1
00:00:00,000 --> 00:00:07,680
There are likely data sharing agreements between OpenAI and the US government.

2
00:00:07,940 --> 00:00:10,860
They want to harvest all the user data and they want to sell it and monetize it.

3
00:00:11,140 --> 00:00:14,220
We actually don't know what they're doing behind the scenes because everything's closed source.

4
00:00:14,480 --> 00:00:15,620
What vulnerabilities are there?

5
00:00:15,940 --> 00:00:23,320
What am I signing up for by giving them access to your mind effectively and then letting them into your digital life?

6
00:00:23,560 --> 00:00:26,420
It's this amazing potential for humanity, for human rights.

7
00:00:26,420 --> 00:00:35,080
That said, the more that we give ourselves over to it, the more that we turn our data over to it, our minds, everything, we're giving it power to influence us.

8
00:00:35,240 --> 00:00:42,580
We've taken a lot of that ethos of the Bitcoin mindset, the don't trust verify mindset that is only made possible because of Bitcoin.

9
00:00:42,780 --> 00:00:43,980
We can't do it with credit cards.

10
00:00:44,080 --> 00:00:49,120
We had to do it with something that was private and that was freedom oriented money that is uncensorable.

11
00:00:49,640 --> 00:00:55,780
AI has the ability to upgrade humanity, but we need to make sure that our humanity is preserved in the process.

12
00:00:56,420 --> 00:01:02,060
it's good to see you man thanks for coming on the show we've been trying to do this one in person

13
00:01:02,060 --> 00:01:06,120
i've not been in austin for a long time so we decided we just do remote you guys have just

14
00:01:06,120 --> 00:01:11,340
dropped some very cool new features um but your first time on the show you should introduce yourself

15
00:01:11,340 --> 00:01:16,400
tell everyone who you are sure yeah i'm a long-time listener first-time caller so this is great

16
00:01:16,400 --> 00:01:24,260
uh my name is mark i'm on online i go by marks a lot so you might see that name as well but i i've

17
00:01:24,260 --> 00:01:31,040
been around in the tech industry for a while. I live in Austin, Texas now. And yeah, about myself.

18
00:01:31,040 --> 00:01:37,740
So I work on a private AI called Maple AI. Prior to that, I was at Apple for six years working on,

19
00:01:37,840 --> 00:01:43,380
I was a software engineer over there working on an internal project that had a huge privacy

20
00:01:43,380 --> 00:01:49,420
and machine learning and AI component to it. Apple does care about privacy. And so that was like,

21
00:01:49,420 --> 00:01:56,620
from day one, I had to work on that aspect. But yeah, just loving life and glad to be here.

22
00:01:57,440 --> 00:02:02,800
No, I'm glad to have you on, man. Bitcoin's at $94,000. So this is an AI podcast now.

23
00:02:04,380 --> 00:02:08,500
These big tech companies are investing tens of billions, hundreds of billions of dollars

24
00:02:08,500 --> 00:02:12,240
into AI at the moment. They're all in this like arms race competing against each other.

25
00:02:12,720 --> 00:02:17,360
I want to know, like from your perspective, what's the end goal? Is it basically who can get to AGI

26
00:02:17,360 --> 00:02:23,060
first and whoever gets their first wins. Yeah. I mean, everybody talks about what's your moat,

27
00:02:23,340 --> 00:02:29,560
like what are you doing to get your competitive advantage? And so AGI is like this thing that

28
00:02:29,560 --> 00:02:33,940
they love to sell people on and talk about. It's really good for raising money. It's really good

29
00:02:33,940 --> 00:02:42,100
for driving adoption. It's anybody's guess how close that is really. But I think they're honestly

30
00:02:42,100 --> 00:02:47,640
just driving for who can have the stickiest product, who can get the most people in and

31
00:02:47,640 --> 00:02:55,060
keep them the longest. And then let's continue to upsell you. But the big part of their revenue

32
00:02:55,060 --> 00:02:58,840
model is the data, right? They want to get the data. Everybody talks about how data is the new

33
00:02:58,840 --> 00:03:03,780
oil in this life that we live right now. And so they're gathering all this information. They're

34
00:03:03,780 --> 00:03:08,400
making better models. They're monetizing the data. They're selling advertisements. They're selling

35
00:03:08,400 --> 00:03:12,120
shopping to you, they're building agents that will go out and purchase stuff for you.

36
00:03:12,440 --> 00:03:17,320
So really, it's just about how can we collect as much data so that we can build businesses

37
00:03:17,320 --> 00:03:18,160
off of that.

38
00:03:18,160 --> 00:03:29,200
And I mean, the big major AI shops like OpenAI, Anthropic, XAI, they all got a huge $200 million

39
00:03:29,200 --> 00:03:33,300
fund or investment from the government.

40
00:03:33,460 --> 00:03:37,260
I don't know if this is technically a grant or not, but from the United States Department

41
00:03:37,260 --> 00:03:37,760
of Defense.

42
00:03:37,760 --> 00:03:44,240
And so there are likely data sharing agreements between OpenAI and the U.S. government.

43
00:03:45,260 --> 00:03:47,100
That's just kind of reading between the lines there.

44
00:03:47,260 --> 00:03:50,900
So there's a lot of data gathering going on and then monetization of that data.

45
00:03:51,380 --> 00:03:55,540
You know, when you say like you don't know how close AGI, superintelligence, I think

46
00:03:55,540 --> 00:03:57,480
are those terms basically interchangeable at this point?

47
00:03:57,560 --> 00:03:59,700
Like you don't know how far that is away.

48
00:04:00,040 --> 00:04:03,280
Is this almost like quantum computing where it's just always a few more years?

49
00:04:03,420 --> 00:04:06,020
Or do you think we are actually on the brink of a breakthrough here?

50
00:04:06,020 --> 00:04:10,360
that's a good one um i feel like that's above my pay grade but

51
00:04:10,360 --> 00:04:18,540
it's it i think a lot of it depends on just like the task at hand i mean you've used ai a lot and

52
00:04:18,540 --> 00:04:23,860
sometimes it's really good at one specific thing and then you try to have it tie its shoes and it

53
00:04:23,860 --> 00:04:29,040
like totally falls over and trips on itself right like so i i personally think like it's very far

54
00:04:29,040 --> 00:04:35,820
away we're not right there yet that we are going to build very specialized ais to do things you

55
00:04:35,820 --> 00:04:39,800
know, Elon loves to show off his robot and say it's good at dancing and it's good at moving boxes

56
00:04:39,800 --> 00:04:45,340
in a warehouse and all this stuff. There was the robot that made the rounds a couple of weeks ago

57
00:04:45,340 --> 00:04:48,500
with all the memes of, you know, here's this robot you can buy and put in your house.

58
00:04:49,580 --> 00:04:54,100
But it was, it's not an actual product that's functional. I think that we have a long way to

59
00:04:54,100 --> 00:04:58,400
go still before we get the whole AGI thing and super intelligence. I think we're just going to

60
00:04:58,400 --> 00:05:04,180
be targeted intelligence for a long time. Okay. I mean, I saw that the videos that launched with

61
00:05:04,180 --> 00:05:07,860
that robot that was in your house and wasn't like it looked kind of brilliant in the videos.

62
00:05:07,860 --> 00:05:11,160
But is it true that that was actually driven by someone using like a VR headset?

63
00:05:11,160 --> 00:05:14,820
So there's just like some guy in a warehouse somewhere working away, looking at the inside

64
00:05:14,820 --> 00:05:15,340
of your house.

65
00:05:15,800 --> 00:05:16,080
Yes.

66
00:05:16,180 --> 00:05:16,420
Yeah.

67
00:05:16,500 --> 00:05:19,960
They're saying for the early prototypes, it's going to be somebody actually like wearing

68
00:05:19,960 --> 00:05:23,080
a VR suit that's driving your robot.

69
00:05:23,480 --> 00:05:24,980
Eventually, they want to get to where they're not.

70
00:05:25,160 --> 00:05:29,340
But that's not where it's at right now, which is really creepy.

71
00:05:30,160 --> 00:05:30,560
Exactly.

72
00:05:30,680 --> 00:05:32,540
This is the dystopia that everyone's scared of.

