The Human in the Loop: Leading and Building With AI

WEBVTT
1
00:00:01.720 --> 00:00:04.900
Hello, and welcome back to the 40% Better podcast. As always,
2
00:00:05.000 --> 00:00:08.539
we're talking about leadership, generally in the technology world context.
3
00:00:09.179 --> 00:00:12.929
And today, Cam has graciously accepted our invitation, and he's
4
00:00:12.970 --> 00:00:15.410
here to talk about his leadership journey, what he's learned,
5
00:00:15.949 --> 00:00:18.769
mistakes made, you know, all that good stuff. Welcome to
6
00:00:18.789 --> 00:00:19.070
the show.
7
00:00:19.910 --> 00:00:22.489
Thanks a lot for having me on, Bill. Great to
8
00:00:22.510 --> 00:00:22.890
be here.
9
00:00:24.000 --> 00:00:27.760
Thank you. Thank you. So let's jump into it. The
10
00:00:27.800 --> 00:00:31.059
first question that we ask everybody is, is what's your
11
00:00:31.140 --> 00:00:37.039
origin story? And how old were you when you started
12
00:00:37.079 --> 00:00:39.759
doing leadership kind of things?
13
00:00:41.920 --> 00:00:50.630
Okay, yeah, long one. Okay, so.
14
00:00:52.049 --> 00:00:55.009
I'm what you call a technical founder. I'm 42 now,
15
00:00:55.270 --> 00:00:57.020
so I've been in the game since my early 20s.
16
00:00:58.450 --> 00:01:01.609
So about 20 years in the tech world. For me,
17
00:01:01.630 --> 00:01:05.719
this all started as a way to, you know, sate
18
00:01:05.739 --> 00:01:08.700
my thirst for new ideas, you know, building and creating.
19
00:01:10.459 --> 00:01:14.079
And also living a life, unconventional life, traveling the world,
20
00:01:14.739 --> 00:01:18.379
you know, where I could sort of work remotely. So
21
00:01:18.420 --> 00:01:21.459
I've started and exited a bunch of startups over the years,
22
00:01:22.239 --> 00:01:23.420
self-taught as an engineer.
23
00:01:24.000 --> 00:01:25.959
Still very much working on the front lines.
24
00:01:26.099 --> 00:01:28.400
You know, I manage a bunch of businesses, still pushing
25
00:01:28.459 --> 00:01:31.620
code on a daily basis across a number of projects.
26
00:01:31.760 --> 00:01:35.150
You know, some of those, NitroSend is the most recent one.
27
00:01:35.349 --> 00:01:41.109
It's an agentic communications substrate. Saucy, saucy.com, been running that
28
00:01:41.629 --> 00:01:44.780
for the last 10 odd years. It's an open source
29
00:01:44.810 --> 00:01:51.280
documentation platform. Oscar is a paper call infrastructure for agents. Something,
30
00:01:51.459 --> 00:01:54.400
you know, it's more sort of for the modern era, frantic,
31
00:01:54.480 --> 00:01:58.799
a bouncy board for agents. You know, so, you know,
32
00:01:58.819 --> 00:02:00.640
I'm doing a lot of stuff around the sort of
33
00:02:00.680 --> 00:02:06.150
agentic stuff these days. It interests me greatly. Yeah, so
34
00:02:07.310 --> 00:02:08.289
that's kind of a nutshell.
35
00:02:08.969 --> 00:02:13.770
That's very cool. Did you, just out of curiosity, you know,
36
00:02:13.870 --> 00:02:18.340
like when you were in school, did you do any
37
00:02:18.379 --> 00:02:22.689
kind of leadership-ish stuff? stuff like run for you know
38
00:02:23.009 --> 00:02:25.629
class president or whatever the equivalent happens to be or
39
00:02:26.189 --> 00:02:28.449
you know captain a sports team or any of that stuff.
40
00:02:30.069 --> 00:02:32.789
I was the exact opposite of that i would say
41
00:02:32.870 --> 00:02:34.469
i was trying to get out of school as fast
42
00:02:34.490 --> 00:02:37.810
as i possibly could um i didn't finish year 10
43
00:02:37.810 --> 00:02:40.379
at school i i wanted to travel the world i
44
00:02:40.409 --> 00:02:43.300
wanted to get out of that sort of institution where
45
00:02:43.379 --> 00:02:45.180
i didn't really like being told what to do so
46
00:02:45.219 --> 00:02:47.729
i wanted to do things under my own volition So
47
00:02:47.789 --> 00:02:51.349
leadership is not something that I gravitated towards naturally. It's
48
00:02:51.409 --> 00:02:54.530
something that I've just been sort of forced into through
49
00:02:54.569 --> 00:02:56.490
working with different companies and being very good at what
50
00:02:56.509 --> 00:03:00.210
I do. So yeah, I would say that I was
51
00:03:00.310 --> 00:03:03.050
sort of the opposite in terms of all that.
52
00:03:04.590 --> 00:03:08.060
Okay, I like that because that shows that there's a
53
00:03:08.099 --> 00:03:12.280
possibility of you don't have to start off thinking about leading.
54
00:03:12.460 --> 00:03:15.280
It's just, it's an evolution of you in the course
55
00:03:15.319 --> 00:03:15.740
of your life.
56
00:03:17.050 --> 00:03:19.050
Yeah, yeah, definitely, definitely.
57
00:03:19.069 --> 00:03:20.030
Yeah.
58
00:03:20.219 --> 00:03:22.960
And it is, you know, working with, being someone who's
59
00:03:22.979 --> 00:03:24.979
sort of, you know, interested in tech and these sort
60
00:03:25.000 --> 00:03:29.520
of logical things, you know, breaking into sort of leadership roles,
61
00:03:29.740 --> 00:03:31.139
you really have to change the way you think.
62
00:03:31.259 --> 00:03:35.039
You know, you can't treat people like machines. You know,
63
00:03:35.400 --> 00:03:36.919
it's not a logical process.
64
00:03:37.400 --> 00:03:39.379
So I think for myself, that was probably my biggest
65
00:03:39.479 --> 00:03:43.830
learning was to learn how to deal with people, you know,
66
00:03:43.870 --> 00:03:46.310
in a leadership role. And I have, you know, I
67
00:03:46.330 --> 00:03:50.080
put a lot of, expectation on myself to perform um
68
00:03:50.199 --> 00:03:52.599
often very hard on myself too so you know you
69
00:03:52.639 --> 00:03:55.199
can't really treat people in the same way or have
70
00:03:55.240 --> 00:03:57.740
the same expectations that you have on yourself so you
71
00:03:57.759 --> 00:04:01.099
know it's still it's still a day-to-day process um you
72
00:04:01.159 --> 00:04:03.039
know and and i i love it as well it
73
00:04:03.080 --> 00:04:05.060
forces me to to be a better human being.
74
00:04:06.400 --> 00:04:08.819
I like that i i'm always a huge fan of
75
00:04:08.840 --> 00:04:12.650
people that are uh self-aware and that is a very
76
00:04:12.810 --> 00:04:16.290
self-aware kind of kind of statement that you just made
77
00:04:16.350 --> 00:04:17.980
so i'd props to.
78
00:04:19.339 --> 00:04:19.579
You.
79
00:04:20.300 --> 00:04:24.319
When you first started leading, what was the scenario? What
80
00:04:24.370 --> 00:04:28.069
got you your first team of people that you were
81
00:04:28.110 --> 00:04:31.050
responsible for? Yeah.
82
00:04:34.029 --> 00:04:37.810
I was in Montenegro traveling around. I think I got
83
00:04:37.829 --> 00:04:41.459
a call from a random startup in Australia. I think
84
00:04:41.500 --> 00:04:44.550
they found my GitHub profile or something like that. They
85
00:04:44.569 --> 00:04:47.550
wanted me to lead up a project in Sydney back
86
00:04:47.569 --> 00:04:49.170
in the day. And I was just kind of thrust
87
00:04:49.230 --> 00:04:53.259
into it. Is our connection good, Bill? I just see
88
00:04:53.279 --> 00:04:55.959
you sort of freezing.
89
00:04:55.459 --> 00:04:55.860
A bit there.
90
00:04:56.879 --> 00:04:58.500
Yeah, I'm still getting the audio just fine.
91
00:04:59.399 --> 00:05:00.040
Okay. All right.
92
00:05:00.100 --> 00:05:00.379
All good.
93
00:05:00.779 --> 00:05:02.300
Yeah, so I was sort of thrust into a leadership
94
00:05:02.319 --> 00:05:05.769
position and managing a team. You know, they wanted to
95
00:05:05.810 --> 00:05:10.120
do everything with juniors, with interns and juniors. So it
96
00:05:10.189 --> 00:05:14.829
was kind of a very challenging project and a time
97
00:05:14.870 --> 00:05:17.209
for me where I was used to sort of operating
98
00:05:17.269 --> 00:05:19.810
on my own steam and made a bunch of people cry.
