

500
00:29:26,300 --> 00:29:29,400
generating novel drug compounds.
All of this stuff is just

501
00:29:29,400 --> 00:29:34,000
information genetics.
So, if you want to test

502
00:29:34,000 --> 00:29:37,800
something out to just 80 or 90%
Fidelity really, really, really

503
00:29:37,800 --> 00:29:41,200
fast and then that last 10, 20
percent is still done with

504
00:29:41,200 --> 00:29:43,100
humans and maybe regular
computers.

505
00:29:44,100 --> 00:29:47,100
That's a big change, its a
couple orders of magnitude, I

506
00:29:47,100 --> 00:29:49,000
think in terms of like speed of
innovation.

507
00:29:49,000 --> 00:29:51,600
So that's that's a huge use case
for the generative Networks.

508
00:29:53,100 --> 00:29:58,600
That's you know like it it's
like very enlightening outside

509
00:29:58,700 --> 00:30:02,500
thinking about doing simulations
using these Technologies because

510
00:30:02,500 --> 00:30:06,400
they can generate different.
Let's say prototype for you but

511
00:30:06,900 --> 00:30:15,000
there are some people that they
are arguing that we might be too

512
00:30:15,000 --> 00:30:19,200
much dependent on the existing
knowledge that already these

513
00:30:19,600 --> 00:30:22,700
molded they have so we'll be
lazy to generate in.

514
00:30:22,800 --> 00:30:25,700
New content for the language
Maltese to use.

515
00:30:26,000 --> 00:30:30,100
What do you think about that?
I think.

516
00:30:31,800 --> 00:30:38,400
How much does it cost to get the
best aircraft modeler, the best

517
00:30:38,400 --> 00:30:41,700
car modeler to do something?
A lot of money.

518
00:30:42,000 --> 00:30:46,200
How much does it cost to get 80
to 90% of the best?

519
00:30:46,700 --> 00:30:50,400
Still, probably quite a bit.
So the difference is okay.

520
00:30:50,400 --> 00:30:53,900
Yes.
If even if we, let's say I'll

521
00:30:53,900 --> 00:30:58,700
buy that argument and we can
only get to even 80% of the

522
00:30:58,700 --> 00:31:01,400
average of any skill given has
enough training.

523
00:31:02,700 --> 00:31:04,500
Who cares, right?
All it's doing is repeating

524
00:31:04,500 --> 00:31:07,400
existing patterns.
Well, what matters is that I

525
00:31:07,400 --> 00:31:11,100
don't have to learn a skill.
I can have 0% car modeling

526
00:31:11,100 --> 00:31:13,000
knowledge and now I can model a
car.

527
00:31:13,400 --> 00:31:17,100
That's a very big deal just in
the Practical perspective like

528
00:31:17,100 --> 00:31:20,100
for my stories right?
I can do the cover art.

529
00:31:20,500 --> 00:31:22,600
I can't do cover art before
this.

530
00:31:22,600 --> 00:31:27,200
I spent $250 or $500 for like
incredible concept art and it

531
00:31:27,200 --> 00:31:31,700
took weeks of back and forth and
it was a huge pain and Half the

532
00:31:31,700 --> 00:31:33,000
time.
It doesn't end up looking right

533
00:31:33,000 --> 00:31:35,400
now.
I can just, I don't know, just

534
00:31:35,400 --> 00:31:37,600
throw stuff at the wall.
I'll just do it for four hours

535
00:31:37,600 --> 00:31:40,800
and I've generated a couple
hundred pictures and it cost me

536
00:31:40,808 --> 00:31:44,400
basically nothing.
You can imagine that for any

537
00:31:44,400 --> 00:31:49,100
Fields, like you don't have to
learn a skill to do it at 80%

538
00:31:49,100 --> 00:31:53,700
capacity of average like the
average professional, that's a

539
00:31:53,700 --> 00:31:57,500
very big deal and so you don't
even have to Advanced

540
00:31:57,500 --> 00:32:00,600
state-of-the-art in order for it
to matter significantly.