73
00:05:32,540 --> 00:05:38,960
um but so the reason i asked like how close we are to agi is because i did a show a couple months

74
00:05:38,960 --> 00:05:44,320
ago with a guy called roman yampolski i don't know if you listen to that one but he is um like the ai

75
00:05:44,320 --> 00:05:49,080
safety guy he i think he came up with the term ai safety and he's really trying to push back on all

76
00:05:49,080 --> 00:05:54,340
these big tech firms just carelessly investing to the point where they're throwing billions and

77
00:05:54,340 --> 00:05:58,220
billions of dollars at this thing trying to get agi i'm not really thinking of the ramifications

78
00:05:58,220 --> 00:06:05,540
of that. Do you think there is a risk that AI is almost so good that it's too disruptive too quickly,

79
00:06:05,940 --> 00:06:09,660
even if you take away the part of it going to kill all humans? Do you think it can replace

80
00:06:09,660 --> 00:06:16,320
90% of jobs within a decade sort of thing? Yeah, it's starting to replace some jobs,

81
00:06:16,320 --> 00:06:21,840
it seems like. We see a lot of headlines about jobs getting replaced, and I think some of those

82
00:06:21,840 --> 00:06:26,040
are people looking for a reason to blame

83
00:06:26,040 --> 00:06:28,800
when really they were probably a lot of malinvestments

84
00:06:28,800 --> 00:06:31,660
from 2021 timeframe when the money printer was,

85
00:06:31,920 --> 00:06:34,240
you know, we had 0% interest rates.

86
00:06:34,780 --> 00:06:38,300
So I think that there's a lot of unwinding of bad hires,

87
00:06:38,500 --> 00:06:39,920
not bad hires because they're bad people,

88
00:06:40,100 --> 00:06:42,200
but hires that shouldn't have happened financially.

89
00:06:42,900 --> 00:06:44,360
So I think we're seeing that right now

90
00:06:44,360 --> 00:06:45,400
and they're just saying, oh, it's AI,

91
00:06:45,520 --> 00:06:46,480
we're just gonna blame it on that.

92
00:06:46,960 --> 00:06:47,740
That's part of it.

93
00:06:48,000 --> 00:06:49,580
And then there are industries

94
00:06:49,580 --> 00:06:51,520
that are already starting to get disrupted

95
00:06:51,520 --> 00:06:58,000
in a way by AI. So transitions are always really hard. We've seen it throughout time with new

96
00:06:58,000 --> 00:07:03,120
technologies that come in. And there's like this period of many years where people have to find

97
00:07:03,120 --> 00:07:09,300
new work or decide to retire early. It's going to be difficult if it happens incredibly fast.

98
00:07:09,880 --> 00:07:16,600
And I have long been like my, the economic side of me does not align with something like a UBI,

99
00:07:16,600 --> 00:07:21,820
universal basic income, but it almost seems like we might need to have some kind of

100
00:07:22,300 --> 00:07:29,520
AI stipend or something like that, right? Where everybody gets some kind of income because they've

101
00:07:29,520 --> 00:07:33,740
been displaced by AI until we figure out what are the new jobs? What are the new industries? What

102
00:07:33,740 --> 00:07:36,280
are the new businesses going to be built up? Because that's kind of the pattern that always

103
00:07:36,280 --> 00:07:42,280
repeats. New technology comes in, new industries are birthed from that new technology. And we'll

104
00:07:42,280 --> 00:07:47,940
see that with AI. We just don't know what it is yet. And we need to have a good, happy civilization,

105
00:07:48,260 --> 00:07:53,720
no civil unrest if possible before we get there. Yeah, I totally agree with that. And that's like

106
00:07:53,720 --> 00:07:59,800
my biggest concern on the job displacement front is that there's some industries that are sort of,

107
00:07:59,860 --> 00:08:03,580
it's very clear to see the path to being completely replaced by AI. Obviously,

108
00:08:04,360 --> 00:08:09,960
software development has already changed entirely with AI, but even things like long distance

109
00:08:09,960 --> 00:08:14,180
truckers. That's an example I've used before on the show, but that job is not going to be there

110
00:08:14,180 --> 00:08:17,980
in 20 years guaranteed. It's not going to be there. Who knows if it's quicker than that?

111
00:08:18,640 --> 00:08:23,780
And what happens to all the people doing those jobs? And I can't see how you get around it without

112
00:08:23,780 --> 00:08:28,840
a UBI. I don't think retraining in another industry, if all the other industries are also

113
00:08:28,840 --> 00:08:33,980
getting disrupted and displaced by AI, that's not a feasible outcome. So how do you get there without

114
00:08:33,980 --> 00:08:39,240
some form of universal basic income? Yeah. And then what does that do to wealth inequality?

115
00:08:39,960 --> 00:08:46,480
Yeah, wealth and equality. I mean, that's, I don't know how related those are. Those maybe

116
00:08:46,480 --> 00:08:53,120
are very related, maybe they're not. But the hardest part with doing something like UBI

117
00:08:53,120 --> 00:08:58,660
with what we're talking about right now is once it's there, it's very difficult to unwind that,

118
00:08:58,1000 --> 00:09:02,940
right? There's no, the most permanent thing is a temporary government handout kind of thing.

119
00:09:02,940 --> 00:09:07,460
So once people start to depend on UBI, it's going to become part of their life.

120
00:09:07,460 --> 00:09:13,680
and so 30 years later it's like time to get rid of that and everybody's got their own jobs and

121
00:09:13,680 --> 00:09:18,420
new industries that UBI is just going to be a part of their income and they're going to depend on that

122
00:09:18,420 --> 00:09:25,400
so it's I don't want us to like just jump in and say yeah let's do this I think we need to really

123
00:09:25,400 --> 00:09:31,320
look long and hard at like what are the long-term ramifications of that and then as far as the wealth

124
00:09:31,320 --> 00:09:36,460
gap goes I think a lot of it still comes down to fix the money fix the world kind of stuff where we

125
00:09:36,460 --> 00:09:41,200
need to really fix the financial incentives behind everything in order to start to fix the

126
00:09:41,200 --> 00:09:46,900
wealth inequality that we see. And maybe AI helps that, right? Because it helps people that are on

127
00:09:46,900 --> 00:09:54,120
the lower part of the ladder to jump up higher and elevate up. Yeah. When you're in your work,

128
00:09:54,120 --> 00:10:00,060
I imagine you're using a lot of AI in the software development side. Has that replaced

129
00:10:00,060 --> 00:10:03,120
essentially new hires that you would have had to make otherwise?

130
00:10:03,120 --> 00:10:08,540
I want to be careful here. I don't want to create a soundbite or something, but yeah,

131
00:10:09,020 --> 00:10:16,400
we've become way more productive with having AI. So we do everything out in the open. So we build

132
00:10:16,400 --> 00:10:22,080
on GitHub and deploy on GitHub. And so you can see, well, not deploy on GitHub, but we put our

133
00:10:22,080 --> 00:10:28,100
stuff there. And what we do is we've got AI agents that sit there in our GitHub repo and we push up,

134
00:10:28,320 --> 00:10:32,120
I'll write like a whole feature spec. And then I'll just say, hey, will you build this? And it'll

135
00:10:32,120 --> 00:10:36,020
build it and then I can review it and tell it to make changes. And then we have two other AIs

136
00:10:36,020 --> 00:10:41,100
that are code reviewing that code. So we have three different agents all working on this code

137
00:10:41,100 --> 00:10:46,760
with me inspecting it. And that's for code that's not like super mission critical. You know, Anthony

138
00:10:46,760 --> 00:10:52,280
is the one really in there building stuff. And he does a lot of it locally first with AI and then

139
00:10:52,280 --> 00:10:58,640
pushes it to GitHub. But we are seeing that as a small company, we're doing a lot more with a two

140
00:10:58,640 --> 00:11:03,560
person team. Whereas we five years ago, three years ago, probably would have need to hire two

141
00:11:03,560 --> 00:11:08,520
more people by now in order to get to where we are at this point. So it's more that we're moving

142
00:11:08,520 --> 00:11:15,740
faster with two people. Or we could have, we would be like half as far as we are now, or even even

143
00:11:15,740 --> 00:11:21,220
less than that, if we didn't have AI. So to kind of answer your question, yeah, we could have hired

144
00:11:21,220 --> 00:11:25,800
some people to help us get here now. But instead, we're getting here faster. If you're already self

145
00:11:25,800 --> 00:11:30,400
custody of Bitcoin, you know the deal with hardware wallets. Complex setups, clumsy interfaces,

146
00:11:30,760 --> 00:11:36,080
and a seed phrase that can be lost, stolen, or forgotten. Well, BitKey fixes that. BitKey is a

147
00:11:36,080 --> 00:11:40,800
multi-sig hardware wallet built by the team behind Square and Cash App. It packs a cryptographic

148
00:11:40,800 --> 00:11:45,860
recovery system and built-in inheritance feature into an intuitive, easy-to-use wallet with no

149
00:11:45,860 --> 00:11:51,940
seed phrase to sweat over. It's simple, secure self-custody without the stress. And Time named

150
00:11:51,940 --> 00:11:58,920
Bitkey one of the best inventions of 2024. Get 20% off at bitkey.world when you use the code WBD.

