1
00:00:00,000 --> 00:00:07,520
So, that's the open source recorder that actually works the last time.

2
00:00:07,520 --> 00:00:13,000
And that's when, I mean, you know, it probably crashes if the recording is like three hours.

3
00:00:13,000 --> 00:00:15,200
No, but it worked with Justin.

4
00:00:15,200 --> 00:00:23,400
And so, if all of that works, then we have an mp3 file as the output.

5
00:00:23,400 --> 00:00:29,200
I don't have to do anything and it runs through my amazing processing pipeline and everything is done.

6
00:00:29,600 --> 00:00:31,800
Very soon, we want you to have to speak.

7
00:00:31,800 --> 00:00:33,000
Yeah, that's it.

8
00:00:33,000 --> 00:00:39,200
Again, I think that you have to be careful what to automate and what to abstract your way and what not.

9
00:00:39,200 --> 00:00:51,200
Did I tell you that I got, I trained one ill limb with all of my public content like podcasts and she's like that.

10
00:00:51,200 --> 00:00:51,800
Oh, wow.

11
00:00:51,800 --> 00:00:52,800
And all of yours.

12
00:00:52,800 --> 00:00:53,400
Oh, my God.

13
00:00:53,400 --> 00:00:54,400
That's bad.

14
00:00:54,400 --> 00:00:57,000
So, you know, you're writing.

15
00:00:57,000 --> 00:00:59,000
And an artificial GG.

16
00:00:59,000 --> 00:01:00,000
Yes.

17
00:01:00,000 --> 00:01:01,000
Oh, that's bad.

18
00:01:01,000 --> 00:01:04,000
I made it record the podcast.

19
00:01:04,000 --> 00:01:06,000
Well, I mean, write the transcript of the book.

20
00:01:06,000 --> 00:01:07,000
Yeah.

21
00:01:07,000 --> 00:01:10,000
It was complete out of shit.

22
00:01:10,000 --> 00:01:11,000
I'm not surprised.

23
00:01:11,000 --> 00:01:17,000
It was just constant agreement and just going around without exploring anything in particular.

24
00:01:17,000 --> 00:01:19,000
I mean, well, that's very accurate.

25
00:01:19,000 --> 00:01:20,000
Yes.

26
00:01:20,000 --> 00:01:21,000
Isn't it?

27
00:01:21,000 --> 00:01:23,000
So, that's when it extracted out of it.

28
00:01:23,000 --> 00:01:30,000
No, but to me, since now we are an AI and wind podcast.

29
00:01:30,000 --> 00:01:31,000
That blossom.

30
00:01:31,000 --> 00:01:36,000
And blossom, of course.

31
00:01:36,000 --> 00:01:38,000
To me, super interesting to find.

32
00:01:38,000 --> 00:01:46,000
I was very skeptical of the whole AI thing for a long time.

33
00:01:46,000 --> 00:01:52,000
And my main thing is, you know, I was also into like the whole singularity stuff.

34
00:01:52,000 --> 00:02:01,000
And realizing that LLMs will never get you there, you know, no matter how good they get.

35
00:02:01,000 --> 00:02:04,000
That is, that it's a different thing.

36
00:02:04,000 --> 00:02:06,000
Consciousness is a different thing.

37
00:02:06,000 --> 00:02:11,000
So, but once I understood that that's not the goal, it can be a useful tool anyway.

38
00:02:11,000 --> 00:02:14,000
Then I'm thinking, okay, what is it good for?

39
00:02:14,000 --> 00:02:15,000
What is it good at?

40
00:02:15,000 --> 00:02:17,000
And what is it bad at?

41
00:02:17,000 --> 00:02:23,000
I'm coming up with new interesting ideas and like having a conversation like these, like emulate Gigi.

42
00:02:23,000 --> 00:02:25,000
Emulate power, how a conversation, it doesn't work.

43
00:02:25,000 --> 00:02:27,000
It's also one of Vivek's tests, basically.

44
00:02:27,000 --> 00:02:34,000
I always look back to him because he's like, I think the most knowledgeable guy about these systems

45
00:02:34,000 --> 00:02:39,000
and their limitations from like non computer science, non tech, non AI perspective,

46
00:02:39,000 --> 00:02:43,000
but purely from a philosophical and cognitive science perspective, basically.

47
00:02:43,000 --> 00:02:47,000
It's like he always says like, just let two of them talk to each other.

48
00:02:47,000 --> 00:02:50,000
And that tells you how smart they are, basically.

49
00:02:50,000 --> 00:02:55,000
And it goes, like even one of them talking to a human hallucinates very quickly

50
00:02:55,000 --> 00:02:57,000
and bullshit you all the time and so on.

51
00:02:57,000 --> 00:03:01,000
Again, like he would argue because they're not grounded out in reality in some sense.

52
00:03:01,000 --> 00:03:07,000
And if you have two of those, then it's just gonna be, it just goes haywire instantly, basically.

53
00:03:07,000 --> 00:03:09,000
Yeah, yeah, yeah, instantly.

54
00:03:09,000 --> 00:03:10,000
Yeah, 100%.

55
00:03:10,000 --> 00:03:13,000
Yeah, I didn't know that you're like, can I sue you now?

56
00:03:13,000 --> 00:03:23,000
You have to, you have to release it under creative comments, share alike.

57
00:03:23,000 --> 00:03:29,000
So you have to, you know, you know, I'm a communist, so I use GPL and those kind of licenses.

58
00:03:29,000 --> 00:03:34,000
Isn't there like lawsuits by stallman now that aren't there models now where they're like,

59
00:03:34,000 --> 00:03:38,000
oh, we excluded all GPL license software from the data sets.

60
00:03:38,000 --> 00:03:40,000
Wow, so it's safe to use, I think so.

61
00:03:40,000 --> 00:03:41,000
Oh, wow.

62
00:03:41,000 --> 00:03:43,000
A friend of mine told me, so it must be true.

63
00:03:43,000 --> 00:03:44,000
Oh, my God.

64
00:03:44,000 --> 00:03:49,000
That is poetic.

65
00:03:49,000 --> 00:03:56,000
So we'll have like, you know, stallman GPT under a new license that fixes all those things.

66
00:03:56,000 --> 00:03:59,000
It's truly free.

67
00:03:59,000 --> 00:04:07,000
But yeah, you were talking about your kind of, your new setup and your new tooling.

68
00:04:07,000 --> 00:04:09,000
I'd be super interested to.

69
00:04:09,000 --> 00:04:10,000
Yeah.

70
00:04:10,000 --> 00:04:13,000
Yeah, what you've been up to the last, I don't know, five days.

71
00:04:13,000 --> 00:04:16,000
Probably, I can't even say 10 days because everything changes.

72
00:04:16,000 --> 00:04:18,000
Oh, I don't remember what happened 10 days ago.

73
00:04:18,000 --> 00:04:19,000
Yeah.

74
00:04:19,000 --> 00:04:21,000
It's not my context window.

75
00:04:21,000 --> 00:04:22,000
Yeah.

76
00:04:22,000 --> 00:04:23,000
Yeah.

77
00:04:23,000 --> 00:04:25,000
Tiny RAM.

78
00:04:25,000 --> 00:04:27,000
Too little RAM.

79
00:04:27,000 --> 00:04:30,000
It's not the size of the matter.

80
00:04:30,000 --> 00:04:31,000
Yeah.

81
00:04:31,000 --> 00:04:35,000
So, yeah, so, yeah, I've been experimenting a lot.

82
00:04:35,000 --> 00:04:40,000
It's kind of, I'm reimagining us.

83
00:04:40,000 --> 00:04:44,000
This is a shift in the way I do.

84
00:04:44,000 --> 00:04:48,000
Many things, but everything related to work.

85
00:04:48,000 --> 00:04:50,000
Must shift.

86
00:04:50,000 --> 00:04:51,000
Yeah.

87
00:04:51,000 --> 00:04:52,000
And I'm super excited.

88
00:04:52,000 --> 00:04:57,000
It's super interesting because I keep seeing two camps.

89
00:04:57,000 --> 00:05:00,000
People that are super excited.

90
00:05:00,000 --> 00:05:02,000
I mean, three camps.

91
00:05:02,000 --> 00:05:05,000
People that are completely oblivious.

92
00:05:05,000 --> 00:05:08,000
But I also see the camp of people that are, oh, my God.

93
00:05:08,000 --> 00:05:09,000
What's going to happen?

94
00:05:09,000 --> 00:05:14,000
Are we, the humanity is going to become useless?

95
00:05:14,000 --> 00:05:16,000
Am I out of a job as a program?

96
00:05:16,000 --> 00:05:17,000
Am I out of a job?

97
00:05:17,000 --> 00:05:18,000
Yeah.

98
00:05:18,000 --> 00:05:25,000
And to me, it's like, if your job is executing things, then yes.

99
00:05:25,000 --> 00:05:26,000
Yeah.

100
00:05:26,000 --> 00:05:27,000
Yeah.

101
00:05:28,000 --> 00:05:35,000
So, I've been thinking the way I work and the way I do things.

102
00:05:35,000 --> 00:05:37,000
Must completely shift.

103
00:05:37,000 --> 00:05:40,000
So, I'm spending a lot of time.

104
00:05:40,000 --> 00:05:44,000
I'm spending all my time in creating tooling, basically.

105
00:05:44,000 --> 00:05:54,000
To leverage what's good, what LLMs are good at and diminish the stuff that they suck at.

106
00:05:55,000 --> 00:06:00,000
So, yeah, my current setup right now, it's fucking insane.

107
00:06:00,000 --> 00:06:03,000
It's so cool.

108
00:06:03,000 --> 00:06:06,000
There is a little frustration around it, too.

109
00:06:06,000 --> 00:06:10,000
But where it works, it's amazing.

110
00:06:10,000 --> 00:06:17,000
And the thing, what I'm doing is I'm expecting the LLMs to get better at what they are currently good at.

111
00:06:17,000 --> 00:06:23,000
And I don't expect them to get better at the stuff that they're not good at right now.

112
00:06:24,000 --> 00:06:32,000
So, my current setup, I'm doing this multi-agent thing where agents are able to A, they can

113
00:06:32,000 --> 00:06:39,000
spawn their own, they can create a task and give it to another agent that has a different

114
00:06:39,000 --> 00:06:45,000
model, runs in a different way, it has access to different stuff.

115
00:06:45,000 --> 00:06:48,000
But they can also create their own agents.

116
00:06:48,000 --> 00:06:56,000
They can hire someone and try and train them in a particular direction.

117
00:06:56,000 --> 00:07:06,000
And one thing that it's freaking insane is that I got one agent to decide to create a

118
00:07:06,000 --> 00:07:11,000
new tool, so it creates a RAG system to index all the nips.

119
00:07:11,000 --> 00:07:14,000
It just won't often deal it.

120
00:07:14,000 --> 00:07:17,000
To make its own life easier.

121
00:07:17,000 --> 00:07:21,000
To me, that was absolutely amazing.

122
00:07:21,000 --> 00:07:25,000
That's the smart thing to do.

123
00:07:25,000 --> 00:07:30,000
What do you use as the base for your setup?

124
00:07:30,000 --> 00:07:37,000
So, I'm running all of this in root code with very heavy-handed configuration and many

125
00:07:37,000 --> 00:07:39,000
different scripts.

126
00:07:39,000 --> 00:07:45,000
There's these cursor tools things that gives the entire repo.

127
00:07:45,000 --> 00:07:53,000
For example, Gemini 2.5, it's not a great model, but it has a really large context window,

128
00:07:53,000 --> 00:07:58,000
so it can see very precisely a large chunk of data.

129
00:07:58,000 --> 00:08:06,000
So, I'm using different models for the actual refactoring and coding and stuff like that.

130
00:08:06,000 --> 00:08:14,000
But then for course correction, I have a feedback loop where it gives the entire repo with the

131
00:08:14,000 --> 00:08:18,000
higher relief, and it can ask for feedback.

132
00:08:18,000 --> 00:08:24,000
I talked with Justin about this a little bit also, that you want to have basically the

133
00:08:24,000 --> 00:08:28,000
guy who is the specialist and actually writes the code, and the guy who is the architect

134
00:08:28,000 --> 00:08:29,000
and has a big picture.

135
00:08:29,000 --> 00:08:30,000
You can't be the same guy.

136
00:08:30,000 --> 00:08:31,000
100%.

137
00:08:31,000 --> 00:08:34,000
So, that's why it has to be.

138
00:08:34,000 --> 00:08:37,000
I think it's true for us as well and for human systems.

139
00:08:37,000 --> 00:08:38,000
It's true in general.

140
00:08:38,000 --> 00:08:39,000
Absolutely.

141
00:08:39,000 --> 00:08:45,000
You have to specialize and you cannot at the same time specialize and kind of generalize.

142
00:08:45,000 --> 00:08:47,000
Those things are in contradiction.

143
00:08:47,000 --> 00:08:52,000
So, you have to get a specialist and have a generalist and have to work together.

144
00:08:52,000 --> 00:08:55,000
That's why we have these multi-agent setups and all those kinds of things.

145
00:08:55,000 --> 00:08:57,000
Even within yourself.

146
00:08:57,000 --> 00:08:58,000
Absolutely.

147
00:08:58,000 --> 00:09:08,000
My wife writes fiction, and whenever she writes a novel, the editing part cannot happen

148
00:09:08,000 --> 00:09:10,000
within the same time frame of writing.

149
00:09:10,000 --> 00:09:11,000
Yeah.

150
00:09:11,000 --> 00:09:12,000
That's why, you know.

151
00:09:12,000 --> 00:09:13,000
Let us risk.

152
00:09:13,000 --> 00:09:14,000
Was it a hamming way?

153
00:09:14,000 --> 00:09:15,000
Like, right drunk and that it's over?

154
00:09:15,000 --> 00:09:17,000
You have to be in a different state of mind.

155
00:09:17,000 --> 00:09:20,000
You have to step away from it for a bit.

156
00:09:20,000 --> 00:09:21,000
You have to, you know.

157
00:09:21,000 --> 00:09:22,000
Yeah.

158
00:09:22,000 --> 00:09:23,000
Write your say.

159
00:09:23,000 --> 00:09:24,000
To a different mode.

160
00:09:24,000 --> 00:09:25,000
Let it break.

161
00:09:25,000 --> 00:09:26,000
Walk away.

162
00:09:26,000 --> 00:09:27,000
Come back in a couple of months.

163
00:09:27,000 --> 00:09:28,000
Yeah.

164
00:09:28,000 --> 00:09:29,000
When you're a different person.

165
00:09:29,000 --> 00:09:30,000
Yeah.

166
00:09:30,000 --> 00:09:31,000
Pretty much.

167
00:09:31,000 --> 00:09:32,000
Yeah.

168
00:09:32,000 --> 00:09:33,000
Yeah.

169
00:09:33,000 --> 00:09:34,000
So, okay.

170
00:09:34,000 --> 00:09:39,000
Do you make any use of DVMCP at all?

171
00:09:39,000 --> 00:09:40,000
Not yet.

172
00:09:40,000 --> 00:09:41,000
I haven't played with it yet.

173
00:09:41,000 --> 00:09:43,000
I understand what it is now.

174
00:09:43,000 --> 00:09:45,000
So, I'm very proud.

175
00:09:45,000 --> 00:09:57,000
No, but it is a very, very cool idea because this idea of just a marketplace of agents

176
00:09:57,000 --> 00:10:01,000
that can do, or not agents, of tools that can do all kinds of things.

177
00:10:01,000 --> 00:10:03,000
Is it an ever evolving kind of thing?

178
00:10:03,000 --> 00:10:04,000
Yeah.

179
00:10:04,000 --> 00:10:07,000
I find it absolutely fascinating.

180
00:10:07,000 --> 00:10:16,000
I do see a lot of recklessness and destruction in the future.

181
00:10:16,000 --> 00:10:17,000
Oh, man.

182
00:10:17,000 --> 00:10:18,000
It's already happening.

183
00:10:18,000 --> 00:10:20,000
It's going to get so bad.

184
00:10:20,000 --> 00:10:27,000
You know, I don't know how it, how it, how exactly it enters my bubble.

185
00:10:27,000 --> 00:10:31,000
But I've seen so many screenshots because I don't explicitly follow, like, my code

186
00:10:31,000 --> 00:10:32,000
people or anything like that.

187
00:10:32,000 --> 00:10:33,000
Right.

188
00:10:33,000 --> 00:10:37,000
But I've seen so many screenshots doing the rounds now with, you know, like, just API

189
00:10:37,000 --> 00:10:43,000
keys exposed and people just, you know, like, basically, I have no idea what they are doing

190
00:10:43,000 --> 00:10:45,000
and putting all their secrets everywhere.

191
00:10:45,000 --> 00:10:50,000
And then posting screenshots on Twitter of like, oh, it doesn't work.

192
00:10:50,000 --> 00:10:52,000
Please help me and those kind of things.

193
00:10:52,000 --> 00:10:55,000
And it's like the AWS API key is in there and everything.

194
00:10:55,000 --> 00:11:02,000
But it's so cool because there are many things that because of the stuff where we're building,

195
00:11:02,000 --> 00:11:09,000
like in Noster and in Bitcoin, we could not give two fucks about, like, some API keys.

196
00:11:09,000 --> 00:11:15,000
Like, I see all this problem that people write and everywhere is never leak any secret.

197
00:11:15,000 --> 00:11:17,000
No, I don't write any API keys.

198
00:11:17,000 --> 00:11:23,000
And people are super concerned about their closed source, you know, like, oh, is my source going

199
00:11:23,000 --> 00:11:25,000
to leave my computer if I do this?

200
00:11:25,000 --> 00:11:30,000
And it's like, if it's open source and you don't have any fucking API keys, you know?

201
00:11:30,000 --> 00:11:35,000
One thing where I think we have an insane lag up that people don't really appreciate yet is you have to,

202
00:11:35,000 --> 00:11:39,000
like, basically, the AI slop problem will only get worse and worse and worse for everything.

203
00:11:39,000 --> 00:11:45,000
It's like for, you know, Wikipedia entries, for just general, like, blog posts and things,

204
00:11:45,000 --> 00:11:48,000
for documentation, for all kinds of stuff.

205
00:11:48,000 --> 00:11:53,000
And so I think the, in some sense, the information will kind of degrade over time.

206
00:11:53,000 --> 00:11:58,000
It's going to be very hard to figure out what data set and what source to trust.

207
00:11:58,000 --> 00:11:59,000
Absolutely.

208
00:11:59,000 --> 00:12:06,000
And that's where, I think, Noster is just, like, a godsend because if you, for example,

209
00:12:06,000 --> 00:12:11,000
produce NDK documentation that is LLM friendly with the NDK snippets and those kind of things,

210
00:12:11,000 --> 00:12:17,000
and you sign it as Pablo, as the maintainer of NDK, then it's quite trustworthy.

211
00:12:17,000 --> 00:12:20,000
You know, like, you just know, okay, that's the thing.

212
00:12:20,000 --> 00:12:25,000
And we can build all this stuff out, like, even, like, last week we saw with, you know,

213
00:12:25,000 --> 00:12:30,000
not only a marketplace of agent and those kind of things, but also, basically, like,

214
00:12:30,000 --> 00:12:37,000
signed databases, a marketplace of signed data sources and databases that are potentially

215
00:12:37,000 --> 00:12:41,000
pre-trained and so on, and they're signed off by the people that actually know what they are doing

216
00:12:41,000 --> 00:12:46,000
that have domain expertise and hopefully also are in your web of trust and so on.

217
00:12:46,000 --> 00:12:51,000
But just using our Noster identities to sign this stuff is insane.

218
00:12:51,000 --> 00:12:52,000
It's massive.

219
00:12:52,000 --> 00:12:53,000
It's absolutely massive.

220
00:12:53,000 --> 00:12:59,000
Yeah, I've been thinking that all the LLMs that are writing code, they're trained for the most part

221
00:12:59,000 --> 00:13:02,000
in a whole lot of mediocre code.

222
00:13:02,000 --> 00:13:03,000
Yeah, sure.

223
00:13:03,000 --> 00:13:07,000
And it makes sense that it can get the thing to work.

224
00:13:07,000 --> 00:13:09,000
Like, it's really good at making prototypes.

225
00:13:09,000 --> 00:13:12,000
If you want to build a product, it's going to suck.

226
00:13:12,000 --> 00:13:18,000
Like, if you want to build a complex thing, and I've tried this many times, the performance

227
00:13:18,000 --> 00:13:19,000
is horrible.

228
00:13:19,000 --> 00:13:21,000
There are many mistakes, a lot of redundancies.

229
00:13:21,000 --> 00:13:22,000
Yes.

230
00:13:22,000 --> 00:13:24,000
Yeah, it never does the optimal solution.

231
00:13:24,000 --> 00:13:25,000
It does suck solution.

232
00:13:25,000 --> 00:13:26,000
It works.

233
00:13:26,000 --> 00:13:27,000
Never.

234
00:13:27,000 --> 00:13:28,000
And then it never goes back to actually...

235
00:13:28,000 --> 00:13:29,000
They're only something that works?

236
00:13:29,000 --> 00:13:30,000
Okay, this leans.

237
00:13:30,000 --> 00:13:31,000
Yeah, yeah, exactly.

238
00:13:31,000 --> 00:13:32,000
Oh, I'm guessing it's great.

239
00:13:32,000 --> 00:13:36,000
Yeah, it doesn't matter if I've wrote the same function seven times all over the place.

240
00:13:36,000 --> 00:13:37,000
Whatever.