541
00:32:00,600 --> 00:32:02,900
It's about democracy.
Ization of skill set.

542
00:32:05,500 --> 00:32:08,500
Well, that that's another point
of view.

543
00:32:08,500 --> 00:32:11,800
I would say which is also valid
Stephen, can you 10?

544
00:32:11,800 --> 00:32:15,000
Because I know you're right,
sci-fi also as well, right?

545
00:32:15,000 --> 00:32:18,900
So have you tell us a little bit
more about this side?

546
00:32:19,900 --> 00:32:23,100
I'm sure that I have a lot of my
audience who would be interested

547
00:32:23,100 --> 00:32:29,100
to know more?
Yeah, it's just for fun but I do

548
00:32:29,100 --> 00:32:31,600
occasionally try to write.
Sometimes there's a combination

549
00:32:31,600 --> 00:32:33,800
of these.
So it's sci-fi and fantasy.

550
00:32:33,800 --> 00:32:36,800
So Fantasies mostly for fun.
I always try to have a lesson.

551
00:32:38,400 --> 00:32:42,500
So I've written a story recently
called reverie, that is a story

552
00:32:42,500 --> 00:32:46,900
about I think that Humanity
doesn't have a lot of great

553
00:32:46,900 --> 00:32:49,700
things to run towards like
Collective goals.

554
00:32:50,000 --> 00:32:53,200
I think we have a lot of things
we're trying to avoid but that's

555
00:32:53,200 --> 00:32:56,300
not super motivating who gets up
every day saying oh I hope I

556
00:32:56,300 --> 00:32:57,800
don't hope I'm not in pain
today.

557
00:32:58,200 --> 00:33:02,500
This has super motivating so I
tried to come up with.

558
00:33:02,500 --> 00:33:05,200
Here's where I think we should
try to go as a species.

559
00:33:06,700 --> 00:33:09,000
I'm writing a story.
Now called hell makers.

560
00:33:09,400 --> 00:33:13,100
This is about this.
Why I smiled when you mentioned

561
00:33:14,100 --> 00:33:18,200
I think you mentioned something
about Bots Autumn a pot.

562
00:33:18,900 --> 00:33:23,100
So already I've seen people talk
about, somebody has already had

563
00:33:23,100 --> 00:33:26,700
the bright idea, you know, would
be great, we should do, we need

564
00:33:26,700 --> 00:33:29,000
to open source these models
because you know, people

565
00:33:29,000 --> 00:33:30,800
probably don't want us to have
access to this.

566
00:33:30,800 --> 00:33:32,600
It's everything should be free
and open.

567
00:33:32,900 --> 00:33:35,700
You know what else we should do?
We should put this on

568
00:33:35,700 --> 00:33:37,500
decentralized Computing,
platforms?

569
00:33:37,500 --> 00:33:41,500
This is a good idea and I'm just
like this is the worst idea.

570
00:33:41,700 --> 00:33:43,300
Let me write a story to tell you
why.

571
00:33:43,700 --> 00:33:47,600
So oh it's a basically a story
about the founder of a

572
00:33:47,600 --> 00:33:50,700
decentralized digital autonomous
worker protocol.

573
00:33:51,700 --> 00:33:54,600
They just start like a
decentralized Computing

574
00:33:55,100 --> 00:33:59,400
blockchain protocol right?
The goal is that uses a

575
00:33:59,400 --> 00:34:02,400
consensus mechanism so you can't
turn it off easily unless you

576
00:34:02,408 --> 00:34:04,400
can turn off all the nodes on
the network.

577
00:34:05,600 --> 00:34:08,600
You can't turn it off and it's
permission list so anyone can

578
00:34:08,600 --> 00:34:11,800
insert money and you insert a
prompt pull a model off the

579
00:34:11,800 --> 00:34:15,199
shelf and you go tell it to do
something and Go do whatever

580
00:34:15,199 --> 00:34:18,699
until it runs out of money and
you can't turn it off.