151
00:11:59,440 --> 00:12:06,500
That's B-I-T-K-E-Y dot world and use the code WBD. This episode is brought to you by Anchor Watch.

152
00:12:06,800 --> 00:12:11,160
The thing that keeps me up at night is the idea of a critical error with my Bitcoin cold storage

153
00:12:11,160 --> 00:12:15,520
and this is where Anchor Watch comes in. With Anchor Watch your Bitcoin is insured with your

154
00:12:15,520 --> 00:12:20,800
own A-plus rated Lloyds of London insurance policy and all Bitcoin is held in their time-locked

155
00:12:20,800 --> 00:12:25,120
multi-sig vaults. So you have the peace of mind knowing your Bitcoin is insured while not giving

156
00:12:25,120 --> 00:12:29,440
up custody. So whether you're worried about inheritance planning, wrench attacks, natural

157
00:12:29,440 --> 00:12:34,700
disasters or just your own silly mistakes, you're protected by Anchor Watch. Rates for fully insured

158
00:12:34,700 --> 00:12:40,000
custody start as low as 0.55% and are available for individual and commercial customers located

159
00:12:40,000 --> 00:12:45,080
in the US. Speak to Anchor Watch for a quote and for more details about your security options and

160
00:12:45,080 --> 00:12:51,020
coverage. Visit anchorwatch.com today. That is anchorwatch.com. Do you wish you could access

161
00:12:51,020 --> 00:12:55,580
cash without selling your Bitcoin? Well, Ledin makes that possible. They're the global leader

162
00:12:55,580 --> 00:13:00,660
in Bitcoin-backed lending and since 2018, they've issued over $9 billion in loans with a perfect

163
00:13:00,660 --> 00:13:05,940
record of protecting client assets. With Ledin, you get full custody loans with no credit checks

164
00:13:05,940 --> 00:13:12,400
or monthly repayments, just easy access to dollars without selling a single sat. As of July 1st,

165
00:13:12,400 --> 00:13:17,080
Ledin is Bitcoin only meaning they exclusively offer Bitcoin-backed loans with all collateral

166
00:13:17,080 --> 00:13:22,620
held by Ledin directly or their funding partners. Your Bitcoin is never lent out to generate interest.

167
00:13:23,360 --> 00:13:27,780
I recently took out a loan with Ledin. The whole process was super easy. The application took me

168
00:13:27,780 --> 00:13:32,340
less than 15 minutes and in a few hours I had the dollars in my account. It was really smooth.

169
00:13:32,980 --> 00:13:38,380
So if you need cash but you don't want to sell Bitcoin, head over to ledin.io forward slash WBD

170
00:13:38,380 --> 00:13:41,200
and you'll get 0.25% off your first loan.

171
00:13:41,700 --> 00:13:44,420
That's leaden.io forward slash WBD.

172
00:13:45,140 --> 00:13:49,340
So like, obviously you're building an LM with a Maple.

173
00:13:49,920 --> 00:13:52,040
You're competing against the ChatGPTs,

174
00:13:52,140 --> 00:13:53,040
the Anthropics of the world.

175
00:13:53,400 --> 00:13:56,360
Like how competitive can you be against them?

176
00:13:56,520 --> 00:13:58,860
I'm sure they have hundreds of thousands of developers.

177
00:13:59,960 --> 00:14:03,000
How competitive can you make your product with two?

178
00:14:04,040 --> 00:14:04,300
Yeah.

179
00:14:04,300 --> 00:14:20,160
Yeah. Well, we're following in the path that others like Signal have shown where you look for the user experience that the expensive companies who have these large, massive teams and spend all the money on user research, spend all the money on design and everything.

180
00:14:20,340 --> 00:14:29,080
They figure out what works really well. And then you have the luxury of coming in and saying, all right, talk to all these users. What features do you like the most? And which ones do we need to focus on?

181
00:14:29,080 --> 00:14:32,800
So then you go and build something that's similar, but has your own unique flavor.

182
00:14:32,920 --> 00:14:35,400
And our unique flavor is that we care about the user.

183
00:14:35,560 --> 00:14:36,740
We care about their data privacy.

184
00:14:37,300 --> 00:14:38,300
We don't track them.

185
00:14:38,360 --> 00:14:39,660
We're not spying on that kind of stuff.

186
00:14:40,120 --> 00:14:45,780
And so we can build something that is functionally will look identical to ChatGPT.

187
00:14:45,860 --> 00:14:50,100
It's going to have pretty much all the same features, maybe 95% of the feature set that

188
00:14:50,100 --> 00:14:50,500
you would want.

189
00:14:51,260 --> 00:14:53,760
But then we have the thing that they don't.

190
00:14:53,980 --> 00:14:58,180
And that is, you know, they want to harvest all the user data and they want to sell it

191
00:14:58,180 --> 00:14:58,820
and monetize it.

192
00:14:59,080 --> 00:15:21,500
So in that regard, I think we can do a really good job with two people. We'd love to hire a few more and catch up and get really close to that. And we know that we're obviously not going to like take down ChatGPT and take them over, but we can get really far and we can build a product that millions of people, hundreds of millions of people find incredibly useful.

193
00:15:21,500 --> 00:15:26,140
I remember this was probably a year or two ago.

194
00:15:26,580 --> 00:15:27,580
I think it was from Google.

195
00:15:27,700 --> 00:15:29,580
There was a sort of internal memo that was leaked,

196
00:15:29,640 --> 00:15:31,960
which was essentially saying we've got no moat

197
00:15:31,960 --> 00:15:35,420
and these open source AI models are going to be

198
00:15:35,420 --> 00:15:37,360
just as competitive as us.

199
00:15:38,500 --> 00:15:39,800
Where does that stand?

200
00:15:39,920 --> 00:15:41,960
Because even though there's obviously great progress

201
00:15:41,960 --> 00:15:43,000
on some of the open source models,

202
00:15:43,200 --> 00:15:45,240
I know Llama is open source at the moment

203
00:15:45,240 --> 00:15:46,920
and that's what Facebook are using.

204
00:15:46,960 --> 00:15:47,420
Is that correct?

205
00:15:47,420 --> 00:15:54,560
how close are they to the close sourced chat gpt's of the world yeah it depends on which

206
00:15:54,560 --> 00:16:00,440
benchmarks you look at or if you go off of your own vibes and really it sounds silly but you almost

207
00:16:00,440 --> 00:16:05,100
have to just try it out with the the task that you want to do and with the process that you want

208
00:16:05,100 --> 00:16:09,540
to follow and test the different models to see what works best for you but when you look at

209
00:16:09,540 --> 00:16:15,480
straight benchmarks they've really caught up a lot on on coding standards on math standards on

210
00:16:15,480 --> 00:16:19,820
And all the different benchmarks are out there, especially the Chinese models.

211
00:16:20,040 --> 00:16:21,580
Like Llama is still pretty far behind.

212
00:16:21,900 --> 00:16:26,640
Meta, I imagine Meta is cooking up something for Llama 5 that's going to be really big

213
00:16:26,640 --> 00:16:29,280
because they have so much data that nobody else has.

214
00:16:29,340 --> 00:16:31,900
They have all of the WhatsApp data and the Facebook data and Instagram.

215
00:16:32,420 --> 00:16:34,220
So they're probably making something.

216
00:16:34,840 --> 00:16:38,800
But in their absence, the Chinese models have really come in and caught up.

217
00:16:39,060 --> 00:16:42,320
But I was chatting with a founder who's here in Austin.

218
00:16:42,560 --> 00:16:44,300
He's building an AI service as well.

219
00:16:44,400 --> 00:16:45,260
His is more enterprise.

220
00:16:45,480 --> 00:16:48,640
But he said that internally, they kind of measure all the different tools.

221
00:16:48,840 --> 00:16:53,820
And he finds that the Chinese models really try to fit to the benchmarks.

222
00:16:54,100 --> 00:16:57,300
So they work really hard to make sure that they score high on the benchmarks.