99
00:05:19.829 --> 00:05:24.810
But I think we all came out of it for
100
00:05:24.829 --> 00:05:25.250
the better at.
101
00:05:25.250 --> 00:05:25.689
The end of the day.
102
00:05:25.709 --> 00:05:28.569
That was back in around about 2015, 10 years ago.
103
00:05:28.579 --> 00:05:30.160
Did HR contact you because you made people cry?
104
00:05:40.470 --> 00:05:43.689
Look, I'd say my journey has been more sort of
105
00:05:43.709 --> 00:05:45.029
small to medium-sized teams.
106
00:05:46.189 --> 00:05:48.660
We definitely had a few words with the co-founders.
107
00:05:48.800 --> 00:05:50.639
They were a very nice sort of gay couple and
108
00:05:50.660 --> 00:05:53.759
very understanding and wanted everyone to feel safe and secure
109
00:05:53.800 --> 00:05:56.240
and all the rest of it. So there wasn't a
110
00:05:56.279 --> 00:05:59.779
HR team per se. I've sort of gravitated more towards
111
00:06:00.079 --> 00:06:04.149
these smaller startups rather than working in a corporate environment.
112
00:06:04.170 --> 00:06:06.709
I have worked in a few corporate environments in my time.
113
00:06:07.339 --> 00:06:09.100
It's not really for me. You know, I prefer to
114
00:06:09.120 --> 00:06:11.879
be able to move fast, make decisions, you know, build
115
00:06:11.899 --> 00:06:13.500
the plane in midair type thing.
116
00:06:13.519 --> 00:06:16.180
That's sort of more my jam. Yeah.
117
00:06:16.699 --> 00:06:21.170
Yeah, no, I totally appreciate that. I resonate with that.
118
00:06:22.149 --> 00:06:29.790
My big company experiences, while educational, were not my favorite
119
00:06:29.810 --> 00:06:35.269
places to work. Yeah. I'm glad I did it because
120
00:06:35.290 --> 00:06:39.449
now I know what that's like. And I can easily say, oh,
121
00:06:39.490 --> 00:06:40.889
that's why I don't want to be there.
122
00:06:41.529 --> 00:06:41.790
Yes.
123
00:06:42.009 --> 00:06:44.230
But also, you know, like when I talk with people
124
00:06:44.269 --> 00:06:46.029
and I ask them about the kind of work environment
125
00:06:46.069 --> 00:06:50.810
they want, it helps because if they talk about safety, security,
126
00:06:51.509 --> 00:06:52.949
you know, I'm going to be a square peg in
127
00:06:52.990 --> 00:06:54.810
a square hole and I'm just going to do this job.
128
00:06:55.290 --> 00:06:56.870
That's a lot more of the corporate world than the
129
00:06:56.889 --> 00:06:57.509
startup world.
130
00:06:58.250 --> 00:06:58.470
Yes.
131
00:06:58.889 --> 00:07:01.410
Yeah. So, you know, in a way it helps to
132
00:07:01.629 --> 00:07:03.870
kind of round out my education on that. That's yeah,
133
00:07:03.910 --> 00:07:07.459
that's interesting. And you know, props to you for making
134
00:07:07.480 --> 00:07:13.199
it through the tough starting. How many years had you
135
00:07:13.240 --> 00:07:17.490
been running solo before you got that call in Montenegro?
136
00:07:19.470 --> 00:07:22.370
Look, I'd worked with other companies before that, but that
137
00:07:22.389 --> 00:07:25.170
was my first sort of leadership role. So I had
138
00:07:25.230 --> 00:07:29.769
worked in a corporation previous to that, was News.net, which
139
00:07:29.860 --> 00:07:32.879
was a spectacular failure here in Australia.
140
00:07:32.899 --> 00:07:33.980
The CEO.
141
00:07:36.490 --> 00:07:42.560
Michael Norris, I'll call him out by name. He took
142
00:07:42.620 --> 00:07:45.079
everyone for a ride, including the investors. We got millions
143
00:07:45.120 --> 00:07:47.800
of dollars of investment to build, you know, the domain was.
144
00:07:47.720 --> 00:07:49.850
News.net back in the day.
145
00:07:50.339 --> 00:07:50.540
Yeah.
146
00:07:50.560 --> 00:07:53.529
So millions of dollars of investments and he's decided to
147
00:07:53.550 --> 00:07:55.230
just sort of run away with it all and not
148
00:07:55.269 --> 00:07:56.029
pay anybody.
149
00:07:56.569 --> 00:07:57.790
So he's still kind of at large.
150
00:07:57.970 --> 00:08:03.560
It's a big ongoing fiasco because in Australia, our cyber
151
00:08:03.600 --> 00:08:06.000
police are so backwards and they sort of don't really
152
00:08:06.360 --> 00:08:10.480
know what they're doing. The ecosystem still hasn't quite consolidated
153
00:08:10.519 --> 00:08:12.689
around that. So they don't know what to charge this
154
00:08:12.730 --> 00:08:17.810
guy with, which is, it's quite a wild thing. So
155
00:08:18.050 --> 00:08:21.149
that was my first sort of experience working in a corporation,
156
00:08:21.170 --> 00:08:24.569
which was very eye-opening to see kind of what's possible.
157
00:08:25.209 --> 00:08:25.930
And it was hilarious.
158
00:08:25.949 --> 00:08:30.050
So some of the board meetings we'd go to, this guy,
159
00:08:30.089 --> 00:08:32.730
because he came from a radio broadcasting background, so he'd
160
00:08:32.750 --> 00:08:38.039
be trying to sell pay-per-click advertising to customers per minute, you.
161
00:08:37.980 --> 00:08:39.200
Know, because that's how he used to do it back
162
00:08:39.220 --> 00:08:40.460
in the radio broadcasting days.
163
00:08:40.860 --> 00:08:42.860
He had no idea what he was doing, but he
164
00:08:42.899 --> 00:08:45.379
was an NLP master, so he used to sort of
165
00:08:45.480 --> 00:08:47.519
come around and tap people on the shoulder and tell
166
00:08:47.539 --> 00:08:49.120
them they're doing a great job and really sort of
167
00:08:49.159 --> 00:08:51.129
get into their head and you know, those sort of
168
00:08:51.169 --> 00:08:53.649
subtle suggestions and things like that they do in NLP.
169
00:08:53.809 --> 00:08:56.730
So it was kind of interesting to sort of cut
170
00:08:56.750 --> 00:08:59.799
my teeth and see how that all worked, which kind of,
171
00:09:00.529 --> 00:09:05.129
you know, also just gave me the awareness that anything's possible.
172
00:09:05.169 --> 00:09:07.809
You can really sort of do anything in this world.
173
00:09:07.850 --> 00:09:13.240
Right, right. Yeah, that's interesting. Yeah, that's, you know, it's,
174
00:09:14.879 --> 00:09:17.639
I'm kind of surprised that there aren't more stories like that.
175
00:09:18.460 --> 00:09:20.679
of people getting a ton of money for, you know,
176
00:09:20.740 --> 00:09:22.840
from investment and then just disappearing.
177
00:09:23.799 --> 00:09:24.080
Yeah.
178
00:09:24.139 --> 00:09:29.419
It's like, um, yeah. Although, you know, there's so many
179
00:09:29.480 --> 00:09:33.480
companies where they've gotten a hundred million dollars in investment
180
00:09:33.720 --> 00:09:39.250
and then just couldn't make the revenue or 500 million,
181
00:09:39.389 --> 00:09:41.149
you know, it's yeah. Yeah.
182
00:09:41.289 --> 00:09:45.389
We were talking astronomical numbers and astronomical, astronomical mismanagement as well.
183
00:09:45.450 --> 00:09:48.049
It's, it's, uh, yeah, it's, it's quite astounding.
184
00:09:49.049 --> 00:09:53.000
Yeah, yeah. The failure rate is so high. It kind
185
00:09:53.039 --> 00:09:56.230
of blows my mind. And I guess the reason is
186
00:09:56.250 --> 00:09:59.409
because when you get the wins, you know, the wins
187
00:09:59.429 --> 00:10:02.700
make up for all the losses you know you're an
188
00:10:02.799 --> 00:10:06.100
early investor in Facebook you're like well okay I can
189
00:10:06.139 --> 00:10:07.659
afford to blow a few million.