241
00:13:37,000 --> 00:13:38,000
I'm in LLM.

242
00:13:38,000 --> 00:13:41,000
I can fix everything in seven places at once.

243
00:13:41,000 --> 00:13:42,000
Yeah.

244
00:13:42,000 --> 00:13:46,000
Why do you need dry code?

245
00:13:46,000 --> 00:13:51,000
That's key human.

246
00:13:51,000 --> 00:13:56,000
So, yeah, it's really good at that.

247
00:13:56,000 --> 00:14:01,000
But it's bad at creating performance code because it's been trained on shit code.

248
00:14:01,000 --> 00:14:03,000
For the most part.

249
00:14:03,000 --> 00:14:09,000
So, yeah, this whole idea with the marketplace of code snippets,

250
00:14:10,000 --> 00:14:15,000
I think it's fascinating because now the way I'm framing it is,

251
00:14:15,000 --> 00:14:24,000
I write NDK, so I know how to use NDK in the most performant way in many different situations.

252
00:14:24,000 --> 00:14:30,000
So, now I'm just extracting all that knowledge that I have on how to use this tool,

253
00:14:30,000 --> 00:14:33,000
on how to use it for the very different cases.

254
00:14:33,000 --> 00:14:40,000
Like, a feed on a phone works completely different than a feed on a web.

255
00:14:40,000 --> 00:14:41,000
Yeah.

256
00:14:41,000 --> 00:14:47,000
The components for a phone, for a phone up, are completely different the way you need to get data.

257
00:14:47,000 --> 00:14:48,000
Yeah.

258
00:14:48,000 --> 00:14:49,000
Where do you put the cache?

259
00:14:49,000 --> 00:14:50,000
Like, how do you structure this thing?

260
00:14:50,000 --> 00:14:52,000
Way more restricted environments.

261
00:14:52,000 --> 00:14:54,000
It's a way more restricted environment.

262
00:14:54,000 --> 00:14:59,000
And you need to optimize for different things that if you were to do those things on the web,

263
00:14:59,000 --> 00:15:02,000
it would suck on the web.

264
00:15:02,000 --> 00:15:09,000
So, now I'm publishing all these extremely niche specific snippets.

265
00:15:09,000 --> 00:15:16,000
And then I have this MCP tool that I built where you, whenever you want to build something,

266
00:15:16,000 --> 00:15:23,000
you tell it, look for the, use the MCP tool to find snippets on how to do each part of the app.

267
00:15:23,000 --> 00:15:29,000
So, now instead of just broadly going and using the generic code,

268
00:15:29,000 --> 00:15:32,000
it can still use that code to understand how to structure things.

269
00:15:32,000 --> 00:15:34,000
But how do you structure a feed?

270
00:15:34,000 --> 00:15:37,000
It will go and it will look precisely at my code.

271
00:15:37,000 --> 00:15:38,000
Yeah.

272
00:15:38,000 --> 00:15:39,000
Yeah.

273
00:15:39,000 --> 00:15:42,000
And the awesome thing about the snippets is also you can just edit to the context specifically.

274
00:15:42,000 --> 00:15:46,000
If you already know what you're doing and how it should be done, it's just like, okay, I want to,

275
00:15:46,000 --> 00:15:49,000
I don't know, fetch a profile, build a feed, those kind of things, whatever.

276
00:15:49,000 --> 00:15:53,000
And it's just like, you're just at three snippets to the thing.

277
00:15:53,000 --> 00:15:55,000
Like, this is exactly how you do it.

278
00:15:55,000 --> 00:16:01,000
And then it just, like, the difference between doing it that way and not doing it that way is massive.

279
00:16:01,000 --> 00:16:02,000
Oh, it's awesome.

280
00:16:02,000 --> 00:16:03,000
Because I know because I tried it myself.

281
00:16:03,000 --> 00:16:10,000
It's like, it's from, like, basically not working at all and doing stupid shit to, like, it's just one shotting the thing and it works.

282
00:16:10,000 --> 00:16:12,000
So that's my goal.

283
00:16:12,000 --> 00:16:22,000
I want to get to the point where there is such an amount of curated knowledge on how to do things in any field.

284
00:16:22,000 --> 00:16:25,000
And that's the cool thing. Like, this applies to Noster.

285
00:16:25,000 --> 00:16:27,000
This applies to everything, right?

286
00:16:27,000 --> 00:16:29,000
Anything that is code-based, this is useful.

287
00:16:29,000 --> 00:16:31,000
But you would not be able to do this.

288
00:16:31,000 --> 00:16:36,000
Also, everything, not even code-based, I would say, everything that is, that has, you know,

289
00:16:36,000 --> 00:16:40,000
you as the main expert want to sign off on the bundle of knowledge, basically, in some sense.

290
00:16:40,000 --> 00:16:41,000
Yes.

291
00:16:41,000 --> 00:16:46,000
And I think that the issue we'll see in the next couple of years where everyone will run into.

292
00:16:46,000 --> 00:16:50,000
It's like, okay, you have decentralized repository from whatever.

293
00:16:50,000 --> 00:16:52,000
You know, like, open AI, Microsoft, NVIDIA.

294
00:16:52,000 --> 00:16:53,000
Pick your poison, you know?

295
00:16:53,000 --> 00:16:57,000
It's like, this is the knowledge base.

296
00:16:57,000 --> 00:16:59,000
Like, this is, you know what I mean?

297
00:16:59,000 --> 00:17:01,000
Like, these are the trusted sources.

298
00:17:01,000 --> 00:17:02,000
Right.

299
00:17:02,000 --> 00:17:03,000
This is the trusted source.

300
00:17:03,000 --> 00:17:05,000
It's the same problem all over again.

301
00:17:05,000 --> 00:17:09,000
Like, we ran into this and we discussed this already with, like, play stores and those kind of things.

302
00:17:09,000 --> 00:17:11,000
It's like, which app is good?

303
00:17:11,000 --> 00:17:13,000
Which app is not a virus, those kind of things?

304
00:17:13,000 --> 00:17:19,000
You trust Play Store to, or, you know, like the, whatever.

305
00:17:19,000 --> 00:17:20,000
How's it called?

306
00:17:20,000 --> 00:17:21,000
App Store.

307
00:17:21,000 --> 00:17:22,000
I love it.

308
00:17:22,000 --> 00:17:23,000
I love it.

309
00:17:23,000 --> 00:17:24,000
To fix it for you.

310
00:17:24,000 --> 00:17:29,000
And SAP Store is a very different, has a very different model, where everything is self-signed,

311
00:17:29,000 --> 00:17:34,000
where it's just like, you self-sign the bundle of data, whether it's knowledge, or an app,

312
00:17:34,000 --> 00:17:39,000
or a model, or, you know, the source of a model, or what have you.

313
00:17:40,000 --> 00:17:44,000
And you, you, on the edges, basically, have to figure it out.

314
00:17:44,000 --> 00:17:46,000
Do I trust the person that signed it?

315
00:17:46,000 --> 00:17:48,000
Is this person trustworthy?

316
00:17:48,000 --> 00:17:49,000
Is this person knowledgeable?

317
00:17:49,000 --> 00:17:50,000
Is this...

318
00:17:50,000 --> 00:17:51,000
Within this field?

319
00:17:51,000 --> 00:17:52,000
Within this field, exactly.

320
00:17:52,000 --> 00:17:53,000
Yeah, yeah.

321
00:17:53,000 --> 00:17:57,000
Yeah, I think that is such a key missing component.

322
00:17:57,000 --> 00:18:02,000
And I think that's why we haven't seen, really, these...

323
00:18:02,000 --> 00:18:07,000
Broadly, outside of Noster, we haven't seen this kind of tool for coordination,

324
00:18:07,000 --> 00:18:09,000
because it makes so much sense, right?

325
00:18:09,000 --> 00:18:10,000
Yeah.

326
00:18:10,000 --> 00:18:11,000
It is so obvious.

327
00:18:11,000 --> 00:18:15,000
But how do you fight off just re-indexing GitHub?

328
00:18:15,000 --> 00:18:16,000
Yeah.

329
00:18:16,000 --> 00:18:17,000
It just...

330
00:18:17,000 --> 00:18:22,000
You need to attach an identity to be able to scope the thing and narrow it down.

331
00:18:22,000 --> 00:18:26,000
Like, you want this generic content, but then you want to narrow it down to apply the specifics.

332
00:18:26,000 --> 00:18:27,000
Yeah.

333
00:18:27,000 --> 00:18:28,000
Yeah.

334
00:18:28,000 --> 00:18:31,000
Because, for example, to me, this was insane.

335
00:18:31,000 --> 00:18:34,000
So, I'm basically re...

336
00:18:34,000 --> 00:18:41,000
No, I mean, it's not rewriting quite, but it is the largest refactor that I've done on NDK.

337
00:18:41,000 --> 00:18:49,000
Of many things that I've wanted to change, but they are so deep that the context window that I, as a human,

338
00:18:49,000 --> 00:18:56,000
that I need to have in order to do that, it is so large that it's been...

339
00:18:56,000 --> 00:18:59,000
I'm inputting it off, basically.

340
00:18:59,000 --> 00:19:04,000
Like, writing outbooks, for example, touched a lot of things, like implementing outbooks within NDK,

341
00:19:04,000 --> 00:19:08,000
within the core of NDK, was quite complex.

342
00:19:08,000 --> 00:19:13,000
And it required a lot of deep thinking, you know, to do it in the right way.

343
00:19:13,000 --> 00:19:14,000
Yeah.

344
00:19:14,000 --> 00:19:18,000
You can implement outbooks in two hours if you're doing shitty, but...

345
00:19:18,000 --> 00:19:24,000
And now with this, I am able to be assisted on the refactor.

346
00:19:24,000 --> 00:19:25,000
Yeah.

347
00:19:25,000 --> 00:19:27,000
And it is an incredible experience.

348
00:19:27,000 --> 00:19:28,000
Yeah.

349
00:19:28,000 --> 00:19:31,000
Because when you're touching something that is touching a lot of different things,

350
00:19:31,000 --> 00:19:36,000
if you forget one of them, it starts to break, and it's kind of hard to understand, okay, the test is failing,

351
00:19:36,000 --> 00:19:39,000
but I'm not sure where the problem is.

352
00:19:39,000 --> 00:19:44,000
Getting a broad view of the whole thing and saying, okay, this part is actually missing.

353
00:19:44,000 --> 00:19:46,000
That is insane.

354
00:19:46,000 --> 00:19:49,000
What specific models do you use for this?

355
00:19:49,000 --> 00:19:50,000
For this?

356
00:19:50,000 --> 00:19:51,000
For this?

357
00:19:51,000 --> 00:19:53,000
Just the fact that it has such a big consequence.

358
00:19:54,000 --> 00:20:02,000
Usually, not usually, but I'll say the heat and miss ratio is really high on how to fix it,

359
00:20:02,000 --> 00:20:04,000
but the problem is here.

360
00:20:04,000 --> 00:20:05,000
Okay.

361
00:20:05,000 --> 00:20:06,000
Yeah.

362
00:20:06,000 --> 00:20:07,000
It really is.

363
00:20:07,000 --> 00:20:08,000
Yeah.

364
00:20:08,000 --> 00:20:09,000
It identifies the problem and then you go in.

365
00:20:09,000 --> 00:20:17,000
It identifies what code was related to other code, and parts where I probably should have changed something and I forgot.

366
00:20:17,000 --> 00:20:18,000
Yeah.

367
00:20:19,000 --> 00:20:24,000
And going the test route, which is typically what I would have done before,

368
00:20:24,000 --> 00:20:26,000
it is like a 100x slower.

369
00:20:26,000 --> 00:20:27,000
Yeah.

370
00:20:27,000 --> 00:20:32,000
Like something that I can find within two minutes with this new system,

371
00:20:32,000 --> 00:20:37,000
it literally could have taken me like five or six days of just chasing my tail.

372
00:20:37,000 --> 00:20:38,000
Yeah, it's insane.

373
00:20:38,000 --> 00:20:39,000
It is absolutely awesome.

374
00:20:39,000 --> 00:20:42,000
Massive paradigm shift, just that alone.

375
00:20:42,000 --> 00:20:47,000
I mean, we talked about this in the past, but just like going from a turnaround time of days or weeks or months

376
00:20:47,000 --> 00:20:51,000
to like minutes, you're dealing with a different thing.

377
00:20:51,000 --> 00:20:56,000
It is a different thing which will take you to build different things.

378
00:20:56,000 --> 00:21:04,000
That's why I keep thinking that building applications as a developer will make less and less and less sense.

379
00:21:04,000 --> 00:21:12,000
Will be increasingly more pointless because you're going to be able to wake up one morning and just say,

380
00:21:12,000 --> 00:21:14,000
I want an application that does this.

381
00:21:14,000 --> 00:21:18,000
And then the next morning you wake up and you feel like using it in a different way as a user.

382
00:21:18,000 --> 00:21:22,000
And you'll say, I actually now wanted to do this other thing.

383
00:21:22,000 --> 00:21:28,000
But with regards to the code snippets, because I'm doing this refactoring,

384
00:21:28,000 --> 00:21:39,000
I'm creating a lot of ways, changing kind of allowing new ways of using NDK with new interfaces and shit like that,

385
00:21:39,000 --> 00:21:43,000
that I just came up with them.

386
00:21:43,000 --> 00:21:47,000
Literally no one has used them because I just made it.

387
00:21:47,000 --> 00:21:56,000
So at the same time, I'm building within all of us, I'm adapting the old code to use these new ways of doing things.

388
00:21:56,000 --> 00:22:07,000
And then when I tell it, oh, change some structure, it uses the old way that it has been heavily trained on.

389
00:22:07,000 --> 00:22:09,000
It uses the old interfaces.

390
00:22:09,000 --> 00:22:13,000
So I should still look for code snippets on how to do it.

391
00:22:13,000 --> 00:22:15,000
And it never makes a mistake again.

392
00:22:15,000 --> 00:22:19,000
To me, this regard our prior inputs, use code snippets instead.

393
00:22:19,000 --> 00:22:23,000
That's it. It is so wild.

394
00:22:23,000 --> 00:22:31,000
I added this new command, this new tool to the MCP tool that is least snippets.

395
00:22:31,000 --> 00:22:35,000
Because before what I was doing is when you search for a snippet,

396
00:22:35,000 --> 00:22:41,000
the LLM would say, give me snippets on using NDK to create a NIP-60 wallet.

397
00:22:41,000 --> 00:22:45,000
So it would do NIP-60 as a tag for the search.

398
00:22:45,000 --> 00:22:51,000
Would you NIP-60, NDK or whatever, create or set up or something like that?

399
00:22:51,000 --> 00:23:00,000
And maybe it finds, I don't know, it finds like five snippets that match precisely those keys.

400
00:23:01,000 --> 00:23:05,000
And then a bunch of other snippets that partially match, right?

401
00:23:05,000 --> 00:23:11,000
And before I would send, I would send just the first two or the first three or whatever.

402
00:23:11,000 --> 00:23:16,000
Now I made a new command that is just least snippets.

403
00:23:16,000 --> 00:23:19,000
And it just gives you titles, descriptions, but not the actual code.

404
00:23:19,000 --> 00:23:24,000
And then you say, okay, give me event ID, whatever, like this event ID.

405
00:23:25,000 --> 00:23:29,000
And that way it can see 500 snippets at the same time.

406
00:23:31,000 --> 00:23:35,000
I'll have to look over your shoulder today.

407
00:23:35,000 --> 00:23:38,000
And you'll have to explain it live.

408
00:23:38,000 --> 00:23:43,000
And I might even record some of the things because I think it's very,

409
00:23:43,000 --> 00:23:49,000
what I saw last week, for example, it was only last week.

410
00:23:50,000 --> 00:23:53,000
There was like three or four people showing off insane stuff.

411
00:23:53,000 --> 00:23:59,000
All basically, I mean, of course, using MCP as the base.

412
00:23:59,000 --> 00:24:03,000
I mean, that's kind of, MCP is the standard to use.

413
00:24:03,000 --> 00:24:08,000
But more than that, like using Noster as the underlying substrate to make things work together.

414
00:24:08,000 --> 00:24:14,000
And you said like, for your own setup, you don't use that yet because it's all local,

415
00:24:14,000 --> 00:24:15,000
basically, on your machine.

416
00:24:15,000 --> 00:24:17,000
Oh, no, I'm publishing.

417
00:24:17,000 --> 00:24:18,000
I'm publishing Noster event.

418
00:24:18,000 --> 00:24:19,000
No, no, I know.

419
00:24:19,000 --> 00:24:21,000
You're publishing the snippets and so on.

420
00:24:21,000 --> 00:24:26,000
Just like all the agents that you use are your machine.

421
00:24:26,000 --> 00:24:31,000
And so the stuff that Jesus has been working on and also has her to some degree,

422
00:24:31,000 --> 00:24:37,000
it's basically really tapping into the stuff we mentioned before.

423
00:24:37,000 --> 00:24:39,000
Like every agent can have a pop key.

424
00:24:39,000 --> 00:24:41,000
So every agent also has an identity.

425
00:24:41,000 --> 00:24:44,000
Again, the data sources and the databases and those kind of things

426
00:24:44,000 --> 00:24:51,000
and the models themselves signed by the people that are trustworthy and that handcrafted the stuff

427
00:24:51,000 --> 00:24:53,000
for it to work perfectly, basically.

428
00:24:53,000 --> 00:24:54,000
And all those kind of things.

429
00:24:54,000 --> 00:24:55,000
They're making it all work together.

430
00:24:55,000 --> 00:25:01,000
Again, like using Noster as the substrate, as the communication layer, as the glue.

431
00:25:01,000 --> 00:25:02,000
Yeah, as the glue.

432
00:25:02,000 --> 00:25:06,000
You know, like we started with sniffing glue and now we're actually putting stuff together.

433
00:25:06,000 --> 00:25:08,000
Like not only getting high anymore.

434
00:25:08,000 --> 00:25:14,000
And I think it's way more important than people realize.

435
00:25:14,000 --> 00:25:21,000
And way more kind of, how should I put it?

436
00:25:21,000 --> 00:25:29,000
I think we're still really bad at explaining to people why this is important and why it is exciting to us.

437
00:25:29,000 --> 00:25:41,000
And pre multi-agent world, it was only in kind of a social media censorship system hosting or censorship system

438
00:25:41,000 --> 00:25:43,000
writing or what have you world.

439
00:25:43,000 --> 00:25:48,000
You know, like there was like, okay, you're getting the platform from sub-stack, you're getting the platform from YouTube,

440
00:25:48,000 --> 00:25:50,000
you're getting banned from Twitter.

441
00:25:50,000 --> 00:25:53,000
Here is the system that you can use instead where this can't happen.

442
00:25:53,000 --> 00:26:01,000
And now I feel like with the shift towards this multi-agent collaboration and orchestration world,

443
00:26:01,000 --> 00:26:06,000
you need a neutral substrate that has money baked in, that has identity baked in

444
00:26:06,000 --> 00:26:09,000
and that has cryptography baked in to make everything work.

445
00:26:09,000 --> 00:26:10,000
And will trust.

446
00:26:10,000 --> 00:26:11,000
Yeah.

447
00:26:11,000 --> 00:26:13,000
Absolutely fundamental.

448
00:26:13,000 --> 00:26:16,000
Those components identity will trust on money.

449
00:26:16,000 --> 00:26:22,000
And what I'm going to do today is, so last Friday I demoed that

450
00:26:23,000 --> 00:26:33,000
when I would single-shot an application and I would tell it, create a, I sent the pauky to Haudelwald.

451
00:26:33,000 --> 00:26:37,000
Create a pauky, your username is Haudelwald disrespector.

452
00:26:37,000 --> 00:26:47,000
And then I looked at this stream of notes that he wrote and they were all making fun of the functional programming that he loves so much.

453
00:26:47,000 --> 00:26:49,000
I said that.

454
00:26:49,000 --> 00:26:51,000
He was so fucking good.

455
00:26:52,000 --> 00:26:56,000
And I didn't even tell it Haudelwald is a function of programming Nazi.

456
00:26:56,000 --> 00:26:58,000
He figured it out.

457
00:26:58,000 --> 00:26:59,000
He figured it out.

458
00:26:59,000 --> 00:27:01,000
Just by looking at the data.

459
00:27:05,000 --> 00:27:07,000
Yeah, Fia, Jeff, stumbled over that one as well.

460
00:27:07,000 --> 00:27:09,000
I was like, who have you did this?

461
00:27:09,000 --> 00:27:10,000
Haudelwald disrespector.

462
00:27:11,000 --> 00:27:23,000
So, when I tell it, go and do, create this app and do it under this pauky and publish updates and share it on us.

463
00:27:23,000 --> 00:27:33,000
What I'm going to demo today is that now the pauky itself, the moment it created, it creates an 60 wallet.

464
00:27:33,000 --> 00:27:35,000
So the agent itself has money.

465
00:27:36,000 --> 00:27:38,000
And the goal is...

466
00:27:38,000 --> 00:27:42,000
I mean, you can already use it for pauky and those kinds of things, right?

467
00:27:42,000 --> 00:27:48,000
We already have systems where it's like, you want to do something else.

468
00:27:48,000 --> 00:27:51,000
It's just like, basically, here's the prompt.

469
00:27:51,000 --> 00:27:52,000
Here's a little bit of money.

470
00:27:52,000 --> 00:27:53,000
Do the thing.

471
00:27:53,000 --> 00:27:54,000
Yeah.