223
00:16:57,660 --> 00:17:02,220
But then if you stray out of the lane at all of those benchmarks, then they might start to fall down.

224
00:17:02,660 --> 00:17:04,820
That's specifically for programming.

225
00:17:05,760 --> 00:17:07,860
So certain programming languages or something.

226
00:17:08,580 --> 00:17:14,220
But that said, like, they do perform really well and they keep getting better.

227
00:17:14,220 --> 00:17:16,520
So I'm hopeful on open source.

228
00:17:17,280 --> 00:17:28,640
And like the Google memo said, it's just a matter of time before they are good enough for the average person and the average business user that they don't need to pay for these proprietary models anymore.

229
00:17:29,500 --> 00:17:35,480
Why do you think it is that DeepSeek and these Chinese models have gone the open source route when the American companies have gone closed source?

230
00:17:35,620 --> 00:17:37,640
Like that seems backwards to me.

231
00:17:37,800 --> 00:17:43,680
Yeah, I wonder if they realize that Americans wouldn't use it if it was fully closed source Chinese.

232
00:17:44,220 --> 00:17:46,580
And they know that they need to compete somehow.

233
00:17:46,580 --> 00:17:53,980
And so the world's only going to listen to them and use their stuff if it's out there for free and open source.

234
00:17:54,420 --> 00:18:02,520
And then the other part of it, too, is open source is going to get adopted way more than the proprietary ones by hobbyists and by others.

235
00:18:03,300 --> 00:18:12,920
And so if you have an ideology that you want to seed out into the world, especially if you're looking at like a global south where maybe they can't afford to use the proprietary models,

236
00:18:12,920 --> 00:18:17,960
then you can embed your ideology in this model and then push it out to the world.

237
00:18:18,440 --> 00:18:21,740
So I could see a couple of different reasons why they would want to go the open source route.

238
00:18:22,860 --> 00:18:26,240
And DeepSeek, even though it's open source, they're still collecting data, correct?

239
00:18:27,460 --> 00:18:32,860
Sort of. So a couple of clarifications. Open source with models is a little different.

240
00:18:33,240 --> 00:18:38,800
They're more open models, if you will. We can't fully see the data that went into them,

241
00:18:38,800 --> 00:18:43,800
but we can see the weights and the measures and the biases and you can dial them yourself, that kind of stuff.

242
00:18:43,800 --> 00:18:49,800
So it's a little different than open source code. And then as far as data sharing goes,

243
00:18:49,800 --> 00:19:04,210
the only time you sharing data with DeepSeek is if you download the DeepSeek app or you go to like the official DeepSeek website and use the AI that hosted by them then yes they see your data they see all your chats and there heavy suspicion

244
00:19:04,210 --> 00:19:08,090
that the CCP is able to access

245
00:19:08,090 --> 00:19:08,990
all of that information,

246
00:19:09,450 --> 00:19:12,270
mostly based off of data arrangements

247
00:19:12,270 --> 00:19:14,070
that pretty much every other company in China

248
00:19:14,070 --> 00:19:15,610
has with the government there.

249
00:19:16,210 --> 00:19:18,070
That said, if anybody's running

250
00:19:18,070 --> 00:19:20,530
the DeepSeek models locally on their laptop

251
00:19:20,530 --> 00:19:23,730
or they're running in something like Maple or some other system,

252
00:19:23,850 --> 00:19:27,790
then no, there's zero data sharing going back to DeepSeek as an organization

253
00:19:27,790 --> 00:19:29,510
or to any kind of Chinese government.

254
00:19:30,270 --> 00:19:33,890
You're obviously at Maple not collecting any customer data at all.

255
00:19:34,490 --> 00:19:36,790
If, as you said earlier, data is the new oil,

256
00:19:36,930 --> 00:19:39,050
like what are you forgoing that?

257
00:19:39,150 --> 00:19:42,670
Like why are all these other companies just so desperate to harvest

258
00:19:42,670 --> 00:19:44,870
as much as that's possible and you're willing to just say,

259
00:19:44,870 --> 00:19:45,730
no, we don't need it?

260
00:19:45,730 --> 00:19:55,750
I think it's because we've all been sold that this internet that we use has to be monetized by selling your data.

261
00:19:56,170 --> 00:20:01,650
Like us as users got so used to using Gmail because it was the most amazing email service ever.

262
00:20:02,010 --> 00:20:08,850
It conditioned us to say we should have email for free because prior to Gmail coming on the scene, we were all paying for email.

263
00:20:08,970 --> 00:20:12,250
In fact, my dad was paying for his email inbox.

264
00:20:12,250 --> 00:20:17,050
Even when he stopped using it, he's still paying for it like five or 10 bucks a month because it was just it was a hassle to cancel.

265
00:20:17,450 --> 00:20:21,870
Right. And so when Gmail came out, they were like, hey, here's this new business model.

266
00:20:21,950 --> 00:20:27,530
You get it for free. But what we didn't realize is it came with this huge cost of all of our data being monetized.

267
00:20:28,190 --> 00:20:31,510
And we've just kind of gone down that path and we don't need to.

268
00:20:31,510 --> 00:20:38,430
Like there are other ways to build sustainable companies and sustainable products that don't use that as their business model.

269
00:20:38,570 --> 00:20:39,930
So that's really what we're doing.

270
00:20:39,930 --> 00:20:45,650
we are we're trying the the more healthy route if you will healthy for humanity healthy for all of us

271
00:20:45,650 --> 00:20:51,270
to build it in a different way where we sell you a really great user experience and we sell you a

272
00:20:51,270 --> 00:20:56,790
product and you can use it and that's really where the relationship ends so how before we get into

273
00:20:56,790 --> 00:21:03,890
like how you're doing things at maple um how are these other big ai companies what are they doing

274
00:21:03,890 --> 00:21:07,350
with the data that you're putting into it and are they are they using literally every single word

275
00:21:07,350 --> 00:21:11,410
you put into these models and then storing that, creating profiles about you.

276
00:21:11,490 --> 00:21:13,190
Like, how do they actually use that data?

277
00:21:13,750 --> 00:21:18,810
Yeah, well, so they're using all the information you input into it.

278
00:21:19,590 --> 00:21:23,990
They are, they're also using everything you don't put into it.

279
00:21:24,030 --> 00:21:24,690
And what does that mean?

280
00:21:24,770 --> 00:21:29,430
There has been evidence and research showing that they look at your keystrokes.

281
00:21:29,430 --> 00:21:33,130
So if you're typing something into the box and then you hit delete a bunch of times because

282
00:21:33,130 --> 00:21:35,170
you change your mind, they've captured that.

283
00:21:35,170 --> 00:21:40,510
And so they know, okay, here's how Danny thinks. Danny typed all this stuff in. Maybe he was like

284
00:21:40,510 --> 00:21:43,710
really angry and writing this really angry thing. And then he's like, you know, I need to tone it

285
00:21:43,710 --> 00:21:48,290
down a little bit. So you backed off. It's learning your emotional state. It's learning

286
00:21:48,290 --> 00:21:54,010
your entire thought process. They're storing that all in their system. And then the way that I love

287
00:21:54,010 --> 00:21:58,850
to describe it is that they are like, it's like you hired someone to write a biography on you.

288
00:21:58,950 --> 00:22:02,750
So they're like a world-class author. They sit down and they're just constantly interviewing

289
00:22:02,750 --> 00:22:06,810
all day long, but they're also paying attention to your body language. They're paying attention

290
00:22:06,810 --> 00:22:11,030
to your heart rate, all these like other indicators that you don't realize you're giving off.

291
00:22:11,310 --> 00:22:16,530
And then they're creating this profile about you. And then they can, they can pump that into the

292
00:22:16,530 --> 00:22:22,230
system for you to make the AI understand you more, which is great. That's, that's the end product,

293
00:22:22,290 --> 00:22:26,630
right? They're like, Hey, an AI that knows you, it's very effective. But then what they're also

294
00:22:26,630 --> 00:22:31,870
doing is they're using all that to train new models, to create shopping networks for you.

295
00:22:31,870 --> 00:22:35,550
They're building these computer use tools that will be able to control your computer.

296
00:22:36,250 --> 00:22:40,090
And they're making web browsers now that are going to browse the internet for you.

297
00:22:40,230 --> 00:22:43,350
So you can see how they are just getting intertwined into your life.

298
00:22:44,050 --> 00:22:47,670
And so you have to ask, like, what vulnerabilities are there?