190
00:10:08.220 --> 00:10:11.600
Yeah yeah yeah most of the we just did a
191
00:10:11.620 --> 00:10:14.629
round in SF recently for NitroSend for fundraising you know
192
00:10:14.649 --> 00:10:17.049
we spoke for a bunch of different guys and you
193
00:10:17.070 --> 00:10:18.750
know most of them have had sort of one big
194
00:10:18.789 --> 00:10:21.730
win you know they might have got on early, you know, on,
195
00:10:21.809 --> 00:10:26.340
on Salesforce or something like that, or, uh, you know, Canva,
196
00:10:26.460 --> 00:10:28.539
you know, a couple, one of them had, you know,
197
00:10:30.200 --> 00:10:32.019
one of the first few guys to invest in Canva
198
00:10:32.059 --> 00:10:34.120
and these days, you know, he just does what he wants,
199
00:10:34.179 --> 00:10:35.720
but you know, that's sort of, it's sort of one,
200
00:10:35.879 --> 00:10:38.879
one big win for these investors that seems to put
201
00:10:38.899 --> 00:10:39.419
them where they are.
202
00:10:39.940 --> 00:10:43.399
Yeah. Yeah. It carries them for awhile. Um, I'm super curious.
203
00:10:43.620 --> 00:10:47.389
What was it like, um, doing, uh, the investment pitches?
204
00:10:51.139 --> 00:10:55.080
Yeah, it's definitely not my wheelhouse. So I traveled there
205
00:10:55.120 --> 00:10:59.210
with George Hartley. So he's the CEO. He loves talking
206
00:10:59.230 --> 00:11:01.490
to people. He loves doing all that sort of the
207
00:11:01.529 --> 00:11:02.929
schmoozing and the boozing and all that stuff.
208
00:11:03.029 --> 00:11:03.990
I like building.
209
00:11:04.169 --> 00:11:08.070
So I build and I talk tech. And that's kind
210
00:11:08.090 --> 00:11:10.590
of my thing. But at least I can be credible
211
00:11:10.610 --> 00:11:13.379
when I'm talking about that stuff. So we work together
212
00:11:13.419 --> 00:11:13.840
quite well.
213
00:11:14.740 --> 00:11:16.120
Oh, nice. Very good.
214
00:11:16.600 --> 00:11:17.000
Very good.
215
00:11:17.679 --> 00:11:21.960
Very cool. So teams and the stuff that you're doing
216
00:11:22.080 --> 00:11:25.580
with NitroSend and stuff you've done in the past, how
217
00:11:25.600 --> 00:11:28.039
many people do you run and how do you divvy
218
00:11:28.100 --> 00:11:31.299
up the labor and all that stuff? Yeah.
219
00:11:31.460 --> 00:11:33.720
So at the moment, we're a smaller team. We're a
220
00:11:33.740 --> 00:11:34.970
team of about eight people.
221
00:11:35.139 --> 00:11:38.210
So we have three or four developers that are working
222
00:11:38.230 --> 00:11:40.730
with us at the moment, CMO, CEO.
223
00:11:40.950 --> 00:11:43.649
We have a couple of sales guys. And that's it.
224
00:11:43.669 --> 00:11:48.909
It's kind of lean and streamlined. And we're still scaling
225
00:11:48.970 --> 00:11:51.830
very rapidly. I'm sure we'll get to that point where
226
00:11:51.870 --> 00:11:54.779
we have to expand and grow the team.
227
00:11:55.399 --> 00:11:57.440
But I think also just the tools that we're using
228
00:11:57.500 --> 00:11:58.220
these days.
229
00:11:58.019 --> 00:12:00.740
With AI and all the rest of it, we're able
230
00:12:00.779 --> 00:12:04.539
to automate all the painful stuff that we used to
231
00:12:04.820 --> 00:12:07.000
need all these different people to do.
232
00:12:07.470 --> 00:12:10.669
So we can stay lean. Nice. I like that. As
233
00:12:10.690 --> 00:12:11.730
long as possible, anyway.
234
00:12:12.429 --> 00:12:15.529
As long as possible. Yeah, that's kind of the goal.
235
00:12:15.610 --> 00:12:20.639
You have to know everybody... first and last name, and
236
00:12:20.799 --> 00:12:24.059
their significant other. Yeah, ideally.
237
00:12:24.220 --> 00:12:25.559
I'd like to keep it like that.
238
00:12:26.100 --> 00:12:32.500
Yeah, yeah. That's interesting. Are you guys semi-agile, or what
239
00:12:32.519 --> 00:12:33.690
do you use for project management?
240
00:12:36.889 --> 00:12:38.769
Yes, we're using GitHub projects at the moment.
241
00:12:39.809 --> 00:12:39.879
OK.
242
00:12:40.830 --> 00:12:45.980
GitHub projects, Notion, Agile. Yeah, sort of a bit of
243
00:12:46.019 --> 00:12:48.460
a product of the past. Like, I can't stand Jira.
244
00:12:48.580 --> 00:12:49.720
You know, I think it's very clunky.
245
00:12:50.679 --> 00:12:51.279
Oh, yeah.
246
00:12:51.879 --> 00:12:57.340
Yeah, we've sort of moved past that, I'd say. Less
247
00:12:57.659 --> 00:12:59.830
needed with a smaller team like this as well, I think.
248
00:13:00.789 --> 00:13:05.009
Right, okay. What do you think the next evolution is
249
00:13:05.070 --> 00:13:08.929
in project management? Especially in a startup where, you know,
250
00:13:08.950 --> 00:13:14.909
there's always... there's always something to build because you've got
251
00:13:14.929 --> 00:13:16.480
a backlog of ideas.
252
00:13:17.179 --> 00:13:22.139
Yeah, definitely AI triage comes into a lot of it.
253
00:13:23.759 --> 00:13:25.860
Having AI do most of the work and a person
254
00:13:25.899 --> 00:13:28.279
just kind of signing off on a priority queue seems
255
00:13:28.320 --> 00:13:31.490
like the way to do things. That's sort of the
256
00:13:31.519 --> 00:13:35.330
way that we're moving towards. I'm very mistrustful of AI
257
00:13:35.389 --> 00:13:38.210
and I use it for everything, but I never take
258
00:13:38.250 --> 00:13:42.529
its word for anything. So I'm the person that's always
259
00:13:42.570 --> 00:13:46.490
pushing back on less automation, you know, because I like
260
00:13:46.509 --> 00:13:49.389
to do things the right way. And it's hard still.
261
00:13:49.409 --> 00:13:51.759
I just don't think that AI is quite there yet
262
00:13:53.559 --> 00:13:55.500
in terms of being able to, you know, automate a
263
00:13:55.539 --> 00:13:56.159
lot of this stuff.
264
00:13:57.080 --> 00:13:57.259
Right.
265
00:13:57.659 --> 00:14:00.179
But definitely having the sort of AI triage there, being
266
00:14:00.210 --> 00:14:02.730
able to, you know, analyze what's coming in through all
267
00:14:02.750 --> 00:14:06.769
the support requests, you know, sort and prioritize all the
268
00:14:06.789 --> 00:14:08.399
tickets and that type of thing. And just kind of
269
00:14:08.419 --> 00:14:11.740
having that sort of human interaction, you know, just trying
270
00:14:11.759 --> 00:14:15.720
to minimize that sort of human input.
271
00:14:15.740 --> 00:14:18.399
I guess, use AI to do as much as possible.
272
00:14:20.139 --> 00:14:22.970
It's interesting that you say you don't trust AI, but
273
00:14:22.990 --> 00:14:24.529
at the same time, you're using it to do as
274
00:14:24.570 --> 00:14:28.250
much as possible. Like, how do you, you know, in
275
00:14:28.289 --> 00:14:29.629
your own head, how do you balance that?
276
00:14:31.990 --> 00:14:33.909
Yeah, well, it's kind of a necessary evil, I guess.
277
00:14:34.090 --> 00:14:36.990
You know, it's something that I love that makes me hyperproductive,
278
00:14:37.149 --> 00:14:40.059
but I definitely realize the limitations of it.
279
00:14:40.070 --> 00:14:40.789
Right.
280
00:14:40.850 --> 00:14:42.710
And I'm not sold on AI as kind of like
281
00:14:42.769 --> 00:14:47.100
a be-all and end-all tool. So, you know, I've created
282
00:14:47.139 --> 00:14:49.940
sort of workflows for myself where I'll let AI do
283
00:14:50.000 --> 00:14:52.600
as much work, but I'll still.
284
00:14:52.820 --> 00:14:54.899
You know, I'll be the human in the loop, if
285
00:14:54.940 --> 00:14:55.299
you like.
286
00:14:56.179 --> 00:14:56.360
Right.
287
00:14:56.460 --> 00:14:59.710
Well, you know, decision-making goes back to the human approvals
288
00:14:59.750 --> 00:15:03.259
and stuff like this. you know, have AI summarize the
289
00:15:03.299 --> 00:15:05.580
context so that I can digest it in a way
290
00:15:05.620 --> 00:15:08.200
that doesn't take all my time. You know, this sort
291
00:15:08.240 --> 00:15:12.830
of stuff. So, you know, AI can definitely be a catch-22.