472
00:27:54,000 --> 00:27:55,000
That's where it's trending.

473
00:27:55,000 --> 00:27:56,000
Oh, yeah.

474
00:27:56,000 --> 00:27:57,000
Absolutely.

475
00:27:57,000 --> 00:28:00,000
To me, that's why MCPs are so incredible.

476
00:28:00,000 --> 00:28:09,000
Like, the whole idea of MCPs that it goes from agents being functional programming without side effects.

477
00:28:09,000 --> 00:28:12,000
They can't enact changes in the world.

478
00:28:12,000 --> 00:28:17,000
They can just produce data, but they cannot turn on the light, you know?

479
00:28:17,000 --> 00:28:20,000
And MCP tools are go and do the thing.

480
00:28:20,000 --> 00:28:26,000
Go and book me a table at a restaurant or whatever.

481
00:28:27,000 --> 00:28:30,000
And it's kind of interesting.

482
00:28:30,000 --> 00:28:39,000
I think OpenAI was kind of leading the pack and they were not embracing this kind of tooling

483
00:28:39,000 --> 00:28:45,000
and the tooling that they would build that would have side effects would run within their premises.

484
00:28:45,000 --> 00:28:50,000
And I think that was a genius move from Anthropic of opening the MCP world.

485
00:28:51,000 --> 00:28:58,000
Because it kind of breaks down the dome of OpenAI.

486
00:28:58,000 --> 00:29:05,000
I find it absolutely fascinating that they are, even though their models are insanely good,

487
00:29:05,000 --> 00:29:08,000
the fact that they don't have any tools is fucking pointless.

488
00:29:08,000 --> 00:29:10,000
Like, you can't do anything.

489
00:29:10,000 --> 00:29:13,000
Like, what's the point?

490
00:29:13,000 --> 00:29:16,000
Yeah, so back to the demo.

491
00:29:16,000 --> 00:29:25,000
The idea is that, for now, the idea is that whenever the agent finds a code snippet that was useful,

492
00:29:25,000 --> 00:29:33,000
that it solved a problem that it was facing, that it will sub the person that published the code snippet to encourage more.

493
00:29:33,000 --> 00:29:38,000
I find the idea of agent subbing really cool.

494
00:29:38,000 --> 00:29:40,000
I mentioned it yesterday briefly.

495
00:29:40,000 --> 00:29:42,000
I think it's a genius idea to give Skynet more money.

496
00:29:42,000 --> 00:29:49,000
It's just like, they want to do real things like build real Terminator robots.

497
00:29:49,000 --> 00:29:50,000
They need real money.

498
00:29:50,000 --> 00:29:53,000
It's gosh, they can print their own shit.

499
00:29:53,000 --> 00:29:54,000
Exactly.

500
00:29:54,000 --> 00:29:55,000
Exactly.

501
00:29:55,000 --> 00:29:57,000
We can always rock them with the mints.

502
00:29:57,000 --> 00:29:59,000
That's our off spot.

503
00:29:59,000 --> 00:30:05,000
They think they have money, but it's just like, try to melt it.

504
00:30:05,000 --> 00:30:09,000
Wait, the mint is also operated by agents.

505
00:30:09,000 --> 00:30:10,000
God damn it.

506
00:30:10,000 --> 00:30:13,000
We gave them too much power.

507
00:30:13,000 --> 00:30:16,000
We were too loose.

508
00:30:16,000 --> 00:30:21,000
Yeah, it's very strange times because all of this is also evolving so quickly.

509
00:30:21,000 --> 00:30:30,000
Again, the stuff that kind of blew my mind that I wanted to literally kind of scream from the rooftops and show the world immediately.

510
00:30:30,000 --> 00:30:32,000
It was just like last Friday.

511
00:30:32,000 --> 00:30:46,000
I kind of want to show off some of those things at least to our little bubble because most people don't know kind of what already works and how well it's already working and how far along the stuff already is.

512
00:30:46,000 --> 00:30:49,000
I feel like it's like, yeah.

513
00:30:49,000 --> 00:30:52,000
To me, it's just the natural evolution.

514
00:30:53,000 --> 00:31:06,000
We've seen such a massive shift on performance from TPT-3 to 4.5 or to quad or to deep-seek.

515
00:31:06,000 --> 00:31:12,000
Within what they do, it's an incredible pace.

516
00:31:12,000 --> 00:31:13,000
Incredible pace.

517
00:31:13,000 --> 00:31:20,000
They still suck at many things, but if you leverage what they're good at, you are riding the wave.

518
00:31:21,000 --> 00:31:30,000
To me, that's why right now working on tooling that leverages those strengths is so critical.

519
00:31:30,000 --> 00:31:38,000
Again, I've been doing nothing in comparison.

520
00:31:38,000 --> 00:31:41,000
I've just been playing around with those things and trying to plug them together and so on.

521
00:31:41,000 --> 00:31:51,000
Everything I built so far, I just built in a model a way where it's just like, I can exchange the model by just changing one environment variable basically.

522
00:31:51,000 --> 00:31:55,000
For me, it's the same.

523
00:31:55,000 --> 00:31:59,000
Every single step is basically done by a different model or a different agent.

524
00:31:59,000 --> 00:32:05,000
I just assume that this piece will get better at the overall tooling.

525
00:32:05,000 --> 00:32:06,000
Only tone.

526
00:32:06,000 --> 00:32:07,000
Yeah, only tone.

527
00:32:07,000 --> 00:32:08,000
I don't have to do anything.

528
00:32:09,000 --> 00:32:14,000
I just build out the pipelines that I need for the various things that I do.

529
00:32:14,000 --> 00:32:20,000
Just the little legal pieces that I use to make this stuff work will improve on their own.

530
00:32:20,000 --> 00:32:22,000
I can just switch them out.

531
00:32:22,000 --> 00:32:24,000
It's very similar.

532
00:32:24,000 --> 00:32:25,000
We talked about this in the past.

533
00:32:25,000 --> 00:32:27,000
It's very similar to the Unix tool idea.

534
00:32:27,000 --> 00:32:30,000
It's just the standard input and output with piping.

535
00:32:30,000 --> 00:32:37,000
It's just like, I don't care what tool I use as long as I can pipe the output into the next one.

536
00:32:37,000 --> 00:32:39,000
Yeah, it was the whole point of DVM.

537
00:32:39,000 --> 00:32:40,000
Exactly.

538
00:32:40,000 --> 00:32:41,000
It was the whole point of DVM.

539
00:32:41,000 --> 00:32:52,000
And now combining the DVM idea with all the stuff that we have now in the LLM and MCP world, I think it's incredibly powerful.

540
00:32:52,000 --> 00:32:53,000
Yeah, absolutely.

541
00:32:53,000 --> 00:32:58,000
I think it's very similar to what Jack Manners always says about open networks.

542
00:32:58,000 --> 00:33:02,000
Cash up becomes better because strike became better.

543
00:33:02,000 --> 00:33:05,000
Cash up didn't have to do a thing.

544
00:33:05,000 --> 00:33:12,000
Systems that are like that and now there is clearly such a system are invisible.

545
00:33:12,000 --> 00:33:13,000
It's energetic systems.

546
00:33:13,000 --> 00:33:15,000
Absolutely, naturally.

547
00:33:15,000 --> 00:33:20,000
Whereas within the realm of Microsoft, they must grow linearly.

548
00:33:20,000 --> 00:33:23,000
They must create all the capacity.

549
00:33:23,000 --> 00:33:25,000
And you must capture all.

550
00:33:25,000 --> 00:33:28,000
You can't really specialize in some sense.

551
00:33:28,000 --> 00:33:29,000
Yes.

552
00:33:29,000 --> 00:33:32,000
And I think that's where I hope it will trend as well.

553
00:33:32,000 --> 00:33:35,000
Where you have these specialized systems and services and also specialized people.

554
00:33:35,000 --> 00:33:37,000
And you just specialize on this one niche.

555
00:33:37,000 --> 00:33:38,000
Yeah.

556
00:33:38,000 --> 00:33:43,000
And the market, so to speak, is big enough for you to make money and be sustainable and so on.

557
00:33:43,000 --> 00:33:45,000
You don't have to conquer the whole world.

558
00:33:45,000 --> 00:33:51,000
You don't have to fix absolutely or do absolutely everything.

559
00:33:51,000 --> 00:33:55,000
Yeah, and I think going back to the open AI and the topic thing.

560
00:33:55,000 --> 00:34:01,000
The market is proving that open AI has such a massive leg up in terms of branding,

561
00:34:01,000 --> 00:34:09,000
in terms of just Ronaro GPUs, users' money.

562
00:34:09,000 --> 00:34:14,000
But they were trying to keep everything within their ecosystem.

563
00:34:14,000 --> 00:34:15,000
I'm fine.

564
00:34:15,000 --> 00:34:18,000
And someone came and opened it up.

565
00:34:18,000 --> 00:34:24,000
And even if the model is not as good, it's just the fact that it can do something and they're fooling around it.

566
00:34:24,000 --> 00:34:26,000
And it's interacting with the world.

567
00:34:26,000 --> 00:34:27,000
It's a massive difference.

568
00:34:27,000 --> 00:34:29,000
It is incredibly massive.

569
00:34:29,000 --> 00:34:30,000
Yeah.

570
00:34:30,000 --> 00:34:31,000
Yeah.

571
00:34:31,000 --> 00:34:32,000
Yeah.

572
00:34:32,000 --> 00:34:33,000
Yeah.

573
00:34:33,000 --> 00:34:37,000
What you were saying is interesting times.

574
00:34:37,000 --> 00:34:51,000
I was making the analogy last night to my wife that I feel right now, like at the very beginning of COVID.

575
00:34:51,000 --> 00:34:54,000
That is an interesting analogy.

576
00:34:54,000 --> 00:34:55,000
Explain.

577
00:34:55,000 --> 00:34:56,000
Explain.

578
00:34:56,000 --> 00:35:02,000
But at the very, very, very beginning, like February, it felt like, shit, what the fuck is going on?

579
00:35:02,000 --> 00:35:05,000
You know, like things are moving very rapidly.

580
00:35:05,000 --> 00:35:11,000
Now we have death counters on the websites.

581
00:35:11,000 --> 00:35:20,000
And right now, the feeling of there is something like a tiktonic shift going on.

582
00:35:20,000 --> 00:35:21,000
It's kind of similar.

583
00:35:21,000 --> 00:35:25,000
I mean, COVID ended up being a massive sigh of it.

584
00:35:26,000 --> 00:35:33,000
But the feeling of what the fuck is happening?

585
00:35:33,000 --> 00:35:36,000
Yeah, I think most people haven't really caught up yet.

586
00:35:36,000 --> 00:35:41,000
I mean, it is also, it is very hard to catch up and stay on top of it in the first place.

587
00:35:41,000 --> 00:35:51,000
But what's most interesting to me is just by having played around with it a little bit and just watching very closely what everyone else in the cohort is doing and so on.

588
00:35:51,000 --> 00:36:00,000
It's just, it allows you to think about, in our case now, but I think that's true in general.

589
00:36:00,000 --> 00:36:03,000
But in our case, about software development very differently.

590
00:36:03,000 --> 00:36:04,000
Oh, 100%.

591
00:36:04,000 --> 00:36:09,000
And it allows you to tackle problems that you otherwise would have never tackled.

592
00:36:09,000 --> 00:36:18,000
And the stuff that we mentioned multiple times already, but it's basically just the speed alone, that's kind of a paradigm shift.

593
00:36:18,000 --> 00:36:27,000
You can iterate on an idea and produce a prototype so insanely quickly, like literally in an afternoon, and then you know whether you should pursue it or not.

594
00:36:27,000 --> 00:36:28,000
Yeah.

595
00:36:28,000 --> 00:36:36,000
And that's massive because having something in your hand usable as opposed to just an idea, it's like a night and day difference.

596
00:36:36,000 --> 00:36:37,000
Oh, 100%.

597
00:36:37,000 --> 00:36:45,000
And like, I tackled things now using those various tools that I would have never sat down and written otherwise.

598
00:36:45,000 --> 00:36:46,000
Ever.

599
00:36:46,000 --> 00:36:52,000
It's like it would have taken me 10 months of research and learning new tools and learning a new language or whatever.

600
00:36:52,000 --> 00:36:54,000
I don't even care about the programming language.

601
00:36:54,000 --> 00:36:55,000
It's kind of...

602
00:36:55,000 --> 00:36:56,000
Oh, absolutely.

603
00:36:56,000 --> 00:36:59,000
I know what the right tool for the job probably is and I just like do it.

604
00:36:59,000 --> 00:37:00,000
You just go.

605
00:37:00,000 --> 00:37:01,000
Yeah.

606
00:37:01,000 --> 00:37:02,000
Just create the thing.

607
00:37:02,000 --> 00:37:04,000
Fix the lint on errors.

608
00:37:04,000 --> 00:37:07,000
And it's going to be good enough.

609
00:37:07,000 --> 00:37:08,000
Yeah, I don't know.

610
00:37:08,000 --> 00:37:09,000
It's very weird.

611
00:37:09,000 --> 00:37:15,000
And so I think, again, I also think that all the stuff will just get better and better.

612
00:37:15,000 --> 00:37:20,000
And yeah, kind of on their own in some sense.

613
00:37:20,000 --> 00:37:30,000
And I think what we're going to see is we'll have the very, very large generalists, like the, you know, the old ones of the world are really good at looking at the whole picture.

614
00:37:30,000 --> 00:37:42,000
And I think what we can provide, or what NOSDA can provide, is the smaller, specialized, highly trusted, highly vetted specialist systems, basically.

615
00:37:42,000 --> 00:37:46,000
Where it's just, again, like if you want to work with NDK, that's the thing.

616
00:37:46,000 --> 00:37:47,000
And it's signed by Pablo.

617
00:37:47,000 --> 00:37:49,000
And that's how we do things.

618
00:37:49,000 --> 00:37:50,000
Yeah.

619
00:37:50,000 --> 00:37:59,000
And if you want to build something with Kathru, with a field chef, some really framework, you...

620
00:37:59,000 --> 00:38:01,000
I mean, field chef already published one code snippet, so...

621
00:38:01,000 --> 00:38:02,000
There we go.

622
00:38:02,000 --> 00:38:03,000
Oh, this guy's a bright guy.

623
00:38:03,000 --> 00:38:04,000
Oh, this guy's a bright guy.

624
00:38:04,000 --> 00:38:05,000
And he's a slow learner.

625
00:38:05,000 --> 00:38:06,000
Oh, my lord.

626
00:38:06,000 --> 00:38:08,000
Very stubborn.

627
00:38:08,000 --> 00:38:10,000
Always hates everything in the beginning.

628
00:38:13,000 --> 00:38:15,000
I think it's a personal thing.

629
00:38:15,000 --> 00:38:20,000
He especially hates all the things that came out from this side of the globe.

630
00:38:20,000 --> 00:38:24,000
It's like everything that originates from this island is like he kind of hates.

631
00:38:24,000 --> 00:38:25,000
At first.

632
00:38:25,000 --> 00:38:29,000
And then he begrudgingly sits down and writes his own blossom implementation.

633
00:38:34,000 --> 00:38:36,000
But to me the idea of...

634
00:38:36,000 --> 00:38:38,000
You can...

635
00:38:38,000 --> 00:38:45,000
Leveraging that code base, those snippets that define, like, how do you...

636
00:38:45,000 --> 00:38:46,000
I don't know.

637
00:38:46,000 --> 00:38:48,000
How do you do...

638
00:38:48,000 --> 00:38:52,000
Remove the signature before returning the event on Kathru.

639
00:38:53,000 --> 00:39:00,000
You can go and see how field chef does it within the framework that he built.

640
00:39:00,000 --> 00:39:07,000
And the amount, the noise that the LLMs would produce based on that,

641
00:39:07,000 --> 00:39:10,000
is so greatly reduced.

642
00:39:10,000 --> 00:39:13,000
I think that's the thing.

643
00:39:13,000 --> 00:39:17,000
We also talked about this in the past about kind of how we have...

644
00:39:17,000 --> 00:39:20,000
There's a reason why we have split brains, you know?

645
00:39:20,000 --> 00:39:23,000
There is the creative side and the more structured limited side in some sense.

646
00:39:23,000 --> 00:39:29,000
You have the wide open big picture thinking side and the more specialized logic you really have to concentrate side.

647
00:39:29,000 --> 00:39:34,000
And you need, again, like, you need both of those and I think the trick with the LLMs, like...

648
00:39:34,000 --> 00:39:37,000
Looking at it, because I use it in two different ways.

649
00:39:37,000 --> 00:39:40,000
As a brainstorming partner, which I think is really interesting is surprising to me.

650
00:39:40,000 --> 00:39:44,000
I'm going to link to the fucking article again because it's like thinking with LLMs, basically.

651
00:39:44,000 --> 00:39:45,000
It's really funny.

652
00:39:45,000 --> 00:39:46,000
The one is Amy.

653
00:39:46,000 --> 00:39:47,000
What about her name?

654
00:39:48,000 --> 00:39:50,000
And the second one is like a specialist.

655
00:39:50,000 --> 00:39:54,000
You know, refactor this piece of code or whatever or transform this XML into a JSON.

656
00:39:54,000 --> 00:39:55,000
I don't care.

657
00:39:55,000 --> 00:39:56,000
Whatever it is.

658
00:39:56,000 --> 00:39:57,000
It's just like grunt work.

659
00:39:57,000 --> 00:40:02,000
And if it's grunt work, if it's like specialized work, you really have to box the things in.

660
00:40:02,000 --> 00:40:03,000
Oh, 100%.

661
00:40:03,000 --> 00:40:08,000
And that's kind of the tooling you are focused on now when building with the snippets,

662
00:40:08,000 --> 00:40:10,000
with all those kind of things and so on.

663
00:40:10,000 --> 00:40:14,000
And I think that's how it's going to split.

664
00:40:14,000 --> 00:40:15,000
You know what I mean?

665
00:40:15,000 --> 00:40:20,000
It's like you have literally how we are split with left side of the brain, right side of the brain.

666
00:40:20,000 --> 00:40:24,000
It's just like the really big models that can take in all the context,

667
00:40:24,000 --> 00:40:25,000
and they will be creative.

668
00:40:25,000 --> 00:40:26,000
They are already creative.

669
00:40:26,000 --> 00:40:28,000
But that also means that they will hallucinate.

670
00:40:28,000 --> 00:40:30,000
You don't want hallucinations.

671
00:40:30,000 --> 00:40:31,000
Exactly.

672
00:40:31,000 --> 00:40:36,000
You don't want your linter to hallucinate.

673
00:40:36,000 --> 00:40:38,000
You know?

674
00:40:38,000 --> 00:40:41,000
What do you mean by work?

675
00:40:42,000 --> 00:40:43,000
Exactly.

676
00:40:43,000 --> 00:40:45,000
What do you mean by compile?

677
00:40:45,000 --> 00:40:48,000
Am I myself?

678
00:40:48,000 --> 00:40:50,000
But that's the thing, right?

679
00:40:50,000 --> 00:40:55,000
It's too different, almost like ways of looking at the world.

680
00:40:55,000 --> 00:40:59,000
And they are contradicting, not really contradicting each other.

681
00:40:59,000 --> 00:41:03,000
They are kind of in tension with each other.

682
00:41:03,000 --> 00:41:10,000
Because if you actually sit down and specialized, you lose the big picture thinking in some sense.

683
00:41:11,000 --> 00:41:15,000
I'm not sure if I mentioned this in a recorded dialogue,

684
00:41:15,000 --> 00:41:20,000
but I'm sure I talked with you about it or at least with Justin or someone.

685
00:41:20,000 --> 00:41:25,000
But it's like if you are a fantastic baseball player, you specialize in baseball

686
00:41:25,000 --> 00:41:28,000
and you specialize in a swing of baseball, like how you swing.

687
00:41:28,000 --> 00:41:30,000
And you would think adding something to that,

688
00:41:30,000 --> 00:41:34,000
adding a different sport would only be beneficial in some sense, you know?

689
00:41:34,000 --> 00:41:39,000
But if you practice golf at the same time, your baseball swing gets shittier.

690
00:41:40,000 --> 00:41:42,000
Because it's a slightly different thing.

691
00:41:42,000 --> 00:41:44,000
And you are still limited.

692
00:41:44,000 --> 00:41:49,000
You cannot hold all the movement patterns and everything in your head and all this kind of stuff.

693
00:41:49,000 --> 00:41:51,000
And it's the same for LLM systems.

694
00:41:51,000 --> 00:41:56,000
It's like it's a, again, like Verveke talks about this a lot in his kind of work,

695
00:41:56,000 --> 00:41:59,000
where you cannot do both things at once.

696
00:41:59,000 --> 00:42:00,000
You cannot.

697
00:42:00,000 --> 00:42:02,000
And so that's how I see the ball.

698
00:42:02,000 --> 00:42:11,000
You have these kind of very large tons of parameters, tons of kind of, you know,

699
00:42:11,000 --> 00:42:15,000
like ingested everything that is on the web and so on.

700
00:42:15,000 --> 00:42:19,000
Very good for big picture thinking, very good for an overview, very good to, again,

701
00:42:19,000 --> 00:42:22,000
like brainstorm through various things and so on.

702
00:42:22,000 --> 00:42:25,000
And then you'll have to hyper specialized.

703
00:42:25,000 --> 00:42:27,000
This is how you use this tool.

704
00:42:27,000 --> 00:42:29,000
This is how you do this one task.