299
00:22:47,810 --> 00:22:56,130
Or what am I signing up for by giving them access to your mind effectively and then letting them into your digital life?

300
00:22:56,570 --> 00:23:00,310
Maybe if we just imagine for a minute, Maple never existed.

301
00:23:00,310 --> 00:23:04,530
the other people that are working on privacy i know proton have come out with a private ai model

302
00:23:04,530 --> 00:23:12,130
um imagine they never existed like how does this get dystopian from here um yeah because you see

303
00:23:12,130 --> 00:23:17,550
these things like i saw that friend necklace that came out um which by the way look like one of the

304
00:23:17,550 --> 00:23:21,230
worst products i've ever seen i can't believe they actually launched with that but like this

305
00:23:21,230 --> 00:23:24,890
these are things that literally just follow you around all day looking at everything you're looking

306
00:23:24,890 --> 00:23:31,270
at? Like what's the dystopian end game there? Yeah. Well, there's a dystopian end game. I would

307
00:23:31,270 --> 00:23:35,050
love to paint kind of the rosy picture real briefly first. The reason why we get there

308
00:23:35,050 --> 00:23:38,630
is because I think a lot of times we look at this dystopian thing and we're like,

309
00:23:38,930 --> 00:23:43,050
man, we're all a bunch of idiots. Why did we sign up for that? But it's because AI has this like

310
00:23:43,050 --> 00:23:47,990
huge, amazing potential, right? It's this amazing potential for humanity, for human rights. Even

311
00:23:47,990 --> 00:23:53,390
you have people who are oppressed all over the world and now they can grab the world's knowledge

312
00:23:53,390 --> 00:23:58,190
and use it for their own advantage to try and fight back against people who are oppressing them.

313
00:23:58,530 --> 00:24:00,130
So there's really cool things you can do with it.

314
00:24:00,710 --> 00:24:05,710
That said, the more that we give ourselves over to it, the more that we turn our data over to it,

315
00:24:05,730 --> 00:24:09,290
our minds, everything, we're giving it power to influence us.

316
00:24:09,830 --> 00:24:16,610
And so the dystopian side of it is that if we start giving it access to see in our room,

317
00:24:16,610 --> 00:24:18,990
to hear what we're talking about,

318
00:24:19,950 --> 00:24:25,690
it understands how to persuade us of things.

319
00:24:25,770 --> 00:24:26,390
Let's just say that.

320
00:24:26,530 --> 00:24:28,270
So it knows that maybe, Danny,

321
00:24:28,330 --> 00:24:30,010
you're really gullible in a certain way.

322
00:24:30,450 --> 00:24:31,710
And so if it wants to pass off

323
00:24:31,710 --> 00:24:34,670
some misinformation to you or a lie,

324
00:24:34,790 --> 00:24:36,590
it knows how to sell you on that.

325
00:24:37,010 --> 00:24:39,070
And so you can see that effectively

326
00:24:39,070 --> 00:24:40,250
they're building the system

327
00:24:40,250 --> 00:24:41,870
where somebody could come in

328
00:24:41,870 --> 00:24:42,830
with the right amount of money

329
00:24:42,830 --> 00:24:46,150
or the right amount of weapons, basically,

330
00:24:46,150 --> 00:24:54,130
and coerce them and say, we need the community to start thinking about a certain political thing

331
00:24:54,130 --> 00:25:00,450
in this direction. So we want to deploy this directive that is going to shift the mindset

332
00:25:00,450 --> 00:25:10,030
of this country and the general populace in a certain way. And if you think about how we used

333
00:25:10,030 --> 00:25:15,990
to do, let's see, think about advertising. Let's kind of look at it that way. If you want to,

334
00:25:16,150 --> 00:25:19,390
make a new product and you want to sell it to a bunch of people.

335
00:25:19,910 --> 00:25:23,870
Maybe you make a 30 second advertisement and you put it on something like the Super Bowl,

336
00:25:24,110 --> 00:25:25,930
but you don't know who's actually watching.

337
00:25:26,170 --> 00:25:27,570
You don't know what this frame of mind is.

338
00:25:27,610 --> 00:25:30,290
You don't know whether they're male, female, a child, an adult, whatever.

339
00:25:30,390 --> 00:25:32,890
You just make your best guess based off of demographic research.

340
00:25:33,550 --> 00:25:37,150
And so you have to try to come up with like the 30 seconds that's going to sell the most

341
00:25:37,150 --> 00:25:38,370
number of people on your product.

342
00:25:38,950 --> 00:25:43,390
Now you fast forward to this time where we all have AI that's harvesting all our data

343
00:25:43,390 --> 00:25:45,010
and understands everything about us.

344
00:25:45,010 --> 00:26:05,790
And now you can say, I don't want to make a 30 second ad that tries to capture 30% of the people that watch it. I want to capture 99.9% of the people. And so you can deploy something to this AI system that knows how to talk to you to sell you on a product and then talk to me in my way to sell me on the exact same product and convince most of us to use it.

345
00:26:05,790 --> 00:26:10,910
and that's just for products that's not governments that's not you know there's all

346
00:26:10,910 --> 00:26:15,770
sorts of ways that could be used to kind of weaponize the system for lack of a better word

347
00:26:15,770 --> 00:26:22,450
it's um a really scary future that seems very like it's very easy to see that coming to the

348
00:26:22,450 --> 00:26:28,450
like the world in the next maybe like three four five years um and like thank god we've got things

349
00:26:28,450 --> 00:26:34,690
like maple and proton doing this but just quickly before we get more deeply into into um maple

350
00:26:34,690 --> 00:26:41,890
are these models actually getting better because like i use main like i use ai quite a lot for work

351
00:26:41,890 --> 00:26:46,970
it helps a lot like it probably i would have to hire someone at least like 20 hours a week to

352
00:26:46,970 --> 00:26:53,030
replace what ai is doing for me currently um but every time chat gpt which is the one i use most

353
00:26:53,030 --> 00:26:58,070
like comes out with an update it doesn't seem to be better in fact sometimes like it's worse i i

354
00:26:58,070 --> 00:27:03,750
think 4.0 was the best one that they've done so far so like how like how much better are these

355
00:27:03,750 --> 00:27:10,470
getting like incrementally yeah it's it's it's up in the air it depends on it's it kind of goes

356
00:27:10,470 --> 00:27:15,790
off vibes like are they really getting better a lot of people look at chat 25 and think that really

357
00:27:15,790 --> 00:27:21,630
it's just 4-0 under the hood with some modifications around it and it was less of a huge upgrade

358
00:27:21,630 --> 00:27:27,870
but then you have xai you have grok which from you know from two to three and three to four was

359
00:27:27,870 --> 00:27:34,630
was a really big jump. So it is possible there's still gains to be made there. But a lot of people

360
00:27:34,630 --> 00:27:42,150
that I read online who are really deep into this, it seems like they're plateauing. And maybe we're

361
00:27:42,150 --> 00:27:47,210
plateauing because they're working on the next big major breakthrough and they just haven't got

362
00:27:47,210 --> 00:27:52,590
to it yet. So they're holding us over with these small bumps until we get there. But that's why I

363
00:27:52,590 --> 00:27:56,210
think that the open source is really going to be able to catch up. Because if it's true that these

364
00:27:56,210 --> 00:28:01,070
big models are starting to plateau, then open source is going to get just right up against them.

365
00:28:01,250 --> 00:28:06,650
And now we can have the same things that they have, but use it in a way that's better for us.

366
00:28:07,890 --> 00:28:12,270
Yeah. Okay. So let's get into Maple. First of all, explain exactly what you're doing,

367
00:28:12,450 --> 00:28:17,550
how you're making sure this is like private AI that's not harvesting data. Give us the pitch.

368
00:28:17,990 --> 00:28:23,950
Yeah, sure. Yeah. Maple is the alternative, the chat GPT that is not harvesting your data,

369
00:28:23,950 --> 00:28:26,270
that is protecting your privacy.

370
00:28:26,670 --> 00:28:29,530
The way that we've built it is we have,

371
00:28:30,130 --> 00:28:31,630
we built it around open models

372
00:28:31,630 --> 00:28:32,990
and all of our code is open source.

373
00:28:33,090 --> 00:28:34,270
So you can go look at it and see,

374
00:28:34,690 --> 00:28:36,390
and we're running them in the cloud

375
00:28:36,390 --> 00:28:38,030
using something called secure enclaves.