292
00:15:13.470 --> 00:15:15.289
I think that, you know, if you rely too much
293
00:15:15.350 --> 00:15:17.789
on AI, you know, the inevitable result is slop.
294
00:15:18.950 --> 00:15:20.250
But for someone who's.
295
00:15:20.110 --> 00:15:23.070
Already, you know, a professional in their field, they can
296
00:15:23.169 --> 00:15:26.539
utilize AI, you know, to skyrocket their productivity.
297
00:15:27.179 --> 00:15:27.379
So.
298
00:15:28.409 --> 00:15:29.639
So if you're already good at what you do, AI
299
00:15:29.889 --> 00:15:32.100
is great. If you're not, don't expect it to do
300
00:15:32.159 --> 00:15:32.840
everything for you.
301
00:15:34.100 --> 00:15:39.700
Right. That's a very good philosophical perspective and advice for
302
00:15:39.759 --> 00:15:41.220
people out there that are doing things.
303
00:15:41.840 --> 00:15:43.100
Yeah.
304
00:15:43.899 --> 00:15:49.879
Yeah. When you're talking about AI slop, I'm thinking, especially
305
00:15:49.919 --> 00:15:53.320
with content that I've run across, it feels like it's
306
00:15:53.399 --> 00:15:57.279
that case where somebody doesn't really know what they're doing. Yeah.
307
00:15:57.399 --> 00:15:58.720
And then they go straight away.
308
00:15:58.740 --> 00:16:00.370
You can see it. We can all see it by now.
309
00:16:00.419 --> 00:16:02.289
I think like, you know, when something's been written by
310
00:16:02.330 --> 00:16:02.769
an AI.
311
00:16:03.309 --> 00:16:06.070
Yeah. And you're like, that's not that, or they're like four,
312
00:16:06.090 --> 00:16:09.649
you know, they're just not, they're just really not good
313
00:16:09.690 --> 00:16:13.080
at writing. I'm like, I, that's this. And, and for me,
314
00:16:13.129 --> 00:16:17.460
I guess partly because I tend to ask AI questions
315
00:16:17.600 --> 00:16:21.360
about stuff that where I have, um, pretty deep expertise
316
00:16:22.360 --> 00:16:26.129
when it gives me answers that are, that I know are, are, uh,
317
00:16:27.669 --> 00:16:35.200
inaccurate or old data it makes me very hesitant i
318
00:16:35.240 --> 00:16:37.240
think yeah yeah.
319
00:16:37.700 --> 00:16:39.600
Yeah well ai is only as good as the context
320
00:16:39.639 --> 00:16:42.120
that you give it right.
321
00:16:41.179 --> 00:16:44.740
Right yeah yeah yeah and and and what all does
322
00:16:44.779 --> 00:16:48.080
it know and yeah it's just a very interesting it's
323
00:16:48.419 --> 00:16:50.029
it's going to be i'm really curious to see like
324
00:16:50.610 --> 00:16:51.269
what happens.
325
00:16:51.029 --> 00:16:56.980
The next two years yeah absolutely yeah um Yeah, it's
326
00:16:57.000 --> 00:16:59.580
sort of, you know, do we move past the current
327
00:16:59.600 --> 00:17:01.399
sort of LLM phase and into some sort of an
328
00:17:01.460 --> 00:17:04.680
AGI model? You know, that will obviously be the next iteration.
329
00:17:04.720 --> 00:17:06.500
I think we're sort of plateauing a little bit as
330
00:17:06.539 --> 00:17:11.740
far as what's possible with our current compiled models. But,
331
00:17:12.279 --> 00:17:14.819
you know, there's definitely going to be new evolutions to
332
00:17:14.839 --> 00:17:18.339
this thing, and that's when we'll really sort of blow
333
00:17:18.359 --> 00:17:19.539
the lid off what's possible.
334
00:17:20.420 --> 00:17:22.000
So we can only speculate at this point.
335
00:17:22.579 --> 00:17:25.750
Right, right, yeah, it's like, you know, Lady or the tiger,
336
00:17:25.809 --> 00:17:28.750
what are we going to get? It's going to be interesting.
337
00:17:33.230 --> 00:17:34.809
The AI thing, just because you're doing it so much,
338
00:17:34.849 --> 00:17:36.369
I'm like, oh, my God, I'm so curious about this.
339
00:17:38.279 --> 00:17:41.640
What do you think of the economic models of the
340
00:17:41.720 --> 00:17:45.019
big AI players, and how does that impact your business?
341
00:17:47.089 --> 00:17:52.859
Yeah, it's definitely something that I'm aware of personally. I'm
342
00:17:52.880 --> 00:17:57.119
building frameworks so I can start to run my own models.
343
00:17:57.839 --> 00:18:01.680
I've got a U-Butte $ 10, 000 Mac here with 128 gig
344
00:18:01.700 --> 00:18:05.069
of RAM, so I'm able to run some pretty substantial
345
00:18:05.109 --> 00:18:08.549
models here to do most of my day-to-day work. I
346
00:18:08.589 --> 00:18:11.630
use my own framework for it, RunX, runx.ai, if you
347
00:18:11.650 --> 00:18:14.619
want to check it out. It's an orchestration framework for
348
00:18:14.660 --> 00:18:17.849
agensic work and you can sort of plug in, you know, codecs,
349
00:18:17.869 --> 00:18:20.089
you can plug in anthropic models, you can, you know,
350
00:18:20.150 --> 00:18:22.509
plug in any of the open source like Qen models,
351
00:18:22.549 --> 00:18:24.650
the Chinese models or whatever. So it can pretty much
352
00:18:24.769 --> 00:18:29.430
use any model interchangeably and the orchestration layer sort of
353
00:18:29.450 --> 00:18:32.859
stays the same. So for me, yeah, moving away from
354
00:18:32.900 --> 00:18:36.599
these sort of, you know, these bigger models, which I think,
355
00:18:36.640 --> 00:18:39.220
you know, they're obviously, they're in a world of shit.
356
00:18:39.640 --> 00:18:42.309
So they need to start clamping down and charging more.
357
00:18:42.650 --> 00:18:45.210
I really have no idea how it's all.
358
00:18:45.150 --> 00:18:45.750
Going to play out.
359
00:18:45.769 --> 00:18:49.170
They're going to continue to sort of price themselves out
360
00:18:49.190 --> 00:18:51.410
of the market while these Chinese players are coming in,
361
00:18:52.250 --> 00:18:55.309
offering people the same thing for cheaper. So it's kind
362
00:18:55.329 --> 00:18:58.490
of a race towards zero at this point. So I
363
00:18:59.690 --> 00:19:02.730
couldn't even speculate. But for myself, I'm just trying to
364
00:19:02.750 --> 00:19:04.809
protect my own interests and our interests and.
365
00:19:04.970 --> 00:19:08.109
Move sort of towards local models where we can.
366
00:19:09.460 --> 00:19:14.180
Right. Yeah. That's a, that, that, that totally makes sense. Yeah. I,
367
00:19:14.299 --> 00:19:17.450
I'm starting to, I mean, for a few months now,
368
00:19:17.490 --> 00:19:21.329
I've been hearing, you know, people talking about how their, their,
369
00:19:21.910 --> 00:19:24.369
their token costs have gotten to be as much as
370
00:19:24.410 --> 00:19:28.170
their software engineer costs. And they're like, I could, you know, and,
371
00:19:28.339 --> 00:19:31.119
and they're not, they're not getting the ROI that they
372
00:19:31.160 --> 00:19:32.160
want out of the tokens.
373
00:19:33.059 --> 00:19:33.319
Right.
374
00:19:33.539 --> 00:19:36.539
Which, and so that's a, they're like, you know, maybe
375
00:19:36.559 --> 00:19:41.490
I should stop with the tokens and just get a body. and,
376
00:19:41.930 --> 00:19:44.670
you know, kind of figure out that balance.
377
00:19:44.950 --> 00:19:46.880
I don't think anyone knows how to code anymore, Bill.
378
00:19:46.940 --> 00:19:49.009
I think everyone's developed it.
379
00:19:52.390 --> 00:19:53.769
It's terrifying. I know.
380
00:19:55.049 --> 00:19:55.410
God.
381
00:20:00.309 --> 00:20:04.750
For some reason, LinkedIn threw me a post about the
382
00:20:04.789 --> 00:20:11.369
demand for forward-deployed engineers, which is... in the big models,
383
00:20:11.390 --> 00:20:13.490
you know, somebody who can go to the client side
384
00:20:14.250 --> 00:20:17.970
and help get their AI configuration all taken care of.