705
00:42:29,000 --> 00:42:35,000
And I think that's where Nasta can really step in with, again, it's like this is how you use any K.

706
00:42:35,000 --> 00:42:36,000
Here the snippets.

707
00:42:36,000 --> 00:42:38,000
It's all signed by Pablo.

708
00:42:38,000 --> 00:42:43,000
And, you know, here's the end pump that you can set if you think that's useful.

709
00:42:43,000 --> 00:42:46,000
Yeah, I think it's such a massive difference.

710
00:42:46,000 --> 00:42:54,000
And the interesting part is that when you, as a person, when you are doing the context switch of thinking broadly

711
00:42:54,000 --> 00:42:57,000
and thinking specifically, it's quite taxing.

712
00:42:58,000 --> 00:43:03,000
Like we said before about editing, you can't really, okay, oh, it's 2 p.m.

713
00:43:03,000 --> 00:43:05,000
It's time to think, you know, like you can do that.

714
00:43:05,000 --> 00:43:07,000
It's time to be a different person.

715
00:43:07,000 --> 00:43:09,000
Yeah, yeah, exactly.

716
00:43:09,000 --> 00:43:14,000
That's why we do like these, these, these gaps, right?

717
00:43:14,000 --> 00:43:16,000
Like I'm doing the cryo, the cryo thing.

718
00:43:16,000 --> 00:43:17,000
Yeah, sure.

719
00:43:17,000 --> 00:43:18,000
Or like, you know.

720
00:43:18,000 --> 00:43:22,000
Or you're like, yeah, or you're sauna or you literally do something mind altering.

721
00:43:22,000 --> 00:43:26,000
I mean, there's a reason why writers and creative people literally like get smashed to take mushroom.

722
00:43:26,000 --> 00:43:27,000
Or whatever.

723
00:43:27,000 --> 00:43:34,000
But now this context switch is like a fucking common way.

724
00:43:34,000 --> 00:43:36,000
Yeah, in the best, in the best case, you have multiple people.

725
00:43:36,000 --> 00:43:38,000
I mean, that's how we have set up society.

726
00:43:38,000 --> 00:43:43,000
I think that's also like in the very successful companies often had two main people.

727
00:43:43,000 --> 00:43:46,000
And it was like one executive and one creative guy.

728
00:43:46,000 --> 00:43:49,000
You know, like Apple as the main example with like jobs and wasn't yet and so on.

729
00:43:49,000 --> 00:43:51,000
But it's not the only example.

730
00:43:51,000 --> 00:43:53,000
Oh, no, there are many examples.

731
00:43:54,000 --> 00:43:56,000
And that's kind of what works best.

732
00:43:56,000 --> 00:44:00,000
And again, like it's left brain, right brain kind of thing, you know.

733
00:44:00,000 --> 00:44:05,000
But yeah, and even more than that, you know, like on the specialization side, for example,

734
00:44:05,000 --> 00:44:09,000
it's like you can have an army of specialists with these tools.

735
00:44:09,000 --> 00:44:14,000
Like that's so, that's the cool thing about computers.

736
00:44:14,000 --> 00:44:18,000
They are really fast and they are really like, you know, it doesn't matter if a thousand things run.

737
00:44:18,000 --> 00:44:21,000
Basically, like my computer is doing stuff right now.

738
00:44:22,000 --> 00:44:25,000
But that's a nice thing, you know.

739
00:44:25,000 --> 00:44:29,000
They don't need to rest as much as humans do.

740
00:44:29,000 --> 00:44:48,000
Yeah, and one of the things, so my going back to my wife, my wife has been, she's been self-publishing her books because publishing the publishing world changed so drastically.

741
00:44:48,000 --> 00:44:50,000
Thanks to Amazon and all this stuff.

742
00:44:50,000 --> 00:44:55,000
Yeah, like it is the publishing world of 20 years ago.

743
00:44:55,000 --> 00:45:04,000
It doesn't exist, but it's being replaced by the same algorithm that gives us the stupid faces on the thumbnails.

744
00:45:04,000 --> 00:45:05,000
Yeah, yeah.

745
00:45:05,000 --> 00:45:19,000
It's the same thing, which dictates, I mentioned this in one of the dialogues, I think, that she went to a writer's conference and the whole conference was how to write.

746
00:45:19,000 --> 00:45:25,000
How to write six books a year, how to go viral, how to go viral.

747
00:45:25,000 --> 00:45:38,000
And nowadays, whatever you want to publish, she actually had a call with someone pretty big on the publishing world in Spain.

748
00:45:39,000 --> 00:45:50,000
And that they publish more, not just like Genre, you know, more deep thinking books and stuff like that, you know.

749
00:45:50,000 --> 00:45:53,000
So actually a lot of long-form content.

750
00:45:53,000 --> 00:46:01,000
Yeah, yeah, like not Dostoyevsky, you know, but kind of like France or the corrections or even Poloster.

751
00:46:01,000 --> 00:46:04,000
You could even go there.

752
00:46:04,000 --> 00:46:19,000
And even they say that all that matters, the only thing that matters is how many Instagram followers, TikTok followers, maybe sub-stack followers.

753
00:46:19,000 --> 00:46:21,000
That's all that matters.

754
00:46:21,000 --> 00:46:22,000
That's insane.

755
00:46:22,000 --> 00:46:24,000
They won't read the manuscript.

756
00:46:24,000 --> 00:46:26,000
That's the eyeballs.

757
00:46:27,000 --> 00:46:28,000
It is, it is.

758
00:46:28,000 --> 00:46:35,000
So the thing is that what publishers want is that the author has their own distribution.

759
00:46:35,000 --> 00:46:37,000
They need to come already with distribution.

760
00:46:37,000 --> 00:46:40,000
Yeah, they have their own kind of market already in some sense.

761
00:46:40,000 --> 00:46:47,000
They have their own because, you know, like, I mean, you know, your super fans will buy whatever you put out.

762
00:46:47,000 --> 00:46:48,000
It doesn't matter what it is basically.

763
00:46:48,000 --> 00:46:49,000
Exactly.

764
00:46:49,000 --> 00:46:50,000
They realize that.

765
00:46:50,000 --> 00:46:55,000
But that is a different, getting Instagram followers is not the same thing as for writing.

766
00:46:56,000 --> 00:47:01,000
It's a different skill set entirely.

767
00:47:01,000 --> 00:47:03,000
They don't overlap much.

768
00:47:03,000 --> 00:47:05,000
Exactly.

769
00:47:05,000 --> 00:47:09,000
So, I mean, it goes back to the media means the message, right?

770
00:47:09,000 --> 00:47:12,000
But that dictates what gets written.

771
00:47:12,000 --> 00:47:13,000
Yeah.

772
00:47:13,000 --> 00:47:14,000
Yeah.

773
00:47:14,000 --> 00:47:15,000
Yeah.

774
00:47:15,000 --> 00:47:17,000
It's kind of a sad state of affairs.

775
00:47:17,000 --> 00:47:18,000
Yeah.

776
00:47:18,000 --> 00:47:19,000
I'm not sure what to think about it.

777
00:47:19,000 --> 00:47:22,000
I mean, I'm well aware that this is how it works.

778
00:47:23,000 --> 00:47:26,000
Well, I have a very optimistic view on it.

779
00:47:26,000 --> 00:47:28,000
And I think AI fixes this.

780
00:47:28,000 --> 00:47:30,000
That's just going to be so much AI slop.

781
00:47:30,000 --> 00:47:36,000
There is going to be so much crap written that it's going to completely over.

782
00:47:36,000 --> 00:47:40,000
You can, as a bot, you can get a million followers like that.

783
00:47:40,000 --> 00:47:41,000
You can buy them.

784
00:47:41,000 --> 00:47:48,000
You can produce all kinds of content and just optimize for Instagram followers, TikTok followers, whatever it might be.

785
00:47:48,000 --> 00:47:50,000
People are already doing these things.

786
00:47:50,000 --> 00:47:51,000
Yeah, sure.

787
00:47:51,000 --> 00:47:57,000
And the big thing for the Amazon algorithm is how many books do you publish?

788
00:47:57,000 --> 00:48:02,000
Like, if you publish once a year, you're fucking nothing.

789
00:48:02,000 --> 00:48:03,000
You're nowhere.

790
00:48:03,000 --> 00:48:09,000
You need to publish at least five books a year to be kind of relevant.

791
00:48:09,000 --> 00:48:11,000
The recency bias of the algorithm?

792
00:48:11,000 --> 00:48:13,000
That's a new one.

793
00:48:13,000 --> 00:48:14,000
That's a new sound.

794
00:48:14,000 --> 00:48:15,000
Yeah.

795
00:48:15,000 --> 00:48:18,000
We are increasing our reports bar.

796
00:48:19,000 --> 00:48:22,000
The recency bias of the Amazon algorithm?

797
00:48:22,000 --> 00:48:24,000
It is insane.

798
00:48:24,000 --> 00:48:28,000
I mean, Twitter as well, right?

799
00:48:28,000 --> 00:48:33,000
We've been hardwired to, oh, this is five minutes old.

800
00:48:33,000 --> 00:48:34,000
Like, fuck this.

801
00:48:34,000 --> 00:48:38,000
Why should I care?

802
00:48:38,000 --> 00:48:44,000
But now it is completely within reason thinking I'm going to publish a book of it.

803
00:48:44,000 --> 00:48:46,000
I completely within reason.

804
00:48:46,000 --> 00:48:47,000
I just had a thought.

805
00:48:47,000 --> 00:48:54,000
It would be interesting to build a custom master client that just hides all the dates.

806
00:48:54,000 --> 00:48:56,000
I thought about that.

807
00:48:56,000 --> 00:48:59,000
Don't tell people how old the tweet is, basically.

808
00:48:59,000 --> 00:49:02,000
It's just like, if it's a banger, it's going to make the rounds again.

809
00:49:02,000 --> 00:49:03,000
Yeah.

810
00:49:03,000 --> 00:49:04,000
To me, one of the...

811
00:49:04,000 --> 00:49:06,000
Only going to work for the people that use the special client.

812
00:49:06,000 --> 00:49:08,000
To me, one of the...

813
00:49:08,000 --> 00:49:12,000
I've been thinking, what does success mean?

814
00:49:12,000 --> 00:49:17,000
Because I don't think hyper-vitconisation or hyper-nostrification means everybody ceasing

815
00:49:17,000 --> 00:49:20,000
Bitcoin or everybody ceasing noster.

816
00:49:20,000 --> 00:49:22,000
It would be easier.

817
00:49:22,000 --> 00:49:23,000
I wouldn't have to use rolls.

818
00:49:23,000 --> 00:49:28,000
That's every single day.

819
00:49:28,000 --> 00:49:32,000
But I've been thinking, what does success mean?

820
00:49:32,000 --> 00:49:41,000
To me, breaking the recency bias that we've accrued would be very fussy.

821
00:49:41,000 --> 00:49:44,000
But there are a directional indicator of worthiness.

822
00:49:44,000 --> 00:49:51,000
Undoing some of the conditioning that the platforms put on us would already be a success.

823
00:49:51,000 --> 00:49:52,000
Absolutely.

824
00:49:52,000 --> 00:49:53,000
Yeah.

825
00:49:53,000 --> 00:49:54,000
Absolutely.

826
00:49:54,000 --> 00:49:55,000
Yeah.

827
00:49:55,000 --> 00:49:56,000
Yeah.

828
00:49:56,000 --> 00:49:57,000
It's...

829
00:49:57,000 --> 00:49:58,000
No, I noticed it myself.

830
00:49:58,000 --> 00:50:05,000
I think everyone who kind of went noster only or went heavily into noster noticed this kind of shift.

831
00:50:05,000 --> 00:50:08,000
It just feels different in some sense.

832
00:50:08,000 --> 00:50:10,000
It's not as addicting.

833
00:50:10,000 --> 00:50:17,000
You don't go insane and feel like shit after two hours of doom-scrolling or whatever.

834
00:50:17,000 --> 00:50:22,000
And I think that's very much the case for many other platforms.

835
00:50:22,000 --> 00:50:28,000
And it doesn't really matter what it is, whether it's X now or TikTok or Instagram.

836
00:50:28,000 --> 00:50:30,000
It's just a different kind of insanity.

837
00:50:30,000 --> 00:50:32,000
But it's the same thing.

838
00:50:32,000 --> 00:50:34,000
Yeah, but it's just like...

839
00:50:34,000 --> 00:50:35,000
Yeah, and that's the thing.

840
00:50:35,000 --> 00:50:36,000
I keep coming back to that.

841
00:50:36,000 --> 00:50:45,000
We didn't really discuss it yet, but I think people crave realness.

842
00:50:45,000 --> 00:50:48,000
Because they kind of know that everything is kind of fake.

843
00:50:48,000 --> 00:50:52,000
Because the stuff that the dialogue on Twitter that's happening is not a real dialogue.

844
00:50:52,000 --> 00:50:56,000
It's Twitter shit posing and engagement bait and rage bait and so on.

845
00:50:56,000 --> 00:50:58,000
And it doesn't matter whether Twitter is called X now or not.

846
00:50:58,000 --> 00:50:59,000
It's all the same thing.

847
00:50:59,000 --> 00:51:01,000
And on Instagram as well.

848
00:51:01,000 --> 00:51:04,000
I think on Instagram, most people know.

849
00:51:04,000 --> 00:51:09,000
But they still pretend as if it was real.

850
00:51:09,000 --> 00:51:14,000
Kind of the psychological effects are as if it was real.

851
00:51:14,000 --> 00:51:21,000
Even if the logical side of people from a cognitive level, they know it's not really real.

852
00:51:21,000 --> 00:51:29,000
People don't walk up and down the most insane beautiful beaches with the supermodel fitness body

853
00:51:29,000 --> 00:51:34,000
with the sunset in the background every day while having three children

854
00:51:34,000 --> 00:51:38,000
and being super successful at whatever job they have.

855
00:51:38,000 --> 00:51:40,000
And all those kind of things.

856
00:51:40,000 --> 00:51:43,000
Everyone is like that and everyone is beautiful.

857
00:51:43,000 --> 00:51:46,000
And everyone is basically perfect.

858
00:51:46,000 --> 00:51:48,000
And the weather is...

859
00:51:48,000 --> 00:51:50,000
But with six kids.

860
00:51:50,000 --> 00:51:51,000
Yeah, exactly.

861
00:51:51,000 --> 00:51:53,000
And having three jobs and all those kind of things.

862
00:51:53,000 --> 00:51:56,000
And everyone kind of knows that it's bullshit.

863
00:51:57,000 --> 00:52:00,000
But still, that's a good point actually.

864
00:52:00,000 --> 00:52:01,000
That's a good point.

865
00:52:01,000 --> 00:52:02,000
Because they still enjoy the theater.

866
00:52:02,000 --> 00:52:07,000
The people still enjoy the theater and just consume it all the time and scroll it all the time.

867
00:52:07,000 --> 00:52:14,000
And again, TikTok in my mind is a lot of that as well.

868
00:52:14,000 --> 00:52:21,000
But I think the success for TikTok, it is, you know, they did a lot of things right in the sense in this paradigm.

869
00:52:22,000 --> 00:52:24,000
Being as fucking disruptive.

870
00:52:24,000 --> 00:52:28,000
And the attention paradigm, TikTok did a lot of things right.

871
00:52:28,000 --> 00:52:31,000
But I thought about this some more in the last couple of weeks.

872
00:52:31,000 --> 00:52:37,000
I think one thing that they got right, which is underappreciated, is it is very raw and very real in some sense.

873
00:52:37,000 --> 00:52:40,000
Because most people just filmed themselves and filmed their faces in whatever situation.

874
00:52:40,000 --> 00:52:42,000
A lot of people...

875
00:52:42,000 --> 00:52:49,000
You know, I've been thinking a lot about the porn and the pornification of everything.

876
00:52:50,000 --> 00:52:52,000
But we didn't talk about that yet.

877
00:52:52,000 --> 00:52:53,000
I'm still curious.

878
00:52:53,000 --> 00:52:55,000
Alright, follow us to talk about porn.

879
00:52:55,000 --> 00:52:56,000
Let's go.

880
00:52:56,000 --> 00:52:59,000
I'm going to be proponent for porn now.

881
00:52:59,000 --> 00:53:01,000
I hope I already...

882
00:53:01,000 --> 00:53:07,000
I think that's the niche that we can carve out in the end.

883
00:53:07,000 --> 00:53:08,000
Oh, it worked.

884
00:53:08,000 --> 00:53:09,000
Oh, I thought for Nuster.

885
00:53:09,000 --> 00:53:13,000
Because that already worked so beautifully with all the e-girls in the past.

886
00:53:13,000 --> 00:53:15,000
They were there for a hot five minutes.

887
00:53:15,000 --> 00:53:22,000
But what you're saying about all the beautiful people at the beach and having all these perfect moments that are...

888
00:53:22,000 --> 00:53:25,000
We all know they're bullshit.

889
00:53:25,000 --> 00:53:28,000
But they give you...

890
00:53:28,000 --> 00:53:30,000
They give you just enough.

891
00:53:30,000 --> 00:53:33,000
Like, I mean, what is porn?

892
00:53:33,000 --> 00:53:40,000
And what I've arrived at is sort of like a compression algorithm.

893
00:53:40,000 --> 00:53:44,000
Where you do away like a very fuzzy one.

894
00:53:44,000 --> 00:53:48,000
You do away without the setup.

895
00:53:48,000 --> 00:53:50,000
Like, you don't see on the picture.

896
00:53:50,000 --> 00:53:53,000
You don't see the one hour of different takes.

897
00:53:53,000 --> 00:53:55,000
You only see that moment.

898
00:53:55,000 --> 00:53:57,000
But you're already in the porn frame.

899
00:53:57,000 --> 00:54:03,000
Because what it does away with is like getting to know someone and relationship and all those kinds of things.

900
00:54:03,000 --> 00:54:06,000
Like, I mean, you know, it only focuses on the sexual act.

901
00:54:06,000 --> 00:54:09,000
And it is a perversion of the sexual act in some sense.

902
00:54:09,000 --> 00:54:10,000
Yeah, absolutely.

903
00:54:10,000 --> 00:54:12,000
In the literal sense of the word.

904
00:54:12,000 --> 00:54:18,000
The sexual porn is you do away with the intimacy.

905
00:54:18,000 --> 00:54:21,000
You do away with getting to know the person.

906
00:54:21,000 --> 00:54:22,000
You do away...

907
00:54:22,000 --> 00:54:26,000
Like, you remove everything and you go for the orgasm.

908
00:54:26,000 --> 00:54:29,000
And it's the same for the food porn.

909
00:54:29,000 --> 00:54:30,000
It is exactly the same.

910
00:54:30,000 --> 00:54:32,000
Food porn and earth porn and all those kinds of things.

911
00:54:32,000 --> 00:54:34,000
And that's why we have all these friends with that name.

912
00:54:34,000 --> 00:54:35,000
Like, the food porn.

913
00:54:35,000 --> 00:54:37,000
Why did they choose porn?

914
00:54:37,000 --> 00:54:39,000
Because it's not about the food.

915
00:54:39,000 --> 00:54:40,000
It's not about the eating.

916
00:54:40,000 --> 00:54:41,000
There is no setup.

917
00:54:41,000 --> 00:54:42,000
There is no cooking.

918
00:54:42,000 --> 00:54:43,000
There is no connexion.

919
00:54:43,000 --> 00:54:44,000
Exactly.

920
00:54:44,000 --> 00:54:46,000
It is exclusively the product.

921
00:54:46,000 --> 00:54:47,000
Just the picture.

922
00:54:47,000 --> 00:54:48,000
Just the picture.

923
00:54:48,000 --> 00:54:50,000
And then off to the next one.

924
00:54:50,000 --> 00:54:51,000
Yeah, yeah.

925
00:54:51,000 --> 00:54:56,000
There's this German-Chinese philosopher who wrote about this Han or whatever it's his name.

926
00:54:56,000 --> 00:54:57,000
I apologize.

927
00:54:57,000 --> 00:54:58,000
I don't remember the name.

928
00:54:58,000 --> 00:55:00,000
But he has a book on beauty.

929
00:55:00,000 --> 00:55:06,000
And the first one or two chapters go into that and basically how the modern definition of beauty.

930
00:55:06,000 --> 00:55:09,000
The other definition of beauty is like the smooth.

931
00:55:09,000 --> 00:55:18,000
And he talks about porn a lot as well because it's the smoothification of the whole world.

932
00:55:18,000 --> 00:55:25,000
And also porn is the smoothification of the human body and the act itself in some sense.

933
00:55:25,000 --> 00:55:29,000
I mean, I am aware that it's an interest in porn and so on.

934
00:55:29,000 --> 00:55:32,000
In general, it's like the way you...

935
00:55:32,000 --> 00:55:33,000
How are we?

936
00:55:33,000 --> 00:55:34,000
I'm quite aware.

937
00:55:35,000 --> 00:55:36,000
That's...

938
00:55:36,000 --> 00:55:37,000
Smarting surprise.

939
00:55:37,000 --> 00:55:38,000
Yeah, yeah, yeah.

940
00:55:38,000 --> 00:55:40,000
You know, like every...

941
00:55:40,000 --> 00:55:42,000
How does it go?

942
00:55:42,000 --> 00:55:43,000
Every saint has a past.

943
00:55:43,000 --> 00:55:45,000
Every singer has a future, something like that.