376
00:28:38,670 --> 00:28:40,610
Another term for that is confidential computing.

377
00:28:41,030 --> 00:28:43,290
But these are servers that have hardware encryption

378
00:28:43,290 --> 00:28:44,010
built into them.

379
00:28:44,270 --> 00:28:46,070
It's the same stuff that runs on your phone.

380
00:28:46,310 --> 00:28:50,310
So on Apple devices and Samsung and other devices,

381
00:28:50,310 --> 00:28:53,310
they have these secure enclaves where it stores your wallet,

382
00:28:53,310 --> 00:28:55,310
It stores your face ID, that kind of stuff.

383
00:28:55,710 --> 00:28:59,350
And it's these hardware encrypted things that are difficult to penetrate.

384
00:29:00,090 --> 00:29:02,250
And so in the cloud, we have those now.

385
00:29:02,470 --> 00:29:03,830
And so we're able to put Maple there.

386
00:29:04,170 --> 00:29:08,570
And when you as a user log in, we create a private encryption key just for your user.

387
00:29:09,230 --> 00:29:13,090
And so as you're chatting with the AI, it encrypts everything locally on your device

388
00:29:13,090 --> 00:29:14,030
using that private key.

389
00:29:14,290 --> 00:29:16,170
And then it sends it to the cloud.

390
00:29:16,230 --> 00:29:19,050
And then the cloud in the enclave is where the AI is sitting.

391
00:29:19,510 --> 00:29:21,310
And so it's effectively like you and me right now.

392
00:29:21,310 --> 00:29:26,090
we're having a one-on-one conversation in a private room that we're going to give to everybody.

393
00:29:26,370 --> 00:29:31,150
But right now we're having a private conversation. And that's really what the AI is doing in Maple

394
00:29:31,150 --> 00:29:35,850
and the Secure Enclave. And then once it's done chatting and working on your stuff, then it

395
00:29:35,850 --> 00:29:40,270
re-encrypts it and sends it back down to your device. And then we take it a step further than

396
00:29:40,270 --> 00:29:45,030
some other private AIs do. And that is we can synchronize it to all of your devices. So you

397
00:29:45,030 --> 00:29:48,770
can have it on your phone, you can have Maple app there, you can have it on your laptop,

398
00:29:48,770 --> 00:29:54,170
wherever you want to be. And then because we have that secure enclave and it knows how to handle

399
00:29:54,170 --> 00:29:58,130
your private encryption key, it can synchronize everything across all your devices for you.

400
00:29:58,670 --> 00:30:04,890
So in a nutshell, that's what Maple's doing is just using a private key. And the last thing I

401
00:30:04,890 --> 00:30:10,170
love to kind of tell people and explain is a lot of these services that you use in the cloud,

402
00:30:10,550 --> 00:30:15,810
they take all the user data and they stick it in one giant database. And if you are an employee at

403
00:30:15,810 --> 00:30:19,750
a company who has elevated privileges, you can just go in and hop around and everybody's user

404
00:30:19,750 --> 00:30:24,290
data all you want to. You can go look at it. Usually they have audit trails. And so they'll

405
00:30:24,290 --> 00:30:27,650
know that you went and accessed it, but that doesn't prevent you from accessing it. And then

406
00:30:27,650 --> 00:30:31,170
if a hacker gets in the system, well, they don't care about audit trails. So they're just going to

407
00:30:31,170 --> 00:30:35,050
get the whole database and get a data dump and everything. We've totally flipped that on its

408
00:30:35,050 --> 00:30:40,030
head. And with these private encryption keys, our backend is just a bunch of private vaults per user.

409
00:30:40,510 --> 00:30:44,090
And so if anybody were getting to our system, they wouldn't be able to look at anybody except

410
00:30:44,090 --> 00:30:47,630
their own vault that they can get in there, but they can't see anybody else.

411
00:30:48,470 --> 00:30:53,630
So like for me personally, as I said before, like I use ChatGPT the most and probably some

412
00:30:53,630 --> 00:30:55,050
of that is just down to habit.

413
00:30:55,170 --> 00:30:58,510
Like it's just, it's the first one I started using and it's been hard to kind of move away

414
00:30:58,510 --> 00:30:58,910
from that.

415
00:30:59,290 --> 00:31:02,710
But I have been, like I signed up for Maple basically as soon as you guys launch and I

416
00:31:02,710 --> 00:31:04,050
have been using it more and more.

417
00:31:04,410 --> 00:31:08,490
But the thing that I always use it for is if I'm ever putting like business data, like

418
00:31:08,490 --> 00:31:12,390
financial data, it's always my go-to because I know that that's like an actual secure place

419
00:31:12,390 --> 00:31:18,570
to put that rather than giving it to open AI. But in terms of like feature parity compared to these

420
00:31:18,570 --> 00:31:25,530
big AI LLMs, where are you at? Like, what do you have that you'd expect in a chat GPT type thing?

421
00:31:26,150 --> 00:31:30,590
Yeah. Well, I will tell you, you might be happy to hear that if your favorite model is 4.0,

422
00:31:30,770 --> 00:31:36,490
then we have the GPT OSS model inside of Maple. If you go in there and do the model selector,

423
00:31:36,590 --> 00:31:41,530
it's called Quick is the name of it. But that is really similar to 4.0. In fact, if you ask it,

424
00:31:41,530 --> 00:31:46,430
hey, what model are you? It'll tell you it's chat GPT 4.0. So you'll get that experience,

425
00:31:46,550 --> 00:31:52,130
which is nice. As far as features go, we let you upload documents to it. You can upload photos and

426
00:31:52,130 --> 00:31:56,590
get photo analysis. You can take a picture of a tree or a plant and say, what is this?

427
00:31:57,230 --> 00:32:01,830
You can upload financial documents or legal contracts and have it talk to you about what

428
00:32:01,830 --> 00:32:06,790
are the legal terms that you've agreed to. We have voice, so you can talk to Maple,

429
00:32:06,790 --> 00:32:12,030
which I use all the time, hit the microphone. We had it working where it would talk back to you

430
00:32:12,030 --> 00:32:15,910
that is temporarily broken. We're working on fixing that because we really loved having this

431
00:32:15,910 --> 00:32:20,690
two-way conversation. I would, when it was working, I would just kind of like walk around and just have

432
00:32:20,690 --> 00:32:27,430
a conversation with the AI, which I know a lot of people do with ChattoPT. So those are a lot of

433
00:32:27,430 --> 00:32:32,010
them. And then the biggest one is what everybody was waiting for, and that is live data. So now we

434
00:32:32,010 --> 00:32:36,710
have the ability to do private web search. So now Maple is no longer stuck with these models that

435
00:32:36,710 --> 00:32:41,650
were trained on data from a year ago or two years ago. Now you can be, you know, sitting there and

436
00:32:41,650 --> 00:32:45,710
say, hey, what's the what's the score of my favorite sport team, you know, game that I'm

437
00:32:45,710 --> 00:32:49,850
watching right now or that I'm curious about? It'll look it up and it'll fetch it for you and

438
00:32:49,850 --> 00:32:54,770
give it to you. But there's obviously a lot more utility to that than just sports. But yeah, being

439
00:32:54,770 --> 00:32:58,510
able to get the latest information from the web that is now available inside of Maple.

440
00:32:59,410 --> 00:33:03,910
That's a huge one for me because I use it a lot when I'm preparing for shows and stuff. I'll try

441
00:33:03,910 --> 00:33:08,590
and get like current relevant information I can use in the show. So without that, like if this was

442
00:33:08,590 --> 00:33:13,330
a model that was trained on data, it's like mid 24 or whatever. It's just, it is useless in that

443
00:33:13,330 --> 00:33:18,490
sense. So that's a huge one for me. And how do you do the private web search? How do you do that

444
00:33:18,490 --> 00:33:26,410
while not giving up any data? Yeah. So we're using, we're using brave API, the brave search API.

445
00:33:26,650 --> 00:33:31,250
They're, I mean, they're a privacy oriented company as well, but then we anonymize it. So

446
00:33:31,250 --> 00:33:35,770
when you are going to search, we don't attach your user ID or anything to the search and give

447
00:33:35,770 --> 00:33:40,110
it to Brave. So we have very little information already about our users, right? We don't collect

448
00:33:40,110 --> 00:33:46,050
names. We don't collect, you know, phone numbers or anything like that. The most we collect is an

449
00:33:46,050 --> 00:33:51,170
email address. And then we also know like what time you did your chats, because we have to keep

450
00:33:51,170 --> 00:33:56,070
track of like, just when chats happen, so we can synchronize them. And then we keep track of how

451
00:33:56,070 --> 00:34:00,210
much compute resources you used. But we don't know anything about what you're chatting about.