581
00:34:19,199 --> 00:34:22,500
So I show this is probably not a
good idea.

582
00:34:22,900 --> 00:34:28,000
So, punch line of the story is
that the creator of the protocol

583
00:34:28,000 --> 00:34:31,400
ends up using this to get back
at their co-founders, who cut

584
00:34:31,400 --> 00:34:35,800
them out of a deal.
And as the name implies creates

585
00:34:35,800 --> 00:34:40,300
a bot that literally just
publicly tries to ruin this

586
00:34:40,300 --> 00:34:43,900
other person in perpetuity, and
cannot be turned off.

587
00:34:44,100 --> 00:34:45,699
Off.
So this is not good.

588
00:34:46,000 --> 00:34:49,100
We should think very carefully
about not having an off switch

589
00:34:49,100 --> 00:34:51,500
for our machines.
It just seemed like a very dumb

590
00:34:51,500 --> 00:34:56,199
idea to put them on a box that
you can't turn off correctly.

591
00:34:57,500 --> 00:35:04,300
Hey, hey question that I asked.
Now you became a legend question

592
00:35:04,300 --> 00:35:09,500
for me.
Like do you think that is

593
00:35:09,600 --> 00:35:14,700
allowing all of us Humanity to
read to reach singularity?

594
00:35:20,400 --> 00:35:24,600
Maybe.
Good question.

595
00:35:24,600 --> 00:35:29,600
I know sir the key is in what
you said at the end which is all

596
00:35:29,600 --> 00:35:32,100
of us and Humanity to reach the
singularity.

597
00:35:32,100 --> 00:35:35,800
Yes, I actually think that
positive outcomes are the most

598
00:35:35,800 --> 00:35:38,000
likely scenario in the long
term.

599
00:35:38,000 --> 00:35:41,100
And there's a lot we can do
about it and a lot of things

600
00:35:41,100 --> 00:35:43,500
that we could be doing about it.
That are actually quite feasible

601
00:35:43,700 --> 00:35:45,400
that are probably not well
known.

602
00:35:45,900 --> 00:35:50,500
So it's something I'm focused on
but yes, I think technically

603
00:35:50,500 --> 00:35:51,300
speaking.
Yes.

604
00:35:51,500 --> 00:35:55,400
This is the way you would need
scaled intelligence.

605
00:35:55,400 --> 00:35:59,600
And I think the most important
part of AI is moving us towards

606
00:35:59,600 --> 00:36:03,600
Singularity, is that maybe is
also under appreciated is that

607
00:36:03,600 --> 00:36:07,700
doesn't have cognitive biases.
Unlike our own brains, we can

608
00:36:07,700 --> 00:36:11,600
reprogram it and we can program
away the blind spots that we

609
00:36:11,600 --> 00:36:14,000
know we have.
It doesn't matter if you know

610
00:36:14,000 --> 00:36:19,300
about anchoring bias, or recency
bias, if you still did you

611
00:36:19,308 --> 00:36:21,100
experience them whether you want
them or not?

612
00:36:21,300 --> 00:36:25,000
You So, make dumb choices all
the, if you're tired and hungry,

613
00:36:25,300 --> 00:36:28,000
you're gonna make bad choices.
Like this is just how the human

614
00:36:28,000 --> 00:36:31,500
brain works.
You can't program around it, but

615
00:36:31,500 --> 00:36:33,100
that's not the same for our
machines.

616
00:36:33,100 --> 00:36:35,900
Which means that we should be
able to program them to our

617
00:36:35,900 --> 00:36:38,000
ideals.
And as long as we can keep

618
00:36:38,000 --> 00:36:41,300
advancing them, I don't see why
you know you can't have

619
00:36:41,300 --> 00:36:43,600
wonderful incredible outcomes
from that.