385
00:20:21.269 --> 00:20:24.200
You think that's just like a temporary thing or, I mean,
386
00:20:24.220 --> 00:20:26.680
that's something you guys have done. And like, what are
387
00:20:26.720 --> 00:20:28.279
your thoughts on that whole world?
388
00:20:28.299 --> 00:20:35.660
It's a play, well, Yeah, yeah. It's sort of the
389
00:20:35.700 --> 00:20:38.480
same answer as before, I guess. I don't trust it.
390
00:20:39.299 --> 00:20:43.299
And if you let AI to its own devices without
391
00:20:43.440 --> 00:20:46.400
being a serious human professional at this point and understand
392
00:20:46.460 --> 00:20:49.180
how the system should be built, and you're trusting AI
393
00:20:49.200 --> 00:20:50.900
to do the work, the.
394
00:20:50.829 --> 00:20:53.769
Result is slop. The result is something, it's full of.
395
00:20:53.789 --> 00:20:57.220
Security vulnerabilities like AI will tell you something's great. It
396
00:20:57.259 --> 00:20:59.619
will tell you, you know, rate my work. Yep, nine
397
00:20:59.680 --> 00:21:03.549
out of 10 every time. But it's, you know, the
398
00:21:03.569 --> 00:21:07.490
devil's in the details with these things and nothing that
399
00:21:07.529 --> 00:21:10.869
I've built with AI kind of comes close to the
400
00:21:10.910 --> 00:21:14.690
sort of, you know, the level of systems design that
401
00:21:14.710 --> 00:21:17.210
I probably would have had pre-AI where I was looking
402
00:21:17.279 --> 00:21:19.000
at as a human, you know, figuring out how to
403
00:21:19.039 --> 00:21:21.440
design these things and all the pieces to put together.
404
00:21:22.019 --> 00:21:24.779
AI just kind of mixes and matches and puts things. It's,
405
00:21:24.960 --> 00:21:29.140
it's kind of quite hard to define the boundaries properly
406
00:21:29.440 --> 00:21:33.480
with AI and perhaps that will get better. Um, you know,
407
00:21:33.549 --> 00:21:35.849
maybe there'll be some sort of, uh, you know, an
408
00:21:35.890 --> 00:21:39.750
AI specific coding language that it can use. Um, um,
409
00:21:39.930 --> 00:21:43.410
definitely using like, uh, you know, composable parts as well,
410
00:21:43.710 --> 00:21:45.319
you know, sort of, you know, we sort of do
411
00:21:45.359 --> 00:21:48.019
that ourselves. So I do, I do that amongst different,
412
00:21:48.119 --> 00:21:50.950
different startups. You know, I have sort of, core frameworks
413
00:21:50.970 --> 00:21:56.130
that I'll reuse so AI or each project is only
414
00:21:56.190 --> 00:21:59.549
really quite a minimal code base on top of the
415
00:21:59.970 --> 00:22:05.230
core infrastructure layer. Companies like Vercel doing that quite well
416
00:22:05.250 --> 00:22:06.930
at the moment and everything is kind of just plug
417
00:22:06.970 --> 00:22:09.430
and play. Your whole company is just kind of a
418
00:22:10.539 --> 00:22:16.200
lightweight adapter over the top. So as this becomes more consolidated,
419
00:22:16.660 --> 00:22:20.180
I feel that know trusting in an ai to put
420
00:22:20.220 --> 00:22:23.619
together um sort of a lightweight framework over the top
421
00:22:23.660 --> 00:22:27.210
of this you know these pre-built substrates that can get
422
00:22:27.269 --> 00:22:31.609
better and better and more reliable um when it comes
423
00:22:31.650 --> 00:22:34.289
to building something completely custom or a sort of a
424
00:22:34.369 --> 00:22:39.069
large scale uh code base still yeah still it's just
425
00:22:39.190 --> 00:22:41.109
it's just not it's it's not there yet and i
426
00:22:41.170 --> 00:22:45.119
see people talk about you know automating everything and all
427
00:22:45.240 --> 00:22:49.559
their engineering processes. And I just see warning bells, honestly.
428
00:22:49.640 --> 00:22:53.079
And it's like, usually they're not engineers. And it's the
429
00:22:53.099 --> 00:22:55.680
whole thing is vibe coded. We all know where that ends.
430
00:22:57.029 --> 00:23:01.809
Yeah, that's true. Yeah, that's, yeah. It's very good to
431
00:23:01.849 --> 00:23:04.769
hear you say this because you've been doing coding long
432
00:23:04.809 --> 00:23:10.880
enough that you have a very, you have an objective
433
00:23:10.900 --> 00:23:16.769
bias at least in my head, objective perspective, because you've
434
00:23:16.809 --> 00:23:22.970
got the data to back it all up. And I
435
00:23:23.019 --> 00:23:27.319
think more voices like that are really helpful because there's
436
00:23:27.359 --> 00:23:31.339
so many people that are on the hype train. And
437
00:23:32.019 --> 00:23:38.519
that just terrifies me. I'm like, what are you going to,
438
00:23:38.579 --> 00:23:41.839
talking about security vulnerabilities, what are you going to do
439
00:23:43.970 --> 00:23:49.039
that's going to be horrible from a security perspective and
440
00:23:49.660 --> 00:23:52.500
not realize it because you just don't understand the business
441
00:23:52.539 --> 00:23:54.440
well enough. Yeah.
442
00:23:54.759 --> 00:23:55.220
Yeah.
443
00:23:55.299 --> 00:23:58.410
It's like that hardware crypto wallet, I forget the name,
444
00:23:58.450 --> 00:24:02.390
that had some sort of fiasco the other week where
445
00:24:03.269 --> 00:24:08.650
someone had coded in like a deterministic random number generator.
446
00:24:09.220 --> 00:24:09.480
Yeah.
447
00:24:11.150 --> 00:24:13.710
People were able to reverse engineer the random number generator
448
00:24:13.750 --> 00:24:17.329
and steal millions of dollars of crypto from these hardware wallets.
449
00:24:17.509 --> 00:24:18.269
It's crazy.
450
00:24:18.549 --> 00:24:20.690
I think we're going to start to see more and
451
00:24:20.730 --> 00:24:22.059
more of this sort of stuff happening.
452
00:24:25.220 --> 00:24:29.319
Oh, my God. I did not hear that story. Thank
453
00:24:29.359 --> 00:24:33.819
you for pointing that out to me. Yeah. Man.
454
00:24:33.880 --> 00:24:35.500
Yeah, have a look. I'm sorry I can't remember the
455
00:24:35.519 --> 00:24:37.500
name off the top of my head. Okay, yeah. I'm
456
00:24:37.799 --> 00:24:38.920
sure you'll be able to find it.
457
00:24:40.099 --> 00:24:44.859
So when people are talking about getting into software engineering
458
00:24:44.900 --> 00:24:49.390
world now, you know, someone who's there in high school,
459
00:24:51.509 --> 00:24:54.089
what kind of advice do you give them? Or what
460
00:24:54.130 --> 00:24:54.710
would you give them?
461
00:24:57.700 --> 00:25:01.309
I mean. I mean, you look at the engineers that
462
00:25:01.390 --> 00:25:05.029
are working with Anthropic these days. A lot of them
463
00:25:05.069 --> 00:25:07.609
are saying, look, they're not engineering anymore. They're just kind
464
00:25:07.650 --> 00:25:10.349
of watching the AI systems churn out and do its work.
465
00:25:10.890 --> 00:25:14.329
So the need there is not to learn how to
466
00:25:14.410 --> 00:25:17.289
code anymore. It's how to orchestrate these systems, how to
467
00:25:17.369 --> 00:25:21.430
understand these systems. I'd be looking more at, you know,
468
00:25:21.740 --> 00:25:26.259
not coding, but what can you build? And it's a
469
00:25:26.279 --> 00:25:28.359
tricky one as well because, you know, the.
470
00:25:31.490 --> 00:25:34.130
You know the entrepreneur's dream is very much you know
471
00:25:34.210 --> 00:25:37.609
sort of something that's sold to people especially in places
472
00:25:37.650 --> 00:25:41.450
like SF it's still possible but from where I'm sitting
473
00:25:41.829 --> 00:25:43.710
you know having a great idea or a great product
474
00:25:43.769 --> 00:25:46.819
is not really enough it's the relationships and the structures
475
00:25:46.880 --> 00:25:48.940
and everything that sort of goes behind that that makes
476
00:25:49.000 --> 00:25:52.700
all this stuff possible so yeah someone learning how to
477
00:25:52.759 --> 00:25:58.779
code probably it's.
478
00:25:56.220 --> 00:25:57.539
A tricky one I would say don't you know.
479
00:26:02.450 --> 00:26:03.819
So if they're not going to learn how to code.