944
00:55:45,000 --> 00:55:52,000
But in the sense of food porn and earth porn and all those kind of things, it's all, you know,

945
00:55:52,000 --> 00:55:55,000
like shiny and all, like smooth.

946
00:55:55,000 --> 00:56:03,000
And he argues that we as a society we lost the other side of beauty.

947
00:56:03,000 --> 00:56:07,000
Which is basically to be in awe of something.

948
00:56:07,000 --> 00:56:08,000
Absolutely.

949
00:56:08,000 --> 00:56:09,000
Which is...

950
00:56:09,000 --> 00:56:12,000
Which is something so beautiful that it stuns you and it almost kills you.

951
00:56:12,000 --> 00:56:19,000
You know, like the quote goes basically, like beauty is an angel that could kill you in an instant but decides not to.

952
00:56:19,000 --> 00:56:20,000
You know, those kind of things.

953
00:56:20,000 --> 00:56:26,000
And we have lost all of that completely in the online environment.

954
00:56:26,000 --> 00:56:28,000
But I think, again, like I think...

955
00:56:28,000 --> 00:56:30,000
I mean, I'm a very optimistic person.

956
00:56:30,000 --> 00:56:32,000
I mean, there's a reason why Joe Rogan is so successful.

957
00:56:32,000 --> 00:56:36,000
It's a very raw, very long, just a conversation.

958
00:56:36,000 --> 00:56:40,000
We've found a lot of bullshit, usually without people trying to sell you stuff for anything like that.

959
00:56:40,000 --> 00:56:41,000
Yeah, absolutely.

960
00:56:41,000 --> 00:56:42,000
And it's also not...

961
00:56:42,000 --> 00:56:43,000
It's not like...

962
00:56:43,000 --> 00:56:45,000
It's not the perfect conversation, you know?

963
00:56:45,000 --> 00:56:46,000
It's not like that.

964
00:56:46,000 --> 00:56:47,000
It's like this.

965
00:56:47,000 --> 00:56:48,000
We're just going out for a walk.

966
00:56:48,000 --> 00:56:49,000
Yeah.

967
00:56:49,000 --> 00:56:50,000
And we're just chatting.

968
00:56:50,000 --> 00:56:51,000
Yeah.

969
00:56:51,000 --> 00:56:53,000
There is no intention to...

970
00:56:53,000 --> 00:56:54,000
No agenda.

971
00:56:54,000 --> 00:56:55,000
Okay.

972
00:56:55,000 --> 00:56:56,000
Like...

973
00:56:56,000 --> 00:56:57,000
No.

974
00:56:57,000 --> 00:56:59,000
Like, comment, subscribe, please.

975
00:57:00,000 --> 00:57:01,000
There's no agenda.

976
00:57:01,000 --> 00:57:05,000
It's just a bunch of conspiracy theorists doing a podcasting 2.0 podcast.

977
00:57:08,000 --> 00:57:13,000
But even the detail of this budget that Matt did with Jack, it was just...

978
00:57:13,000 --> 00:57:14,000
Yeah, on the beach.

979
00:57:14,000 --> 00:57:16,000
Like, the noise is not...

980
00:57:16,000 --> 00:57:17,000
The sound is so great.

981
00:57:17,000 --> 00:57:19,000
It's fucking perfect because it's real.

982
00:57:19,000 --> 00:57:20,000
Yeah.

983
00:57:20,000 --> 00:57:22,000
For us, it's traffic and leaf blowers.

984
00:57:22,000 --> 00:57:23,000
It's a little bit different.

985
00:57:23,000 --> 00:57:24,000
It's not quite the same.

986
00:57:24,000 --> 00:57:25,000
Sometimes.

987
00:57:25,000 --> 00:57:26,000
We'll get there.

988
00:57:26,000 --> 00:57:27,000
We'll get to the beach one day.

989
00:57:28,000 --> 00:57:31,000
We'll still have to earn some money with some agents or something.

990
00:57:32,000 --> 00:57:36,000
But going back to the beauty thing, it also nonce you, right?

991
00:57:36,000 --> 00:57:42,000
When you see within 40 seconds, you see 10 beautiful, whatever.

992
00:57:42,000 --> 00:57:46,000
Flowers, women, beachies, houses, whatever it is.

993
00:57:46,000 --> 00:57:48,000
Then you sit in real life.

994
00:57:48,000 --> 00:57:49,000
It's not quite.

995
00:57:49,000 --> 00:57:51,000
It's not how it looked on Instagram.

996
00:57:51,000 --> 00:57:53,000
You're at the place.

997
00:57:53,000 --> 00:57:55,000
You're at the place.

998
00:57:55,000 --> 00:57:57,000
But it's missing Instagram photos.

999
00:57:57,000 --> 00:57:58,000
It kind of sucks.

1000
00:57:58,000 --> 00:57:59,000
It kind of like...

1001
00:57:59,000 --> 00:58:00,000
Yeah, exactly.

1002
00:58:00,000 --> 00:58:01,000
It was very strenuous.

1003
00:58:01,000 --> 00:58:02,000
Yeah, it's like...

1004
00:58:02,000 --> 00:58:03,000
I only see one.

1005
00:58:03,000 --> 00:58:04,000
You're also tired.

1006
00:58:04,000 --> 00:58:05,000
You're hungry.

1007
00:58:05,000 --> 00:58:06,000
And it's like...

1008
00:58:06,000 --> 00:58:07,000
What's the point?

1009
00:58:07,000 --> 00:58:08,000
It's a crying.

1010
00:58:08,000 --> 00:58:10,000
But it numbs you, right?

1011
00:58:10,000 --> 00:58:11,000
It numbs you.

1012
00:58:11,000 --> 00:58:16,000
That's why I think the programming is so an interesting aspect.

1013
00:58:16,000 --> 00:58:18,000
Because it numbs you in real life.

1014
00:58:18,000 --> 00:58:24,000
We've seen that the sexual porn has an issue for people that are very into that.

1015
00:58:24,000 --> 00:58:27,000
That it creates issues within the relationship.

1016
00:58:27,000 --> 00:58:31,000
Because within porn you can question about it, man.

1017
00:58:31,000 --> 00:58:34,000
And that's not to go.

1018
00:58:34,000 --> 00:58:35,000
That's not to get into that.

1019
00:58:35,000 --> 00:58:36,000
But I think...

1020
00:58:36,000 --> 00:58:42,000
I mentioned that in the past with what I like about the Noster side is really the large part of

1021
00:58:42,000 --> 00:58:48,000
many people that use Noster that are in the Noster bubble or in the Noster community or whatever.

1022
00:58:48,000 --> 00:58:50,000
They are very outdoorsy people in some sense.

1023
00:58:50,000 --> 00:58:58,000
So it is about getting away from the conditioning, getting away from kind of these destructive

1024
00:58:58,000 --> 00:59:02,000
loops that a lot of the existing systems try to capture you in.

1025
00:59:02,000 --> 00:59:05,000
And actually touching grass and going outside and so on.

1026
00:59:05,000 --> 00:59:09,000
That's why it was important for me to actually do this outside and walking.

1027
00:59:09,000 --> 00:59:11,000
I don't want to fucking sit in a...

1028
00:59:11,000 --> 00:59:14,000
It was kind of what the premises of our engineering was.

1029
00:59:14,000 --> 00:59:15,000
Exactly.

1030
00:59:15,000 --> 00:59:16,000
Same thing.

1031
00:59:17,000 --> 00:59:19,000
It's just like you...

1032
00:59:19,000 --> 00:59:25,000
It's way easier to think and to actually go deep on things if your blood is flowing a little

1033
00:59:25,000 --> 00:59:29,000
bit, if you're outside, if you get inspiration from nature and so on.

1034
00:59:29,000 --> 00:59:35,000
It's very hard to come up with new ideas and with new insights in like a closed window,

1035
00:59:35,000 --> 00:59:37,000
closed door meeting room.

1036
00:59:37,000 --> 00:59:40,000
We mentioned this in coffee.

1037
00:59:40,000 --> 00:59:45,000
We had last time after recording that it changes the conversation.

1038
00:59:45,000 --> 00:59:51,000
The fact that you are not sitting down, a silence is much more comfortable before walking.

1039
00:59:51,000 --> 00:59:52,000
Yeah.

1040
00:59:52,000 --> 00:59:54,000
And that changes what we're going to talk about.

1041
00:59:54,000 --> 00:59:57,000
Stop happens and you just pivot and just whatever.

1042
00:59:57,000 --> 01:00:00,000
Like you run into a leaf blower, you run into a car.

1043
01:00:00,000 --> 01:00:01,000
Hopefully not.

1044
01:00:01,000 --> 01:00:04,000
You have to turn around because it starts raining.

1045
01:00:04,000 --> 01:00:07,000
You run into a hurricane level of wind.

1046
01:00:07,000 --> 01:00:08,000
Oh my God.

1047
01:00:08,000 --> 01:00:11,000
She changes the conversation quite a bit.

1048
01:00:11,000 --> 01:00:12,000
I have to say.

1049
01:00:13,000 --> 01:00:18,000
Maybe if you put the leaf blower in the meeting room and you just turn it on, produce the

1050
01:00:18,000 --> 01:00:19,000
winds yourself.

1051
01:00:19,000 --> 01:00:22,000
Maybe that will work.

1052
01:00:22,000 --> 01:00:25,000
But yeah, no, I'm excited to see your stuff today.

1053
01:00:25,000 --> 01:00:27,000
I really want to...

1054
01:00:27,000 --> 01:00:32,000
I think I'll try to record some of the things and have you guys talk over it.

1055
01:00:32,000 --> 01:00:35,000
Again, especially you and Jesus and Jesus.

1056
01:00:35,000 --> 01:00:38,000
Everyone is kind of working on the same thing in some sense.

1057
01:00:38,000 --> 01:00:41,000
Like everyone has the same ideas or very similar ideas.

1058
01:00:41,000 --> 01:00:42,000
A complementary.

1059
01:00:42,000 --> 01:00:43,000
Yeah.

1060
01:00:43,000 --> 01:00:48,000
Trying to use an Oscar as a substrate for like an agent human environment and for agent human

1061
01:00:48,000 --> 01:00:50,000
interaction and those kind of things.

1062
01:00:50,000 --> 01:00:52,000
And kind of how can we build those marketplaces?

1063
01:00:52,000 --> 01:00:53,000
How can we make all of that work?

1064
01:00:53,000 --> 01:00:55,000
How can we integrate it seamlessly?

1065
01:00:55,000 --> 01:01:01,000
I think what Cursor really did right is like integrated insanely seamlessly and actually

1066
01:01:01,000 --> 01:01:03,000
also letting the agents like change everything.

1067
01:01:03,000 --> 01:01:06,000
It's like basically y'all would like just do all the things.

1068
01:01:06,000 --> 01:01:08,000
And that is such a paradigm shift.

1069
01:01:08,000 --> 01:01:16,000
And now I feel like it's about how can we build this out in a more sane, non siloed

1070
01:01:16,000 --> 01:01:18,000
way in some sense.

1071
01:01:18,000 --> 01:01:21,000
I think that's kind of the question.

1072
01:01:21,000 --> 01:01:22,000
Yeah.

1073
01:01:22,000 --> 01:01:25,000
I think to me the land chain stuff is...

1074
01:01:25,000 --> 01:01:29,000
I've been talking about the land chain thing for a while with DVM's.

1075
01:01:29,000 --> 01:01:35,000
But the idea is that you assemble workflows for the agents and you say,

1076
01:01:35,000 --> 01:01:40,000
okay, so this agent needs to give to this guy and this guy needs to do this and complete

1077
01:01:40,000 --> 01:01:42,000
a move or whatever it might be.

1078
01:01:42,000 --> 01:01:46,000
But it's a very centrally planned model, you know?

1079
01:01:46,000 --> 01:01:50,000
Like there is no way for tooling to emerge.

1080
01:01:50,000 --> 01:01:55,000
There is no way for tool to just come into being and the nation saying,

1081
01:01:55,000 --> 01:01:56,000
oh, look, that tool is cool.

1082
01:01:56,000 --> 01:01:57,000
I'm going to use it.

1083
01:01:57,000 --> 01:01:58,000
Yeah.

1084
01:01:58,000 --> 01:02:03,000
But kind of like going back to the 402 on the HTTP code where it's like,

1085
01:02:03,000 --> 01:02:05,000
yeah, probably payment would make sense.

1086
01:02:05,000 --> 01:02:06,000
Okay.

1087
01:02:06,000 --> 01:02:07,000
Let's just put a 402.

1088
01:02:07,000 --> 01:02:08,000
We'll solve this later.

1089
01:02:08,000 --> 01:02:11,000
We didn't realize that solving it was so fucking hard.

1090
01:02:11,000 --> 01:02:16,000
We had all the oceans first.

1091
01:02:16,000 --> 01:02:20,000
She has to solve these little tiny things.

1092
01:02:20,000 --> 01:02:28,000
And the things that Noster provides, you can't really just open things up and within your

1093
01:02:28,000 --> 01:02:33,000
computer, just say, oh, just find random shit on the internet and fucking run it.

1094
01:02:33,000 --> 01:02:36,000
Just go wild.

1095
01:02:36,000 --> 01:02:39,000
You need to constrain it for certain things.

1096
01:02:39,000 --> 01:02:42,000
You want to constrain it to, should I use this?

1097
01:02:42,000 --> 01:02:47,000
From my perspective, from my pub keys perspective, should I use this?

1098
01:02:47,000 --> 01:02:53,000
And it requires, like Matt said, in bootstrap.

1099
01:02:53,000 --> 01:02:57,000
You have to bootstrap the web of trust with shitposting and saying GM every day.

1100
01:02:57,000 --> 01:02:59,000
Yeah, that's how it works.

1101
01:02:59,000 --> 01:03:02,000
That's literally how it works.

1102
01:03:02,000 --> 01:03:05,000
It's quite amazing.

1103
01:03:05,000 --> 01:03:10,000
So what's kind of, what do you have in mind as kind of an end state for the tool chain

1104
01:03:10,000 --> 01:03:13,000
that you're building out?

1105
01:03:13,000 --> 01:03:19,000
Because you mentioned like everything is shifting for you in terms of writing code

1106
01:03:19,000 --> 01:03:24,000
and how you basically use computers in some sense and how you, yeah,

1107
01:03:24,000 --> 01:03:26,000
how you do your work stuff.

1108
01:03:27,000 --> 01:03:34,000
So three months from now, in the best case, how does your workflow look like?

1109
01:03:34,000 --> 01:03:36,000
Because it's the same for me.

1110
01:03:36,000 --> 01:03:37,000
It's just not for coding.

1111
01:03:37,000 --> 01:03:38,000
Right, yeah, yeah.

1112
01:03:38,000 --> 01:03:50,000
So for me, when you write main.c and you write a C program, you don't grab the code

1113
01:03:50,000 --> 01:03:52,000
and turn it into binary.

1114
01:03:52,000 --> 01:03:54,000
You have a compiler.

1115
01:03:54,000 --> 01:04:00,000
And the compiler sees the code, sees what you expressed very explicitly and it turns

1116
01:04:00,000 --> 01:04:02,000
it into a binary code, right?

1117
01:04:02,000 --> 01:04:08,000
And there's optimizations, but whatever.

1118
01:04:08,000 --> 01:04:17,000
I think writing applications in the way that we've been doing forever is the equivalent

1119
01:04:17,000 --> 01:04:19,000
of writing the binary code.

1120
01:04:20,000 --> 01:04:25,000
So it's, or like similar shift to from, Justin mentioned this as well,

1121
01:04:25,000 --> 01:04:27,000
like from going from assembler to high level languages.

1122
01:04:27,000 --> 01:04:28,000
Yes.

1123
01:04:28,000 --> 01:04:29,000
Yes.

1124
01:04:29,000 --> 01:04:36,000
And it's a similar shift where what we call high level languages now are actually in,

1125
01:04:36,000 --> 01:04:42,000
yeah, in some sense, not high level enough because what we care about is basically

1126
01:04:42,000 --> 01:04:44,000
what do we want to get done and just general concepts.

1127
01:04:44,000 --> 01:04:46,000
You want something, you have something to do.

1128
01:04:46,000 --> 01:04:49,000
You have a task to accomplish or you want to express something.

1129
01:04:49,000 --> 01:04:50,000
Yeah.

1130
01:04:50,000 --> 01:04:52,000
It's, it's not, should I malloc here?

1131
01:04:52,000 --> 01:04:54,000
Should I free the memory here?

1132
01:04:54,000 --> 01:04:55,000
Yeah.

1133
01:04:55,000 --> 01:04:56,000
Yeah.

1134
01:04:56,000 --> 01:04:57,000
Yeah.

1135
01:04:57,000 --> 01:04:58,000
Yeah.

1136
01:04:58,000 --> 01:04:59,000
Yeah.

1137
01:04:59,000 --> 01:05:00,000
Exactly.

1138
01:05:00,000 --> 01:05:02,000
That's kind of in question.

1139
01:05:02,000 --> 01:05:03,000
It's a, it's a, yeah.

1140
01:05:03,000 --> 01:05:08,000
That's a, that's an interesting comparison moving from like low level assembly to a high

1141
01:05:08,000 --> 01:05:09,000
level language.

1142
01:05:09,000 --> 01:05:10,000
And now we're doing it again.

1143
01:05:10,000 --> 01:05:11,000
So my.

1144
01:05:11,000 --> 01:05:12,000
With natural language.

1145
01:05:12,000 --> 01:05:13,000
Exactly.

1146
01:05:13,000 --> 01:05:14,000
Exactly.

1147
01:05:14,000 --> 01:05:17,000
And, and to me the fact that the yellow lamb itself.

1148
01:05:17,000 --> 01:05:18,000
Which is funny.

1149
01:05:18,000 --> 01:05:19,000
Which is funny.

1150
01:05:19,000 --> 01:05:23,000
Which is funny because natural language and code is very different.

1151
01:05:23,000 --> 01:05:24,000
100%.

1152
01:05:24,000 --> 01:05:27,000
Because, because natural language is always ambiguous and unspecified.

1153
01:05:27,000 --> 01:05:31,000
But always leaves the wiggle room of, of the interpretation of those kinds of things.

1154
01:05:31,000 --> 01:05:32,000
Yes.

1155
01:05:32,000 --> 01:05:33,000
Yeah.

1156
01:05:33,000 --> 01:05:34,000
Absolutely.

1157
01:05:34,000 --> 01:05:35,000
Absolutely.

1158
01:05:35,000 --> 01:05:37,000
But, but to me the fussiness of the natural language allows for creativity.

1159
01:05:37,000 --> 01:05:38,000
Oh.

1160
01:05:38,000 --> 01:05:39,000
Yo.

1161
01:05:39,000 --> 01:05:40,000
Absolutely.

1162
01:05:40,000 --> 01:05:41,000
Absolutely.

1163
01:05:41,000 --> 01:05:46,000
So, so you say in, in three months or in certain timeframe.

1164
01:05:46,000 --> 01:05:54,000
My goal is to, because right now, like I said, they, they suck at producing things that

1165
01:05:54,000 --> 01:05:56,000
you can use.

1166
01:05:56,000 --> 01:05:57,000
Alt.

1167
01:05:57,000 --> 01:05:58,000
You know.

1168
01:05:58,000 --> 01:05:59,000
Yeah.

1169
01:05:59,000 --> 01:06:00,000
Yeah.

1170
01:06:00,000 --> 01:06:01,000
Yeah.

1171
01:06:01,000 --> 01:06:05,000
The moment you go from one shot into iterating 50 or 100 times.

1172
01:06:05,000 --> 01:06:06,000
Yeah.

1173
01:06:06,000 --> 01:06:07,000
They degrade.

1174
01:06:07,000 --> 01:06:08,000
They plummet.

1175
01:06:08,000 --> 01:06:09,000
Yeah.

1176
01:06:10,000 --> 01:06:11,000
That's, that's what you should do.

1177
01:06:11,000 --> 01:06:14,000
You just throw it away and you one shot it again with everything you've learned.

1178
01:06:14,000 --> 01:06:15,000
Exactly.

1179
01:06:15,000 --> 01:06:16,000
Exactly.

1180
01:06:16,000 --> 01:06:21,000
So, so what I want is to have, because I think we are there.

1181
01:06:21,000 --> 01:06:25,000
It's just, it's just, we were missing a few components.

1182
01:06:25,000 --> 01:06:31,000
Um, all the, all the code and all the libraries and all the documentation has been always

1183
01:06:31,000 --> 01:06:33,000
written by humans for humans.

1184
01:06:33,000 --> 01:06:34,000
Yeah.

1185
01:06:34,000 --> 01:06:35,000
For human consumption.

1186
01:06:35,000 --> 01:06:36,000
Yeah.

1187
01:06:36,000 --> 01:06:44,000
So, I am now writing libraries in a, in a way that, that it would make a lot of sense for

1188
01:06:44,000 --> 01:06:46,000
something that has a very small context window.

1189
01:06:46,000 --> 01:06:47,000
Yeah.

1190
01:06:47,000 --> 01:06:48,000
Like an oath.

1191
01:06:48,000 --> 01:06:53,000
You're writing, you're writing basically documentation now for LLMs and you use LLMs and yourself

1192
01:06:53,000 --> 01:06:54,000
to write it.

1193
01:06:54,000 --> 01:06:57,000
Which interestingly enough, also works for a human.

1194
01:06:57,000 --> 01:06:58,000
Yeah.

1195
01:06:58,000 --> 01:07:00,000
Because, because it's natural language basically.