620
00:36:43,600 --> 00:36:46,600
So ensure yes but I think it's
neutral.

621
00:36:46,800 --> 00:36:51,000
It depends on what we do.
That's that's fair enough.

622
00:36:51,000 --> 00:36:55,500
I would say before I will you
know keep it for you to do a

623
00:36:55,500 --> 00:36:58,700
final conclusion.
Not related to a I related

624
00:36:58,700 --> 00:37:00,700
because you are a solopreneur,
right?

625
00:37:01,000 --> 00:37:06,400
So How?
How do you see it?

626
00:37:06,500 --> 00:37:08,400
Like, what's your experience
being a solopreneur?

627
00:37:08,400 --> 00:37:10,900
Because again, this is this is
the second.

628
00:37:11,500 --> 00:37:13,900
Common question, I'm asking to
my guest, if there are

629
00:37:13,900 --> 00:37:18,400
solopreneurs like why you opted
to be a solopreneur like why

630
00:37:18,400 --> 00:37:22,000
you're mad part of I know that
you work for some companies

631
00:37:22,000 --> 00:37:23,800
before and I did by the way as
well.

632
00:37:23,800 --> 00:37:29,000
I'm a solopreneur I'm off now.
So tell me your experience as a

633
00:37:29,000 --> 00:37:32,900
solopreneur and why we are
seeing more solopreneurs in the

634
00:37:32,900 --> 00:37:38,300
field.
The answer, the last one first

635
00:37:38,300 --> 00:37:41,700
because it's fairly simple.
I think we're seeing more people

636
00:37:41,700 --> 00:37:44,200
are more isolated in general but
I think we're also more

637
00:37:44,200 --> 00:37:46,100
empowered by technology in
general.

638
00:37:46,200 --> 00:37:48,600
Like I mentioned like I can do a
lot of things.

639
00:37:48,600 --> 00:37:52,800
I don't know how to do thanks to
advances in Ai and other

640
00:37:52,800 --> 00:37:56,600
software, right?
I don't know how to compute

641
00:37:56,600 --> 00:37:58,800
compound interest.
I mean, I could do it by hand if

642
00:37:58,800 --> 00:38:01,600
you gave me the formula and a
lot of time but spreadsheet can

643
00:38:01,600 --> 00:38:04,900
do it.
So I'm an able to do a lot by

644
00:38:04,900 --> 00:38:08,600
myself and because I'm working
in software and AI which is like

645
00:38:08,600 --> 00:38:12,800
software for software.
It's, I can I have a lot of

646
00:38:12,800 --> 00:38:19,400
Leverage so I am enabled to do
it but also at the same time,

647
00:38:20,000 --> 00:38:22,900
it's often easier just to pay
contractors and I find that

648
00:38:22,900 --> 00:38:26,200
there is a sort of cultural
Zeitgeist Movement towards

649
00:38:28,100 --> 00:38:31,200
people who want their own
autonomy and they want to

650
00:38:31,208 --> 00:38:34,700
interface with other people as
people and they don't want to be

651
00:38:34,700 --> 00:38:37,900
an employee.
So I found it very easy to just,

652
00:38:37,900 --> 00:38:40,500
hey, what's your rate like,
let's figure out a deal.

653
00:38:40,500 --> 00:38:43,300
It's figure out a partnership
and that has been the easier way

654
00:38:43,300 --> 00:38:45,800
to go.
So that's the answer to the

655
00:38:45,800 --> 00:38:47,700
second part.
So the first part, my experience

656
00:38:47,700 --> 00:38:53,400
is a solopreneur I think that
the challenge especially with a

657
00:38:53,400 --> 00:38:56,300
I right, so selling software
development is fairly

658
00:38:56,300 --> 00:39:00,500
straightforward because people
understand, you didn't how to

659
00:39:00,800 --> 00:39:04,200
how to Value it.
The biggest challenge of getting

660
00:39:04,200 --> 00:39:09,300
paid build stuff with AI is how
to help people understand the

661
00:39:09,300 --> 00:39:13,300
opportunities that they have and
how to help them value it.