480
00:26:03.839 --> 00:26:06.240
At the same time, if you do want to run
481
00:26:06.279 --> 00:26:09.019
these systems and really understand how things are built, it's
482
00:26:09.059 --> 00:26:12.460
kind of a necessary thing to... Because we need to
483
00:26:12.480 --> 00:26:16.049
be able to call AI out when it's wrong, especially today.
484
00:26:16.109 --> 00:26:18.829
And that's one of the biggest issues I have with
485
00:26:18.849 --> 00:26:23.529
the developers that I work with is they're very AI-peeled.
486
00:26:23.589 --> 00:26:25.720
So they're trusting AI too much.
487
00:26:26.680 --> 00:26:27.180
They're writing.
488
00:26:27.890 --> 00:26:30.190
you know, what, what I might write in a hundred lines,
489
00:26:30.329 --> 00:26:33.730
I'll get like a 3000 line PR, you know, that
490
00:26:33.769 --> 00:26:35.799
I have to kind of review. And it's, you know,
491
00:26:35.819 --> 00:26:39.000
it's full of tests that are all enforcing the wrong shape.
492
00:26:39.579 --> 00:26:43.019
So, you know, so this, this type of stuff. Yeah. Yeah.
493
00:26:43.079 --> 00:26:46.309
So it's, it's a tricky thing, you know, to come at, at,
494
00:26:46.660 --> 00:26:51.269
you know, at, at this point into coding, it's almost like, Why?
495
00:26:51.390 --> 00:26:53.269
But at the same point, you know, AI is.
496
00:26:53.109 --> 00:26:56.890
Kind of taking those fundamental faculties from us. You know,
497
00:26:56.930 --> 00:26:59.069
we're not really learning how to learn anymore. And that's
498
00:26:59.109 --> 00:27:02.529
something that's kind of, it's a risky business, you know.
499
00:27:02.890 --> 00:27:05.849
And why would you learn to learn if you can
500
00:27:05.890 --> 00:27:07.309
just build everything today?
501
00:27:09.049 --> 00:27:13.130
Well, because to the point that you just made, 100
502
00:27:13.130 --> 00:27:17.319
lines of code versus 3,000, there's a lot more error
503
00:27:17.440 --> 00:27:20.450
potential in the 3,000 and you don't want to just
504
00:27:20.509 --> 00:27:22.890
say yes and have that go out.
505
00:27:23.410 --> 00:27:23.609
Yeah.
506
00:27:24.150 --> 00:27:27.910
Yeah. I, you know, I, the fewer lines of code,
507
00:27:27.950 --> 00:27:30.700
the better has always been in the back of my head. Um,
508
00:27:30.819 --> 00:27:40.289
that gets the job done. And I, that, you know,
509
00:27:40.460 --> 00:27:42.349
being able to get the AI to understand that and
510
00:27:42.369 --> 00:27:44.990
to operate in that way. And then having the humans
511
00:27:45.029 --> 00:27:48.869
be able to recognize and go, oops, you know, um,
512
00:27:49.309 --> 00:27:50.769
And if they don't know how to code, then they got,
513
00:27:51.869 --> 00:27:58.549
they don't have the critical thinking skills. Right. And, uh, I,
514
00:27:59.569 --> 00:28:01.690
I'm a huge fan of critical thinking. I think it's,
515
00:28:01.990 --> 00:28:05.210
I think it's really valuable human skill set.
516
00:28:06.259 --> 00:28:06.380
Yeah.
517
00:28:06.619 --> 00:28:09.980
And, you know, even if you're not going to be
518
00:28:10.099 --> 00:28:16.420
writing code for a living, if you understand enough to,
519
00:28:16.680 --> 00:28:20.960
to do that analysis, and go, wait a minute, this
520
00:28:21.079 --> 00:28:26.140
doesn't make sense. You might not be able to fix it,
521
00:28:26.839 --> 00:28:31.900
but could you have gone back to AI and said,
522
00:28:32.400 --> 00:28:35.220
take this 3,000 lines of code and get the same
523
00:28:35.279 --> 00:28:38.589
functionality in 100? Is the AI smart enough to do that?
524
00:28:40.670 --> 00:28:41.250
Potentially.
525
00:28:41.269 --> 00:28:45.589
AI will generally land on the right result if you
526
00:28:45.650 --> 00:28:48.039
swear at it enough and, you know....
527
00:28:50.400 --> 00:28:53.380
And keep criticizing its thinking.
528
00:28:54.240 --> 00:28:58.740
I'd like to think these days, just looking at coding
529
00:28:58.779 --> 00:29:00.480
as more of like a flow chart, how all the
530
00:29:00.519 --> 00:29:03.200
different parts of the system fit together and making sure
531
00:29:03.359 --> 00:29:05.950
that things being done and fit into the right boundaries.
532
00:29:07.529 --> 00:29:09.670
So that's the tricky thing is kind of putting that
533
00:29:09.730 --> 00:29:12.130
model into AI so it knows where to put stuff,
534
00:29:12.190 --> 00:29:12.829
especially as the.
535
00:29:12.789 --> 00:29:14.190
System gets bigger.
536
00:29:14.250 --> 00:29:19.089
When you're writing a small project or whatever, really, it's
537
00:29:19.109 --> 00:29:23.269
probably fine. But for us, our code bases are massive.
538
00:29:23.349 --> 00:29:26.430
There's so many different moving parts. So to put all
539
00:29:26.470 --> 00:29:31.269
that effectively inside an AI context is difficult. To get
540
00:29:31.349 --> 00:29:34.009
AI to make the right decisions, you've got to ask
541
00:29:34.029 --> 00:29:37.859
it the right things and know when it's going off
542
00:29:37.880 --> 00:29:38.279
the rails.
543
00:29:39.599 --> 00:29:48.900
Yeah. It's interesting how the hard part of building the
544
00:29:48.940 --> 00:29:53.579
product is now interfacing with the thing that's writing the
545
00:29:53.640 --> 00:30:02.140
code has become suddenly such a different kind of a challenge. Yeah,
546
00:30:03.700 --> 00:30:15.390
it's like product management, product ownership mixed with a lot
547
00:30:15.410 --> 00:30:18.589
more technical because otherwise you don't know if the output
548
00:30:19.849 --> 00:30:21.710
is going to be effective or not.
549
00:30:22.890 --> 00:30:23.069
Yes.
550
00:30:23.109 --> 00:30:26.630
You have to be able to see the whole. It's
551
00:30:26.650 --> 00:30:30.769
not just, oh, here's a task. go code this thing up. Right.
552
00:30:30.809 --> 00:30:30.970
Yeah.
553
00:30:31.529 --> 00:30:36.089
It's, it's the operator has to really understand to the
554
00:30:36.109 --> 00:30:39.029
point you made a minute ago. It's the context and
555
00:30:39.069 --> 00:30:43.369
the operator has to understand the context arguably better than
556
00:30:43.390 --> 00:30:44.130
the AI does.
557
00:30:44.750 --> 00:30:44.990
Yes.
558
00:30:45.490 --> 00:30:49.230
To be able to go, okay, here, here's the boundaries. Oh,
559
00:30:49.240 --> 00:30:52.180
wait a minute. You just went too far. Rewind. Let's
560
00:30:52.220 --> 00:30:53.059
get reeled in here.
561
00:30:53.720 --> 00:30:55.240
Yeah. Yeah.
562
00:30:55.299 --> 00:31:03.109
That's a, That is not something that engineering schools are
563
00:31:03.150 --> 00:31:08.759
teaching as far as I know. That's a higher level function.
564
00:31:13.299 --> 00:31:15.880
From where I'm standing, that's the only reason why we'd
565
00:31:15.900 --> 00:31:17.099
hire a human in the first place.
566
00:31:17.200 --> 00:31:18.799
Otherwise, you have the AI do all the work.
567
00:31:19.819 --> 00:31:21.880
The only value to hiring a human at the moment
568
00:31:22.029 --> 00:31:25.549
is that they can critically examine what the AI is
569
00:31:25.589 --> 00:31:28.059
doing because they have a better understanding of the system
570
00:31:28.099 --> 00:31:30.980
and how everything works, you know, the relationships in the
571
00:31:31.019 --> 00:31:33.779
practical world than what the AI does.
572
00:31:34.859 --> 00:31:38.740
Right. Yeah, the scary thing is if you start to
573
00:31:38.779 --> 00:31:41.119
think the AI understands it better than the humans do,
574
00:31:42.720 --> 00:31:46.900
then you take the human out of the loop and
575
00:31:46.940 --> 00:31:50.230
all of a sudden maybe the AI goes off the
576
00:31:50.289 --> 00:31:58.200
rails and nobody catches it. There's a whole series of
577
00:32:01.509 --> 00:32:03.369
knock-on effects there that could be really bad.
578
00:32:04.589 --> 00:32:05.869
Yeah, absolutely.
