1
00:00:00,000 --> 00:00:08,080
Welcome to Pleb Chain Radio, a live show brought to you by Plebs for Plebs, which focuses

2
00:00:08,080 --> 00:00:10,940
on the intersection of Nostra and Bitcoin protocols.

3
00:00:11,700 --> 00:00:16,700
Join QW and Avi as they run down the weekly news and developments, breaking down the current

4
00:00:16,700 --> 00:00:21,600
thing and the future frontier with the foundation of decentralization, the builders, thinkers,

5
00:00:21,840 --> 00:00:22,780
doers and plebs.

6
00:00:26,500 --> 00:00:28,360
All right, we are live.

7
00:00:28,360 --> 00:00:35,460
Welcome, gentle plebs, to the lightning-laced airwaves. It is Friday the 5th of September.

8
00:00:35,920 --> 00:00:44,160
We're in September already, QW. Look at that. And it is episode 128 of Pleb Chain Radio.

9
00:00:44,760 --> 00:00:52,000
It's exactly 4pm on the dot on the East Coast of the United States at the time of recording.

10
00:00:52,000 --> 00:00:54,420
we have a fun show ahead of you today

11
00:00:54,420 --> 00:00:56,580
Alex Gleason, the Gleasonator

12
00:00:56,580 --> 00:00:58,560
also the creator of

13
00:00:58,560 --> 00:01:00,320
Soapbox and Shakespeare

14
00:01:00,320 --> 00:01:02,460
is here to break

15
00:01:02,460 --> 00:01:04,420
all of that down for us

16
00:01:04,420 --> 00:01:06,020
and a reminder folks

17
00:01:06,020 --> 00:01:08,900
if you are listening to the show on Apple or Spotify

18
00:01:08,900 --> 00:01:10,560
first of all thank you for listening

19
00:01:10,560 --> 00:01:12,640
but I would urge you to hit pause

20
00:01:12,640 --> 00:01:15,000
and switch over to the Fountain Podcasting app

21
00:01:15,000 --> 00:01:16,560
where you can earn some sats

22
00:01:16,560 --> 00:01:18,840
and support the value for value revolution

23
00:01:18,840 --> 00:01:19,780
as you listen

24
00:01:19,780 --> 00:01:21,120
and while you're there

25
00:01:21,120 --> 00:01:29,080
we would appreciate it if you hit that subscribe button it helps us keep our mics warm and the show

26
00:01:29,080 --> 00:01:36,840
ad free this show is streamed live on zap.stream and a burgeoning list of Noster clients that

27
00:01:36,840 --> 00:01:45,280
support streaming such as Amethyst, Noster, Fountain and in the recent past QW we had a bit

28
00:01:45,280 --> 00:01:52,920
for Misfire last week with the denial of service attack on Zap.stream. But as of this week for us,

29
00:01:53,600 --> 00:01:59,460
streamed live on Primal as well. I see it, Avi. I see it. We're live. We are live. What's up,

30
00:01:59,560 --> 00:02:07,920
Primal? We made it. That's good to know. Yeah. It is good to know, QW. And you've been paying

31
00:02:07,920 --> 00:02:12,100
attention to Bitcoin and Noster culture in the last week. What have you witnessed?

32
00:02:12,100 --> 00:02:18,440
everyone's a streamer, Avi. Everyone's testing their own stream. I'm seeing a lot of the pop-ups

33
00:02:18,440 --> 00:02:24,180
because I mean, I follow a lot of people. I mean, I got skin in the game with the friend zone,

34
00:02:24,280 --> 00:02:33,840
the social graph. So I mean, just right now there's, well, we got Odell streaming. We got

35
00:02:33,840 --> 00:02:41,580
Walker streaming. We got the grinder doing his gaming. I'm seeing a lot of testing. So it's,

36
00:02:41,580 --> 00:02:44,720
I think everyone's just playing with the new toys, the new shiny toys.

37
00:02:45,120 --> 00:02:50,500
And that's specifically primal users because, you know, there's a good amount of primal users.

38
00:02:50,700 --> 00:02:52,460
So everyone's playing with the shiny toys.

39
00:02:52,460 --> 00:02:58,960
And I'm noticing people like Sergio testing things, the Bitcoin bartender.

40
00:02:59,920 --> 00:03:00,840
It's just great to see.

41
00:03:01,180 --> 00:03:07,900
Obviously, you know, we're seeing that 24-hour stream from Scardust.

42
00:03:07,900 --> 00:03:13,840
it's just it's neat to see everything kind of come together in a smooth fashion so

43
00:03:13,840 --> 00:03:19,640
it's something where people are like hey how do I do that so now they're trying it and I've

44
00:03:19,640 --> 00:03:24,720
noticed a lot of people just watching a silly stream I think everyone just testing the chat

45
00:03:24,720 --> 00:03:31,360
testing the zap just testing the features and that's what I mean like million said I think it

46
00:03:31,360 --> 00:03:36,640
was in Prague Nostra is just one big experiment right now and that's essentially what everyone's

47
00:03:36,640 --> 00:03:44,280
doing. We're experimenting. So it's, it's fun. And looks like we just got a 42 set zap on primal

48
00:03:44,280 --> 00:03:49,360
possibly. So, you know, they're going to be testing while we're, we're, we're talking here,

49
00:03:49,400 --> 00:03:56,300
Avi. And what's more poetic than that beyond Shakespeare, but totally Nostra. That's the

50
00:03:56,300 --> 00:04:03,560
Nostra culture. Just test away. So yeah, that's what, that's what I'm noticing. And dare I say,

51
00:04:03,560 --> 00:04:11,760
Butzikol Avi. I'm not going to go into it, but this week was overtaken by an interesting probiotic

52
00:04:11,760 --> 00:04:19,580
pleb suggestion. So there you go. I'm sorry for saying it.

53
00:04:21,660 --> 00:04:23,580
And what have you got for our sermon today, KW?

54
00:04:24,580 --> 00:04:30,140
Well, you know, experimenting, experimenting, and I've experimented. I've done a little bit

55
00:04:30,140 --> 00:04:37,900
of vibe coding. I'm kind of a, I'm like a Luddite, I feel like is in regards to moving along with AI.

56
00:04:38,080 --> 00:04:42,180
Obviously you were kind of a first mover. You can, and I asked you about AI. I'm like, Hey,

57
00:04:42,180 --> 00:04:47,720
which, which AI should I use for this, that, and you have, you know, you have, you were pretty much

58
00:04:47,720 --> 00:04:53,800
doing it for your job at one time too, at least applying it. So you've kind of been my, my Miyagi

59
00:04:53,800 --> 00:05:03,760
in a way. And I come from a very hands-on service-led industry, construction. It's tough

60
00:05:03,760 --> 00:05:11,020
for me to wrap my, just believe in it, I suppose. But I've tried it. I use it now. I'm starting to

61
00:05:11,020 --> 00:05:17,600
integrate it. I'm starting to use it within our business. So I'm coming around. But when it comes

62
00:05:17,600 --> 00:05:23,080
to this vibe coding, and I see a lot on Nostra because Nostra is a place of experimenting. You

63
00:05:23,080 --> 00:05:28,360
get a lot of people that we appreciate the coders, we appreciate the developers. Are we there yet,

64
00:05:28,460 --> 00:05:35,820
Avi? Are we there yet? Has vibe coding truly arrived or is it a time-consuming bug-filled

65
00:05:35,820 --> 00:05:43,120
experiment? Does the final product equal the cost or is there even a final polished product?

66
00:05:43,720 --> 00:05:47,620
I want to believe anyone could be a coder. I want to believe I am as well.

67
00:05:47,620 --> 00:05:56,300
but if I go to the driving range once, am I now a golfer? And that's one of the big things. It's

68
00:05:56,300 --> 00:06:02,920
like, you know, I'm a coder. I mean, you can be, it's a meme in a way, but I feel like the

69
00:06:02,920 --> 00:06:07,580
development just keeps getting better. It gets better. But at the same time, there's a graveyard

70
00:06:07,580 --> 00:06:13,640
of proof of concepts, essentially, abandoned where riddled with in recent history. I've seen

71
00:06:13,640 --> 00:06:19,440
last one month, you know, you'll see, oh, look at this mini app, this mini app, you try it. And then

72
00:06:19,440 --> 00:06:24,660
the person that developed it's onto the next one. And maybe that's a contagion of how exciting this

73
00:06:24,660 --> 00:06:30,360
is. But I don't see that. I see me using it. And then I'm like, hey, what happened here? What

74
00:06:30,360 --> 00:06:34,720
happened to this bug? And then the person that developed it's onto the next thing. So there's

75
00:06:34,720 --> 00:06:40,440
certainly a problem there. But, you know, I like to think that AI development, just like we've seen

76
00:06:40,440 --> 00:06:46,540
with anything, it will get better and that problem will solve itself. But, you know, is it today?

77
00:06:46,880 --> 00:06:51,040
Are we truly there yet, Avi? I don't think we're truly there yet,

78
00:06:51,180 --> 00:06:56,280
Kitabi. And, you know, we've talked about this a few times on the show. And I expressed the view

79
00:06:56,280 --> 00:07:03,360
that the vast majority of Vibe-coded apps that are coming out now, it's akin to the meme coin

80
00:07:03,360 --> 00:07:09,680
mania, right? You just keep, it's trying to catch attention. Most of them fail. They enter the

81
00:07:09,680 --> 00:07:11,760
graveyard of abandonware.

82
00:07:12,580 --> 00:07:14,960
But there might be one or two that do succeed.

83
00:07:15,060 --> 00:07:19,360
If I'm not mistaken, BitChat was largely Vibe-coded.

84
00:07:19,580 --> 00:07:22,880
BitChat is Jack Dorsey's app.

85
00:07:22,960 --> 00:07:26,060
We've talked about this as well on the last couple of shows.

86
00:07:27,240 --> 00:07:28,040
If I'm not mistaken.

87
00:07:28,040 --> 00:07:28,940
It's a good point, though.

88
00:07:29,860 --> 00:07:34,140
Successful Vibe-coded apps done by actual coders

89
00:07:34,140 --> 00:07:36,900
have a great chance of success, correct?

90
00:07:38,440 --> 00:07:39,040
Well, yes.

91
00:07:39,040 --> 00:07:41,880
They can actually fix things and they understand what's going on.

92
00:07:43,000 --> 00:07:43,240
Yeah.

93
00:07:43,240 --> 00:07:43,680
I don't know.

94
00:07:43,820 --> 00:07:43,960
Yeah.

95
00:07:44,740 --> 00:07:46,840
And I know it's going to help a lot.

96
00:07:47,000 --> 00:07:48,820
So, you know, I want to learn more.

97
00:07:48,960 --> 00:07:54,240
I'm going to be a contrarian in this during this episode because I think Alex, he's got a great mind.

98
00:07:54,240 --> 00:07:56,540
And I think we can talk about this.

99
00:07:57,040 --> 00:07:58,680
So, yeah.

100
00:07:58,680 --> 00:07:59,520
Let's do it.

101
00:08:00,020 --> 00:08:00,300
Yeah.

102
00:08:00,420 --> 00:08:03,340
Welcome to the show, the Gleasonator.

103
00:08:03,720 --> 00:08:08,260
And Alex is actually sending us 100 sats every time we call them the Gleasonator.

104
00:08:08,260 --> 00:08:10,880
Hello, hello. Happy to be here.

105
00:08:11,560 --> 00:08:17,740
I hope I didn't offend you with my contrarian view when we opened the sermon there.

106
00:08:18,260 --> 00:08:19,680
You're about to get glicinated, buddy.

107
00:08:20,960 --> 00:08:30,540
I hope so. And I think there's a lot of people like me that maybe don't understand or maybe can't see the future because we have no experience in the past.

108
00:08:30,540 --> 00:08:39,680
So, you know, I like to take that view and, you know, sprinkle a little fairy dust on it, I suppose.

109
00:08:41,460 --> 00:08:42,680
No, absolutely.

110
00:08:44,780 --> 00:08:45,580
Yeah, go ahead.

111
00:08:46,220 --> 00:08:47,840
Yeah, and you've done a show with Avi.

112
00:08:48,040 --> 00:08:51,740
Avi, did you want to plug your other show while we're here?

113
00:08:52,480 --> 00:08:55,260
Well, you know, it will be an interesting one for folks to listen to.

114
00:08:55,380 --> 00:08:58,320
This is on my AI-focused podcast called Acceleration.

115
00:08:58,320 --> 00:09:04,200
uh it's the sixth episode i think alex we spoke two or three months ago and i think it'll be an

116
00:09:04,200 --> 00:09:09,760
interesting one for folks to listen to because uh it's at this point a historical artifact

117
00:09:09,760 --> 00:09:18,140
we talked about ai developments that are completely out of date as of today uh but uh but alex maybe

118
00:09:18,140 --> 00:09:24,120
we start where we usually do on the show we're just talking a little bit about your journey to

119
00:09:24,120 --> 00:09:30,440
NOSTA and well actually decentralized social media because that's something you've been big in

120
00:09:30,440 --> 00:09:37,900
how did that journey begin and how did that lead you to NOSTA yeah so I want to just say

121
00:09:37,900 --> 00:09:42,780
acceleration is the name of the game right now and I also have so many thoughts about

122
00:09:42,780 --> 00:09:49,740
what has been said that I want to pick apart but in terms of how I got started so you know

123
00:09:49,740 --> 00:09:57,640
back in the day, I always thought I was going to be an artist. And so I was working on creating a

124
00:09:57,640 --> 00:10:02,680
book. And this kind of got me down into this rabbit hole of copyright and open source. And

125
00:10:02,680 --> 00:10:08,220
I ended up realizing my talents were better spent doing programming. But I've always kind of taken

126
00:10:08,220 --> 00:10:15,920
this, you know, artist's point of view on the things that I build. And another big motif, I guess,

127
00:10:15,920 --> 00:10:21,640
of my life has been freedom. You know, seeking freedom and fighting for freedom, not just for

128
00:10:21,640 --> 00:10:28,620
myself, but for other people too. And, and so I've been involved in open source for a really long

129
00:10:28,620 --> 00:10:36,980
time. And I knew that social media is a big area to, for improvement in terms of user freedom.

130
00:10:37,620 --> 00:10:42,680
And so that's why I was really excited when I started seeing solutions like Mastodon come out,

131
00:10:42,680 --> 00:10:45,780
because I'm like, oh, wow, you know, this is maybe the solution to the problem.