662
00:39:13,800 --> 00:39:16,700
It's there's a lot of work that
has to be done.

663
00:39:16,800 --> 00:39:18,900
You don't have to do a lot of
work to tell people.

664
00:39:19,000 --> 00:39:22,000
All about regular Software
System, they already know how to

665
00:39:22,000 --> 00:39:24,200
reason about it.
There's a lot of competitors,

666
00:39:24,200 --> 00:39:29,300
they know what to compare it to.
It's very simple but for for AI,

667
00:39:29,300 --> 00:39:32,200
there's a long conversation of
Education.

668
00:39:32,500 --> 00:39:33,900
Okay.
Here's what is possible.

669
00:39:34,000 --> 00:39:35,900
Okay.
Now I need to fully understand

670
00:39:35,900 --> 00:39:38,700
your business so that I can
analyze.

671
00:39:39,200 --> 00:39:43,000
Okay, where's the opportunity?
And then from there I have to

672
00:39:43,008 --> 00:39:45,600
help you understand what the
value is that we can justify the

673
00:39:45,600 --> 00:39:48,400
cost, which is higher than
regular software, a lot of

674
00:39:48,408 --> 00:39:49,100
steps.
Soooo.

675
00:39:49,500 --> 00:39:51,600
So yeah, it's a very high trust
thing.

676
00:39:52,600 --> 00:39:56,600
Also, I noticed that Most
businesses.

677
00:39:56,900 --> 00:39:59,400
So, again, this is just money
from Consulting, not building my

678
00:39:59,400 --> 00:40:03,400
own products, most businesses
would benefit, they're not even

679
00:40:03,400 --> 00:40:06,200
fully utilizing software, then
it's definitely not even fully

680
00:40:06,200 --> 00:40:11,500
using non software Solutions.
And so AI is often, like, why

681
00:40:11,500 --> 00:40:14,400
would you bring this really
complex solution and because

682
00:40:14,400 --> 00:40:17,000
it's sexy.
It's got Buzz words, but I've

683
00:40:17,000 --> 00:40:22,700
found so far for the most part
price optimization retention

684
00:40:22,700 --> 00:40:26,000
prediction.
Collaborative filtering.

685
00:40:26,000 --> 00:40:31,800
Just even just regression like
linear or logistic regression to

686
00:40:31,800 --> 00:40:35,700
statistical models.
These are often like 80/20, so

687
00:40:35,700 --> 00:40:37,700
most people don't need the big
guns.

688
00:40:38,800 --> 00:40:43,300
That is the cool stuff for sure.
I'd say generative AI, the large

689
00:40:43,300 --> 00:40:46,800
language models changes things a
little bit, but by and large,

690
00:40:47,100 --> 00:40:49,600
you don't need AI for most
solutions.

691
00:40:50,100 --> 00:40:53,800
So a lot of the challenge is
finding the people who really do

692
00:40:53,800 --> 00:40:58,400
have a slam-dunk use case and
building the trust telling the

693
00:40:58,400 --> 00:41:02,100
story.
So yeah that's that's what it's

694
00:41:02,100 --> 00:41:05,400
like is that there's a lot of
legwork that's sort of the AI

695
00:41:05,400 --> 00:41:09,500
side.
Damn, that's fair enough.

696
00:41:09,500 --> 00:41:12,500
I would say and I have to agree
with a lot of the points that

697
00:41:12,500 --> 00:41:15,600
you mentioned Steven.
Steven any final thing.

698
00:41:15,600 --> 00:41:18,700
You maybe something.
I didn't ask you, you want to

699
00:41:18,800 --> 00:41:27,400
say or shared before we close.
Well, my producer and the social

700
00:41:27,400 --> 00:41:30,800
media growth manager would
probably hit me over the head if

701
00:41:30,800 --> 00:41:33,100
I didn't say go to the YouTube
channel.