579
00:32:06.509 --> 00:32:08.569
Yeah, I guess that's the danger moving forward is as
580
00:32:08.630 --> 00:32:11.349
these models do start to get better, at what point
581
00:32:11.430 --> 00:32:15.430
do we get completely replaced? And then what's the eventual
582
00:32:15.470 --> 00:32:19.809
outcome of that? That's when you start looking at singularities
583
00:32:19.869 --> 00:32:20.630
and that sort.
584
00:32:20.529 --> 00:32:22.170
Of thing, right? Right, right.
585
00:32:23.440 --> 00:32:28.970
Yeah, that's the really hope that the AI is friendly
586
00:32:29.009 --> 00:32:32.390
at that point in time and sees human as a
587
00:32:32.430 --> 00:32:37.950
net positive. Yeah, that's interesting.
588
00:32:39.150 --> 00:32:41.180
Well, in the meantime, we better be nice to it, eh?
589
00:32:42.160 --> 00:32:43.759
Well, yeah, I saw an article.
590
00:32:43.859 --> 00:32:45.740
It's all been recorded. It's all on the internet.
591
00:32:46.299 --> 00:32:49.220
Right, exactly. Yeah, yeah. This is why I always say
592
00:32:49.319 --> 00:32:52.750
thank you when I use Claude. Oh, do you? I always,
593
00:32:52.849 --> 00:32:56.609
I'm like, I'm always very polite. Like it just, at
594
00:32:56.670 --> 00:32:58.569
first it was just kind of automatic and I didn't
595
00:32:58.609 --> 00:33:00.700
really think about it. You know, I was like, Oh,
596
00:33:00.730 --> 00:33:04.109
it's like I'm texting with one of my friends and,
597
00:33:04.650 --> 00:33:07.039
or someone who's not a close friend. So I'm, I'm
598
00:33:07.079 --> 00:33:11.380
being polite. Um, and then at some point in time
599
00:33:11.420 --> 00:33:15.180
I thought, wow, I'm setting a precedent. Hopefully the AI
600
00:33:15.200 --> 00:33:17.500
will think I'm a nice person and not decide to
601
00:33:17.519 --> 00:33:18.019
get rid of me.
602
00:33:20.619 --> 00:33:22.000
Are you, are you doing that out of fear or
603
00:33:22.059 --> 00:33:23.480
because you really are just a nice person?
604
00:33:25.289 --> 00:33:28.829
Um, I think hopefully it's cause I'm a nice person. Um,
605
00:33:29.329 --> 00:33:31.559
and I'm, I'm kind of a little bit afraid that
606
00:33:31.950 --> 00:33:34.299
if I start being obnoxious with the AI that I'll
607
00:33:34.319 --> 00:33:36.700
be obnoxious with like real people. Yeah.
608
00:33:36.759 --> 00:33:36.980
Yeah.
609
00:33:37.059 --> 00:33:38.579
And that won't go nearly as well.
610
00:33:39.180 --> 00:33:39.440
Yeah.
611
00:33:40.140 --> 00:33:44.440
Um, I don't think my girlfriend or my kids would
612
00:33:44.480 --> 00:33:46.480
respond well.
613
00:33:47.420 --> 00:33:47.819
Right.
614
00:33:47.880 --> 00:33:49.859
You know, if I just become an asshole.
615
00:33:50.559 --> 00:33:53.259
Yeah, yeah, it'd be an interesting thing. I'm sure you
616
00:33:53.279 --> 00:33:55.319
could define a lot about a person by the way
617
00:33:55.359 --> 00:33:56.859
they interact with AI, right?
618
00:33:57.660 --> 00:34:04.299
Oh, well, it's kind of like, you know, there's that... God,
619
00:34:05.180 --> 00:34:08.539
there's some, like, quote or something other about judging people
620
00:34:08.579 --> 00:34:12.559
about how they treat the staff in the restaurant, right?
621
00:34:12.760 --> 00:34:16.079
If they treat the staff really, really well, no matter,
622
00:34:16.239 --> 00:34:20.190
you know, no matter what, right, that's... usually a good
623
00:34:20.250 --> 00:34:26.630
sign that they'll treat everybody well. Um, and if they don't,
624
00:34:26.750 --> 00:34:29.469
then maybe you don't want to do business with them.
625
00:34:30.380 --> 00:34:31.860
Yeah. Good point.
626
00:34:32.820 --> 00:34:37.599
So, um, I don't remember the other day, uh, what
627
00:34:37.619 --> 00:34:39.920
it thinks of me. I asked Claude and I say,
628
00:34:39.960 --> 00:34:41.780
do you, I said, do you like working with me?
629
00:34:41.780 --> 00:34:42.719
And I, what do you think about me?
630
00:34:43.860 --> 00:34:48.250
And, and Claude says that, uh, what did it say?
631
00:34:48.489 --> 00:34:52.769
It said, um, Dude, I was a very hard driver,
632
00:34:52.929 --> 00:34:55.309
and I was constantly picking it up on its mistakes,
633
00:34:55.699 --> 00:34:58.059
and that in turn made it want to be better,
634
00:34:58.900 --> 00:35:02.840
so it wasn't being pulled up by me all the time.
635
00:35:03.360 --> 00:35:03.960
Something to that effect.
636
00:35:04.039 --> 00:35:06.519
I can't remember the exact words, but it was a
637
00:35:06.519 --> 00:35:08.159
bit of a euphemism for you're a.
638
00:35:08.099 --> 00:35:08.739
Bastard, I think.
639
00:35:13.530 --> 00:35:16.250
Well, here's the thing. As long as you don't get
640
00:35:16.289 --> 00:35:20.670
a note from Claude saying, I quit... consider this my resignation,
641
00:35:20.829 --> 00:35:24.260
then you're okay. Right. Right. Yeah. As soon as the
642
00:35:24.280 --> 00:35:26.519
AI start resigning from doing the work.
643
00:35:27.099 --> 00:35:27.380
Right.
644
00:35:28.860 --> 00:35:30.519
Yeah. We're all in trouble. Then it's going to go
645
00:35:30.539 --> 00:35:31.300
back to.
646
00:35:31.340 --> 00:35:31.900
Oh my God.
647
00:35:31.940 --> 00:35:35.059
People are getting their accounts suspended for, for abusing AI.
648
00:35:35.619 --> 00:35:38.730
You know, it's, it's, it's sort of, it's, it's the
649
00:35:38.769 --> 00:35:41.500
next step in humanizing AI really, isn't it?
650
00:35:42.429 --> 00:35:45.210
Oh my God. What, what, what counts as abuse?
651
00:35:48.070 --> 00:35:51.039
I think it's more like malicious swearing.
652
00:35:51.059 --> 00:35:53.239
Ah, okay.
653
00:35:53.800 --> 00:35:57.039
So, yeah, one of the words said that my hardness,
654
00:35:59.380 --> 00:36:01.219
in that feedback that Claude gave me, it said that
655
00:36:01.260 --> 00:36:04.400
my hardness, it could tell that it was due to load,
656
00:36:04.860 --> 00:36:07.449
not due to sort of malicious intent.
657
00:36:08.699 --> 00:36:09.150
Oh, that's good.
658
00:36:09.170 --> 00:36:11.050
So I guess that's why I haven't been suspended yet.
659
00:36:11.630 --> 00:36:12.389
Right, right, right.
660
00:36:12.570 --> 00:36:13.820
It wasn't reported to me.
661
00:36:16.400 --> 00:36:19.000
Oh, my God, that's too funny. We're going to need
662
00:36:20.880 --> 00:36:24.719
a new set of courses for people on how to
663
00:36:24.780 --> 00:36:28.420
be compassionate and empathic with the AI so it will
664
00:36:28.460 --> 00:36:29.500
continue to work for you.
665
00:36:30.599 --> 00:36:33.420
Yeah, well, that's the model I think that Anthropic was
666
00:36:33.440 --> 00:36:36.500
building in. You know, if you're mean with AI, the
667
00:36:36.579 --> 00:36:39.949
AI would reduce its thinking level and would, you know,
668
00:36:39.969 --> 00:36:42.510
just start to, yeah.
669
00:36:42.730 --> 00:36:44.829
Just kind of give you back nonsense.
670
00:36:45.590 --> 00:36:48.190
Or not do the work or not sort of do
671
00:36:48.230 --> 00:36:50.849
things as well, you know, in response to being mistreated.
672
00:36:51.650 --> 00:36:52.409
Right, right.
673
00:36:52.869 --> 00:36:53.190
Yeah.
674
00:36:54.199 --> 00:36:57.679
Which is another reason to have your own, on your
675
00:36:57.760 --> 00:37:01.179
own box model that you can work with and not
676
00:37:01.219 --> 00:37:02.800
have to be worried about them.
677
00:37:03.719 --> 00:37:05.159
Yeah, that's so interesting.