132
00:10:45,940 --> 00:10:49,900
So I jumped into that and I started just building stuff on it.

133
00:10:49,980 --> 00:10:54,600
I created Soapbox, which at the time was a fork of Mastodon.

134
00:10:54,600 --> 00:10:58,960
But now really today it's more the name of our organization than anything.

135
00:10:59,180 --> 00:11:03,200
But that word, Soapbox, has meant different things throughout the years.

136
00:11:04,500 --> 00:11:08,400
But then I went on from there to become the head of engineering at Truth Social,

137
00:11:08,620 --> 00:11:10,200
which is a crazy twist of events.

138
00:11:10,280 --> 00:11:11,440
I still can't believe that happened.

139
00:11:11,440 --> 00:11:14,100
And it feels like a dream to me when I think back on it now.

140
00:11:15,080 --> 00:11:26,800
And that was just a crazy timing situation where I was in the right place at the right time, I guess, or the wrong place, depending on what your perspective is on political matters.

141
00:11:27,880 --> 00:11:30,660
But it was a really big opportunity for me.

142
00:11:31,180 --> 00:11:36,520
And so I was happy to help, especially because I didn't think that Donald Trump should have been banned off of Twitter.

143
00:11:36,520 --> 00:11:49,592
even though I don support his policies in other ways that just my point point of view as a principled like freedom advocate So yeah so I helped build Truth Social

144
00:11:49,592 --> 00:11:57,652
I eventually quit when I realized that our visions were not aligned anymore because I really still wanted to do open source decentralized freedom tech.

145
00:11:57,752 --> 00:12:00,092
And it was clear to me that that's not what they wanted to do.

146
00:12:00,172 --> 00:12:07,112
They wanted to do a conservative hug box where they get to control, you know, speech of their political opponents.

147
00:12:07,692 --> 00:12:15,592
So once I realized that that's what was happening, which is probably obvious to anyone else looking in on this thing, then I'm like, all right, I'm out.

148
00:12:16,592 --> 00:12:21,252
And so from there, I had been doing stuff in ActivityPub and the Fediverse and Mastodon.

149
00:12:21,252 --> 00:12:29,392
But as I was changing that part of my life and career, I decided that now would be the best time to jump into Noster.

150
00:12:29,392 --> 00:12:35,132
and so the first thing I did was I built a bridge between ActivityPub and Noster which I know that

151
00:12:35,132 --> 00:12:42,012
Avi and many other people on Noster hates but I think that I think that it's been a great success

152
00:12:42,012 --> 00:12:48,112
in many ways because it has shown that when you have two decentralized systems that they can be

153
00:12:48,112 --> 00:12:53,372
linked together in a permissionless way and and the only reason a person would dislike this is

154
00:12:53,372 --> 00:12:59,792
because of like culture clash between these, you know, these differing groups. Um, so that was

155
00:12:59,792 --> 00:13:04,812
sort of my first foray into it. And then I, I started developing more Noster software and

156
00:13:04,812 --> 00:13:13,232
eventually led down the path of building stuff with AI. Um, because of, I mean, it got to a point

157
00:13:13,232 --> 00:13:18,672
where I just couldn't ignore the acceleration anymore. Um, and so I'm like, all right, I got

158
00:13:18,672 --> 00:13:26,332
to embrace this or die uh and i at first it was a little bit of doom and gloom about it but but

159
00:13:26,332 --> 00:13:33,292
then as i started building stuff on i'm like you know this is actually really fun um and so my some

160
00:13:33,292 --> 00:13:40,872
of my most recent work involves mk stack stack cli and shakespeare and so when we talked last time i

161
00:13:40,872 --> 00:13:45,972
i think stacks was released but shakespeare was not released now shakespeare.diy is a thing

162
00:13:45,972 --> 00:13:53,212
And I'm working on a second version of it currently that I'm really excited to tell you guys about.

163
00:13:53,432 --> 00:13:56,892
So that's basically it. That's my whole life story up until up until here.

164
00:13:57,572 --> 00:13:59,292
Let me know what. Yeah, go ahead.

165
00:13:59,292 --> 00:14:04,652
I want to pull the string on that true social because I agree with you.

166
00:14:04,752 --> 00:14:09,912
It was not about, you know, it wasn't about believing in a person's point of view.

167
00:14:10,112 --> 00:14:15,772
It was believing his freedom of speech and how truth social came about was,

168
00:14:15,972 --> 00:14:23,832
was really people were getting censored or taken down on X or social media platforms around the main ones.

169
00:14:24,312 --> 00:14:27,772
And then you had these pop-up ones like Parler, Gab.

170
00:14:28,112 --> 00:14:29,332
Gab got inundated.

171
00:14:29,432 --> 00:14:34,812
You couldn't even use the thing because there were so many people going there and maybe it was getting attacked.

172
00:14:35,432 --> 00:14:42,992
Parler, that was an amazing thing because it was getting, I think Amazon Web Service took down Parler.

173
00:14:42,992 --> 00:14:43,332
Exactly.

174
00:14:43,332 --> 00:14:47,072
So it was this technology.

175
00:14:47,312 --> 00:14:49,632
I mean, it wasn't just one person.

176
00:14:49,632 --> 00:14:51,452
It exposed the fragility of the internet.

177
00:14:52,032 --> 00:14:52,652
Exactly.

178
00:14:53,352 --> 00:15:01,132
And I don't think, and how much these conglomerates or these corporations were actually working, meshed with the government.

179
00:15:01,692 --> 00:15:04,752
And I think that's just going to be what we're going to see in the EU.

180
00:15:05,032 --> 00:15:11,572
We're going to see around the world if you don't continue to have these other options keep popping up.

181
00:15:11,572 --> 00:15:18,072
And it was funny because all of a sudden Truth came out and, you know, it was a president of the United States.

182
00:15:18,672 --> 00:15:24,252
So somewhere along the line, he was a little bit more untouchable when it came to that.

183
00:15:24,312 --> 00:15:24,952
I don't know.

184
00:15:25,032 --> 00:15:33,772
But why do you think, and we don't have to get into it, maybe just a quick reason, but why do you think Gab and Parler didn't work out but Truth did?

185
00:15:34,812 --> 00:15:36,392
I'm not sure the truth worked out.

186
00:15:36,392 --> 00:15:40,252
I don't think it's working out, but I'm saying it was actually functional, right?

187
00:15:40,252 --> 00:15:42,872
Because everything else was getting just blown up.

188
00:15:43,532 --> 00:15:45,972
Or maybe it was a little bit of the Elon purchase.

189
00:15:45,972 --> 00:15:48,152
Because I was involved.

190
00:15:48,892 --> 00:15:49,132
Yeah.

191
00:15:50,652 --> 00:15:51,412
Well played.

192
00:15:53,012 --> 00:15:53,632
All right.

193
00:15:53,712 --> 00:15:54,952
Well, we don't need to go into that.

194
00:15:55,252 --> 00:15:59,232
It was just when you started talking about that, I was like, you know, that was all cause and effect.

195
00:15:59,372 --> 00:16:01,292
And it was interesting times.

196
00:16:01,512 --> 00:16:02,692
I can say that.

197
00:16:04,272 --> 00:16:06,952
I actually want to pull on the thread a little.

198
00:16:06,952 --> 00:16:12,432
not on the truth part, but what you said, Alex, about the fragility of the internet. I don't think

199
00:16:12,432 --> 00:16:18,012
people were fully exposed or were exposed to the full fragility of the internet, because what they

200
00:16:18,012 --> 00:16:25,592
saw with, you know, in early 2021, when all of this was going down, was just one layer of fragility,

201
00:16:25,592 --> 00:16:31,812
which is, oh, wow, Amazon, it's not just that a person can be banned from a platform, a new

202
00:16:31,812 --> 00:16:38,672
platform can pop up and Amazon can ban them. But there are other layers of censorship that can

203
00:16:38,672 --> 00:16:44,412
happen in the internet, such as, we didn't see this, but potentially the App Store level, right?

204
00:16:44,492 --> 00:16:50,052
Google and Apple could have delisted the app if Amazon didn't. And then even let's say Amazon or

205
00:16:50,052 --> 00:16:56,672
Azure or whoever the storage provider is, they didn't block them, the App Store didn't block

206
00:16:56,672 --> 00:16:59,132
then there's still the DNS level risk, right?

207
00:16:59,252 --> 00:17:01,672
So there are multiple layers of fragility.

208
00:17:02,032 --> 00:17:04,432
So given that, Alex,

209
00:17:04,472 --> 00:17:07,812
do you think we can have a true censorship resistant

210
00:17:07,812 --> 00:17:10,472
communications medium?

211
00:17:11,632 --> 00:17:13,352
I mean, I think Noster is a pretty good bet

212
00:17:13,352 --> 00:17:17,832
because it's not really using the existing infrastructure

213
00:17:17,832 --> 00:17:19,692
the way it's intended to be used, right?

214
00:17:19,712 --> 00:17:22,132
And that's an interesting thing about new technologies

215
00:17:22,132 --> 00:17:25,432
is we're like building and building upon

216
00:17:25,432 --> 00:17:30,652
these things that have been done in the past, but not necessarily in the way that they were

217
00:17:30,652 --> 00:17:37,012
intended to be used. And that's when I think that things can get really powerful. So on Noster,

218
00:17:37,852 --> 00:17:48,272
DNS, that's a potential layer that can be threatened, right? But since the actual data

219
00:17:48,272 --> 00:17:53,452
that people are sending over the Noster network is content addressable, it can be uploaded to any

220
00:17:53,452 --> 00:17:59,952
number of relays on different URLs. So like if a relay gets shut down on like a DNS level,

221
00:18:01,052 --> 00:18:06,592
it's like probably recoverable. Probably that data already exists on a different relay on a

222
00:18:06,592 --> 00:18:15,172
different URL or that relay operator can just spin up a new URL. So I think because it is like

223
00:18:15,172 --> 00:18:21,732
sort of designed and built in a way that's not, wasn't using things in ways that they weren't

224
00:18:21,732 --> 00:18:23,812
necessarily intended to be used originally.

225
00:18:24,752 --> 00:18:31,272
That we're sort of operating within this system to the degree that we have to in order to

226
00:18:32,152 --> 00:18:35,312
be accessible at all to the outside world.

227
00:18:35,492 --> 00:18:41,412
But the design of it is in a fairly censorship resistant way.

228
00:18:43,132 --> 00:18:47,172
So do you think you could have, so let's say all three of those layers clamped down,

229
00:18:47,172 --> 00:18:52,132
DNS level, App Store level, and server level.

230
00:18:53,132 --> 00:18:56,052
What would the Nostra's path be in that scenario?

231
00:18:58,572 --> 00:19:04,332
Well, App Store level, you can get around that by just publishing PWA on the web.

232
00:19:05,752 --> 00:19:16,312
You can get around that by just creating new apps under different names and different organizations that essentially do the same thing.

233
00:19:17,172 --> 00:19:20,012
But the App Store situation is really bad.

234
00:19:20,572 --> 00:19:27,272
I feel like people are voting for this every time they are, you know, buying and using an iPhone.

235
00:19:27,652 --> 00:19:29,832
And Android is really not that much better.

236
00:19:30,492 --> 00:19:36,472
But we need people to be, like, saying, no, maybe this is unrealistic.

237
00:19:36,812 --> 00:19:38,672
I think everyone should just install Graphene OS.

238
00:19:40,112 --> 00:19:41,632
I know that this is unrealistic.

239
00:19:42,052 --> 00:19:45,932
But I wish that people would do a little bit more in their day-to-day.

240
00:19:45,932 --> 00:19:50,932
Is it really that hard to not use centralized social media platforms?

241
00:19:51,832 --> 00:19:54,832
And so if people aren't even willing to do that,

242
00:19:54,892 --> 00:19:56,512
then I'm willing to blame them a little bit

243
00:19:56,512 --> 00:19:59,372
for the things that are harder for them to control too.

244
00:20:00,352 --> 00:20:01,892
But what were the other layers?

245
00:20:02,392 --> 00:20:04,612
DNS, App Store, and then what was the last one you mentioned?

246
00:20:05,092 --> 00:20:08,412
Well, just the storage and compute layer, right?

247
00:20:08,552 --> 00:20:11,512
AWS, Azure, Google Cloud.

248
00:20:12,432 --> 00:20:13,052
Oh, right.

249
00:20:13,412 --> 00:20:14,372
Hosting providers.

250
00:20:14,372 --> 00:20:20,472
So I currently have a server rack in my living room.

251
00:20:21,412 --> 00:20:31,012
And I think that people have just not been creative enough in their ways that they can work around web hosting limitations.

252
00:20:31,812 --> 00:20:33,272
Because you definitely can.

253
00:20:35,452 --> 00:20:36,272
Right.

254
00:20:37,372 --> 00:20:42,812
So local hosting, progressive web apps, and then...

255
00:20:42,812 --> 00:20:44,092
Just run your own server, Avi.

256
00:20:44,192 --> 00:20:44,672
Come on, bro.

257
00:20:45,672 --> 00:20:47,732
You got to get the purity stack going.

258
00:20:51,332 --> 00:20:55,552
So were there lessons from truth that led to Soapbox?

259
00:20:55,992 --> 00:20:58,152
I mean, really, what brought you to Soapbox?

260
00:20:58,332 --> 00:21:02,032
What was your grand – what was the big instigation?

261
00:21:03,432 --> 00:21:05,792
So it was because I really liked –

262
00:21:05,792 --> 00:21:07,072
It's taken multiple forms, correct?

263
00:21:07,072 --> 00:21:10,872
So maybe talk about the name itself.

264
00:21:10,872 --> 00:21:17,072
So it's because I really liked Mastodon, but I felt like Mastodon was doing it wrong.

265
00:21:17,872 --> 00:21:26,612
And they have these cookie cutter Mastodon sites that all say Mastodon, and they're trying to make it like it's all one platform from a branding and marketing perspective.

266
00:21:26,812 --> 00:21:27,972
And I thought, no, that's wrong.

267
00:21:28,392 --> 00:21:33,992
Actually, we should be targeting different communities of people with a custom brand.

268
00:21:33,992 --> 00:21:47,772
And so it started out literally as a fork of Mastodon, and then the technology stack has evolved over time to be broken up into a bunch of different pieces and be operable on many different types of architectures.

269
00:21:48,372 --> 00:21:58,812
But the vision of it was just like, hey, build your own social network and have it be connected to this decentralized internet, this future of the web.