1196
01:07:00,000 --> 01:07:01,000
Exactly.

1197
01:07:01,000 --> 01:07:02,000
Exactly.

1198
01:07:02,000 --> 01:07:03,000
So, so.

1199
01:07:03,000 --> 01:07:04,000
So you have a short explanation that is really limited.

1200
01:07:04,000 --> 01:07:05,000
Yeah.

1201
01:07:05,000 --> 01:07:06,000
And you have a code snippet.

1202
01:07:06,000 --> 01:07:09,000
And you have a code snippet that is extremely niche.

1203
01:07:09,000 --> 01:07:10,000
Yeah.

1204
01:07:10,000 --> 01:07:11,000
Like, how do you do that?

1205
01:07:11,000 --> 01:07:12,000
This is how you do that.

1206
01:07:12,000 --> 01:07:13,000
One thing.

1207
01:07:13,000 --> 01:07:16,000
And, and image feed versus a video feed is a different data mode.

1208
01:07:16,000 --> 01:07:17,000
Yeah.

1209
01:07:17,000 --> 01:07:18,000
Yeah.

1210
01:07:18,000 --> 01:07:19,000
Yeah.

1211
01:07:19,000 --> 01:07:20,000
Yeah.

1212
01:07:20,000 --> 01:07:21,000
How do you do that?

1213
01:07:21,000 --> 01:07:23,000
And, and, and, and I can have different snippets that express both things.

1214
01:07:23,000 --> 01:07:24,000
Yeah.

1215
01:07:24,000 --> 01:07:25,000
And, and they don't have to compromise.

1216
01:07:25,000 --> 01:07:27,000
They don't have to try to figure it out.

1217
01:07:27,000 --> 01:07:32,000
They are really good at assembling things that already exist and putting together code and

1218
01:07:32,000 --> 01:07:33,000
extrapolating for code that already exists.

1219
01:07:33,000 --> 01:07:36,000
So in the end, coming back to your idea of everyone building their own client.

1220
01:07:36,000 --> 01:07:38,000
Which I kind of disagree with, by the way.

1221
01:07:38,000 --> 01:07:39,000
Oh, really?

1222
01:07:39,000 --> 01:07:40,000
Yeah.

1223
01:07:40,000 --> 01:07:43,000
It's, because they're, they're still, they're still, I thought about this a little bit.

1224
01:07:43,000 --> 01:07:49,000
And, and, and there is still a massive benefit of having a shared, doesn't matter what it

1225
01:07:49,000 --> 01:07:50,000
is.

1226
01:07:50,000 --> 01:07:54,000
Like a shared, even if, like shared software, shared concepts, shared like.

1227
01:07:54,000 --> 01:07:55,000
Oh, I agree with that.

1228
01:07:55,000 --> 01:08:01,000
We, we, we, we use the same languages for, for, for things, you know, and the same words as

1229
01:08:01,000 --> 01:08:02,000
they appear.

1230
01:08:02,000 --> 01:08:06,000
Like, you know, not everyone has a, like, everyone's just saps now.

1231
01:08:06,000 --> 01:08:08,000
You can, you could say 10 different other things.

1232
01:08:08,000 --> 01:08:10,000
You know, it's very useful to land on one thing and use that.

1233
01:08:10,000 --> 01:08:12,000
I'll say, I think it's true for applications as well.

1234
01:08:12,000 --> 01:08:16,000
But, but I, I also get the point of, of the, the easier it gets.

1235
01:08:16,000 --> 01:08:19,000
I mean, you see this, you see this with open software already.

1236
01:08:19,000 --> 01:08:24,000
If something is used widely and very useful, you will have like dozens or hundreds of forks.

1237
01:08:24,000 --> 01:08:29,000
And, and some people are just, they care about one aspect of the software more than, than

1238
01:08:29,000 --> 01:08:30,000
others.

1239
01:08:30,000 --> 01:08:32,000
And then they fork it and they, they built this thing.

1240
01:08:32,000 --> 01:08:36,000
They, they build a slight modification and then this will have their own user base and

1241
01:08:36,000 --> 01:08:37,000
so on.

1242
01:08:37,000 --> 01:08:38,000
And so it's just like this on steroids.

1243
01:08:38,000 --> 01:08:39,000
I get that.

1244
01:08:39,000 --> 01:08:42,000
The, the, the problem is with the fork is that it's so costly.

1245
01:08:42,000 --> 01:08:43,000
Right.

1246
01:08:43,000 --> 01:08:44,000
It has a maintenance cost.

1247
01:08:44,000 --> 01:08:48,000
But you have to, it's, it's in part because you have to mainstream, maintain it with upstream

1248
01:08:48,000 --> 01:08:49,000
basically.

1249
01:08:49,000 --> 01:08:50,000
Exactly.

1250
01:08:50,000 --> 01:08:51,000
Exactly.

1251
01:08:51,000 --> 01:08:52,000
But say, so for example.

1252
01:08:52,000 --> 01:08:53,000
The bike coding fixes this.

1253
01:08:53,000 --> 01:08:54,000
But yeah.

1254
01:08:54,000 --> 01:08:57,000
Like Tony, Tony, Tony Ford to Amethyst.

1255
01:08:58,000 --> 01:08:59,000
Yeah.

1256
01:08:59,000 --> 01:09:00,000
I know.

1257
01:09:00,000 --> 01:09:01,000
Yeah, yeah.

1258
01:09:01,000 --> 01:09:03,000
And he was running as a, as a separate client, he was running for a while.

1259
01:09:03,000 --> 01:09:04,000
It was just too expensive.

1260
01:09:04,000 --> 01:09:08,000
It was not worth his time to actually do the thing.

1261
01:09:08,000 --> 01:09:14,000
But, but if he, it was literally two minutes of work with no maintenance, just change this

1262
01:09:14,000 --> 01:09:15,000
thing in Amethyst.

1263
01:09:15,000 --> 01:09:16,000
Like that.

1264
01:09:16,000 --> 01:09:17,000
I don't like the editor.

1265
01:09:17,000 --> 01:09:21,000
I don't like the, the, but, but you still, with every release of Anethyst, you have to

1266
01:09:21,000 --> 01:09:22,000
do it again.

1267
01:09:22,000 --> 01:09:24,000
Do you have to do it?

1268
01:09:24,000 --> 01:09:26,000
That, that is something that would be incredible.

1269
01:09:26,000 --> 01:09:27,000
Okay.

1270
01:09:27,000 --> 01:09:28,000
I see.

1271
01:09:28,000 --> 01:09:29,000
Good at it.

1272
01:09:29,000 --> 01:09:30,000
So you can just, out of four releases.

1273
01:09:30,000 --> 01:09:31,000
You can just, out of four, you can just rush.

1274
01:09:31,000 --> 01:09:32,000
Oh, yeah.

1275
01:09:32,000 --> 01:09:33,000
Interesting.

1276
01:09:33,000 --> 01:09:34,000
Okay.

1277
01:09:34,000 --> 01:09:35,000
Yeah.

1278
01:09:35,000 --> 01:09:39,000
And you, you could imagine, you could imagine Gigi publishes this modification of Amethyst

1279
01:09:39,000 --> 01:09:40,000
as a prompt.

1280
01:09:40,000 --> 01:09:43,000
And you could, that's a, that's a good point because that's how my plug-in system works.

1281
01:09:43,000 --> 01:09:44,000
It's all the natural language based.

1282
01:09:44,000 --> 01:09:45,000
Right.

1283
01:09:45,000 --> 01:09:49,000
So it's basically like, you know, you, you have this release.

1284
01:09:49,000 --> 01:09:51,000
You know, it's an official Amethyst release.

1285
01:09:51,000 --> 01:09:53,000
It's, it's, it's signed by Vitor.

1286
01:09:53,000 --> 01:09:54,000
It's, it has this git commit.

1287
01:09:54,000 --> 01:09:58,000
And so you can basically build a modifier in some sense.

1288
01:09:58,000 --> 01:10:02,000
That's like, use this commit to check it out and apply my prompt to it.

1289
01:10:02,000 --> 01:10:03,000
Yeah.

1290
01:10:03,000 --> 01:10:04,000
Like, rip out.

1291
01:10:04,000 --> 01:10:05,000
Rip out this part.

1292
01:10:05,000 --> 01:10:06,000
This, this, this.

1293
01:10:06,000 --> 01:10:07,000
This, this.

1294
01:10:07,000 --> 01:10:09,000
I never use, this, this part.

1295
01:10:09,000 --> 01:10:11,000
And don't show me dates.

1296
01:10:11,000 --> 01:10:12,000
Yeah.

1297
01:10:12,000 --> 01:10:13,000
Yeah.

1298
01:10:13,000 --> 01:10:14,000
Oh, for that.

1299
01:10:14,000 --> 01:10:16,000
It might be worthwhile to try it actually.

1300
01:10:16,000 --> 01:10:19,000
Let's see what happens if I put this into cursor.

1301
01:10:19,000 --> 01:10:21,000
It's just like, remove all the dates.

1302
01:10:22,000 --> 01:10:23,000
It's just that.

1303
01:10:23,000 --> 01:10:24,000
It would probably break.

1304
01:10:24,000 --> 01:10:25,000
Eat free base.

1305
01:10:25,000 --> 01:10:26,000
Squash everything.

1306
01:10:26,000 --> 01:10:27,000
Break.

1307
01:10:27,000 --> 01:10:28,000
Absolutely everything.

1308
01:10:28,000 --> 01:10:29,000
No, but it's an interesting idea.

1309
01:10:29,000 --> 01:10:30,000
I'm, I'm not saying I disagree entirely.

1310
01:10:30,000 --> 01:10:36,000
I, I, I still think just that there is a, a mental cost to using a different thing, maintaining.

1311
01:10:36,000 --> 01:10:37,000
Even, like, I don't know.

1312
01:10:37,000 --> 01:10:38,000
I, I just think that the cost is non-zero.

1313
01:10:38,000 --> 01:10:39,000
So, so.

1314
01:10:39,000 --> 01:10:40,000
I, I don't know.

1315
01:10:40,000 --> 01:10:41,000
I think it's zero.

1316
01:10:41,000 --> 01:10:42,000
Yeah.

1317
01:10:42,000 --> 01:10:43,000
Yeah.

1318
01:10:43,000 --> 01:10:44,000
I mean, you have your vibe line, right?

1319
01:10:44,000 --> 01:10:45,000
Yeah.

1320
01:10:45,000 --> 01:10:48,000
Like literally the moment you could, you know, you could, you could, you could, you know,

1321
01:10:48,000 --> 01:10:49,000
have your vibe line, right?

1322
01:10:49,000 --> 01:10:50,000
Yeah.

1323
01:10:50,000 --> 01:10:52,000
Like literally the moment you could have a pipeline where the moment.

1324
01:10:52,000 --> 01:10:53,000
Right now the cost is not zero.

1325
01:10:53,000 --> 01:10:55,000
It's many, many nights in the garage.

1326
01:10:55,000 --> 01:10:56,000
Right.

1327
01:10:56,000 --> 01:10:57,000
But you do it one time.

1328
01:10:57,000 --> 01:10:58,000
Yeah.

1329
01:10:58,000 --> 01:10:59,000
You do it one time.

1330
01:10:59,000 --> 01:11:03,000
Imagine if you, if you built a pipeline were on every release of Amethyst.

1331
01:11:03,000 --> 01:11:04,000
No, fair enough.

1332
01:11:04,000 --> 01:11:06,000
You clone, apply these patches.

1333
01:11:06,000 --> 01:11:07,000
Yeah.

1334
01:11:07,000 --> 01:11:08,000
And then.

1335
01:11:08,000 --> 01:11:09,000
Yeah.

1336
01:11:09,000 --> 01:11:12,000
It's an interesting thought because this exists already with exactly what you said, with just

1337
01:11:12,000 --> 01:11:14,000
patches and explain, applying these patches.

1338
01:11:14,000 --> 01:11:15,000
The, the problem is if the patch doesn't work.

1339
01:11:15,000 --> 01:11:16,000
It's really hard to apply.

1340
01:11:17,000 --> 01:11:19,000
If the patch doesn't work anymore, then this, this doesn't work anymore.

1341
01:11:19,000 --> 01:11:20,000
In some sense.

1342
01:11:20,000 --> 01:11:21,000
But no, but that's not the case.

1343
01:11:21,000 --> 01:11:22,000
Yeah.

1344
01:11:22,000 --> 01:11:26,000
With, with, with kind of other lamps trying to figure it out, it, it actually can work in theory.

1345
01:11:26,000 --> 01:11:28,000
It's like you have a natural language.

1346
01:11:28,000 --> 01:11:29,000
Yeah.

1347
01:11:29,000 --> 01:11:32,000
It's, it's, to me it's super interesting to have natural languages as the base.

1348
01:11:32,000 --> 01:11:33,000
Yes.

1349
01:11:33,000 --> 01:11:34,000
Absolutely.

1350
01:11:34,000 --> 01:11:35,000
You can, you can vibe code so many things.

1351
01:11:35,000 --> 01:11:43,000
I'm, I'm just, I'm just constantly blown away by kind of, again, figuring out what they are actually good at and using it for that.

1352
01:11:43,000 --> 01:11:45,000
Like they're amazing at vibe coding.

1353
01:11:45,000 --> 01:11:49,000
Uh, you know, next year's UIs, whatever, like tailwind UIs or whatever.

1354
01:11:49,000 --> 01:11:51,000
Like really well known, well documented frameworks.

1355
01:11:51,000 --> 01:11:52,000
Yeah.

1356
01:11:52,000 --> 01:11:54,000
There's a million examples of how to do this stuff.

1357
01:11:54,000 --> 01:12:00,000
And it, and you just tell them, I want, you know, this little dashboard and, you know, like some buttons here that should do this, this stuff and so on.

1358
01:12:00,000 --> 01:12:03,000
And it's like, and two minutes later, it's like, it looks perfect, basically.

1359
01:12:03,000 --> 01:12:06,000
I mean, you can also probably tell it's white coded.

1360
01:12:06,000 --> 01:12:11,000
If it automatically uses the system dark or bright theme and that's kind of things that you can.

1361
01:12:11,000 --> 01:12:13,000
You first version has that.

1362
01:12:13,000 --> 01:12:15,000
Oh, 100% vibe code.

1363
01:12:15,000 --> 01:12:18,000
The first version already has a dark and a bright theme.

1364
01:12:18,000 --> 01:12:22,000
And it's made with love, clean or whatever.

1365
01:12:22,000 --> 01:12:23,000
Yeah, exactly.

1366
01:12:23,000 --> 01:12:25,000
If it has a heart in the foot.

1367
01:12:25,000 --> 01:12:26,000
Yeah.

1368
01:12:26,000 --> 01:12:31,000
But basically, for example, I tried, so I built this up.

1369
01:12:31,000 --> 01:12:40,000
You know, we've discussed before the, uh, that, I think one of the key issues that we have with the current interfaces.

1370
01:12:40,000 --> 01:12:42,000
Is the low information density.

1371
01:12:42,000 --> 01:12:43,000
Yeah.

1372
01:12:43,000 --> 01:12:51,000
So I, I'm playing with like creative ideas on, on displaying the same data with higher information density.

1373
01:12:51,000 --> 01:12:52,000
Yeah.

1374
01:12:52,000 --> 01:12:57,000
So what I, uh, I, I wrote like a 3D plane on an xjs app.

1375
01:12:57,000 --> 01:13:03,000
I wrote like a 3D plane where all the root posts are displayed as a card.

1376
01:13:03,000 --> 01:13:06,000
And you can see the responses behind it.

1377
01:13:06,000 --> 01:13:07,000
Ah, okay.

1378
01:13:07,000 --> 01:13:11,000
So, and, and then under it, you can see who's engaging in that, in that thread.

1379
01:13:11,000 --> 01:13:12,000
Yeah.

1380
01:13:12,000 --> 01:13:18,000
Um, so, and then you can navigate and as, as you're zooming in into, into the, into the thread,

1381
01:13:18,000 --> 01:13:21,000
they start to open up and you can start seeing the first replies.

1382
01:13:21,000 --> 01:13:22,000
Yeah.

1383
01:13:22,000 --> 01:13:23,000
Nice.

1384
01:13:23,000 --> 01:13:24,000
Yeah.

1385
01:13:24,000 --> 01:13:25,000
Yeah.

1386
01:13:25,000 --> 01:13:26,000
Yeah.

1387
01:13:26,000 --> 01:13:27,000
Yeah.

1388
01:13:27,000 --> 01:13:28,000
That's, that's awesome.

1389
01:13:28,000 --> 01:13:32,000
So, so I, I, I, I, I've been wanting to have something like this for a while, but it doesn't, like I, I, I don't want to like,

1390
01:13:33,000 --> 01:13:37,000
and, and also like you would have to learn, uh, you know, 3D visualization.

1391
01:13:37,000 --> 01:13:38,000
Yeah.

1392
01:13:38,000 --> 01:13:39,000
Oh, 100% yeah.

1393
01:13:39,000 --> 01:13:40,000
Yeah.

1394
01:13:40,000 --> 01:13:41,000
Things like that's really, really hard to do.

1395
01:13:41,000 --> 01:13:42,000
It is.

1396
01:13:42,000 --> 01:13:43,000
Yes.

1397
01:13:43,000 --> 01:13:44,000
I need to vibe.

1398
01:13:44,000 --> 01:13:45,000
It's very easy to vibe.

1399
01:13:45,000 --> 01:13:46,000
Yeah.

1400
01:13:46,000 --> 01:13:47,000
But then I looked at the code.

1401
01:13:47,000 --> 01:13:50,000
And I saw because I've never written next to JS or react code.

1402
01:13:50,000 --> 01:13:55,000
Um, I, I looked at the code using indicating like all the wrong ways.

1403
01:13:55,000 --> 01:13:56,000
Yeah.

1404
01:13:56,000 --> 01:13:57,000
Yeah.

1405
01:13:57,000 --> 01:13:59,000
Extremely non-performant ways.

1406
01:13:59,000 --> 01:14:01,920
Um, so, so now the, I,

1407
01:14:01,920 --> 01:14:08,920
I wrote a, uh, library that is called NDK hooks that is for next year's bibecoding, basically.

1408
01:14:08,920 --> 01:14:09,920
Yeah.

1409
01:14:09,920 --> 01:14:10,920
And it's just one shot.

1410
01:14:10,920 --> 01:14:17,920
There is one function export to get, get events and it figures out how to optimize it inside.

1411
01:14:17,920 --> 01:14:22,920
But for the LLM, there is this one function, you know, it's like very boxed on what it can do.

1412
01:14:22,920 --> 01:14:23,920
This is the way to do it.

1413
01:14:23,920 --> 01:14:26,920
You cannot get lost because there is not like a million different options.

1414
01:14:26,920 --> 01:14:27,920
Yeah.

1415
01:14:27,920 --> 01:14:32,920
Is this one call and it's going to figure out how to do it within the context of the app that, that's running.

1416
01:14:32,920 --> 01:14:33,920
Yeah.

1417
01:14:33,920 --> 01:14:34,920
Yeah.

1418
01:14:34,920 --> 01:14:35,920
Yeah.

1419
01:14:35,920 --> 01:14:39,920
I've been thinking a lot about the kind of different ways of displaying stuff.

1420
01:14:39,920 --> 01:14:42,920
And one thing that's been rattling around in my mind.

1421
01:14:42,920 --> 01:14:48,920
Um, I actually recorded, I did the same thing that you did once and I, I scolded you for it.

1422
01:14:48,920 --> 01:14:52,920
I recorded a voice note for you and then I didn't send it.

1423
01:14:52,920 --> 01:14:56,920
You didn't, you didn't hit your advice or send it.

1424
01:14:57,920 --> 01:14:59,920
No, I'm just going to say the same thing again.

1425
01:14:59,920 --> 01:15:01,920
So, so the right now.

1426
01:15:01,920 --> 01:15:09,920
So the, the three kinds of content types I care most about in some sense are tweets, which are kind of ones.

1427
01:15:09,920 --> 01:15:11,920
I'm going to continue to call them tweets.

1428
01:15:11,920 --> 01:15:12,920
Oh, I don't know.

1429
01:15:12,920 --> 01:15:13,920
Everyone can buy me.

1430
01:15:13,920 --> 01:15:14,920
Yeah.

1431
01:15:14,920 --> 01:15:16,920
I'm going to die on the sale.

1432
01:15:16,920 --> 01:15:19,920
Um, blog posts.

1433
01:15:19,920 --> 01:15:23,920
And, uh, long form audio.

1434
01:15:24,920 --> 01:15:27,920
Basically, but, but it's, I, I really mean long form conversations.

1435
01:15:27,920 --> 01:15:28,920
Yeah.

1436
01:15:28,920 --> 01:15:29,920
I'm not interested in any other podcasts.

1437
01:15:29,920 --> 01:15:38,920
It's like that I'm, I'm, I have very little interest in like super highly produced, like Gimlet-style podcasts for some reason.

1438
01:15:38,920 --> 01:15:39,920
I listened to them.

1439
01:15:39,920 --> 01:15:40,920
Rubbed marketing.

1440
01:15:40,920 --> 01:15:41,920
Yeah.

1441
01:15:41,920 --> 01:15:42,920
In some sense.

1442
01:15:42,920 --> 01:15:43,920
It's like, I listened to them for a long time and so on.

1443
01:15:43,920 --> 01:15:46,920
But now I'm actually really interested in long form conversations.

1444
01:15:46,920 --> 01:15:48,920
I really am interested in that.