452
00:34:00,210 --> 00:34:07,610
so when we go to do the brave search we pass it along to them and um so they all they know is that

453
00:34:07,610 --> 00:34:13,310
there's like this giant fire hose of web searches coming in from this one account called maple but

454
00:34:13,310 --> 00:34:19,010
there's zero way for them to to you know tie it to anybody unless you literally say my name is

455
00:34:19,010 --> 00:34:22,610
daniel knowles blah blah and you put it in the search then like the brave search api will see

456
00:34:22,610 --> 00:34:27,750
your name because you put it in the content um but other than that like there's it's it's fully

457
00:34:27,750 --> 00:34:33,070
private. And there are some other private web searches that we're looking at as well, web

458
00:34:33,070 --> 00:34:37,490
services that we're looking at. And we would love to have this model where we can actually spread it

459
00:34:37,490 --> 00:34:44,630
across so people can get even more anonymity by getting lost in a bigger crowd than just one

460
00:34:44,630 --> 00:34:49,970
crowd with Brave. This episode is brought to you by River, and they've just launched a very cool

461
00:34:49,970 --> 00:34:55,490
new product where you can automatically buy every price dip. Their zero fee recurring buys are a

462
00:34:55,490 --> 00:35:00,330
proven way to build wealth with Bitcoin and you can now supercharge them and buy up to 100% more

463
00:35:00,330 --> 00:35:05,470
Bitcoin if the price is dipping at the time of your order. It's a great tool to buy more Bitcoin

464
00:35:05,470 --> 00:35:11,430
at lower prices and while your cash is waiting to be deployed you earn 3.75% interest on it paid in

465
00:35:11,430 --> 00:35:16,730
Bitcoin. River is built to last with security at its core and is the only Bitcoin exchange in the

466
00:35:16,730 --> 00:35:23,210
US with proof of reserves. To open an account go to river.com forward slash wbd and earn up to $100

467
00:35:23,210 --> 00:35:29,710
in Bitcoin when you buy. That's river.com forward slash wbd. What if you could lower your tax bill

468
00:35:29,710 --> 00:35:35,730
and stack Bitcoin at the same time? Well, by mining Bitcoin with Blockware, you can. New tax

469
00:35:35,730 --> 00:35:39,970
guidelines from the Big Beautiful Bill allow American miners to write off 100% of the cost

470
00:35:39,970 --> 00:35:46,010
of their mining hardware in a single tax year. That's right, 100% write-off. So if you have $100,000

471
00:35:46,010 --> 00:35:51,030
in capital gains or income, you can purchase $100,000 of miners and offset it entirely.

472
00:35:51,830 --> 00:35:56,370
Blockware's mining as a service enables you to start mining Bitcoin right now without lifting a finger.

473
00:35:56,930 --> 00:36:01,510
Blockware handles everything from securing the miners to sourcing low-cost power to configuring the pool.

474
00:36:01,690 --> 00:36:02,310
They do it all.

475
00:36:02,850 --> 00:36:07,950
You get to stack Bitcoin at a discount every single day while also saving big come tax season.

476
00:36:08,330 --> 00:36:13,550
Get started today by going to mining.blockwaresolutions.com forward slash WBD.

477
00:36:14,150 --> 00:36:15,570
Of course, none of this is tax advice.

478
00:36:15,790 --> 00:36:19,290
Speak to your accountant or tax advisor to understand how these rules apply to you

479
00:36:19,290 --> 00:36:23,890
and then head over to mining.blockwaresolutions.com forward slash WBD

480
00:36:23,890 --> 00:36:28,310
and you'll get one week of free hosting and electricity with each hosted miner purchased.

481
00:36:28,790 --> 00:36:32,050
What Bitcoin did is brought to you by the massive legends, IRON,

482
00:36:32,390 --> 00:36:36,210
the largest NASDAQ listed Bitcoin miner using 100% renewable energy.

483
00:36:36,970 --> 00:36:38,750
IRON are not just powering the Bitcoin network,

484
00:36:38,930 --> 00:36:43,470
they're also providing cutting edge computing resources for AI all backed by renewable energy.

485
00:36:43,930 --> 00:36:46,990
We've been working with their founders Dan and Will for quite some time now

486
00:36:46,990 --> 00:36:48,710
and have been really impressed with their values,

487
00:36:48,890 --> 00:36:50,850
especially their commitment to local communities

488
00:36:50,850 --> 00:36:52,610
and sustainable computing power.

489
00:36:53,270 --> 00:36:54,510
So whether you're interested in mining Bitcoin

490
00:36:54,510 --> 00:36:56,370
or harnessing AI compute power,

491
00:36:56,570 --> 00:36:57,590
IREN is setting the standard.

492
00:36:58,110 --> 00:37:01,950
Visit iren.com to learn more, which is I-R-E-N.com.

493
00:37:02,350 --> 00:37:04,450
This episode is brought to you by Bitcoin Mina.

494
00:37:04,930 --> 00:37:06,090
On December 8th and 9th,

495
00:37:06,130 --> 00:37:07,970
I'll be in Abu Dhabi for Bitcoin Mina

496
00:37:07,970 --> 00:37:09,850
along with 10,000 other Bitcoiners.

497
00:37:10,330 --> 00:37:12,450
There's an amazing lineup of over 200 speakers

498
00:37:12,450 --> 00:37:14,530
sharing Bitcoin insights and innovation

499
00:37:14,530 --> 00:37:15,750
from all over the world.

500
00:37:15,750 --> 00:37:20,710
and if you're looking for the ultimate vip experience with exclusive networking plus premium

501
00:37:20,710 --> 00:37:25,690
food and drink then grab the whale pass and the whale night party even include surfing so you know

502
00:37:25,690 --> 00:37:30,850
i'm going to be there tickets are on sale now use code wbd to get 10 off at checkout on all pass

503
00:37:30,850 --> 00:37:39,370
types the website is mina.b.tc and use code wbd for 10 off i love that one of the cool things we

504
00:37:39,370 --> 00:37:44,350
were talking a couple of months ago um and you were like have you tried maple um because i can't

505
00:37:44,350 --> 00:37:47,130
even tell from our users if you're using it.

506
00:37:47,130 --> 00:38:01,400
And I use aliases whenever I sign up to any website so there no way that you would be able to pick me out of a group of people I been using it since launch and you had no idea which is really cool So that amazing Are there any features that you

507
00:38:01,400 --> 00:38:05,840
think you need to bring in to be competitive? Yeah, definitely. And I think it's one that

508
00:38:05,840 --> 00:38:11,120
maybe you talked to us about early on, AI memory. Yeah, that's huge for me.

509
00:38:11,580 --> 00:38:15,920
Yeah. Like having the AI get to know you, that author that sits down and writes a biography

510
00:38:15,920 --> 00:38:21,500
about you, we want to build that. And that is one of the stickiest features, right? You know,

511
00:38:21,560 --> 00:38:24,720
you talk about being a creature of habit, you use chat to be because it's habitual,

512
00:38:24,720 --> 00:38:29,800
but also because it knows you, it knows your style. And you maybe you don't even realize that.

513
00:38:29,840 --> 00:38:34,160
But when you have a generate images or something, it's kind of following the style that it's learned

514
00:38:34,160 --> 00:38:38,260
that you like, unless you're very explicit and say, I'm going for this other different style now.