270
00:22:00,252 --> 00:22:02,832
So that was really the initial idea.

271
00:22:02,832 --> 00:22:10,792
and and i think now with ai it actually is still the same fundamental idea of hey build your own

272
00:22:10,792 --> 00:22:17,072
thing and have it be connected with you know this decentralized web with shakespeare um it's just

273
00:22:17,072 --> 00:22:24,612
that you know in prior in previous version of soapbox you have like a theme uh customizer user

274
00:22:24,612 --> 00:22:29,252
interface and you'd have a color color swatches that you can click and you can click and drag

275
00:22:29,252 --> 00:22:32,292
to change from like green to a blue to a yellow,

276
00:22:32,472 --> 00:22:33,992
and you can upload a custom logo.

277
00:22:34,792 --> 00:22:36,572
And that was really the best we could do

278
00:22:36,572 --> 00:22:38,772
in terms of user level customization.

279
00:22:39,772 --> 00:22:42,892
But now I'm seeing a potential future

280
00:22:42,892 --> 00:22:47,432
where we don't even need a user interface like that

281
00:22:47,432 --> 00:22:49,432
to customize a website anymore.

282
00:22:49,892 --> 00:22:50,952
It's just a text field.

283
00:22:50,952 --> 00:22:52,932
And then you tell the AI exactly

284
00:22:52,932 --> 00:22:54,832
what type of customizations you want

285
00:22:54,832 --> 00:22:57,052
and what type of website you want.

286
00:22:57,052 --> 00:23:01,172
and then it can just create that on the spot for you

287
00:23:01,172 --> 00:23:03,052
and create something completely custom

288
00:23:03,052 --> 00:23:04,632
beyond what we could ever do

289
00:23:04,632 --> 00:23:07,372
with buttons and toggles and sliders

290
00:23:07,372 --> 00:23:09,932
in like a customizer UI.

291
00:23:10,292 --> 00:23:13,292
So I'm still really excited about this idea

292
00:23:13,292 --> 00:23:16,372
of building social platforms

293
00:23:16,372 --> 00:23:19,312
catered towards specific communities or groups

294
00:23:19,312 --> 00:23:31,664
and Shakespeare is just a new way of doing that And then there was Ditto at one point right Alex

295
00:23:33,224 --> 00:23:34,484
Ditto is still around.

296
00:23:34,664 --> 00:23:38,584
Ditto is like the main way that I interact with Noster.

297
00:23:39,744 --> 00:23:46,804
And so Ditto is the idea of kind of just taking all of the work that I had built on Mastodon

298
00:23:46,804 --> 00:23:49,904
and on the Fediverse and transferring that to Noster.

299
00:23:49,904 --> 00:24:03,364
So it's the same user experience as the user experience of people running their own social platforms on their own domain name and, you know, customizing it for a particular audience.

300
00:24:03,744 --> 00:24:08,504
But it's still, you know, toggles and sliders and color pickers.

301
00:24:08,744 --> 00:24:19,304
And I think that this idea is going to just be something that becomes lost in history once you can prompt an AI.

302
00:24:19,904 --> 00:24:30,144
And so I, you know, switching over to developing on Shakespeare has kind of taken me a step back in a way from where I was with Ditto because I had this like really polished product.

303
00:24:31,164 --> 00:24:39,844
But I'm trying to get sort of back there to where people are able to build these really polished products.

304
00:24:39,844 --> 00:24:47,044
And it seems like we've actually come very far with it a lot more than you might expect.

305
00:24:49,904 --> 00:24:59,284
so after that then alex you got and we're building up the timeline here getting to shakespeare but

306
00:24:59,284 --> 00:25:08,644
you got involved in the and other stuff collective uh which led to uh stacks and mk stack so what

307
00:25:08,644 --> 00:25:14,944
for folks who are unfamiliar what is the and other stuff collective and what is happening with it

308
00:25:14,944 --> 00:25:24,264
today? Yeah. So it's, it's a group of people, um, involving, uh, me, Rabble, Callie, Jeff,

309
00:25:24,264 --> 00:25:36,564
uh, Jack, Liz. Um, and, and we're just trying to sort of approach how to make Noster win from a

310
00:25:36,564 --> 00:25:46,684
different angle than we were doing before. And so it's sort of like, you know, this consortium,

311
00:25:46,684 --> 00:25:52,964
I guess, of people where we're kind of touching base more often on what each other are doing

312
00:25:52,964 --> 00:25:57,144
and trying to collaborate as much as we can so that we can share resources

313
00:25:57,144 --> 00:26:00,944
and ultimately try to make Noster win.

314
00:26:00,944 --> 00:26:10,264
and your involvement in that consortium or collective or what you will is leading the ai

315
00:26:10,264 --> 00:26:17,524
pillar right correct so so uh mine is the communities pillar and the way i perceive

316
00:26:17,524 --> 00:26:23,584
of it is that there there is this connection and overlap between communities and ai

317
00:26:23,584 --> 00:26:32,904
to where my community's vision is something that would be fulfilled through AI tooling.

318
00:26:34,464 --> 00:26:38,404
Did you check into Rabble this week, what he was doing for the collective?

319
00:26:40,464 --> 00:26:41,324
What do you mean?

320
00:26:41,964 --> 00:26:47,944
Oh, last I saw he was wearing a cat blow-up outfit at Broadway.

321
00:26:49,064 --> 00:26:50,404
Oh, yeah, that's awesome.

322
00:26:50,404 --> 00:26:54,024
Maybe he'd be on the community pillar too there.

323
00:26:56,904 --> 00:26:57,544
Nice.

324
00:26:57,744 --> 00:26:58,904
Sorry to interject, Avi.

325
00:27:00,404 --> 00:27:01,984
No, no, that was good.

326
00:27:03,104 --> 00:27:09,264
So as part of that, then, Alex, you worked on Stacks and MKStack.

327
00:27:10,744 --> 00:27:17,804
Could you give a simple overview of what Stacks is and what MKStack is?

328
00:27:18,604 --> 00:27:19,204
Sure.

329
00:27:19,204 --> 00:27:33,144
So Stacks is the idea that we can make an AI able to create an application with a much higher level of performance.

330
00:27:33,724 --> 00:27:37,484
And by performance, I mean quality of the output, not necessarily speed.

331
00:27:37,924 --> 00:27:45,504
Probably it's going to take, you know, 10 minutes to develop an initial version of an application in any case.

332
00:27:45,504 --> 00:27:58,484
But when you don't use a stack, if you just go raw on like an empty directory in like an AI coding tool, your results may vary.

333
00:27:58,484 --> 00:28:15,644
And the idea of stacks is that we, like a developer, will have an idea of a particular architecture that they want, a particular programming language, a particular paradigm with some desired output, right?

334
00:28:15,744 --> 00:28:26,284
So in the case of MKStack, the desired output is a Noster web client that people can access by visiting a URL in a web browser.

335
00:28:26,284 --> 00:28:47,284
And so that stack is one that involves Node.js, NPM, package.json, React, Tailwind CSS, Vite, ViteSt, and sort of connecting all of those pieces together in a template.

336
00:28:47,284 --> 00:28:55,664
template that has, it's a template, it's a folder full of files that already have, you know, some

337
00:28:55,664 --> 00:29:02,644
code in place, but not a fully functioning application, just enough code to where an LLM

338
00:29:02,644 --> 00:29:07,224
can look at it and kind of say, okay, I understand what's going on here. This looks like, you know,

339
00:29:07,224 --> 00:29:13,484
projects I've seen in my training data. And then most importantly, there's an agents.md file,

340
00:29:13,484 --> 00:29:19,984
which is a big file that contains a bunch of context information for the llm to train it to

341
00:29:19,984 --> 00:29:26,784
do things in the in the best way so the way that comes about is first we start with these just these

342
00:29:26,784 --> 00:29:32,404
files and we have no agents.md file right and that's our initial version of our stack and then

343
00:29:32,404 --> 00:29:39,504
i iterate upon this by by typing a prompt as you know i clone the stack i type a prompt it may be

344
00:29:39,504 --> 00:29:46,304
build me a site like twitter on noster build me a site like instagram on noster build me a noster

345
00:29:46,304 --> 00:29:53,724
badges site these types of prompts and then i'll just see what did the llm do wrong if anything

346
00:29:53,724 --> 00:29:58,744
um and if if there's anything that i felt like it did wrong or could be improved

347
00:29:58,744 --> 00:30:05,264
then i have it go and add information to the agents.md file clarifying how to do this thing

348
00:30:05,264 --> 00:30:12,604
in a correct way. So like, for example, querying data from Nostra relays, right? At first,

349
00:30:13,064 --> 00:30:18,084
there wasn't really any guidance in the agents file of how to do that. And so it wouldn't do it

350
00:30:18,084 --> 00:30:24,304
correctly. So I added, you know, little pieces of code in there to abstract it as much as possible

351
00:30:24,304 --> 00:30:29,344
to create minimal surface area for the LLM to where the LLM only has to care about calling a

352
00:30:29,344 --> 00:30:35,584
particular function and then in the agents.md file we just say we have a section called querying

353
00:30:35,584 --> 00:30:42,464
noster data and then we we have code examples saying here's how you can how you can import and

354
00:30:42,464 --> 00:30:48,464
call this function in order to get noster data and so now when people run this stack and then they say

355
00:30:48,464 --> 00:30:54,864
build me a site like like twitter on noster it has the information it needs for how to actually

356
00:30:54,864 --> 00:31:03,324
query a feed of people that you follow because it it just has that little bit of information so

357
00:31:03,324 --> 00:31:09,984
is this iterative process of having a basically a project template and a context file and then

358
00:31:09,984 --> 00:31:15,404
and then trying different prompts with it seeing what it gets stuck on and then improving that

359
00:31:15,404 --> 00:31:22,824
context file and and the end result is something that is both generalized in in a sense and

360
00:31:22,824 --> 00:31:28,344
specialized in a sense. It's specialized in the actual stack sense of these are the technologies

361
00:31:28,344 --> 00:31:32,704
that are being used for this particular purpose, right? A Nostra Web Client. But it's also

362
00:31:32,704 --> 00:31:38,104
generalized in a sense that because this is an LLM, you can build any type of Nostra Web Client

363
00:31:38,104 --> 00:31:47,904
that you want to build. And so that's stacks in a nutshell. So agents.md sounds like a living,

364
00:31:47,904 --> 00:31:49,824
breathing context file.

365
00:31:50,864 --> 00:31:51,264
Exactly.

366
00:31:52,624 --> 00:31:56,764
Now, is that within the context of a given project,

367
00:31:57,144 --> 00:31:58,984
or is that once it's updated,

368
00:31:59,104 --> 00:32:02,024
is it available to anyone trying to build on MKStack?

369
00:32:02,604 --> 00:32:04,964
So it's within the context of a specific project.

370
00:32:05,224 --> 00:32:07,544
So once you clone an MK,

371
00:32:07,744 --> 00:32:10,383
once you clone MKStack into your own project,

372
00:32:10,864 --> 00:32:12,824
it copies that agents.md file,

373
00:32:12,824 --> 00:32:14,764
and it also gives you the opportunity

374
00:32:14,764 --> 00:32:17,364
to add and update it as well.

375
00:32:17,904 --> 00:32:25,164
Because you're kind of taking a snapshot of the current version of MKStack at that particular time, including all of the files it references.

376
00:32:25,544 --> 00:32:30,883
And so the context file has to be congruent with the files that are on there as well.

377
00:32:31,044 --> 00:32:39,604
So there's no way to upgrade your MKStack to like a newer version of MKStack because that's just not how it works, right?

378
00:32:39,604 --> 00:32:53,404
If there's improvements that we've made, ways that you can get updates to them are to say to the LLM, hey, go look at this MKStack thing and see if there's anything valuable that could be added in here.

379
00:32:53,584 --> 00:32:55,084
Or just vibe it again.

380
00:32:56,184 --> 00:33:00,383
Use the same prompts that you used before and just vibe it again.

381
00:33:02,684 --> 00:33:07,664
So MKStack is a NOSTA-focused stack.

382
00:33:07,664 --> 00:33:10,364
you can have non-NOSTA focus stacks as well, correct?

383
00:33:10,744 --> 00:33:11,924
Absolutely, anything.

384
00:33:13,364 --> 00:33:15,404
Now, are there examples of people?

385
00:33:15,904 --> 00:33:18,304
I know there was one when we last spoke, Alex.

386
00:33:19,244 --> 00:33:22,644
Was it a treasure hunt or geocaching?

387
00:33:22,804 --> 00:33:23,984
Yeah, yeah, yeah, treasures.

388
00:33:24,364 --> 00:33:27,264
Treasures.2 is a really fun one.

389
00:33:27,324 --> 00:33:30,224
So that's an MKStack app made by Chad.

390
00:33:30,984 --> 00:33:32,824
And I really like this one

391
00:33:32,824 --> 00:33:36,064
because this is like for so many reasons.

392
00:33:36,064 --> 00:33:43,124
um you know you could say it's pretty busy doesn't it in what way uh where are you have

393
00:33:43,124 --> 00:33:49,624
you been rummaging the fields geocaching yes yes you're right yes it's had me busy traveling the

394
00:33:49,624 --> 00:33:54,284
world and looking for good spots to plant you know little boxes of toys that's essentially what

395
00:33:54,284 --> 00:34:00,164
geocaches geocaches are is you you like find a good spot to put a little box of toys and then

396
00:34:00,164 --> 00:34:07,444
whoever finds it later they're supposed to take a toy and leave a toy um and so it's just this

397
00:34:07,444 --> 00:34:12,644
great opportunity to go outside right because you could say that one of the biggest problems

398
00:34:12,644 --> 00:34:19,704
with social media is the outrage uh culture and this is literally like taking the power of

399
00:34:19,704 --> 00:34:26,484
computers and then gearing it towards people touching grass um and and that's something that

400
00:34:26,484 --> 00:34:27,744
I love about it.

401
00:34:30,184 --> 00:34:32,424
Yeah, so I think Treasures is going to

402
00:34:32,424 --> 00:34:34,304
solve social media.

403
00:34:35,944 --> 00:34:38,444
Walk us through how Treasures

404
00:34:38,444 --> 00:34:40,444
would work. Let's say I join.

405
00:34:42,004 --> 00:34:42,664
So do you

406
00:34:42,664 --> 00:34:44,644
want to find, do you want

407
00:34:44,644 --> 00:34:46,604
to go on an adventure and find

408
00:34:46,604 --> 00:34:48,624
a treasure that exists, or do you

409
00:34:48,624 --> 00:34:50,204
want to be a creator and

410
00:34:50,204 --> 00:34:51,764
create your own treasure?