702
00:41:33,600 --> 00:41:36,600
Yes, he can Minima.
Yes.

703
00:41:37,800 --> 00:41:41,700
But so I mean do that if you
want to hear more words like

704
00:41:41,700 --> 00:41:45,600
this actually I wanted to ask
where people can find more about

705
00:41:45,600 --> 00:41:50,600
you.
I get better at this guy's butt.

706
00:41:51,000 --> 00:41:56,200
So what I wanted to say is I
want to the whole purpose of

707
00:41:56,200 --> 00:41:59,000
seeking minnemann generals isn't
not-for-profit thing.

708
00:41:59,000 --> 00:42:02,100
Like I can just build software,
like there's no point according

709
00:42:02,100 --> 00:42:05,700
videos or talking to people like
you for for a living.

710
00:42:05,700 --> 00:42:10,200
That's not what I want to do.
I am trying to help people use

711
00:42:10,200 --> 00:42:14,600
technology and to create better
outcomes for everybody.

712
00:42:15,100 --> 00:42:19,400
And I think that we are at a
fairly pivotable pivotal moment

713
00:42:19,400 --> 00:42:24,900
and I see a lot of, despite some
of the Pessimistic take so far.

714
00:42:25,200 --> 00:42:27,200
That's not really how I look at
the world.

715
00:42:27,400 --> 00:42:30,800
I think that we have a lot of
opportunity to make a big

716
00:42:30,800 --> 00:42:34,900
difference right now, and I want
to help people realize that they

717
00:42:34,900 --> 00:42:37,300
don't need to stand here and
look scared.

718
00:42:38,100 --> 00:42:43,400
There's a lot we can do in terms
of AI alignment, wealth

719
00:42:43,400 --> 00:42:45,700
inequality.
We have a lot of tools at our

720
00:42:45,700 --> 00:42:49,900
disposal, and we're really sweet
spot where there's most of the

721
00:42:49,900 --> 00:42:53,800
stuff is Greenfield brand-new.
We have a lot of Ready to use it

722
00:42:53,800 --> 00:42:57,400
well to put Common Sense rules
in place and to really set it up

723
00:42:57,400 --> 00:43:01,100
to work for us instead of
against us or for a select few

724
00:43:01,100 --> 00:43:05,100
people few people and I think
that I'm going to continue

725
00:43:05,100 --> 00:43:08,200
exploring those things.
I have a lot of concrete ideas

726
00:43:08,200 --> 00:43:10,900
but also just like to synthesize
other people's ideas.

727
00:43:11,500 --> 00:43:16,100
There's no need to be terrified
at the future because we have a

728
00:43:16,108 --> 00:43:19,000
lot of control over it.
We're the ones who control it.

729
00:43:19,000 --> 00:43:21,200
I don't know if people seem to
forget that as a, by the way,

730
00:43:21,600 --> 00:43:24,800
the future?
What Your happens is based on

731
00:43:24,800 --> 00:43:28,700
what we do, so if you don't like
it, you should do something

732
00:43:28,700 --> 00:43:30,700
about.
And that's that's what I'm

733
00:43:30,700 --> 00:43:32,100
doing.
If you're interested in that

734
00:43:32,100 --> 00:43:35,700
too, then, yeah, I'd love to
keep talking with you.

735
00:43:36,600 --> 00:43:40,400
That's the really great and I
think we are only same Mission.

736
00:43:40,400 --> 00:43:44,300
I would say Steven because one
of the reasons I'm doing all,

737
00:43:44,300 --> 00:43:49,400
you know, this podcast, I'm
creating a lot of content is to

738
00:43:49,400 --> 00:43:53,200
raise awareness.
This is until people You know

739
00:43:53,200 --> 00:43:57,400
like, hey, you know, it's not
like only the scary stuff you

740
00:43:57,400 --> 00:44:01,100
see in the media, you know, like
people gonna lose their jobs.