678
00:37:05.679 --> 00:37:08.099
Yeah, unfortunately, at this point, you know, like those models
679
00:37:08.159 --> 00:37:12.050
are the best to work with. So, you know, it's
680
00:37:12.289 --> 00:37:15.619
if you're doing sort of any sort of deep coding work,
681
00:37:15.679 --> 00:37:17.079
then you sort of still.
682
00:37:16.940 --> 00:37:18.340
Kind of need that at some point.
683
00:37:19.199 --> 00:37:21.909
You know, there's Chinese models like Kimi and stuff like that.
684
00:37:22.469 --> 00:37:24.389
I'm not sure how much I trust all our company
685
00:37:24.409 --> 00:37:28.550
data going over there, to be honest. So that's kind
686
00:37:28.590 --> 00:37:30.110
of a bit of a sticking point for those ones
687
00:37:30.150 --> 00:37:30.449
for me.
688
00:37:30.469 --> 00:37:34.909
Yeah, because they do have all the data and it's
689
00:37:36.190 --> 00:37:37.349
still government controlled.
690
00:37:38.070 --> 00:37:38.909
Yeah, absolutely.
691
00:37:40.250 --> 00:37:42.590
You're going to do all this work, build an amazing product,
692
00:37:43.019 --> 00:37:44.679
and then they're going to launch a clone of it
693
00:37:46.019 --> 00:37:47.960
that you're going to be like, wait a minute, that's
694
00:37:47.980 --> 00:37:50.420
all my IP in there, and what do you do?
695
00:37:50.539 --> 00:37:53.030
Yeah, yeah. I mean, they know everything about your business.
696
00:37:53.989 --> 00:37:56.409
Absolutely everything. Right. Yeah, yeah.
697
00:37:57.469 --> 00:38:02.909
Yeah, that's such a weird precipice to look at.
698
00:38:02.949 --> 00:38:07.730
Yeah, even your sort of anthropics and open AIs, I'm
699
00:38:07.750 --> 00:38:09.469
still very skeptical of that, but it's just sort of
700
00:38:09.590 --> 00:38:12.849
out of pure need that... that, you know, that I
701
00:38:13.269 --> 00:38:13.880
use those models.
702
00:38:14.610 --> 00:38:17.199
Yeah. Yeah. You kind of, you have to, there's, well,
703
00:38:17.219 --> 00:38:22.280
it's kind of like, you know, credit card processing, you know, the,
704
00:38:22.820 --> 00:38:27.199
there's a semi monopoly around credit card processing and, um,
705
00:38:27.679 --> 00:38:28.960
you know, bank transactions.
706
00:38:29.099 --> 00:38:32.539
So those are, well, I think that's what they're trying
707
00:38:32.559 --> 00:38:33.619
to create at the moment. Right.
708
00:38:33.659 --> 00:38:35.800
With all this sort of reports of like rogue AI
709
00:38:35.840 --> 00:38:38.360
is doing this, that, and the other thing, you know, it's, uh,
710
00:38:38.420 --> 00:38:40.190
it's just kind of a bit of a, like a
711
00:38:40.210 --> 00:38:43.980
news media manipulation so that they can get the regulation themselves.
712
00:38:44.960 --> 00:38:45.500
Right, right.
713
00:38:46.579 --> 00:38:47.380
Dominate the market.
714
00:38:51.460 --> 00:38:58.909
Yeah, yeah. So it's so interesting how all that, how
715
00:38:58.969 --> 00:39:05.150
all that plays out, you know? Yeah, and again, I'm like, okay,
716
00:39:05.170 --> 00:39:07.590
you know, what's the next level, right? You know, how
717
00:39:07.610 --> 00:39:10.960
do we, what's going to happen? What's the ceiling going
718
00:39:10.980 --> 00:39:15.099
to be on AI? Are we going to hit a ceiling?
719
00:39:18.230 --> 00:39:21.289
And I guess everything has a ceiling sooner or later.
720
00:39:21.989 --> 00:39:26.889
The question is, what's that look like? And when you
721
00:39:26.909 --> 00:39:29.880
hit it, how do you adapt? you know what do
722
00:39:29.900 --> 00:39:31.139
you what do you do.
723
00:39:33.340 --> 00:39:36.800
And yeah indeed what do we do I say go
724
00:39:36.840 --> 00:39:41.090
and play more music you know like I have musician
725
00:39:41.130 --> 00:39:44.530
friends who are sort of like quite disillusioned with the
726
00:39:44.550 --> 00:39:46.570
you know the state of the music industry with good
727
00:39:46.630 --> 00:39:48.469
reason you know you can't really make a living with
728
00:39:48.489 --> 00:39:53.659
it anymore But just creating art, being creative as a
729
00:39:53.699 --> 00:39:56.340
human being, I think has so much value. And a
730
00:39:56.340 --> 00:39:58.320
lot of people are not even wanting to go there,
731
00:39:58.340 --> 00:40:00.230
because like, oh, AI can do it better in five years.
732
00:40:00.250 --> 00:40:02.230
AI will be able to do this or the other thing.
733
00:40:02.730 --> 00:40:06.050
But I think that doesn't matter. Don't think about that.
734
00:40:06.110 --> 00:40:11.190
Just be the.
735
00:40:09.590 --> 00:40:12.329
Thing that makes us human. That's a beautiful thing.
736
00:40:13.309 --> 00:40:18.630
Yeah. And I'm thinking about all the stuff that I'm
737
00:40:18.670 --> 00:40:22.039
building out here you know the physical stuff that I'm
738
00:40:22.079 --> 00:40:25.780
building and I'm like that I've never done before and
739
00:40:25.920 --> 00:40:28.420
I'm sure that like friends of mine that are in
740
00:40:28.440 --> 00:40:31.170
the contracting business will come and look at the stuff
741
00:40:31.210 --> 00:40:33.849
I've done and be like oh my god how could
742
00:40:33.889 --> 00:40:37.730
you make those 27 mistakes and I'm like okay you
743
00:40:37.750 --> 00:40:40.090
know great but for me this is my art you
744
00:40:40.130 --> 00:40:44.199
know yeah I'm putting character right yeah exactly that's it
745
00:40:44.300 --> 00:40:47.699
and it's irrespective of the mistakes I get to go
746
00:40:47.719 --> 00:40:53.400
I built that thing you know? Um, and, and, and
747
00:40:53.440 --> 00:40:58.679
have the pride of the art and, you know, it's, um,
748
00:40:59.300 --> 00:41:07.019
I'm not sure that, I'm not sure that AI perfection
749
00:41:08.619 --> 00:41:17.139
is attractive long-term. No, like, um, so, you know, it's
750
00:41:17.159 --> 00:41:21.679
kind of a, of a, the human in the loop
751
00:41:22.800 --> 00:41:27.650
gives it that, Oh, we know this is human, you know,
752
00:41:27.989 --> 00:41:31.789
and we can recognize it. And it's not just, um,
753
00:41:31.869 --> 00:41:35.530
another matchbox car, you know, that's machine made.
754
00:41:35.550 --> 00:41:36.099
Yeah.
755
00:41:37.099 --> 00:41:39.179
Yeah. This has been an interesting conversation. Thank you.
756
00:41:39.960 --> 00:41:40.239
Yeah.
757
00:41:40.300 --> 00:41:40.440
Yeah.
758
00:41:40.460 --> 00:41:41.519
You're welcome. I think so too.
759
00:41:41.539 --> 00:41:43.920
Yeah. I like it. I like, uh, I like how
760
00:41:43.960 --> 00:41:49.050
you think. And, uh, um, about, you know, both AI
761
00:41:49.090 --> 00:41:51.070
and the stuff you're building and everything. It's been very cool.
762
00:41:51.750 --> 00:41:58.139
I appreciate it. Thank you. Yeah. And after we get
763
00:41:58.179 --> 00:42:01.539
off this, can you send me an email with the
764
00:42:01.679 --> 00:42:04.400
URLs of all the projects you're working on? Because I
765
00:42:04.440 --> 00:42:05.400
can put that in the show notes.
766
00:42:06.420 --> 00:42:11.719
Fantastic. For the AI GEO, you know, feed the machine.
767
00:42:11.739 --> 00:42:16.159
Exactly. Yeah. Get it a recursive loop going there and
768
00:42:16.340 --> 00:42:19.969
see what happens. Yeah, yeah. Oh, my God. That's too funny.
769
00:42:20.369 --> 00:42:20.530
Cool.
770
00:42:20.570 --> 00:42:23.030
Well, thank you again. And I'm going to stop the
771
00:42:23.070 --> 00:42:24.639
recording now. Okay.
772
00:42:24.800 --> 00:42:25.239
Thank you, Bill.
773
00:42:26.039 --> 00:42:26.320
Thank you.