411
00:34:52,484 --> 00:34:54,584
Let's just say it's me and my son, and we're

412
00:34:54,584 --> 00:34:55,684
going to go chase some treasures.

413
00:34:56,484 --> 00:35:04,224
Okay. So there's a map and you're going to zoom in on this map to nearby where you live.

414
00:35:05,064 --> 00:35:14,755
And then you going to see markers of treasure chests on the map near where you live hopefully And if you don see any then this is your call to action to put one there

415
00:35:14,755 --> 00:35:18,416
so that the next person who searches can find one.

416
00:35:19,436 --> 00:35:22,876
And so, you know, we have GPS coordinates, longitude and latitude,

417
00:35:22,876 --> 00:35:28,715
and you basically just navigate to where that location is of that particular treasure that you want to find.

418
00:35:28,715 --> 00:35:45,515
But also, you can click on those little treasure icons and it'll show you a preview of what the location looks like, a description of the treasure from the author, you know, find logs where people will go and they'll say like, oh, I found this or I didn't find this.

419
00:35:45,715 --> 00:35:50,215
And, you know, basically a review system for each of these geocaches.

420
00:35:51,055 --> 00:35:52,436
So it's pretty cool.

421
00:35:52,835 --> 00:35:58,515
You can just use it, you know, as a way to kind of go on an adventure without necessarily needing to spend money.

422
00:35:58,515 --> 00:36:04,155
you know because like like do you ever feel like i wish i want to go like i want to go do something

423
00:36:04,155 --> 00:36:12,175
but all there is is like the grocery store or like you know walmart or whatever uh like society is

424
00:36:12,175 --> 00:36:17,916
built in this way to where you have to spend money to do anything fun or entertaining and

425
00:36:17,916 --> 00:36:24,175
this is just like an excuse to just go on an adventure there's some pretty solid people

426
00:36:24,175 --> 00:36:30,416
watching at Walmart though. Let's be real. Well, put a treasure behind a Walmart then.

427
00:36:32,115 --> 00:36:39,235
I don't know if I want to go behind the Walmart. Oh man, but no, no, good point. So like if someone

428
00:36:39,235 --> 00:36:48,356
like Avi who might live in a rural, not very dense pop populated area, he can put his little

429
00:36:48,356 --> 00:36:52,696
town on the map just by authoring treasures, huh?

430
00:36:54,376 --> 00:36:55,055
Yep, exactly.

431
00:36:55,055 --> 00:37:02,155
And I feel like this is a good example of an app that would not have been developed if

432
00:37:02,155 --> 00:37:06,115
it wasn't for AI and LLM-generated code.

433
00:37:06,235 --> 00:37:12,175
It's built with MKStack, and it is a maintained, high-quality application.

434
00:37:13,135 --> 00:37:15,015
So I'm really excited about it.

435
00:37:15,055 --> 00:37:18,035
It's a great use case for MKStack and for AI.

436
00:37:18,035 --> 00:37:24,015
These are the types of things that anyone can build now if they just have the best idea.

437
00:37:24,975 --> 00:37:31,555
Because, you know, I think that there are a lot of things that people are building that they're just experimenting with.

438
00:37:31,715 --> 00:37:33,856
And accelerationism is the word, right?

439
00:37:33,916 --> 00:37:40,515
Like earlier in the discussion, we were talking about this idea of people building these apps and abandoning them.

440
00:37:41,615 --> 00:37:45,975
But I think that that's more of a benefit than anything.

441
00:37:45,975 --> 00:37:52,495
like now we have the opportunity for people to create apps that they abandon and it allows us to

442
00:37:52,495 --> 00:37:58,175
get very fast feedback of what is good and what's not good about this app and maybe it's maybe it's

443
00:37:58,175 --> 00:38:05,035
right to abandon it um you know but but like we just wouldn't have had the opportunity to even

444
00:38:05,035 --> 00:38:11,335
create that in the first place previously without ai um and so treasures is just an example of one

445
00:38:11,335 --> 00:38:16,956
that sticks because it's a really good idea for numerous reasons. And the implementation is really

446
00:38:16,956 --> 00:38:27,715
good too. So that was Stacks and MKStack. And then that's where, when you and I last spoke, Alex,

447
00:38:28,295 --> 00:38:33,535
the clock stopped. Since then, obviously, a lot has happened. There's been an evolution

448
00:38:33,535 --> 00:38:39,876
to Shakespeare. Do you want to talk about that? So you had Stacks and MKStack.

449
00:38:39,876 --> 00:38:43,515
what was the lead up to Shakespeare from there?

450
00:38:43,975 --> 00:38:44,155
Sure.

451
00:38:44,675 --> 00:38:49,396
So Stacks CLI is a terminal application,

452
00:38:49,396 --> 00:38:52,155
which means you have to open the terminal of your computer

453
00:38:52,155 --> 00:38:54,876
and you have to type in a scary text prompt

454
00:38:54,876 --> 00:38:58,155
that probably most people in the world aren't going to be using.

455
00:38:58,396 --> 00:38:59,696
This is more if you're a developer.

456
00:39:01,135 --> 00:39:05,635
But it always has been the vision to make this something

457
00:39:05,635 --> 00:39:07,916
that empowers regular people.

458
00:39:07,916 --> 00:39:24,575
And, you know, there are other applications like Bolt v0, Lovable, that we wanted to – and those are all like anti-freedom subscription-based proprietary closed platforms.

459
00:39:25,775 --> 00:39:29,035
But the UX is really great.

460
00:39:29,035 --> 00:39:35,416
So Stacks is like a first sort of pass at seeing can we do this?

461
00:39:35,416 --> 00:39:40,675
Can we make high-quality AI-generated apps but do it in a command line?

462
00:39:41,376 --> 00:39:51,555
Whereas Shakespeare is the same or very similar experience to Stacks, but done through a graphical web interface.

463
00:39:52,215 --> 00:39:54,295
And Shakespeare is at shakespeare.diy.

464
00:39:54,936 --> 00:39:57,815
The version that's there right now is called Act 1.

465
00:39:57,815 --> 00:40:06,615
uh it has a particular architecture that involves the uh communication with a server a server which

466
00:40:06,615 --> 00:40:12,715
is running in my living room um and so the like network requests and stuff are running off of my

467
00:40:12,715 --> 00:40:18,995
server and i am currently working on an act two of shakespeare which is a different architecture

468
00:40:18,995 --> 00:40:25,995
i think i'm maybe getting ahead of myself a little bit but um the basic idea of shakespeare is

469
00:40:25,995 --> 00:40:31,356
is that regular people can develop Nostra apps with AI.

470
00:40:33,115 --> 00:40:40,195
So you're saying when I developed my Hashman game on MKStacks,

471
00:40:40,376 --> 00:40:41,856
I should have been using Shakespeare?

472
00:40:43,675 --> 00:40:46,835
Because my story is I used it.

473
00:40:46,835 --> 00:40:52,476
I made like a little Pac-Man mining themed app or game.

474
00:40:52,476 --> 00:40:58,535
and I kind of, I think I ruined it like three quarters of the way through. I didn't know what

475
00:40:58,535 --> 00:41:03,235
the hell to do after that. So I started like telling the computer to back it up to when it,

476
00:41:03,655 --> 00:41:09,195
you know, to a certain, you know, prompt. You were using MK stack? Yeah. And you're using it

477
00:41:09,195 --> 00:41:15,315
through the command line. Yeah. Well, no, well, I was telling it. So it wasn't Shakespeare though.

478
00:41:15,416 --> 00:41:21,476
So it wasn't MK stack. Wasn't, I was just telling it a game. I mean, I was like $130 deep into this

479
00:41:21,476 --> 00:41:27,335
thing. Cause it was like, I was running credits. The bigger it got, the more I was like just simple

480
00:41:27,335 --> 00:41:32,856
task. It was, it was, it was taking a little longer. Uh, and I'm like, all right. And then I,

481
00:41:32,956 --> 00:41:38,235
I was like excited. And then I was like, Oh, and then I went back to being excited again.

482
00:41:38,615 --> 00:41:43,115
It was an interesting thing, but I felt like I was a little bit lost, but maybe I should have

483
00:41:43,115 --> 00:41:49,615
been using Shakespeare. Uh, but it just came out at the time. So it wasn't fully working properly.

484
00:41:49,615 --> 00:42:00,896
So I just went to the next thing. So I guess my question is, if someone like me just wants to try things out, you recommend going to Shakespeare, Shakespeare versus MKStack.

485
00:42:01,015 --> 00:42:06,376
If you have zero coding, zero computer background, what do you recommend?

486
00:42:07,335 --> 00:42:11,255
Yes, but to be honest, I'm not sure that it would have made your experience better in that case.

487
00:42:11,735 --> 00:42:18,416
If you're comfortable enough using a terminal, I think that Stack CLI is still the best way to do it.

488
00:42:18,416 --> 00:42:23,055
um, which maybe is not the best thing for me to say from like a marketing perspective, but

489
00:42:23,055 --> 00:42:30,695
like right now that Shakespeare as it is, is geared towards people who are like, have a phobia

490
00:42:30,695 --> 00:42:39,075
of the terminal. Um, and in terms of just like the project getting bigger and getting out of control,

491
00:42:39,635 --> 00:42:48,095
um, it's a known issue. And it, I actually don't think it's a limitation of the LLM. I think it's

492
00:42:48,095 --> 00:42:53,436
a limitation of the design of the tools right now. And like, it's a thing that we're trying

493
00:42:53,436 --> 00:42:59,255
to push the boundaries on and figure out how to improve and just haven't exactly solved that yet.

494
00:42:59,255 --> 00:43:04,456
So like, it's not necessarily going to be opening eyes going to come around the corner and make such

495
00:43:04,456 --> 00:43:09,396
a more powerful LLM. It's going to do this. It's like, I'm going to figure out how to rewire the

496
00:43:09,396 --> 00:43:17,456
internals of, of stack CLI or Shakespeare by extension to make those problems be less noticeable

497
00:43:17,456 --> 00:43:21,835
and to deal with larger code bases more effectively.

498
00:43:22,155 --> 00:43:24,856
So it's a big engineering technical challenge.

499
00:43:24,856 --> 00:43:31,295
I think that, like, I personally am as far at the forefront of it

500
00:43:31,295 --> 00:43:34,476
as I could be right now in terms of, like, what everybody else is doing.

501
00:43:34,635 --> 00:43:36,476
And we're all just kind of trying to figure out right now

502
00:43:36,476 --> 00:43:40,896
what are the next steps in order to make this process be more efficient,

503
00:43:41,075 --> 00:43:45,995
more cost-effective, more productive, and have better outcomes.

504
00:43:47,456 --> 00:43:54,095
But if I play Hashman right now, is it actually the server in your living room?

505
00:43:55,456 --> 00:43:58,595
So the server in my living room is what produced it.

506
00:43:59,476 --> 00:44:00,035
Okay.

507
00:44:00,135 --> 00:44:02,515
It's also what serves it.

508
00:44:02,655 --> 00:44:07,936
But the design of the stack is to produce HTML, CSS, and JavaScript.

509
00:44:08,255 --> 00:44:16,396
And so the majority of that code is, well, all of that code actually is just being executed in your browser on the client side once it gets downloaded.

510
00:44:16,396 --> 00:44:19,595
But yes, it gets downloaded from the server in my living room.

511
00:44:20,275 --> 00:44:23,476
And then once it's on your browser, your browser is executing all of that code.

512
00:44:29,095 --> 00:44:35,896
Alex, you know, when we spoke, we were in the, from an AI model standpoint itself, right?

513
00:44:36,135 --> 00:44:38,315
Forget the tooling around it for a second.

514
00:44:40,775 --> 00:44:44,495
We were in the throes of a pure acceleration in June.

515
00:44:44,495 --> 00:44:51,456
like models were just, you know, the next version was, you know, above and beyond anything imaginable

516
00:44:51,456 --> 00:44:58,155
with the previous version, right? We'd gone from opening eyes 01 to 03 and there was 04, 04 mini

517
00:44:58,155 --> 00:45:06,155
had come out just around then. Google Gemini 2.5 was crushing it. Claude Sonnet 4.0 had come out

518
00:45:06,155 --> 00:45:14,376
just around then and it was you the the acceleration was palpable do you feel in the last three months

519
00:45:14,376 --> 00:45:20,775
or so we've hit a bit of a plateau because that's you know the jump the jump from march or you know

520
00:45:20,775 --> 00:45:29,835
from february to june and then june to september it feels a lot less i agree i think gpt5 highlighted

521
00:45:29,835 --> 00:45:36,015
that a lot because they were like calling it the manhattan project like altman was saying oh my god

522
00:45:36,015 --> 00:45:38,055
this is going to destroy the human race and stuff.

523
00:45:38,055 --> 00:45:41,575
But then it ended up being so disappointing to many people.

524
00:45:43,055 --> 00:45:44,055
And yeah,

525
00:45:44,135 --> 00:45:44,396
I mean,

526
00:45:44,456 --> 00:45:45,476
it's hard to know.

527
00:45:45,595 --> 00:45:46,695
I don't know what they know.

528
00:45:46,795 --> 00:45:51,936
I don't have any insider info on what like the big AI giants are doing behind

529
00:45:51,936 --> 00:45:52,675
closed doors.

530
00:45:53,075 --> 00:45:58,055
But what it certainly seems like on the outside and what my intuition is,

531
00:45:58,055 --> 00:46:01,275
is that the LLM side is plateauing.

532
00:46:02,515 --> 00:46:05,856
And that kind of makes me happy in a way because I feel like,

533
00:46:05,856 --> 00:46:07,315
okay, maybe we have time here.

534
00:46:07,315 --> 00:46:11,575
Like we're not going to just go straight into like apocalypse from this.

535
00:46:12,456 --> 00:46:15,795
People are going to have time to adapt to what's going on here.

536
00:46:15,936 --> 00:46:17,456
It's going to keep being fun.

537
00:46:18,195 --> 00:46:19,795
Preparation time for apocalypse.

538
00:46:20,295 --> 00:46:20,535
Yeah.

539
00:46:21,675 --> 00:46:23,575
It's not going to ruin what I'm doing.

540
00:46:25,495 --> 00:46:26,376
And I,

541
00:46:27,396 --> 00:46:28,715
and maybe this is just a cope.