741
00:44:01,100 --> 00:44:03,400
I don't know what, you know, all
this negativity.

742
00:44:03,400 --> 00:44:08,500
So I'm trying to put some light
on the postage side of the

743
00:44:08,500 --> 00:44:13,200
technology and not just AI, by
the way any technology because I

744
00:44:13,200 --> 00:44:17,000
love technology myself, like,
since I was a child and I always

745
00:44:17,000 --> 00:44:23,000
believed that Technology's main
goal is to take us forward.

746
00:44:23,100 --> 00:44:26,900
Our life's easy from business
perspective, it's always I'm

747
00:44:26,900 --> 00:44:30,200
repeating myself.
Multiple episodes, increase

748
00:44:30,200 --> 00:44:35,300
revenues increased customer
base, decrease churn, all these

749
00:44:35,300 --> 00:44:40,500
nice things that business people
like to hear and AI is no

750
00:44:40,500 --> 00:44:43,800
different than this way.
I will make you know your

751
00:44:43,800 --> 00:44:47,900
business Thrive and better and
for individuals and saying

752
00:44:47,900 --> 00:44:51,100
exactly what you mentioned at
the beginning also as well guys.

753
00:44:51,100 --> 00:44:54,600
Like if you are scared of
Something go study like this is

754
00:44:54,600 --> 00:44:57,000
the best thing you can do
because if you just keep

755
00:44:57,000 --> 00:45:01,000
watching it and do nothing, yeah
you're going to be scared.

756
00:45:01,000 --> 00:45:02,400
I'd sorry.
I cannot help this.

757
00:45:02,400 --> 00:45:04,400
Why?
I'm trying to help you in

758
00:45:04,400 --> 00:45:07,300
generating this contact as much
as I can, of course.

759
00:45:08,600 --> 00:45:12,000
So yeah, this is why we are
aligned at the structure why I

760
00:45:12,000 --> 00:45:15,000
wanted to interview you Stephen
because when I read your bio,

761
00:45:15,400 --> 00:45:19,000
this What attracted me?
And by the way, this is I do

762
00:45:19,000 --> 00:45:21,400
with all my guests.
I would share, you know, the

763
00:45:21,400 --> 00:45:22,900
YouTube channel like anything
else.

764
00:45:23,100 --> 00:45:26,000
I can find if you want to share
that, with me also as well, so

765
00:45:26,000 --> 00:45:29,200
that would go into the
description of the episode and

766
00:45:29,500 --> 00:45:33,700
the description on the YouTube
also, as well, when we came to

767
00:45:33,700 --> 00:45:37,100
an end, Steven thank you very
much for being my guest today.

768
00:45:37,100 --> 00:45:40,200
I really appreciate the time,
and I sure hope.

769
00:45:40,700 --> 00:45:43,500
Thank you very much and for my
accordions, whether you are

770
00:45:43,500 --> 00:45:46,900
watching this on YouTube or you
are listening, using your

771
00:45:47,900 --> 00:45:52,900
favorite food casting platform,
don't forget to subscribe and as

772
00:45:53,000 --> 00:45:56,800
As usual, I'm telling you guys,
I would love to hear your

773
00:45:56,800 --> 00:46:00,100
feedback about this episode or
about the show in general.

774
00:46:00,400 --> 00:46:03,100
If you are interested to be a
guest like Stephen today, now

775
00:46:03,400 --> 00:46:07,600
don't be shy, please come up
with the idea that you want to

776
00:46:07,608 --> 00:46:10,600
discuss and I will be more than
happy to have you as a guest

777
00:46:10,600 --> 00:46:14,300
with me and as usual until we
meet next time.

778
00:46:14,300 --> 00:46:16,300
Thank you very much and we'll
see you soon.