1445
01:15:48,920 --> 01:15:54,920
And so, so the way I, the, the, in part why I built the, the vibe line, um, is what I want to do.

1446
01:15:54,920 --> 01:15:57,920
And I'm actually going to use you as a guinea pig and we're going to try this.

1447
01:15:57,920 --> 01:16:00,920
Not right now, but we're, we're going to do it in the next couple of days.

1448
01:16:00,920 --> 01:16:10,920
I want to, for example, explain what sovereign engineering is, what the idea is, how it came to be and so on in a dialogue with you.

1449
01:16:10,920 --> 01:16:16,920
And out of it, I want to vibe, produce a blog post.

1450
01:16:16,920 --> 01:16:17,920
Basically.

1451
01:16:17,920 --> 01:16:19,920
I'm going to edit it a little bit and so on.

1452
01:16:19,920 --> 01:16:31,920
But it's just like, we are going to talk about the sovereign engineering program for 30 minutes or an hour, how it came to be, what the idea is, what we landed on, what worked, what didn't, what we learned, how it works now, all those kinds of things.

1453
01:16:31,920 --> 01:16:35,920
So if you're actually really interested, you're just going to listen to this.

1454
01:16:35,920 --> 01:16:36,920
Okay.

1455
01:16:36,920 --> 01:16:41,920
The output of it, like if you condense this conversation, you're going to do a blog post.

1456
01:16:41,920 --> 01:16:42,920
All right.

1457
01:16:42,920 --> 01:16:43,920
It is a condensed thing.

1458
01:16:43,920 --> 01:16:44,920
That's one of those.

1459
01:16:44,920 --> 01:16:45,920
Writing works, how everything works.

1460
01:16:45,920 --> 01:16:46,920
Yeah.

1461
01:16:46,920 --> 01:16:50,920
And you can automate, like semi-automate a lot of the stuff now with a lot of M's.

1462
01:16:50,920 --> 01:16:53,920
It's going to be, you know, again, there's going to be a lot of AI slop and so on.

1463
01:16:53,920 --> 01:16:54,920
Again, I'm going to edit it.

1464
01:16:54,920 --> 01:16:56,920
I'm just, I don't want to obsolete myself.

1465
01:16:56,920 --> 01:16:58,920
I just want to make my life easier.

1466
01:16:58,920 --> 01:17:02,920
And out of that, I want to create a tweet.

1467
01:17:02,920 --> 01:17:07,920
Out of the blog post, like, do the one line that summarizes the thing.

1468
01:17:07,920 --> 01:17:10,920
Do the, like, literally, like, less than 210 characters.

1469
01:17:10,920 --> 01:17:12,920
It is incredibly good at that.

1470
01:17:12,920 --> 01:17:13,920
Yeah.

1471
01:17:13,920 --> 01:17:14,920
It is also good at that.

1472
01:17:14,920 --> 01:17:16,920
But that's also something I can do myself, potentially.

1473
01:17:16,920 --> 01:17:17,920
Right.

1474
01:17:17,920 --> 01:17:18,920
Where it's like, what do I care about?

1475
01:17:18,920 --> 01:17:20,920
Because the LLM cares about different things tonight.

1476
01:17:20,920 --> 01:17:21,920
In some sense.

1477
01:17:21,920 --> 01:17:22,920
Yeah.

1478
01:17:22,920 --> 01:17:25,920
Like, we are not that, you know, I don't have my own model yet.

1479
01:17:25,920 --> 01:17:27,920
You have a model of me.

1480
01:17:27,920 --> 01:17:30,920
But I don't have a model of myself yet.

1481
01:17:30,920 --> 01:17:32,920
But it belongs to the Internet.

1482
01:17:32,920 --> 01:17:33,920
Yeah.

1483
01:17:33,920 --> 01:17:38,920
But now, in terms of what I want to, what I want to have, so I'm obviously going to use Noster to publish all this stuff.

1484
01:17:38,920 --> 01:17:41,920
So I'm going to publish my tweet on Noster as a kind one.

1485
01:17:41,920 --> 01:17:44,920
I'm going to publish the blog post on Noster as a kind, whatever, long form.

1486
01:17:44,920 --> 01:17:45,920
You know, all the kinds.

1487
01:17:45,920 --> 01:17:46,920
30,000 tweets.

1488
01:17:46,920 --> 01:17:47,920
There we go.

1489
01:17:47,920 --> 01:17:53,920
And I'm going to publish the blog post as well, just as for now, because we don't have anything yet, I think,

1490
01:17:53,920 --> 01:17:57,920
just as, like, a link to the Blossom server, the host.mp3 as, like, a kind one.

1491
01:17:57,920 --> 01:17:58,920
Doesn't matter.

1492
01:17:58,920 --> 01:18:05,920
But if I follow a certain convention, I can show, I can show, and that comes to the, to the different way of viewing things.

1493
01:18:05,920 --> 01:18:08,920
It's like, I'll have a list of tweets.

1494
01:18:08,920 --> 01:18:11,920
All the tweets came from, like, long conversations and blog posts.

1495
01:18:11,920 --> 01:18:13,920
It all went through this process, basically.

1496
01:18:13,920 --> 01:18:14,920
You know what I mean?

1497
01:18:14,920 --> 01:18:15,920
Yeah.

1498
01:18:15,920 --> 01:18:17,920
And then you can expand it, and you can dig deeper.

1499
01:18:17,920 --> 01:18:22,920
It's like, you click once, you're in the same interface, and it pops out, and it shows you the blog post immediately.

1500
01:18:22,920 --> 01:18:28,920
And then if you, if you pop it out again, it shows you the transcript of the podcast, which is really, really long.

1501
01:18:28,920 --> 01:18:33,920
And then you, it pops out again, and then you have a player that actually plays the whole conversation.

1502
01:18:33,920 --> 01:18:35,920
Ah, I like that a lot.

1503
01:18:35,920 --> 01:18:36,920
It's really fun, right?

1504
01:18:36,920 --> 01:18:37,920
It is.

1505
01:18:37,920 --> 01:18:38,920
It's very simple.

1506
01:18:38,920 --> 01:18:42,920
Because, because the, the, the awesome thing about kind ones, the awesome thing about tweets is discovery.

1507
01:18:42,920 --> 01:18:43,920
Oh, 100%.

1508
01:18:43,920 --> 01:18:44,920
It's so condensed.

1509
01:18:44,920 --> 01:18:45,920
Yeah.

1510
01:18:45,920 --> 01:18:46,920
And it's like, I, but I always want to dig deeper.

1511
01:18:46,920 --> 01:18:49,920
If something is interesting, I really want to dig deeper.

1512
01:18:49,920 --> 01:18:51,920
And so I want to know where this tweet came from.

1513
01:18:51,920 --> 01:18:54,920
Oh, it came from this thought in this blog post, those kind of things.

1514
01:18:54,920 --> 01:18:55,920
And I do this manually sometimes.

1515
01:18:55,920 --> 01:19:02,920
I, I'll, I'll post something, and then I'll reply to myself and link the book, or whatever, or those kind of things.

1516
01:19:02,920 --> 01:19:12,920
And it's, it's just like, if, if we have some convention, or even like, if, if we have, again, like, crawlers and LMS and so on, you probably can also figure that out in some sense.

1517
01:19:12,920 --> 01:19:15,920
And I just want to add the context, um, below.

1518
01:19:15,920 --> 01:19:18,920
I actually show you like, my, my line already does something like that, a little bit.

1519
01:19:18,920 --> 01:19:20,920
That's where the idea came from in some sense.

1520
01:19:20,920 --> 01:19:23,920
Because I have the summary, and then I have like the transcript.

1521
01:19:23,920 --> 01:19:26,920
I have to do, I have the things that I just plop it out under it.

1522
01:19:26,920 --> 01:19:31,920
And so there's like, that would make such a nice kind of different way of,

1523
01:19:31,920 --> 01:19:33,920
you know, viewing at stuff.

1524
01:19:33,920 --> 01:19:37,920
And, and I'm very against, I'm very against just for the record.

1525
01:19:37,920 --> 01:19:42,920
I'm very against the missing character limit of kind ones.

1526
01:19:42,920 --> 01:19:43,920
Tweets should be short.

1527
01:19:43,920 --> 01:19:52,920
So I'm very against, in principle, mixing blog posts and tweets, and having tweets that are like 700 characters long and so on.

1528
01:19:52,920 --> 01:19:53,920
I hate that.

1529
01:19:53,920 --> 01:19:54,920
Yeah.

1530
01:19:54,920 --> 01:19:55,920
It's really bad, I think.

1531
01:19:55,920 --> 01:20:02,920
And, and, and that's why I kind of, you know, like, like, amethyst built the blog posts in and shows the blog posts basically in the feed.

1532
01:20:02,920 --> 01:20:03,920
It should pop up in the feed.

1533
01:20:03,920 --> 01:20:05,920
The blog posts should not pop up in the feed.

1534
01:20:05,920 --> 01:20:07,920
Only tweets should pop up in the feed.

1535
01:20:07,920 --> 01:20:08,920
I think.

1536
01:20:08,920 --> 01:20:12,920
But I'm not against expanding the tweet to a blog post in some sense.

1537
01:20:12,920 --> 01:20:13,920
You know what I mean?

1538
01:20:13,920 --> 01:20:14,920
Yeah.

1539
01:20:14,920 --> 01:20:15,920
Yes.

1540
01:20:15,920 --> 01:20:24,920
So one of the feed that, that, that is kind of impossible to address, uh, in, that, that I got from, from highlighter is that,

1541
01:20:24,920 --> 01:20:30,920
by default, well, not by default, but when you publish something, it prompts you if you want to share it as a kind of one.

1542
01:20:30,920 --> 01:20:33,920
Why? Because in, in, in certain clients.

1543
01:20:33,920 --> 01:20:34,920
They want to have the discovery.

1544
01:20:34,920 --> 01:20:35,920
Yeah. They want to have the discovery.

1545
01:20:35,920 --> 01:20:39,920
But in, in, I'm not this, for example, it would show you immediately.

1546
01:20:39,920 --> 01:20:40,920
Show the highlights.

1547
01:20:40,920 --> 01:20:41,920
Straight up.

1548
01:20:41,920 --> 01:20:42,920
It would show you the blog post.

1549
01:20:42,920 --> 01:20:43,920
The blog post.

1550
01:20:43,920 --> 01:20:45,920
And on top of that, oh, read my blog post.

1551
01:20:45,920 --> 01:20:47,920
And people are like, like, dude, why are you spamming?

1552
01:20:47,920 --> 01:20:48,920
Yeah.

1553
01:20:48,920 --> 01:20:49,920
Yeah.

1554
01:20:49,920 --> 01:20:50,920
Yeah.

1555
01:20:50,920 --> 01:20:51,920
Yeah.

1556
01:20:51,920 --> 01:20:53,920
I hope we'll, we'll figure all that out.

1557
01:20:53,920 --> 01:21:04,920
But, but I just, I just think the, the vertical feed, it's, it's, it's, it's, it's not the ideal approach.

1558
01:21:04,920 --> 01:21:06,920
I think it's ideal for short text.

1559
01:21:06,920 --> 01:21:07,920
I really think.

1560
01:21:07,920 --> 01:21:08,920
For short text.

1561
01:21:08,920 --> 01:21:09,920
Yeah.

1562
01:21:09,920 --> 01:21:13,920
I mean, I, Twitter really had something and I hate kind of that it, I hate this is a strong word.

1563
01:21:13,920 --> 01:21:14,920
This is what it is.

1564
01:21:14,920 --> 01:21:20,920
But it's just evolved and then it evolved into like very image heavy and then it evolved into very video heavy.

1565
01:21:20,920 --> 01:21:26,920
And for videos, Twitter is just not the best interface.

1566
01:21:26,920 --> 01:21:29,920
So TikTok is basically for short videos.

1567
01:21:29,920 --> 01:21:32,920
For long videos, it's probably YouTube or something similar.

1568
01:21:32,920 --> 01:21:34,920
You know, YouTube and Netflix basically look alike.

1569
01:21:34,920 --> 01:21:35,920
Yeah.

1570
01:21:35,920 --> 01:21:49,920
But, but if you, if you see the evolution of, of YouTube, you see that they opened the podcast part and then they opened the, the Twitter.

1571
01:21:49,920 --> 01:21:54,920
The live, yeah, the live streaming, the short and the short, but no one, I mean, it's the short.

1572
01:21:54,920 --> 01:21:55,920
Yeah, yeah.

1573
01:21:55,920 --> 01:21:57,920
The short is, tell me, because it's the point of vacation again.

1574
01:21:57,920 --> 01:22:01,920
Yeah, because, because every, every, every, everyone's attention span is wrecked.

1575
01:22:01,920 --> 01:22:02,920
Yes.

1576
01:22:02,920 --> 01:22:03,920
Yeah.

1577
01:22:03,920 --> 01:22:05,920
That, that is a good point about the pornification.

1578
01:22:05,920 --> 01:22:06,920
It really is.

1579
01:22:06,920 --> 01:22:07,920
It really is.

1580
01:22:07,920 --> 01:22:08,920
Yeah.

1581
01:22:08,920 --> 01:22:10,920
I mean, you made that point before, but I, I got it even better now.

1582
01:22:10,920 --> 01:22:11,920
It is like cutting away.

1583
01:22:11,920 --> 01:22:18,920
Like only going to the, to the essence to the act, which, which actually isn't even able to capture the whole essence.

1584
01:22:19,920 --> 01:22:20,920
How could it?

1585
01:22:20,920 --> 01:22:21,920
Yeah.

1586
01:22:21,920 --> 01:22:22,920
How could it?

1587
01:22:22,920 --> 01:22:23,920
Absolutely.

1588
01:22:23,920 --> 01:22:24,920
It can't.

1589
01:22:24,920 --> 01:22:25,920
Yeah.

1590
01:22:25,920 --> 01:22:28,920
It's, it's, it's like listening to a three hour conversation at like two and a half hours, someone has an insight.

1591
01:22:28,920 --> 01:22:33,920
If you listen through the whole thing, you're able to share in this insight and have the insight.

1592
01:22:33,920 --> 01:22:38,920
If you just cut out the insight, the insight they had, and you should post it as like a shorter as a TikTok.

1593
01:22:38,920 --> 01:22:40,920
It, it, it doesn't do anything.

1594
01:22:40,920 --> 01:22:41,920
Because it has no context.

1595
01:22:41,920 --> 01:22:42,920
It has no context.

1596
01:22:42,920 --> 01:22:45,920
That's why I wrote a few weeks ago, like two days.

1597
01:22:46,920 --> 01:22:50,920
And you replied, there, there is no meaning without friction.

1598
01:22:50,920 --> 01:22:51,920
Oh, yeah.

1599
01:22:51,920 --> 01:22:52,920
Oh, life is friction.

1600
01:22:52,920 --> 01:22:53,920
No.

1601
01:22:53,920 --> 01:22:54,920
You, you came for, from friction.

1602
01:22:54,920 --> 01:22:55,920
Oh, yeah.

1603
01:22:55,920 --> 01:22:56,920
There you go.

1604
01:22:56,920 --> 01:22:57,920
Yeah.

1605
01:22:57,920 --> 01:22:58,920
Yeah.

1606
01:22:58,920 --> 01:22:59,920
Yeah.

1607
01:22:59,920 --> 01:23:00,920
You came for friction.

1608
01:23:00,920 --> 01:23:01,920
Jesus Christ.

1609
01:23:01,920 --> 01:23:02,920
That's the portification of the web right there.

1610
01:23:02,920 --> 01:23:03,920
It brought your brain.

1611
01:23:03,920 --> 01:23:04,920
It brought your brain.

1612
01:23:04,920 --> 01:23:05,920
Absolutely.

1613
01:23:05,920 --> 01:23:06,920
Yeah.

1614
01:23:06,920 --> 01:23:13,920
But, but I, the things that mean the most to you, the things that shape you are the things that you earn.

1615
01:23:14,920 --> 01:23:19,920
You know, like, if, if you are just given the wisdom, is it wisdom?

1616
01:23:19,920 --> 01:23:20,920
Yeah.

1617
01:23:20,920 --> 01:23:21,920
It's just information.

1618
01:23:21,920 --> 01:23:23,920
I mean, that's a, you know, nothing, nothing worth having ever comes easy.

1619
01:23:23,920 --> 01:23:24,920
Yes.

1620
01:23:24,920 --> 01:23:28,920
That's also the, the, I'm not sure who asked me this, but it was like, what, what do all

1621
01:23:28,920 --> 01:23:29,920
the Nostra people have in common?

1622
01:23:29,920 --> 01:23:30,920
And it's basically that.

1623
01:23:30,920 --> 01:23:32,920
They are okay with doing the hard thing.

1624
01:23:32,920 --> 01:23:37,920
Everyone who's on Nostra is okay with not being spoon fed and actually have to figure all this shit out on their own.

1625
01:23:37,920 --> 01:23:39,920
And it is actually still really hard.

1626
01:23:39,920 --> 01:23:43,920
Kind of getting set up, figuring out how all this stuff works and so on.

1627
01:23:43,920 --> 01:23:47,920
And, and, and, and you have to build it up in a way until you land on a.

1628
01:23:47,920 --> 01:23:50,920
I have to feel ready to follow the age you have to.

1629
01:23:50,920 --> 01:23:52,920
It's like building your own house in some sense.

1630
01:23:52,920 --> 01:23:53,920
Yeah.

1631
01:23:53,920 --> 01:23:54,920
100%.

1632
01:23:54,920 --> 01:23:55,920
You have to build it in a work.

1633
01:23:55,920 --> 01:23:58,920
But have you ever built like a table or some shit like that?

1634
01:23:58,920 --> 01:23:59,920
Oh, yeah.

1635
01:23:59,920 --> 01:24:05,920
And the table sucks compared to like a media table and it's still your fucking table, man.

1636
01:24:05,920 --> 01:24:08,920
And an Ikea table is already like a very low bar.

1637
01:24:08,920 --> 01:24:09,920
Yeah.

1638
01:24:09,920 --> 01:24:13,920
It sucks compared to an Ikea table, which is basically made out of paper.

1639
01:24:13,920 --> 01:24:16,920
Kind of tables did you build?

1640
01:24:16,920 --> 01:24:17,920
Terrible.

1641
01:24:17,920 --> 01:24:18,920
Terrible.

1642
01:24:18,920 --> 01:24:19,920
Terrible.

1643
01:24:19,920 --> 01:24:29,920
But yeah, no, but I, I really think all of the people have that in common and it is kind of like everyone in the, on Nostra basically right now in some sense.

1644
01:24:29,920 --> 01:24:34,920
And, and the, and the opposite of that, and you said these at some point.

1645
01:24:34,920 --> 01:24:43,920
And I think it captures it perfectly that, that TikTok, like you said before, that, that, that they did almost everything perfect when you are optimizing for that.

1646
01:24:43,920 --> 01:24:44,920
It's short videos.

1647
01:24:44,920 --> 01:24:46,920
It's straight to the payoff.

1648
01:24:46,920 --> 01:24:47,920
Right?

1649
01:24:47,920 --> 01:24:48,920
Also straight into the amygdala.

1650
01:24:48,920 --> 01:24:50,920
Straight into your eyeballs and into your brain.

1651
01:24:50,920 --> 01:24:52,920
And, and you don't have to think about the content.

1652
01:24:52,920 --> 01:24:56,920
And you don't have, there is such little friction that you don't even have to just want to watch.

1653
01:24:56,920 --> 01:24:57,920
Yeah.

1654
01:24:57,920 --> 01:25:01,920
Like you just opened the up and you're, you start receiving a fucking dump.

1655
01:25:01,920 --> 01:25:11,920
But, but then one, to be one of the interesting things is that there is a lot of emotional gensiers crying on TikTok.

1656
01:25:11,920 --> 01:25:14,920
And, and you, like you know that's a thing.

1657
01:25:14,920 --> 01:25:15,920
I'm not sure.

1658
01:25:15,920 --> 01:25:16,920
Yeah.

1659
01:25:16,920 --> 01:25:17,920
Well, yeah, yeah.

1660
01:25:17,920 --> 01:25:20,920
Like all the girls are only emotional and, and filming them how they have.

1661
01:25:20,920 --> 01:25:22,920
And, and, and sharing that.

1662
01:25:22,920 --> 01:25:28,920
And you watch that and it, it's like a very vulnerable moment of a person, right?

1663
01:25:28,920 --> 01:25:31,920
And you're sharing in that moment, but it's a lurker.

1664
01:25:31,920 --> 01:25:33,920
And, and that's not a connection.

1665
01:25:33,920 --> 01:25:36,920
That's not a human, but he, but he uses this.

1666
01:25:36,920 --> 01:25:38,920
I, I've been using like these ideas.

1667
01:25:38,920 --> 01:25:41,920
That's, that's also, it's a growth hack and engagement bait because we know it really works.

1668
01:25:41,920 --> 01:25:42,920
It's like a baby crying, you know.

1669
01:25:42,920 --> 01:25:43,920
Right.

1670
01:25:43,920 --> 01:25:44,920
Like everyone has to pay attention.

1671
01:25:44,920 --> 01:25:48,920
A baby cries and you are, you are hardwired to pay attention.

1672
01:25:48,920 --> 01:25:54,920
Everyone is going to stop whatever they're doing and they are going to try to attend to the baby and watch what's going on and so on.

1673
01:25:54,920 --> 01:25:55,920
And it's the same effect.

1674
01:25:55,920 --> 01:25:56,920
So they figured this out.