542
00:46:28,896 --> 00:46:29,396
I don't know,

543
00:46:29,476 --> 00:46:32,595
but I feel like the AI tooling is where,

544
00:46:32,595 --> 00:46:33,595
where it is right now,

545
00:46:33,595 --> 00:46:35,075
because this is where you can,

546
00:46:35,075 --> 00:46:37,976
make most of the progress.

547
00:46:39,595 --> 00:46:44,035
And the way I envision it is like the LLM is a brain,

548
00:46:44,215 --> 00:46:46,095
but it's only part of a brain.

549
00:46:46,356 --> 00:46:48,235
I'm not sure which part of the brain it is.

550
00:46:48,235 --> 00:46:51,795
It's maybe small pieces of parts of a brain.

551
00:46:52,447 --> 00:46:57,107
Um, but it doesn't have memories in order to give it memories.

552
00:46:57,107 --> 00:47:03,547
You have to build just regular code, not some crazy, fancy, like, you know, tensor, uh,

553
00:47:03,567 --> 00:47:07,347
machine learning stuff, just regular code, um, to give it memories.

554
00:47:07,767 --> 00:47:09,347
You can, you can give it a body.

555
00:47:09,407 --> 00:47:10,627
I mean, you can put it in a robot.

556
00:47:10,847 --> 00:47:17,367
You can give it tools to, to see, to walk, to, you know, articulate parts of its body.

557
00:47:17,367 --> 00:47:24,767
And those are all just regular code things that essentially wrap this LLM core brain.

558
00:47:25,687 --> 00:47:28,827
So we can expand the brain itself with code.

559
00:47:29,167 --> 00:47:30,827
We can give it a body with code.

560
00:47:31,927 --> 00:47:34,067
The body doesn't necessarily have to be physical.

561
00:47:34,287 --> 00:47:36,267
The body can be virtual too, right?

562
00:47:36,307 --> 00:47:46,747
It can extend into computer systems, into infrastructure, into applications like Stack CLI and like Shakespeare.

563
00:47:46,747 --> 00:47:52,407
Like, Shakespeare is essentially giving a body to this core brain of this LLM.

564
00:47:52,567 --> 00:48:00,947
And I think that right now, at least, this is where we're seeing the most rapid progress and innovation is not on the core brain LLM side.

565
00:48:01,107 --> 00:48:03,187
It's on building the body for the brain.

566
00:48:04,127 --> 00:48:09,507
You know, I tend to agree with that, at least at this particular point in time, Alex.

567
00:48:09,847 --> 00:48:13,527
I reserve the right to change my mind because of the acceleration.

568
00:48:13,607 --> 00:48:14,367
Yeah, exactly.

569
00:48:14,367 --> 00:48:21,627
but if that is true that's kind of disappointing because i remember when

570
00:48:21,627 --> 00:48:29,707
open ai released the full o3 model uh right not the o3 mini but the full i think that was sometime

571
00:48:29,707 --> 00:48:35,747
in april or may and it was just this otherworldly experience just like is how is this possible

572
00:48:35,747 --> 00:48:42,207
how is this level of deep research possible how is this level of reasoning possible and then

573
00:48:42,207 --> 00:48:47,267
And I think that was the peak of AI models, right?

574
00:48:47,267 --> 00:48:48,927
The full O3.

575
00:48:49,367 --> 00:48:52,587
And GPT-5 is, honestly, even on thinking mode,

576
00:48:53,147 --> 00:48:57,147
is not as good as O3 was in April, in my mind.

577
00:48:57,767 --> 00:48:58,207
I agree.

578
00:48:58,427 --> 00:49:00,047
So a little disappointing.

579
00:49:03,227 --> 00:49:07,047
You know, it's not, I mean, from a real world standpoint,

580
00:49:07,207 --> 00:49:08,907
I mean, I'm in a family business.

581
00:49:09,027 --> 00:49:11,387
We've been around for 68 years.

582
00:49:11,387 --> 00:49:19,007
we're very slow movers one thing I'm seeing a lot because I basically built our website

583
00:49:19,007 --> 00:49:25,427
I do I get all the the referrals from the website and I'm starting to see a lot of

584
00:49:25,427 --> 00:49:31,827
where who referred you and it's grok or chat gpt and I'm starting to see those in the real world

585
00:49:31,827 --> 00:49:36,187
which is kind of like because you talk to anybody at work you're like you know about ai they look at

586
00:49:36,187 --> 00:49:41,227
you like well at least in my work they have no idea what I'm talking about but you're starting

587
00:49:41,227 --> 00:49:46,467
to see this kind of happening, people are actually using it. Google's getting kind of removed from it

588
00:49:46,467 --> 00:49:54,007
all. And I think it's great. I want to get on top of it. But I can see a world in 10 years

589
00:49:54,007 --> 00:50:00,407
where everyone's just saying grok. Or if you go on Twitter right now, everything is at grok. What's

590
00:50:00,407 --> 00:50:05,827
this mean? At grok. I mean, we start to lose our mind a little bit, our at least critical thinking,

591
00:50:05,827 --> 00:50:09,247
where we just completely just turn it over to AI.

592
00:50:09,847 --> 00:50:11,647
But at the same time, as a business,

593
00:50:11,827 --> 00:50:15,647
I want to make sure AI knows that I'm the person to talk to.

594
00:50:15,927 --> 00:50:18,507
I want them to send them over my way.

595
00:50:18,967 --> 00:50:21,067
So I'm trying to kind of figure that out,

596
00:50:21,067 --> 00:50:25,027
how to develop my SEO, develop my website

597
00:50:25,027 --> 00:50:27,887
to kind of cater to these robots, I suppose,

598
00:50:28,307 --> 00:50:29,887
because I see the trend building.

599
00:50:31,547 --> 00:50:34,187
So just my real world, you know,

600
00:50:34,187 --> 00:50:39,247
The real world adoption is not here yet, but we're starting to see it kind of sprinkle in.

601
00:50:40,267 --> 00:50:40,767
It's interesting.

602
00:50:41,027 --> 00:50:42,927
I do think Google is in trouble.

603
00:50:43,587 --> 00:50:48,667
And I think they know that they're in trouble too, which is why they're investing so much in AI solutions and in Gemini.

604
00:50:49,787 --> 00:50:54,167
But yeah, web search is not going to be the same anymore.

605
00:50:55,087 --> 00:50:55,307
Yeah.

606
00:50:55,367 --> 00:51:01,307
If you don't have the description right at the top, the AI response, you're going to be left behind.

607
00:51:01,307 --> 00:51:06,867
And if you can't control that AI response, I mean, people pay good money to be at the top.

608
00:51:06,867 --> 00:51:14,067
But if above that is the AI response and it's actually a good response, I don't know.

609
00:51:14,227 --> 00:51:15,147
You can't pay for that.

610
00:51:16,047 --> 00:51:23,487
Yeah, it worries me too because I do a lot of contributions that are not within the mainstream.

611
00:51:24,127 --> 00:51:30,727
For example, I have chosen to host the vast majority of code that I have written on GitLab instead of GitHub.

612
00:51:31,307 --> 00:51:34,827
And that's because GitLab fits with my ethos of open source.

613
00:51:35,727 --> 00:51:47,527
But I wonder, like, am I limiting the spread of, like, my coding style and my, like, libraries that I've developed here by not putting them on GitHub?

614
00:51:48,207 --> 00:51:53,347
Because the LLMs are probably scanning GitHub and maybe not scanning GitLab.

615
00:51:53,467 --> 00:51:55,767
And there's lots of things like this where it's...

616
00:51:55,767 --> 00:51:59,847
Yeah, I've looked into that because I'm trying to figure out where are they pulling all this from.

617
00:51:59,847 --> 00:52:04,407
and you kind of have to be there. It's like another way of like, I got to,

618
00:52:04,407 --> 00:52:07,527
I got to be in that little honeypot of information.

619
00:52:08,087 --> 00:52:11,967
It's like bullying people into centralization and also bullying like

620
00:52:11,967 --> 00:52:15,407
conservatism to some degree, because there's, there's even things like,

621
00:52:15,767 --> 00:52:17,627
I think it's slowing the adoption of,

622
00:52:17,727 --> 00:52:22,207
of new technologies because the AI is trained on the last version of the

623
00:52:22,207 --> 00:52:24,367
technology. And previously, you know,

624
00:52:24,427 --> 00:52:27,227
humans would just learn the new version and upgrade everything. Now,

625
00:52:27,227 --> 00:52:32,647
like there's now that things are going to be maintained by AIs humans don't care as much

626
00:52:32,647 --> 00:52:38,847
about developer experience because if the AI is fine developing it then it's kind of okay if the

627
00:52:38,847 --> 00:52:46,267
code is a little bit messy so so like there is there's definitely some like tech stack stagnation

628
00:52:46,267 --> 00:52:51,927
that is that is happening this kind of a problem where I I think for all of the acceleration AI is

629
00:52:51,927 --> 00:53:02,187
doing, it's also slowing certain things down. And speaking of the buggy coded apps,

630
00:53:02,467 --> 00:53:07,067
is there any concerns or cautions about the rise in vibe coded apps?

631
00:53:08,287 --> 00:53:13,367
From a security standpoint? I mean, just from any standpoint, yeah, but security as well.

632
00:53:13,367 --> 00:53:20,327
What was that app with all the girls on it that apparently was vibe coded?

633
00:53:20,327 --> 00:53:28,047
i mean my my perception of it is like it's like self-driving cars almost to where like yeah there

634
00:53:28,047 --> 00:53:34,387
there's a genuine fear here i get it but like if you just look at traffic accidents the vast

635
00:53:34,387 --> 00:53:41,167
majority of them are caused by humans being stupid behind the wheel and like it's really bad when

636
00:53:41,167 --> 00:53:47,667
there is a malfunction that like is a you know an obvious bug in a self-driving car that should

637
00:53:47,667 --> 00:53:52,447
have been avoided that maybe a human wouldn't have done that but still overall when you look

638
00:53:52,447 --> 00:53:57,807
at it statistically it's just way more likely that a human is going to crash a car than a computer

639
00:53:57,807 --> 00:54:08,727
and and so i think in in a lot of cases we're seeing the same for for ai generated code like

640
00:54:08,727 --> 00:54:13,707
yes maybe maybe now there's people who are writing code who wouldn't have been writing

641
00:54:13,707 --> 00:54:19,127
code before and just due to the sheer volume of code being generated that wasn't generated before

642
00:54:19,127 --> 00:54:24,727
we're seeing more incidences of it but i bet if you would look at it on the scale of like a

643
00:54:24,727 --> 00:54:30,047
percentage wise and i don't have any evidence of this i'm just you know saying my speculation

644
00:54:30,047 --> 00:54:36,747
which is yeah it's just that i i would guess that you would see that there is less security

645
00:54:36,747 --> 00:54:44,467
incidents because ai is very good at figuring out like you know identifying tiny little problems

646
00:54:44,467 --> 00:54:51,787
that a human wouldn't necessarily be able to see or you know humans get tired um you know and like

647
00:54:51,787 --> 00:54:59,107
we have we have many different responsibilities uh potentially families like we only so many hours

648
00:54:59,107 --> 00:55:06,227
in the day and and ai just doesn't care about that so like it'll find your your little tiny sql

649
00:55:06,227 --> 00:55:13,307
injection vulnerability um that you know that you you missed yeah i'll tell you what

650
00:55:13,307 --> 00:55:22,847
yeah go ahead kw okay uh i was just gonna i mean you mentioned autonomous vehicles one almost ran

651
00:55:22,847 --> 00:55:26,727
me off the road i'm in phoenix we have tons of waymos they're all over my neighborhood

652
00:55:26,727 --> 00:55:32,507
one almost ran me off the road one night and i'm like what the hell so i tried to call waymo and

653
00:55:32,507 --> 00:55:33,287
It's just robots.

654
00:55:33,887 --> 00:55:35,527
They're just, I can't even talk to anybody.

655
00:55:35,687 --> 00:55:41,787
So like the liability, it's a weird feeling when no one's accountable, but a robot, you

656
00:55:41,787 --> 00:55:42,247
know what I mean?

657
00:55:42,267 --> 00:55:43,367
It's just really weird.

658
00:55:44,187 --> 00:55:49,327
Well, I'm not going to stand for Waymo or try to like defend big tech corporations, but

659
00:55:49,327 --> 00:55:52,107
I've had humans, you know, drive me off the road too.

660
00:55:52,887 --> 00:55:53,327
Yeah.

661
00:55:53,327 --> 00:55:57,847
But if, if someone vibe coded something and it actually just, just accidentally stole

662
00:55:57,847 --> 00:56:02,167
your data and spread it all over the world, you're like, sorry, bro, it was vibe coded.

663
00:56:02,167 --> 00:56:09,527
But humans do that too. I get what you're saying though. It's a different feeling when it's a robot.

664
00:56:09,527 --> 00:56:10,547
It's a different feeling. Yeah.

665
00:56:10,947 --> 00:56:24,827
Well, there's also, and you mentioned percentage, Alex, and I think I tend to agree with that. Again, I don't have the stats to back it. It's probably a very low percentage of these apps that have vulnerabilities. But I don't think the percentage is the entire picture. It's also severity.

666
00:56:24,827 --> 00:56:27,587
you could have a tiny percentage

667
00:56:27,587 --> 00:56:29,327
of

668
00:56:29,327 --> 00:56:30,667
wipecoded apps that have vulnerabilities

669
00:56:30,667 --> 00:56:33,487
but the vulnerability itself could be incredibly

670
00:56:33,487 --> 00:56:35,447
severe and so this ties

671
00:56:35,447 --> 00:56:37,087
back to the human

672
00:56:37,087 --> 00:56:38,467
part of this as well

673
00:56:38,467 --> 00:56:41,407
part of me worries

674
00:56:41,407 --> 00:56:41,967
that

675
00:56:41,967 --> 00:56:45,247
while wipecoding

676
00:56:45,247 --> 00:56:47,347
is phenomenal for

677
00:56:47,347 --> 00:56:48,487
existing developers

678
00:56:48,487 --> 00:56:51,247
because it 10x is 100x is their

679
00:56:51,247 --> 00:56:53,327
productivity and we've seen this

680
00:56:53,327 --> 00:56:54,747
we've seen testaments to this

681
00:56:54,747 --> 00:57:02,607
but it also seems to it can possibly light a fire uh to the fuel of the dunning kruger effect

682
00:57:02,607 --> 00:57:08,427
in non-developers who suddenly see a little app that you know a couple of prompts somehow

683
00:57:08,427 --> 00:57:12,527
miraculously work and now they think they're developers and they start building financial

684
00:57:12,527 --> 00:57:18,527
infrastructure or health infrastructure right dealing with personal data and that

685
00:57:18,527 --> 00:57:44,130
maybe there hasn been a severe incident yet but that just given human nature and the Dunning effect I don know I kind of see the opposite as being true with that If anything it product managers who their normal job is to just go and boss people around without knowing how to code at all are kind of fucking around and finding out right now

686
00:57:44,130 --> 00:57:56,450
Because they're like, you know, going and typing something in to an AI and then realizing all of the implications of what their request actually entails when they didn't realize that before.

687
00:57:56,450 --> 00:58:05,410
And so once surfacing that and making it so people can see like, oh, actually, this involves this and this and this and this.

688
00:58:05,470 --> 00:58:13,850
And you didn't you didn't realize that before. I think like it helps people who are incompetent maybe become a little bit more competent than they were previously.

689
00:58:14,130 --> 00:58:22,330
well that is the optimistic case right that you people realize how little they know

690
00:58:22,330 --> 00:58:28,570
when they start vibe coding um but it is but i think the the converse is also possible people

691
00:58:28,570 --> 00:58:34,950
assume that they're geniuses now because they got a they got clawed to you know build them a little

692
00:58:34,950 --> 00:58:39,870
game or whatever it is and and then the next thing they do is they build a banking app

693
00:58:39,870 --> 00:58:43,870
hey man it wasn't a little game avi it was a huge game

694
00:58:43,870 --> 00:58:52,330
massive idea huge breaking don't build a banking app kid i'm begging you

695
00:58:52,330 --> 00:58:57,950
uh no i wasn't talking about your game i was just you know make a general statement but here's an

696
00:58:57,950 --> 00:59:03,870
here's another thing and this i ran into and i'm semi-technical i've written handwritten code back

697
00:59:03,870 --> 00:59:11,330
in the day and I spent a lot of time on, even before MK, in fact, I was vibe coding before the

698
00:59:11,330 --> 00:59:19,970
term became popular parlance, right? You know, using replet and those things. And so that was

699
00:59:19,970 --> 00:59:25,570
maybe starting December of last year, all the way up to maybe two or three months ago. And the thing

700
00:59:25,570 --> 00:59:33,450
that eventually made me slow down is I was just burning money. It was, yes, when you get it to

701
00:59:33,450 --> 00:59:38,250
work, it's great. But as you start ramping up the complexity, you start running into debug loops,

702
00:59:38,630 --> 00:59:43,150
the doom loops. And before you know it, and I was using things like Routster and

703
00:59:43,150 --> 00:59:49,250
PPQ, where you're actually spending sats on this. And I was just watching those sats evaporate.

704
00:59:49,930 --> 00:59:54,030
You know, I went through a week stretch when I was spending 50, 60 bucks a day.

705
00:59:54,030 --> 00:59:58,910
and so that made me hit the pause button.

706
00:59:59,810 --> 01:00:03,910
Is that something that you're thinking about, Alex?

707
01:00:04,010 --> 01:00:07,010
Like how do we make this more sustainable?

708
01:00:08,770 --> 01:00:11,030
Yeah, so there's a few things

709
01:00:11,030 --> 01:00:13,290
that I've been thinking about with that.

710
01:00:13,710 --> 01:00:17,050
So one is that I know that in Staxon and Shakespeare right now,

711
01:00:17,110 --> 01:00:18,070
it's not optimized.

712
01:00:18,750 --> 01:00:22,630
There are ways to optimize that I've kind of been noticing

713
01:00:22,630 --> 01:00:24,490
as I've been building this stuff.

714
01:00:25,410 --> 01:00:28,030
And it's things like, you know,

715
01:00:28,150 --> 01:00:31,670
just compressing the context window at certain points

716
01:00:31,670 --> 01:00:34,770
within the flow of messages.

717
01:00:35,210 --> 01:00:39,550
Because what you're paying for most of the time

718
01:00:39,550 --> 01:00:40,670
is the input tokens.

719
01:00:41,330 --> 01:00:43,590
And that can get expensive really fast,

720
01:00:43,710 --> 01:00:45,350
especially on big projects.

721
01:00:46,150 --> 01:00:48,750
The other thing is just the cost of the model itself.

722
01:00:48,750 --> 01:00:58,310
And for that, we can thank China for creating open source models that are extremely efficient, energy efficient.

723
01:00:59,350 --> 01:01:08,530
And their situation there is very interesting right now because they have unlimited energy access, essentially, compared to us.

724
01:01:08,650 --> 01:01:10,110
But they have a chip shortage.

725
01:01:11,330 --> 01:01:16,110
So they're kind of being driven to create more efficient LLMs.

726
01:01:16,110 --> 01:01:21,810
LLMs. And the way they're doing it is with a mixture of expert models, where in the current

727
01:01:21,810 --> 01:01:26,590
American models, there are these giant monoliths that if you ask it a question about cooking advice,

728
01:01:26,930 --> 01:01:31,390
it has to activate all of the world's knowledge to answer your question about cooking advice.

729
01:01:32,090 --> 01:01:39,770
Whereas in the mixture of expert models, it has a gate that will decide that you're asking it

730
01:01:39,770 --> 01:01:46,830
about cooking and then it'll route you to its cooking like internals where they train what they

731
01:01:46,830 --> 01:01:52,610
call experts on particular topics and then they combine responses from multiple experts and so

732
01:01:52,610 --> 01:02:03,090
this actually has a very real uh you know consequence on um on the cpu gpu resource usage

733
01:02:03,090 --> 01:02:09,090
result so designing models in a more efficient way is a huge one also designing the tools in a

734
01:02:09,090 --> 01:02:13,050
different way is a huge one. And they're both areas that I'm exploring and that I think the

735
01:02:13,050 --> 01:02:19,610
world is exploring right now. Is this 4D chess by Trump, Alex, that he's squeezing China with the

736
01:02:19,610 --> 01:02:27,370
export restrictions, forcing them to innovate on the edges, basically, with scraps, with five-year-old

737
01:02:27,370 --> 01:02:34,990
chips? I don't know that Trump is capable of thinking beyond 2D or 1D, but it certainly seems

738
01:02:34,990 --> 01:02:36,410
like that in the output.

739
01:02:37,890 --> 01:02:38,850
It's 8D

740
01:02:38,850 --> 01:02:40,430
checkers, guys. Come on.

741
01:02:42,730 --> 01:02:44,370
The other issue is, I mean,

742
01:02:44,690 --> 01:02:45,870
you know, Claude

743
01:02:45,870 --> 01:02:48,190
4.0, I think there's

744
01:02:48,190 --> 01:02:50,650
4.1 now, right? Both on

745
01:02:50,650 --> 01:02:52,790
Sonnet and Opus, and those are probably the best

746
01:02:52,790 --> 01:02:54,290
coding models out there.

747
01:02:54,770 --> 01:02:55,770
Prohibitively expensive.

748
01:02:58,430 --> 01:02:58,850
Yeah, because

749
01:02:58,850 --> 01:03:00,650
there are those...

750
01:03:00,650 --> 01:03:01,170
Go ahead.

751
01:03:02,170 --> 01:03:04,090
It's the monolith model.

752
01:03:04,090 --> 01:03:11,870
so a model that i would really recommend looking into is glm by zai um and that is a mixture of

753
01:03:11,870 --> 01:03:16,130
extra experts model out of china maybe be careful what you type into it but then again

754
01:03:16,130 --> 01:03:20,970
like american models you have the same privacy concerns typing stuff into it so

755
01:03:20,970 --> 01:03:28,970
um glm won't write you a story about tiananmen square and uh you know american models have their

756
01:03:28,970 --> 01:03:35,010
own restrictions, but it's like 20 times cheaper using GLM, and it's really, really good at coding.

757
01:03:35,710 --> 01:03:41,290
It has some occasional hiccups in its infrastructure because it seems like they're

758
01:03:41,290 --> 01:03:46,070
struggling to scale to the demand. But aside from that, it's a really great model I would

759
01:03:46,070 --> 01:03:51,790
highly recommend if Claude is too expensive. Well, here's the other issue with American models.

760
01:03:52,430 --> 01:03:58,590
And thank you for that, by the way, Alex, GLM. I think folks who've been struggling paying the

761
01:03:58,590 --> 01:04:00,590
builds with Claude, I should certainly look into that.

762
01:04:00,910 --> 01:04:03,950
DeepSea Carbon isn't that terrible either, right?

763
01:04:04,250 --> 01:04:06,870
Another mixture of experts model.

764
01:04:07,510 --> 01:04:08,410
I don't know.

765
01:04:08,970 --> 01:04:09,490
Okay.

766
01:04:10,730 --> 01:04:10,970
Go ahead.

767
01:04:11,170 --> 01:04:13,150
Well, this is just a small detail,

768
01:04:13,330 --> 01:04:15,110
but for people using OpenRouter,

769
01:04:17,290 --> 01:04:19,750
for open source models specifically,

770
01:04:20,210 --> 01:04:21,830
there are different providers, right?

771
01:04:21,830 --> 01:04:24,110
When it's an open source model, anyone can provide it.

772
01:04:24,110 --> 01:04:34,270
And I noticed that some providers actually return lower quality results for the same model when you're using OpenRouter.

773
01:04:34,410 --> 01:04:41,690
And OpenRouter will just randomly route you to any of these providers without concern of the quality of the output.

774
01:04:42,210 --> 01:04:48,850
So yes, DeepSeq R1 from provider A can actually have inferior results to DeepSeq R1 from provider B.

775
01:04:49,310 --> 01:04:50,270
Just a thing to note.

776
01:04:51,230 --> 01:04:52,010
That's interesting.

777
01:04:52,010 --> 01:04:57,830
Maybe it's because Provider A is using their own computer infrastructure and it's not up to scratch.

778
01:04:58,810 --> 01:04:59,630
Could be, yeah.

779
01:04:59,770 --> 01:05:02,430
Yeah, and the configuration of it matters a lot too, apparently.

780
01:05:02,730 --> 01:05:07,210
So they'll misconfigure it and then it'll have an impact on the result.

781
01:05:08,350 --> 01:05:08,690
Got it.

782
01:05:09,250 --> 01:05:15,810
But you mentioned OpenRata and that was sort of leading into my other question, which is, I mean, American models, yeah, maybe they're more powerful.

783
01:05:15,810 --> 01:05:20,710
they're certainly more expensive, but they have, if you're using them for these coding apps,

784
01:05:21,730 --> 01:05:27,710
you have to get API access and to get direct API access, either from Anthropic or from OpenAI

785
01:05:27,710 --> 01:05:32,970
or from anywhere else, it's a full rectal exam, right? You have to take a photograph of yourself,

786
01:05:33,150 --> 01:05:39,410
hold your license. It's the second time we've discussed rectal things, Avi.

787
01:05:40,150 --> 01:05:41,410
I'm starting to trend here.

788
01:05:41,410 --> 01:05:45,310
So this is, the difference is you were using it literally, Kedabi.

789
01:05:45,390 --> 01:05:46,010
This is a metaphor.

790
01:05:47,830 --> 01:05:50,630
And it's terrible.

791
01:05:50,750 --> 01:05:53,510
So then you have something like OpenRouter, which in theory,

792
01:05:54,250 --> 01:05:56,650
well, actually you have KYC with OpenRouter as well.

793
01:05:57,370 --> 01:06:00,230
But you can then use something like Routester

794
01:06:00,230 --> 01:06:02,750
that kind of sits in the middle of you and OpenRouter.

795
01:06:04,990 --> 01:06:06,030
PPQ is the same, right?

796
01:06:06,790 --> 01:06:08,130
So those are no KYC.

797
01:06:08,130 --> 01:06:11,250
You can pay in Bitcoin or in eCash in the case of Routester.

798
01:06:11,410 --> 01:06:14,450
and then they route through OpenRouter,

799
01:06:14,970 --> 01:06:18,990
which, so you're arbitraging the KYC stuff

800
01:06:18,990 --> 01:06:21,050
that OpenRouter would do,

801
01:06:21,110 --> 01:06:23,030
or if you're going direct to the model providers,

802
01:06:23,430 --> 01:06:24,030
they would do.

803
01:06:24,770 --> 01:06:26,190
Is that, so are you seeing,

804
01:06:26,870 --> 01:06:29,990
or actually, can you even track the usage

805
01:06:29,990 --> 01:06:31,790
on Shakespeare and MKStack,

806
01:06:31,790 --> 01:06:33,550
which provider they're using,

807
01:06:33,610 --> 01:06:37,670
whether they're using Routes to PPQ or just direct API?

808
01:06:37,670 --> 01:06:46,790
so um on the current version of shakespeare uh all of the ai traffic is going through the server

809
01:06:46,790 --> 01:06:52,490
in my living room and that is hooked up to open router on my end so i have an open router api key

810
01:06:52,490 --> 01:06:58,870
in there and so that's where all the traffic is going and in the new version of shakespeare that

811
01:06:58,870 --> 01:07:06,310
i'm currently working on that is very experimental um the user's browser connects directly to third

812
01:07:06,310 --> 01:07:13,030
party AI providers. So you would essentially go and get an API key from like OpenAI or Anthropic

813
01:07:13,030 --> 01:07:18,750
or whatever. And then you could install multiple keys inside of your Shakespeare client and then

814
01:07:18,750 --> 01:07:25,210
use that to access models across any provider. And in that case, we wouldn't be able to figure

815
01:07:25,210 --> 01:07:30,370
out which models people are using. It's a more privacy friendly, like offline centric approach

816
01:07:30,370 --> 01:07:31,370
in the newer version.

817
01:07:33,830 --> 01:07:36,130
But the people themselves would have to KYC

818
01:07:36,130 --> 01:07:38,490
if they are going direct to Anthropic.

819
01:07:38,830 --> 01:07:45,010
Yeah, so I feel like I've heard of this happening,

820
01:07:45,110 --> 01:07:47,710
but I'm not sure that I've ever had to do that.

821
01:07:49,610 --> 01:07:51,810
Yeah, I believe my wife was telling me

822
01:07:51,810 --> 01:07:53,430
she had to do it on OpenAI,

823
01:07:53,810 --> 01:07:56,750
of scanning her face 360 degrees and stuff.