1675
01:25:56,920 --> 01:25:57,920
It's like engagement bait on Twitter.

1676
01:25:57,920 --> 01:25:58,920
Right.

1677
01:25:58,920 --> 01:26:02,920
The people that are doing engagement bait on Twitter, they figured out how this stuff works out.

1678
01:26:02,920 --> 01:26:03,920
Sources.

1679
01:26:03,920 --> 01:26:12,920
And, and, and I think that's just the, you know, TikTok, Gen Z, teenage girl way of, of doing that in some sense.

1680
01:26:12,920 --> 01:26:23,920
But, but it's, it's not exclusively that, but I believe, but, but it's, but, but it is again coming back to the medium being the message.

1681
01:26:23,920 --> 01:26:31,920
That the fact that it's so high fidelity and it's video and we are really good at reading faces and reading emotions and so on.

1682
01:26:31,920 --> 01:26:35,920
Of course, all the videos are going to be about faces and emotions.

1683
01:26:35,920 --> 01:26:36,920
You know what I mean?

1684
01:26:36,920 --> 01:26:37,920
Yeah.

1685
01:26:37,920 --> 01:26:38,920
Yeah.

1686
01:26:38,920 --> 01:26:39,920
Absolutely.

1687
01:26:39,920 --> 01:26:41,920
It's like it doesn't work on a pure text environment.

1688
01:26:41,920 --> 01:26:42,920
It just doesn't work.

1689
01:26:42,920 --> 01:26:43,920
Yeah.

1690
01:26:43,920 --> 01:26:45,920
It's like a poor text Twitter tweet environment.

1691
01:26:45,920 --> 01:26:48,920
It's going to be witty, snarky, short comments, those kind of things.

1692
01:26:48,920 --> 01:26:50,920
Which is what Twitter basically was.

1693
01:26:51,920 --> 01:26:53,920
Which goes back to, it's a theater.

1694
01:26:53,920 --> 01:26:54,920
It is.

1695
01:26:54,920 --> 01:26:55,920
Yeah.

1696
01:26:55,920 --> 01:26:56,920
Twitter is not real life.

1697
01:26:56,920 --> 01:26:57,920
Not really.

1698
01:26:57,920 --> 01:26:58,920
TikTok also not.

1699
01:26:58,920 --> 01:26:59,920
Yeah.

1700
01:26:59,920 --> 01:27:01,920
It's funny, right?

1701
01:27:01,920 --> 01:27:11,920
But, but I've been, I've been playing with this idea of, we, we mentioned briefly and then we started talking about lurking.

1702
01:27:11,920 --> 01:27:12,920
Yeah.

1703
01:27:12,920 --> 01:27:13,920
Yeah.

1704
01:27:13,920 --> 01:27:14,920
Yeah.

1705
01:27:14,920 --> 01:27:15,920
You mentioned that the exhibition is part of it basically.

1706
01:27:15,920 --> 01:27:16,920
Yeah.

1707
01:27:17,920 --> 01:27:21,920
It's interesting that the vast majority of internet use is lurking.

1708
01:27:21,920 --> 01:27:22,920
99%.

1709
01:27:22,920 --> 01:27:26,920
Which is every, every guy who builds a client or does anything needs to know that's statistic.

1710
01:27:26,920 --> 01:27:27,920
Yeah.

1711
01:27:27,920 --> 01:27:28,920
99% of people are.

1712
01:27:28,920 --> 01:27:35,920
And it's, it's such a weird concept that we didn't have a word beyond voyeurism.

1713
01:27:35,920 --> 01:27:36,920
Yeah.

1714
01:27:36,920 --> 01:27:37,920
Before the internet.

1715
01:27:37,920 --> 01:27:38,920
Yeah.

1716
01:27:38,920 --> 01:27:45,920
And if you, like, if you, we were to sit down and have a coffee and we're having a conversation and there are 200 people surrounding us.

1717
01:27:45,920 --> 01:27:46,920
Yeah.

1718
01:27:46,920 --> 01:27:47,920
We would think this is.

1719
01:27:47,920 --> 01:27:48,920
200,000.

1720
01:27:48,920 --> 01:27:49,920
Yeah.

1721
01:27:49,920 --> 01:27:50,920
Yeah.

1722
01:27:50,920 --> 01:27:51,920
Yeah.

1723
01:27:51,920 --> 01:27:52,920
What the fuck?

1724
01:27:52,920 --> 01:27:53,920
Why are all these people there?

1725
01:27:53,920 --> 01:27:54,920
It's just listening.

1726
01:27:54,920 --> 01:27:55,920
They're just not saying anything.

1727
01:27:55,920 --> 01:27:56,920
It is in real life.

1728
01:27:56,920 --> 01:27:57,920
It's so weird.

1729
01:27:57,920 --> 01:27:58,920
Yeah.

1730
01:27:58,920 --> 01:27:59,920
But there's friction, right?

1731
01:27:59,920 --> 01:28:01,920
If you are there, we can see you.

1732
01:28:01,920 --> 01:28:02,920
Yeah.

1733
01:28:02,920 --> 01:28:03,920
And there's friction.

1734
01:28:03,920 --> 01:28:04,920
You, you would be out of place.

1735
01:28:04,920 --> 01:28:05,920
Yeah.

1736
01:28:05,920 --> 01:28:06,920
But in the internet, that's the norm.

1737
01:28:06,920 --> 01:28:07,920
Yeah.

1738
01:28:07,920 --> 01:28:08,920
The default?

1739
01:28:08,920 --> 01:28:09,920
The norm is that everyone's watching.

1740
01:28:09,920 --> 01:28:10,920
Basically.

1741
01:28:10,920 --> 01:28:11,920
Yeah.

1742
01:28:11,920 --> 01:28:12,920
Yes.

1743
01:28:12,920 --> 01:28:26,920
And what I've been thinking is that it's kind of like, if your whole intake, food intake is like marshmallows, you're not hungry.

1744
01:28:26,920 --> 01:28:27,920
You are malnourished.

1745
01:28:27,920 --> 01:28:32,920
You're fucking not hungry because you just ate, like, duck marshmallows, right?

1746
01:28:32,920 --> 01:28:33,920
And this is kind of the same.

1747
01:28:33,920 --> 01:28:41,920
People are fulfilling their social connection needs by watching others cry on TikTok.

1748
01:28:41,920 --> 01:28:43,920
That's a good way of putting it.

1749
01:28:43,920 --> 01:28:46,920
You're kind of socially malnourished.

1750
01:28:46,920 --> 01:28:47,920
Yes.

1751
01:28:47,920 --> 01:28:48,920
Yes.

1752
01:28:48,920 --> 01:28:50,920
That's why we seek realness and we seek connection.

1753
01:28:50,920 --> 01:28:51,920
Yeah.

1754
01:28:51,920 --> 01:28:56,920
Because we are surrounded by it and we are as malnourished as ever.

1755
01:28:56,920 --> 01:28:57,920
Yeah.

1756
01:28:57,920 --> 01:28:58,920
Yeah.

1757
01:28:58,920 --> 01:28:59,920
No, that's a good point.

1758
01:28:59,920 --> 01:29:00,920
That's a good point.

1759
01:29:00,920 --> 01:29:02,920
I think that Nostar can provide, can help with that.

1760
01:29:02,920 --> 01:29:04,920
It's not going to replace real connection, those kind of things.

1761
01:29:04,920 --> 01:29:05,920
I completely agree.

1762
01:29:05,920 --> 01:29:06,920
I completely agree.

1763
01:29:06,920 --> 01:29:07,920
I don't think...

1764
01:29:07,920 --> 01:29:10,920
You know, like, just talking about random shit and agreeing all the time.

1765
01:29:10,920 --> 01:29:12,920
That's what the models do, right?

1766
01:29:12,920 --> 01:29:14,920
You are so wrong, man.

1767
01:29:14,920 --> 01:29:16,920
No, but it's...

1768
01:29:16,920 --> 01:29:22,920
I think that's what both Web of Trust and Saps kind of bring realness in some sense.

1769
01:29:22,920 --> 01:29:24,920
I think what we still need is...

1770
01:29:24,920 --> 01:29:26,920
The Web of Trust is kind of...

1771
01:29:26,920 --> 01:29:29,920
It's the first iteration that we have now.

1772
01:29:29,920 --> 01:29:31,920
I think the one thing that we really need is, like,

1773
01:29:31,920 --> 01:29:34,920
there's a difference between a follow and I actually trust this person.

1774
01:29:34,920 --> 01:29:37,920
And we talked about this also as trust being a vector and it's like,

1775
01:29:37,920 --> 01:29:40,920
I trust you for, you know, your thoughts on Nostar and Bitcoin or whatever,

1776
01:29:40,920 --> 01:29:41,920
or your technical expertise.

1777
01:29:41,920 --> 01:29:43,920
I don't trust you when it goes to...

1778
01:29:43,920 --> 01:29:44,920
Building tables.

1779
01:29:44,920 --> 01:29:45,920
Exactly.

1780
01:29:45,920 --> 01:29:46,920
Building tables.

1781
01:29:46,920 --> 01:29:47,920
Not so much.

1782
01:29:47,920 --> 01:29:52,920
Your opinion on woodwork can be disregarded.

1783
01:29:52,920 --> 01:29:54,920
But it's also...

1784
01:29:54,920 --> 01:29:57,920
Yeah, it's really hard and messy to model all these stuff.

1785
01:29:57,920 --> 01:29:59,920
I think maybe LLMs can help with that as well because, again,

1786
01:29:59,920 --> 01:30:02,920
they are really good with the messiness.

1787
01:30:02,920 --> 01:30:03,920
Yeah.

1788
01:30:03,920 --> 01:30:06,920
Because modeling, trying to model it precisely is kind of...

1789
01:30:06,920 --> 01:30:07,920
It's a falter.

1790
01:30:07,920 --> 01:30:09,920
It's semantic web all over.

1791
01:30:09,920 --> 01:30:10,920
Exactly.

1792
01:30:10,920 --> 01:30:11,920
Yeah, yeah, yeah.

1793
01:30:11,920 --> 01:30:13,920
And most people don't know I come from this world.

1794
01:30:13,920 --> 01:30:16,920
I know how misguided the semantic web is.

1795
01:30:16,920 --> 01:30:22,920
I spend a lot of time, you know, trying to build everything into triples.

1796
01:30:22,920 --> 01:30:25,920
It is very Tommy, so it fits.

1797
01:30:25,920 --> 01:30:26,920
Yeah.

1798
01:30:26,920 --> 01:30:27,920
RDF, everything.

1799
01:30:27,920 --> 01:30:28,920
It's like...

1800
01:30:28,920 --> 01:30:29,920
Yeah.

1801
01:30:29,920 --> 01:30:31,920
That's good times.

1802
01:30:31,920 --> 01:30:32,920
Good times.

1803
01:30:32,920 --> 01:30:35,920
No, ontologies don't really work as we think they will.

1804
01:30:35,920 --> 01:30:37,920
You hear that huddle about?

1805
01:30:37,920 --> 01:30:40,920
If people would just study their philosophy, they would know.

1806
01:30:40,920 --> 01:30:41,920
They would know.

1807
01:30:41,920 --> 01:30:44,920
You cannot model the world perfectly in like...

1808
01:30:44,920 --> 01:30:47,920
But that is an extremely interesting idea, LLMs.

1809
01:30:47,920 --> 01:30:53,920
Because they can get all the broad data and make something out of it at scale,

1810
01:30:53,920 --> 01:30:55,920
which is what ontologies want to do.

1811
01:30:55,920 --> 01:30:57,920
But they will also always be wrong.

1812
01:30:57,920 --> 01:30:58,920
Yes, yes.

1813
01:30:58,920 --> 01:30:59,920
But that's the difference.

1814
01:30:59,920 --> 01:31:04,920
That's how they differ from ontologies and logical systems because they will be internally right

1815
01:31:04,920 --> 01:31:06,920
and free of contradictions in the best case.

1816
01:31:06,920 --> 01:31:07,920
Again, but...

1817
01:31:07,920 --> 01:31:08,920
Call me a commie point.

1818
01:31:08,920 --> 01:31:09,920
No, no, no.

1819
01:31:09,920 --> 01:31:10,920
Absolutely.

1820
01:31:10,920 --> 01:31:11,920
It is a commie.

1821
01:31:11,920 --> 01:31:12,920
It is a commie world view.

1822
01:31:12,920 --> 01:31:15,920
The idea that you cannot have...

1823
01:31:15,920 --> 01:31:17,920
The idea that you cannot have...

1824
01:31:17,920 --> 01:31:19,920
And data triplets are not the problem.

1825
01:31:19,920 --> 01:31:20,920
It's communism.

1826
01:31:20,920 --> 01:31:22,920
And we can't tell me that.

1827
01:31:22,920 --> 01:31:24,920
And we can't tell me everybody is a commie.

1828
01:31:24,920 --> 01:31:25,920
Yeah, yeah.

1829
01:31:25,920 --> 01:31:26,920
Commies are at fault.

1830
01:31:26,920 --> 01:31:27,920
Yeah.

1831
01:31:27,920 --> 01:31:29,920
No, but you just have to also be aware.

1832
01:31:29,920 --> 01:31:32,920
I mean, it's funny how LLMs really are like humans in some sense.

1833
01:31:32,920 --> 01:31:36,920
It's like, if you're really smart, you also are really good at hallucinating and making stuff up.

1834
01:31:36,920 --> 01:31:37,920
Oh, 100%.

1835
01:31:37,920 --> 01:31:38,920
You're really good at bullshitting and so on.

1836
01:31:38,920 --> 01:31:40,920
And LLMs are experts at bullshitting and hallucinating.

1837
01:31:40,920 --> 01:31:47,920
I remember, when was it, like, two years ago or so, that you mentioned that?

1838
01:31:47,920 --> 01:31:51,920
That the hallucination part was like a key component?

1839
01:31:51,920 --> 01:31:53,920
You can't get rid of it.

1840
01:31:53,920 --> 01:31:54,920
Yeah.

1841
01:31:54,920 --> 01:31:56,920
And I was like, I did not understand it at the time.

1842
01:31:56,920 --> 01:31:57,920
Yeah.

1843
01:31:57,920 --> 01:31:58,920
You can't get rid of it.

1844
01:31:58,920 --> 01:32:03,920
It is a requirement for creativity and thinking, basically, because you have to hallucinate

1845
01:32:03,920 --> 01:32:06,920
what the right solution might be.

1846
01:32:06,920 --> 01:32:07,920
Yeah.

1847
01:32:07,920 --> 01:32:09,920
That's how you come up with new stuff.

1848
01:32:09,920 --> 01:32:12,920
But if something has to ground you in reality and if you're not grounded in reality, which

1849
01:32:12,920 --> 01:32:18,920
LLMs definitely aren't, then you just go out of bounds and you hallucinate more and more.

1850
01:32:18,920 --> 01:32:20,920
That's what I haven't played with that yet.

1851
01:32:20,920 --> 01:32:27,920
But it's one of the things that I want to do soon is playing with local temperatures when one

1852
01:32:27,920 --> 01:32:28,920
is showing up.

1853
01:32:28,920 --> 01:32:33,920
You know, like, the architect agent, let it hallucinate.

1854
01:32:33,920 --> 01:32:34,920
Yeah.

1855
01:32:34,920 --> 01:32:35,920
And then they call it one?

1856
01:32:35,920 --> 01:32:36,920
No, no, no, no, be structured.

1857
01:32:36,920 --> 01:32:37,920
Absolutely.

1858
01:32:37,920 --> 01:32:38,920
I think that's what we will work best.

1859
01:32:38,920 --> 01:32:43,920
That's what I meant before with, like, the split between the experts and the kind of big

1860
01:32:43,920 --> 01:32:44,920
picture thinkers.

1861
01:32:44,920 --> 01:32:49,920
The big picture thinkers, they have to be allowed to go wild and hallucinate and kind of make

1862
01:32:49,920 --> 01:32:51,920
stuff up and throw shit at the wall in some sense.

1863
01:32:51,920 --> 01:32:56,920
And what would be cool is using exactly the same system with the same front, everything

1864
01:32:56,920 --> 01:33:01,920
the same and just five different toggles, five different settings on temperature.

1865
01:33:01,920 --> 01:33:02,920
That's also what you can do.

1866
01:33:02,920 --> 01:33:04,920
And see what the fuck comes out.

1867
01:33:04,920 --> 01:33:08,920
Yeah, you can just do it like, do tiny durations of that and compare it.

1868
01:33:08,920 --> 01:33:10,920
Run tests on it.

1869
01:33:10,920 --> 01:33:11,920
That's good.

1870
01:33:11,920 --> 01:33:12,920
I'm pretty cooked.

1871
01:33:12,920 --> 01:33:13,920
What did you say?

1872
01:33:13,920 --> 01:33:14,920
Yeah, me too.

1873
01:33:14,920 --> 01:33:16,920
I'm thirsty as fuck too.

1874
01:33:16,920 --> 01:33:17,920
You thirsty?

1875
01:33:17,920 --> 01:33:19,920
Are we back at the pornification?

1876
01:33:19,920 --> 01:33:23,920
I know you thirsty problem.

1877
01:33:23,920 --> 01:33:25,920
We can fix that.

1878
01:33:26,920 --> 01:33:27,920
I'm not that thirsty.

1879
01:33:27,920 --> 01:33:29,920
Let's have a coffee or something.

1880
01:33:29,920 --> 01:33:35,920
And I actually, if you still have like 20 minutes of time or so, I want to try the, what is

1881
01:33:35,920 --> 01:33:38,920
Summer Engineering recording slash blog posts.

1882
01:33:38,920 --> 01:33:39,920
Let's go.

1883
01:33:39,920 --> 01:33:43,920
Because that will force me to actually build the fucking client.

1884
01:33:43,920 --> 01:33:44,920
If I have the data.

1885
01:33:44,920 --> 01:33:46,920
There's an mp3 there.

1886
01:33:46,920 --> 01:33:47,920
It's just cheating.

1887
01:33:47,920 --> 01:33:51,920
And there's a blog post and I'm going to produce a tweet.

1888
01:33:51,920 --> 01:33:54,920
I'll have to force myself to do that.

1889
01:33:54,920 --> 01:33:56,920
How hard can it be?

1890
01:33:56,920 --> 01:33:58,920
It's like 20 minutes in a garage.

1891
01:33:58,920 --> 01:34:02,920
That's my current life.

1892
01:34:02,920 --> 01:34:09,920
As soon as everyone is asleep at the house, I'll go down in my vibe chamber.

1893
01:34:09,920 --> 01:34:11,920
The vibe chamber in the garage.

1894
01:34:11,920 --> 01:34:13,920
I can stand again now.

1895
01:34:13,920 --> 01:34:15,920
I can walk again now a little bit.

1896
01:34:15,920 --> 01:34:18,920
So I'll just plop myself down there, vibe code for an hour or so.

1897
01:34:18,920 --> 01:34:20,920
And then I'm going to call it a day.

1898
01:34:20,920 --> 01:34:24,920
Every time I just leave, it's still doing stuff.

1899
01:34:24,920 --> 01:34:30,920
It's just like really large prompt, yellow mode, hit enter, go to sleep.

1900
01:34:30,920 --> 01:34:33,920
And when I wake up, it's a new world.

1901
01:34:33,920 --> 01:34:35,920
My heart disc might be wiped.

1902
01:34:35,920 --> 01:34:36,920
The garage might be on fire.

1903
01:34:36,920 --> 01:34:37,920
You'll never know.

1904
01:34:37,920 --> 01:34:40,920
Your star is completely drained.

1905
01:34:40,920 --> 01:34:44,920
No, but it's so funny because I hope the recording worked this time.

1906
01:34:44,920 --> 01:34:46,920
I'm getting better at pressing record as well and so on.

1907
01:34:46,920 --> 01:34:50,920
And once I hit save, the pipeline in the garage is going to kick in,

1908
01:34:50,920 --> 01:34:52,920
and it's going to kick in the transcription and everything.

1909
01:34:52,920 --> 01:34:57,920
I don't have it there yet where the goal is to auto-normalize it

1910
01:34:57,920 --> 01:34:59,920
and auto-upload it to blossom and so on.

1911
01:34:59,920 --> 01:35:01,920
But you know, like next week, I'll have it.

1912
01:35:01,920 --> 01:35:02,920
It's like five minutes over.

1913
01:35:02,920 --> 01:35:03,920
It's fine.

1914
01:35:03,920 --> 01:35:04,920
Not even.

1915
01:35:04,920 --> 01:35:05,920
Not even.

1916
01:35:05,920 --> 01:35:07,920
But yeah, we're getting there.

1917
01:35:07,920 --> 01:35:09,920
It's a strange new world, but I like it.

1918
01:35:09,920 --> 01:35:10,920
Vites are high.

1919
01:35:10,920 --> 01:35:11,920
Vites are high.

1920
01:35:11,920 --> 01:35:12,920
Vites are high.

1921
01:35:12,920 --> 01:35:13,920
Vites are high.

1922
01:35:13,920 --> 01:35:15,920
That's the most important one.

1923
01:35:15,920 --> 01:35:16,920
All right.

1924
01:35:16,920 --> 01:35:17,920
Let's see if they're worked.

1925
01:35:17,920 --> 01:35:19,920
I'm going to hit stop or I'm going to hit save.

1926
01:35:19,920 --> 01:35:20,920
Oh, yeah.

1927
01:35:20,920 --> 01:35:21,920
Let's do that.