824
01:07:56,830 --> 01:08:08,173
It crazy But the Chinese providers don care about this So if you want to go with GLM and you want it to be 20 times cheaper and you don want to scan your face you can you can certainly do that

825
01:08:08,173 --> 01:08:17,233
um i believe that ppq is not an open router uh proxy i believe they are a proxy directly to the

826
01:08:17,233 --> 01:08:24,773
upstream providers which is very difficult to do um i at one point i was trying to build like an

827
01:08:24,773 --> 01:08:31,633
open router alternative and i kind of just gave up on it because like instead i just make it so

828
01:08:31,633 --> 01:08:37,233
that the user configures their keys directly with the provider because these ai big ai companies

829
01:08:37,233 --> 01:08:43,373
like anthropic just don't want to work with with you um they'll just ignore it and you can't expect

830
01:08:43,373 --> 01:08:51,273
them to do anything different so uh i really hate anthropic i i i dislike open router a lot although

831
01:08:51,273 --> 01:08:55,613
I dislike them a little bit less after realizing what they deal with.

832
01:08:56,153 --> 01:08:58,733
I mainly dislike that they have the word open in their name,

833
01:08:58,873 --> 01:09:02,093
despite being closed source, similar to open AI.

834
01:09:03,233 --> 01:09:05,093
Yeah, closed source and full KYC.

835
01:09:06,713 --> 01:09:07,413
Yeah, exactly.

836
01:09:07,853 --> 01:09:16,113
But, you know, I'm feeling a little bit more bullish on the idea of local LLMs

837
01:09:16,113 --> 01:09:19,013
than I was, you know, three months ago.

838
01:09:19,013 --> 01:09:27,873
it's unfortunately to run something like glm you still need an upfront cost of about a million

839
01:09:27,873 --> 01:09:34,753
dollars just in graphics cards and then once you have your million dollars worth of graphics cards

840
01:09:34,753 --> 01:09:40,293
then you need the actual server infrastructure and electricity to put them into computers and

841
01:09:40,293 --> 01:09:50,613
run them. But we're seeing a lot of advancements in the realm of Ollama, as well as things like

842
01:09:50,613 --> 01:09:57,753
Web LLM that allow you to run tiny little models inside of a web browser. And even now, newer phones

843
01:09:57,753 --> 01:10:05,453
are starting to ship with LLMs built into the phone. And Google and Apple are creating specialized

844
01:10:05,453 --> 01:10:10,973
hardware to allow this to be possible to you can run these little models offline and they're not

845
01:10:10,973 --> 01:10:18,073
bad they're maybe not going to be like excellent coding models but i'm really excited for for the

846
01:10:18,073 --> 01:10:23,653
use case of moderation specific specifically like imagine you're in a noster client and you want to

847
01:10:23,653 --> 01:10:28,133
say i don't want to see this particular type of of speech i don't want to see any of this bitcoin

848
01:10:28,133 --> 01:10:34,853
stuff anymore right like our best thing we have right now is to just block keywords but imagine

849
01:10:34,853 --> 01:10:39,173
that you could just have a prompt to say, don't show me any of this Bitcoin crap. And then your

850
01:10:39,173 --> 01:10:44,753
posts all get run through a local LLM on your phone. And it is actually powerful enough to

851
01:10:44,753 --> 01:10:50,093
filter that out for you or spam or like, you know, whatever else it is. If you're like a liberal,

852
01:10:50,233 --> 01:10:55,653
maybe you don't want hate speech. You can, you can feel free to define that however you want in

853
01:10:55,653 --> 01:11:01,173
your own personal terms. And I think that this is something that could, you know, fix a lot of

854
01:11:01,173 --> 01:11:09,713
problems for people using Noster. So I think the future is bright for people who care about freedom

855
01:11:09,713 --> 01:11:21,713
and want to use AI. And at the same time, you know, I've always kind of had an enemy

856
01:11:21,713 --> 01:11:28,453
as I've been building stuff in this, you know, fighting for freedom, right? In decentralized

857
01:11:28,453 --> 01:11:34,073
open source space. And in the past, it has been big tech companies like Google and Facebook and

858
01:11:34,073 --> 01:11:40,133
Twitter. And now that they're being eclipsed by these AI giants, I think that our biggest enemy

859
01:11:40,133 --> 01:11:51,913
is open AI, Anthropic, to some smaller degree, open router. And so what I'm looking forward to

860
01:11:51,913 --> 01:11:58,353
most is just being able to use local models, being able to do things on your own machine,

861
01:11:58,453 --> 01:12:03,573
for users to have as much control as possible over that experience and then be able to

862
01:12:03,573 --> 01:12:08,213
essentially to build things that can destabilize the social order,

863
01:12:09,153 --> 01:12:16,993
which is the most fun use case for AI-generated apps, in my opinion. Maybe we can overthrow some

864
01:12:16,993 --> 01:12:22,713
regimes. Well, I think we will run into computational reality. And I think we're

865
01:12:22,713 --> 01:12:29,933
already seeing that, Alex, which is we were, I would say, maybe September of last year to about

866
01:12:29,933 --> 01:12:37,913
June of this year, we were in the golden age of venture capital subsidized compute, which is how

867
01:12:37,913 --> 01:12:44,073
we were able to, we were able to get these top of the line models from open air, anthropic, and

868
01:12:44,073 --> 01:12:53,293
what have you at relatively cheap prices i think reality is going to catch up the the vc subsidies

869
01:12:53,293 --> 01:12:58,633
are going to go away and we'll have to pay the true cost of electricity for these things

870
01:12:58,633 --> 01:13:04,253
which if it if and if that is the case and i if you're an individual running local infrastructure

871
01:13:04,253 --> 01:13:08,693
you're doing that anyway there's no one subsidizing you there's no sugar daddy vc who's

872
01:13:08,693 --> 01:13:15,493
paying your bills. And I wonder if that will hamper some of the things that you're talking

873
01:13:15,493 --> 01:13:20,113
about, which I think is the ideal world, right? We do want to disrupt the social order with

874
01:13:20,113 --> 01:13:27,813
decentralized these individual nodes running their own compute and so on. But I also fear that

875
01:13:27,813 --> 01:13:33,453
reality, economic reality, right? Thermodynamic reality will catch up.

876
01:13:33,453 --> 01:13:54,793
Yeah, I think we're in a bubble for sure. But I think when it bursts, it's going to disproportionately affect the big AI companies like Anthropic, Google, OpenAI. And so that's why it's important to start trying to find solutions now for open source and especially locally runnable models.

877
01:13:54,793 --> 01:14:04,113
yeah well i have another pet peeve or or actually recent grievance with vipe coded

878
01:14:04,113 --> 01:14:11,953
apps alex and and that's this which it's ui saturation uh and you'd mentioned this earlier

879
01:14:11,953 --> 01:14:17,993
on right that and this is not not particularly about mk stack i i ran into this when i was using

880
01:14:17,993 --> 01:14:24,053
replet late last year and early this year is the ui the front-end frameworks that they all use

881
01:14:24,053 --> 01:14:27,973
ShadCN, Tailwind CSS, and it's all the same, right?

882
01:14:28,233 --> 01:14:29,753
Essentially the same set of libraries

883
01:14:29,753 --> 01:14:31,953
that they're using to generate the components,

884
01:14:32,373 --> 01:14:35,013
generate the layout of these apps.

885
01:14:35,653 --> 01:14:37,293
All of these apps are looking the same now.

886
01:14:37,753 --> 01:14:39,413
And with more Vibe-coded apps,

887
01:14:39,573 --> 01:14:43,053
you're saturated with almost identical UIs.

888
01:14:43,133 --> 01:14:44,613
Yes, you can toggle dark mode

889
01:14:44,613 --> 01:14:46,053
to make it look slightly different,

890
01:14:46,193 --> 01:14:49,213
but the icons look the same and so on.

891
01:14:49,213 --> 01:14:52,673
Do you see that changing anytime soon

892
01:14:52,673 --> 01:14:57,333
that we'll have more variability in these user interfaces?

893
01:14:58,853 --> 01:15:01,053
Yeah, so I have a few thoughts on that.

894
01:15:01,653 --> 01:15:04,093
In one sense, I think that that's a win more than anything.

895
01:15:04,713 --> 01:15:07,173
Like we've just over the years figured out

896
01:15:07,173 --> 01:15:09,073
what the best user interfaces look like.

897
01:15:09,893 --> 01:15:12,253
I mean, there have been people arguing

898
01:15:12,253 --> 01:15:13,393
since the beginning of the web

899
01:15:13,393 --> 01:15:15,093
that HTML is supposed to be semantic

900
01:15:15,093 --> 01:15:17,453
and that maybe we don't even need styles.

901
01:15:17,573 --> 01:15:19,693
Maybe it should be like an operating system UI kit

902
01:15:19,693 --> 01:15:21,913
where you put elements

903
01:15:21,913 --> 01:15:24,313
and then the browser decides how to display those elements.

904
01:15:25,873 --> 01:15:32,313
And obviously, as an artist, you know, there's, I also see the point of view that we should

905
01:15:32,313 --> 01:15:34,273
have unique and creative websites.

906
01:15:34,893 --> 01:15:38,073
But I don't think every website needs to be unique and creative.

907
01:15:38,193 --> 01:15:42,833
I think it kind of depends on whether what is being built is more of an application or

908
01:15:42,833 --> 01:15:47,593
more of, you know, a creative project.

909
01:15:48,313 --> 01:15:49,673
And sometimes there's both.

910
01:15:49,673 --> 01:15:56,993
And definitely, you know, there can be room to have both in the same site or in the same project.

911
01:15:57,213 --> 01:16:02,753
But I don't really think it's a limitation of like Tailwinds, ShadCN.

912
01:16:03,953 --> 01:16:14,773
I have seen LLMs be able to just write custom style sheets that sort of make those types of sites look a lot more unique and interesting.

913
01:16:16,513 --> 01:16:18,013
Especially lovable right now.

914
01:16:18,013 --> 01:16:21,833
I hate to shill them in any way because they're like my biggest competitor,

915
01:16:22,593 --> 01:16:26,073
but they have figured out like, like, for example,

916
01:16:26,073 --> 01:16:30,133
if I go in and I say, build me a spooky website, it's going to have like,

917
01:16:30,273 --> 01:16:33,013
you know, uh, sort of suns,

918
01:16:33,733 --> 01:16:36,713
Halloween type of sunset with like, um,

919
01:16:37,033 --> 01:16:42,633
orange sky and creepy trees poking into it and like a custom font being

920
01:16:42,633 --> 01:16:46,993
selected. And it's got like pumpkins and stuff. And, and so, you know,

921
01:16:46,993 --> 01:16:50,013
they're advancing into that territory right now.

922
01:16:50,113 --> 01:16:52,133
That's where Lovable has grown in the past,

923
01:16:52,273 --> 01:16:53,833
you know, three, six months or whatever.

924
01:16:54,313 --> 01:16:57,593
The way that I have grown MKStack is not in that way.

925
01:16:58,273 --> 01:17:01,533
For me, it's been more about making sure

926
01:17:01,533 --> 01:17:04,993
that a Nostra application can actually connect to Nostra

927
01:17:04,993 --> 01:17:06,433
and do Nostra things for real.

928
01:17:06,633 --> 01:17:08,753
So that rather than it just being like a demo site,

929
01:17:08,793 --> 01:17:09,733
it's something that's functional.

930
01:17:11,333 --> 01:17:15,173
But the fact that they are able to do it right now

931
01:17:15,173 --> 01:17:16,433
proves to me that it's doable.

932
01:17:16,993 --> 01:17:20,013
And so it's something to look into on the tooling side.

933
01:17:20,853 --> 01:17:24,353
I'm not worried about it, basically, is the gist of that.

934
01:17:27,353 --> 01:17:27,833
Cool.

935
01:17:28,393 --> 01:17:30,713
So you mentioned Shakespeare 2.0.

936
01:17:30,913 --> 01:17:32,293
What can we expect there?

937
01:17:34,053 --> 01:17:37,193
So this is a complete overhaul of the architecture of Shakespeare.

938
01:17:38,193 --> 01:17:42,333
In Shakespeare 1, which is what's currently on Shakespeare.diy,

939
01:17:43,273 --> 01:17:46,233
in many ways, I consider that to be a success.

940
01:17:46,993 --> 01:17:53,813
People have been able to build all kinds of stuff with it, including people who may not have thought that they could do it.

941
01:17:54,533 --> 01:17:56,033
People who are not programmers.

942
01:17:57,033 --> 01:18:02,133
But there's also been a number of bugs that have been difficult to overcome in that system.

943
01:18:04,213 --> 01:18:06,513
As any power user of it probably knows.

944
01:18:06,753 --> 01:18:16,333
And it has also made me question whether this architecture makes sense for the kind of organization that we are.

945
01:18:16,333 --> 01:18:30,093
Because we are not a VC funded company like Lovable that can have staff engineers monitoring the production server infrastructure 24-7, making sure that there's no bugs in that.

946
01:18:30,093 --> 01:18:46,356
And I feel like to be really fair to the open source community and to people who want to use this thing the best user experience is going to be one that I can deploy and I don necessarily have to be constantly checking and babysitting it to make sure it not going to go down

947
01:18:46,356 --> 01:18:55,896
And so this new architecture is one in which all of the files of your code are actually stored inside of your browser storage.

948
01:18:55,896 --> 01:19:04,436
the end goal is that you will push your code to Noster Git servers.

949
01:19:05,656 --> 01:19:08,976
This is something that Dan Conway Dev has been working on

950
01:19:08,976 --> 01:19:10,396
that I'm really excited about.

951
01:19:10,836 --> 01:19:12,956
It's permissionless Noster Git servers.

952
01:19:13,096 --> 01:19:14,536
All you need is a Noster identity,

953
01:19:15,116 --> 01:19:20,936
and then you can just instantly push code to what is essentially a Git relay.

954
01:19:20,936 --> 01:19:30,856
and so in a way a bigger picture of what this can do is to be a replacement for github in the ai

955
01:19:30,856 --> 01:19:35,556
age so that's something i'm really excited about is people being able to share projects

956
01:19:35,556 --> 01:19:40,676
none of the code would be hosted on my server anymore it would be hosted in the user's browser

957
01:19:40,676 --> 01:19:44,036
and then if they want to synchronize it with their other devices they would push it to nostril

958
01:19:44,036 --> 01:19:49,636
get relays and then clone it back down on their other devices the user's browser connects directly

959
01:19:49,636 --> 01:19:57,536
to remote AI services rather than to my thing that intercepts them so that they can decide

960
01:19:57,536 --> 01:19:59,096
which AI service they want to use.

961
01:19:59,276 --> 01:20:01,736
They can switch between providers.

962
01:20:01,876 --> 01:20:03,596
They can use multiple providers at once.

963
01:20:04,136 --> 01:20:10,136
It's really just all about user freedom and control and open source and making something

964
01:20:10,136 --> 01:20:15,676
that I know that even if I'm not constantly maintaining it, that's still going to continue

965
01:20:15,676 --> 01:20:21,456
to be useful to people because essentially the code is running on their own computer and on their

966
01:20:21,456 --> 01:20:29,136
own device. And so I think that if I can get some of these last pieces working, which have been a

967
01:20:29,136 --> 01:20:34,356
very difficult technical challenge, then I think people are going to like and appreciate this

968
01:20:34,356 --> 01:20:39,936
version of Shakespeare a lot more. And it's going to be a lot more fun for us to build upon because

969
01:20:39,936 --> 01:20:44,136
it's going to be one code base and sort of split out into these smaller ones that have to negotiate

970
01:20:44,136 --> 01:20:52,396
The complexity will be lower in areas that are important, even though it'll be higher in areas that will result in users having freedom.

971
01:20:53,776 --> 01:20:56,116
And so, yeah, I'm excited about it.

972
01:20:56,136 --> 01:21:05,716
Just trying to overcome some of the technical hurdles because we're really pushing the limits of what has been possible for web browsers to do ever.

973
01:21:06,336 --> 01:21:08,016
And nobody else is doing this.

974
01:21:08,016 --> 01:21:12,096
Like Lovable, V0, they're all keeping your files.

975
01:21:12,256 --> 01:21:13,536
They're doing everything on their servers.

976
01:21:14,136 --> 01:21:16,756
Nobody is trying to do AI completely in the browser.

977
01:21:18,396 --> 01:21:23,876
And so I think it would be a huge win for user freedom and just for AI in general.

978
01:21:26,216 --> 01:21:28,336
That sounds exciting, Alex.

979
01:21:28,496 --> 01:21:30,736
When can we expect Shakespeare 2.0?

980
01:21:33,416 --> 01:21:34,556
It's hard to say.

981
01:21:34,896 --> 01:21:38,576
I thought that it would be done by now already.

982
01:21:39,296 --> 01:21:43,656
But there were some parts that I thought were going to be so hard,

983
01:21:43,656 --> 01:21:44,696
And then they turned out to be easy.

984
01:21:44,776 --> 01:21:47,336
And right now I'm really stuck on making type checking work.

985
01:21:48,076 --> 01:21:52,376
And so I have a person that I have asked to help me.

986
01:21:53,116 --> 01:21:56,576
And, you know, hopefully between us, we'll figure something out.

987
01:21:57,316 --> 01:21:59,836
But it's in experimental territory.

988
01:22:01,616 --> 01:22:04,636
A couple of weeks, maybe, I hope.

989
01:22:07,536 --> 01:22:08,996
I heard a problem, Asabi.

990
01:22:09,956 --> 01:22:11,456
So let's see here, two weeks.

991
01:22:12,636 --> 01:22:13,476
Stay tuned.

992
01:22:13,656 --> 01:22:14,756
You heard it here first.

993
01:22:14,956 --> 01:22:18,056
September 19th, Shakespeare 2.0.

994
01:22:18,136 --> 01:22:18,896
Oh, my God.

995
01:22:20,596 --> 01:22:23,936
I'm going to build a little chain countdown where I take a little.

996
01:22:23,936 --> 01:22:24,716
It may be.

997
01:22:24,836 --> 01:22:25,476
I'm racing.

998
01:22:25,476 --> 01:22:26,076
Every day.

999
01:22:26,396 --> 01:22:27,656
I'm racing every day.

1000
01:22:27,776 --> 01:22:28,656
I know people want it.

1001
01:22:28,716 --> 01:22:32,276
Our team really wants it so they can move forward on marketing it and stuff.

1002
01:22:33,216 --> 01:22:37,156
So I'm really trying as hard as I possibly can to get it out as fast as possible.

1003
01:22:37,356 --> 01:22:40,036
It's just I'm moving mountains over here.

1004
01:22:41,856 --> 01:22:42,736
What's up, Avi?

1005
01:22:43,656 --> 01:22:46,536
No, I was saying that.

1006
01:22:46,536 --> 01:22:47,016
You were saying something.

1007
01:22:48,116 --> 01:22:50,896
Every time we have a developer as our guest,

1008
01:22:51,016 --> 01:22:52,636
we ask them their favorite question.

1009
01:22:53,116 --> 01:22:53,416
When?

1010
01:22:54,096 --> 01:22:54,936
That's what we do.

1011
01:22:57,256 --> 01:22:57,456
Yeah.

1012
01:22:58,056 --> 01:22:58,756
Huge break.

1013
01:22:58,756 --> 01:22:58,976
Sue.

1014
01:22:58,976 --> 01:22:59,936
You heard it here first.

1015
01:23:02,616 --> 01:23:04,156
Sue is always the answer.

1016
01:23:06,016 --> 01:23:07,656
In construction, it's always,

1017
01:23:07,796 --> 01:23:09,056
we can start in two weeks.

1018
01:23:09,356 --> 01:23:11,456
That's our motto.

1019
01:23:13,056 --> 01:23:13,576
Great.

1020
01:23:13,656 --> 01:23:19,096
what else are you working on Alex anything outside of Shakespeare else

1021
01:23:19,096 --> 01:23:27,716
um not in particular that's kind of the thing yeah well Chad is working on treasures and well

1022
01:23:27,716 --> 01:23:32,076
you've been you've been hiding treasures I've been hiding treasures yeah the proof of work

1023
01:23:32,076 --> 01:23:38,896
action there out on the real world connecting the universe with the real real world I made a

1024
01:23:38,896 --> 01:23:44,376
number of vibe coded apps um noster hub is one that i thought is really cool that's kind of what

1025
01:23:44,376 --> 01:23:51,156
inspired me to start doing get stuff on noster um and yeah a lot of people are just making really

1026
01:23:51,156 --> 01:23:57,276
cool stuff um you know chad who created treasures is pretty prolific he's done a few other things i

1027
01:23:57,276 --> 01:24:03,776
really am excited about uh another one he he's made a magic the gathering themed noster site

1028
01:24:03,776 --> 01:24:08,196
where people can go and they can build Magic the Gathering decks

1029
01:24:08,196 --> 01:24:10,476
and share them on Noster and comment on them and stuff.

1030
01:24:10,556 --> 01:24:13,776
I think that's super cool and could be grown in a lot of ways.

1031
01:24:13,936 --> 01:24:17,276
And also that's a huge example of communities for me.

1032
01:24:18,016 --> 01:24:21,516
And so these are the types of things I would love to see people building more

1033
01:24:21,516 --> 01:24:28,216
is unique social experiences that are catered to a particular niche.

1034
01:24:28,856 --> 01:24:32,716
And my hope is that these types of things are what's going to make Noster win.

1035
01:24:33,776 --> 01:24:36,116
Are you a Magic the Gathering enjoyer?

1036
01:24:36,836 --> 01:24:37,336
Absolutely.

1037
01:24:38,116 --> 01:24:40,836
I'm a green-red player, personally.

1038
01:24:41,376 --> 01:24:41,716
Okay.

1039
01:24:41,856 --> 01:24:45,356
I still got my black deck from 1993, I think.

1040
01:24:45,496 --> 01:24:45,916
Nice.

1041
01:24:47,136 --> 01:24:48,056
Oh, man.

1042
01:24:48,296 --> 01:24:54,856
It's funny because Magic the Gathering, I think, is the one game that comes up the most in our show, randomly.

1043
01:24:55,456 --> 01:25:00,156
Between, like, Lynn Alden, Druid.

1044
01:25:00,156 --> 01:25:05,576
uh we he wasn't he actually built a bitcoin magic the gathering game that's going to be

1045
01:25:05,576 --> 01:25:11,896
releasing soon uh it's just interesting uh is your app on no straps.com i'll go ahead um

1046
01:25:11,896 --> 01:25:20,836
so it's called surveil uh let me make sure i get the right domain surveil surveilled up cards

1047
01:25:20,836 --> 01:25:26,496
that's how you can find magic the gathering on no stir uh but but yeah when you're playing magic

1048
01:25:26,496 --> 01:25:31,776
the gathering you're essentially executing a program in your head which is really interesting

1049
01:25:31,776 --> 01:25:36,316
like it's been proven that magic the gathering is turing complete and so i think it's very

1050
01:25:36,316 --> 01:25:40,676
appealing to programmers for that reason because when you're playing magic you're essentially

1051
01:25:40,676 --> 01:25:46,756
creating a state on the board and then you're you're triggering actions that have interactions

1052
01:25:46,756 --> 01:25:55,656
with other cards on the board which then make the state become changed again and so it's a

1053
01:25:55,656 --> 01:26:01,016
constantly just changing state um and and in terms of bitcoin i've seen some really interesting ones

1054
01:26:01,016 --> 01:26:09,196
like people using a stack of magic the gathering cards is a is a mnemonic uh key and and so like

1055
01:26:09,196 --> 01:26:15,096
because these cards have unique names they just mapped uh magic the gathering card names to the

1056
01:26:15,096 --> 01:26:22,896
like bip 32 is that is that the right one uh language uh dictionary so there's really interesting

1057
01:26:22,896 --> 01:26:25,296
overlaps and interactions with this game.

1058
01:26:26,456 --> 01:26:28,376
But yeah, this is the type of stuff I want to see.

1059
01:26:29,116 --> 01:26:29,316
Yeah.

1060
01:26:29,476 --> 01:26:31,396
Magic the Gathering and then StarCraft.

1061
01:26:31,776 --> 01:26:32,456
The two.

1062
01:26:32,716 --> 01:26:34,556
The big two that I hear.

1063
01:26:35,796 --> 01:26:38,476
We need more gaming on Noster.

1064
01:26:38,636 --> 01:26:40,676
I'd love to battle Pokemon over Noster.

1065
01:26:40,976 --> 01:26:44,576
That's another vibe-coded app idea I kind of have in the back of my head.

1066
01:26:45,576 --> 01:26:49,036
Upload your Pokemon from your Game Boy into Noster and then battle them.

1067
01:26:50,736 --> 01:26:52,116
I think it could be a lot of fun.

1068
01:26:52,116 --> 01:26:56,296
So scan your real world card and it transitions to a digital form.

1069
01:26:56,556 --> 01:26:57,216
There you go.

1070
01:26:57,356 --> 01:26:57,636
NFT.

1071
01:27:01,636 --> 01:27:03,256
Where were we, Avi?

1072
01:27:03,416 --> 01:27:03,976
What happened?

1073
01:27:07,476 --> 01:27:09,876
So, yeah, anything else?

1074
01:27:10,336 --> 01:27:12,456
Any call to action for listeners?

1075
01:27:14,836 --> 01:27:15,676
Try it.

1076
01:27:15,676 --> 01:27:19,976
I mean, Shakespeare.diy, that's a good place to start.

1077
01:27:19,976 --> 01:27:24,196
if you want to try vibe coding something.

1078
01:27:24,436 --> 01:27:26,236
Stack CLI is another one.

1079
01:27:26,356 --> 01:27:27,276
If you're comfortable in the terminal,

1080
01:27:27,396 --> 01:27:28,776
that's what I would recommend right now.

1081
01:27:29,756 --> 01:27:31,096
And just build stuff.

1082
01:27:31,796 --> 01:27:36,296
I feel like we're in a new renaissance of the internet.

1083
01:27:36,956 --> 01:27:39,016
And what matters most right now is ideas.

1084
01:27:39,516 --> 01:27:41,536
And all of these vibe coded apps

1085
01:27:41,536 --> 01:27:42,696
that people have abandoned,

1086
01:27:42,976 --> 01:27:45,376
it's great that we now have the ability to do it.

1087
01:27:46,136 --> 01:27:49,756
It's just that they either weren't the best idea

1088
01:27:49,756 --> 01:27:52,456
or they weren't the best prompt of that idea.

1089
01:27:53,056 --> 01:27:59,596
So that's my call to action is let's make NOSTER win by trying our ideas.

1090
01:28:01,576 --> 01:28:07,856
And always give the NOSTER devs, the Shakespeare team feedback

1091
01:28:07,856 --> 01:28:09,476
because we're in building mode.

1092
01:28:10,036 --> 01:28:12,196
I think that's one of the biggest things in our NOSTER community

1093
01:28:12,196 --> 01:28:18,296
is always speak up if you want a feature and maybe it'll be heard.

1094
01:28:18,296 --> 01:28:20,996
So, well, thank you very much, Alex.

1095
01:28:20,996 --> 01:28:22,216
Thanks for coming on the show.

1096
01:28:22,276 --> 01:28:22,856
Thank you, guys.

1097
01:28:23,276 --> 01:28:25,216
Listener of your, the Gleasonator.

1098
01:28:25,976 --> 01:28:31,676
So it was nice to have a backseat on some of the AI talk that you and Avi shared.

1099
01:28:32,456 --> 01:28:33,796
And I appreciate you.

1100
01:28:34,036 --> 01:28:35,036
Thank you for your time.

1101
01:28:35,376 --> 01:28:36,476
And anything else, Avi?

1102
01:28:37,756 --> 01:28:39,436
No, thank you, Alex.

1103
01:28:39,556 --> 01:28:42,436
As always, good chatting with you.

1104
01:28:42,496 --> 01:28:47,016
I think we managed to avoid a lot of the topics that we discussed last time.

1105
01:28:47,016 --> 01:28:47,756
So this is good.

1106
01:28:48,296 --> 01:28:49,336
I agree.

1107
01:28:49,856 --> 01:28:50,856
Always moving, Bobby.

1108
01:28:51,436 --> 01:28:54,956
The technology is way beyond what you talked about last time.

1109
01:28:55,676 --> 01:28:58,436
Thank you, everyone, for tuning in to Pleb Chain Radio.

1110
01:28:58,736 --> 01:29:04,336
If you appreciate us as much as we appreciate you, listen on Fountain App, consider hitting the subscribe button.

1111
01:29:04,796 --> 01:29:06,876
Until next week, gentle plebs, farewell.

1112
01:29:08,176 --> 01:29:08,776
Goodbye.