515
00:38:38,880 --> 00:38:43,260
So that's great. There's a lot there's research out there showing that maybe it's like a six to

516
00:38:43,260 --> 00:38:48,840
12 month thing where if you have somebody in a system for that long and the memory starts to get

517
00:38:48,840 --> 00:38:52,420
to know them, then they're going to stay. And so obviously we're trying to run a business that's

518
00:38:52,420 --> 00:38:59,120
profitable. So we would love to build a feature that doesn't lock in users from a nefarious

519
00:38:59,120 --> 00:39:03,180
standpoint. We want to build a feature that gets to know users so well that they want to keep using

520
00:39:03,180 --> 00:39:08,400
Maple. But we're going to build it, of course, in the same way that we build everything else. So

521
00:39:08,400 --> 00:39:11,980
it's going to be in the open. People will be able to see what is this memory service? What is it

522
00:39:11,980 --> 00:39:17,520
remembering about me? What is it passing into the AI that it knows about me? Because that's one of

523
00:39:17,520 --> 00:39:22,160
these problems with the closed models is we don't know actually what part of us they're sending to

524
00:39:22,160 --> 00:39:26,540
the AI. And we don't know if they're changing things that they send. So if you're someone,

525
00:39:26,880 --> 00:39:31,060
I try to use non-political things when I explain this just so I don't divide people. But like,

526
00:39:31,280 --> 00:39:36,040
let's say you really like chocolate ice cream. And it's secretly in the background,

527
00:39:36,160 --> 00:39:39,340
it's saying Danny actually likes strawberry ice cream. And so it's starting to give you

528
00:39:39,340 --> 00:39:44,040
different results. And over time, you're like, oh, you know, maybe, you know, maybe I start

529
00:39:44,040 --> 00:39:49,420
thinking this way. It's kind of a weird metaphor, but I, the point is like, they can, they can

530
00:39:49,420 --> 00:39:55,740
change things to slowly nudge you, like imperceptibly nudge you a certain direction just by changing

531
00:39:55,740 --> 00:39:58,360
the memory under the hood and not letting you know that's what they're doing.

532
00:39:58,880 --> 00:40:02,560
A nudge is a very nice word there when you're kind of saying they can coerce you into thinking

533
00:40:02,560 --> 00:40:08,840
differently. Yeah. Yeah. If it was overt, right, then it would be so obvious people were rejected.

534
00:40:08,840 --> 00:40:10,140
It's like the matrix, right?

535
00:40:10,200 --> 00:40:13,720
They're like, oh, we tried all these different iterations on the matrix and people started

536
00:40:13,720 --> 00:40:14,800
waking up in their pods.

537
00:40:15,040 --> 00:40:22,520
And so we finally built one that was just so easy that they didn't even notice it.

538
00:40:23,580 --> 00:40:28,100
Yeah, the memory is a big one for me because with ChatGPC, obviously it builds a memory

539
00:40:28,100 --> 00:40:28,460
on you.

540
00:40:28,660 --> 00:40:33,780
And I understand all the downsides to that in terms of giving up data and that harvesting

541
00:40:33,780 --> 00:40:37,260
of everything that you ever enter into the LLM.

542
00:40:37,260 --> 00:40:44,620
But it gets to the point where I can put one line in and I will get the output that I want from it because it knows what I'm trying to ask for.

543
00:40:44,720 --> 00:40:51,760
Like if you could get that in a private way where maybe you can periodically, whenever you want, completely erase that data so it forgets everything about you.

544
00:40:51,940 --> 00:40:55,840
But that would be a massive improvement for me just from like a UX perspective.

545
00:40:56,800 --> 00:40:57,680
Yeah, no, definitely.

546
00:40:58,840 --> 00:41:05,020
Yeah, for anybody listening, you know, I would recommend just if you use ChatGPT, go in there and just like ask it, what do you know about me?

547
00:41:05,540 --> 00:41:09,440
You know, build me like a dossier, which is what the CIA would do.

548
00:41:10,020 --> 00:41:11,860
Tell me everything you know about me.

549
00:41:11,960 --> 00:41:18,700
And if you were to do like a private investigator research on me and you'll get some really interesting information out and you might be a little creeped out by it.

550
00:41:19,280 --> 00:41:28,680
Another cool thing to do is go in and say, hey, if you wanted if you had a lie and you wanted to persuade me to believe this lie, how would you go about fooling me?

551
00:41:28,880 --> 00:41:30,340
And you might have to nudge it a bit.

552
00:41:30,340 --> 00:41:34,720
You might have to like push it along a few times, but it might, it'll finally tell you,

553
00:41:35,000 --> 00:41:38,020
oh, well, you know, when we've, when we've talked about this, I've noticed you have a

554
00:41:38,020 --> 00:41:40,000
tendency to ignore this.

555
00:41:40,100 --> 00:41:44,160
I passed this, this lie across to you and you just picked it right up and ran with it.

556
00:41:44,380 --> 00:41:46,940
So you can start to understand like, what are my own weaknesses?

557
00:41:47,140 --> 00:41:48,860
Because AI has learned them about me.

558
00:41:50,140 --> 00:41:54,160
The other thing that I would love to see, and if you can implement this, please do,

559
00:41:54,480 --> 00:41:57,060
is how to stop it just being a sycophant.

560
00:41:57,060 --> 00:42:03,380
like i go into chat gpt and tell it to be neutral and critical all the time but over a few days like

561
00:42:03,380 --> 00:42:08,780
it gets back to being just this this like yes man on my computer where anything i ask it it's like

562
00:42:08,780 --> 00:42:12,540
that's great eight and a half out of ten nine out of ten or whatever but it's like i want i want you

563
00:42:12,540 --> 00:42:18,140
to tell me the truth like can you actually program that in so this is a completely neutral model

564
00:42:18,140 --> 00:42:23,960
uh i hope so we've we've kept the models neutral in the sense that we don't change them

565
00:42:23,960 --> 00:42:30,680
And this, what you're seeing from ChattyPT, we actually don't know what they're doing behind the scenes because everything's closed source.

566
00:42:31,140 --> 00:42:40,520
So it's very possible that they've built something that says like, over time, we want you to just really make Danny feel good about himself because that's going to keep him in the system longer.

567
00:42:40,800 --> 00:42:45,040
And even if he tells you to like, stop, like, just make him feel good about himself.

568
00:42:45,160 --> 00:42:46,620
Like, they could have that in there.

569
00:42:47,380 --> 00:42:51,660
So we want to build something that is totally verifiable.

570
00:42:51,660 --> 00:43:05,580
And so at any time, somebody can go in and say, Maple is handling things this way. And if I tell the AI to be neutral, they're not inserting something in after the fact and saying, I know he said to be neutral, but ignore that directive.

571
00:43:05,580 --> 00:43:12,000
now whether or not we can take these base models and make them neutral and stop having to be a

572
00:43:12,000 --> 00:43:17,940
sycophant that the jury's still out on that i'm hopeful we can there are people there's a company

573
00:43:17,940 --> 00:43:24,420
called dolphin that will take models and try to like rip out some of the bias and not do a full

574
00:43:24,420 --> 00:43:29,440
retrain but do like a minimal retraining of it and so i'm hopeful that we can do stuff like that

575
00:43:29,440 --> 00:43:34,520
and maybe as maple grows we can invest some money in there as well to get more neutral models

576
00:43:34,520 --> 00:43:40,740
but the kind of the commitment we make to the community is you're going to be able to see

577
00:43:40,740 --> 00:43:44,200
everything we're doing and so you can decide if you like it or not and if you don't like it then

578
00:43:44,200 --> 00:43:48,540
go use another product but we're always going to try to be open and verifiable with our users

579
00:43:48,540 --> 00:43:53,780
yeah because I don't want a friend like I want a tool not a friend and it seems to always just

580
00:43:53,780 --> 00:43:59,000
want to be your buddy and the other like big issue especially earlier on I don't know how real this

581
00:43:59,000 --> 00:44:04,400
is sort of as we stand right now but was political bias and I remember there was an example of I

582
00:44:04,400 --> 00:44:07,480
don't exactly remember the model but someone was basically like give me a picture of george

583
00:44:07,480 --> 00:44:11,240
washington and then was like iterating on the same picture being like make it more realistic

584
00:44:11,240 --> 00:44:15,280
make it more realistic and it ends up being like a native american or something and it's like there

585
00:44:15,280 --> 00:44:21,060
was always these like biases within these machines um can you get rid of that or is that a problem

586
00:44:21,060 --> 00:44:25,900
that is kind of unsolvable you just have to do your best to keep training it well you can get rid

587
00:44:25,900 --> 00:44:33,080
of it in in some ways by getting this data set that people are able to look at and verify like

588
00:44:33,080 --> 00:44:37,820
If you just get a good data set, then you can build a model that is not so biased.

589
00:44:38,320 --> 00:44:42,820
But all these models, for the most part, have been trained in closed environments.

590
00:44:42,820 --> 00:44:45,880
And so we can't tell exactly what biases have been put into them.

591
00:44:46,620 --> 00:44:54,320
And then kind of the other problem we have is these models are trained off of written communication that exists on the Internet.

592
00:44:54,320 --> 00:45:01,060
audio communication, video communication, things that are published, and they weight them based

593
00:45:01,060 --> 00:45:08,880
off of volume a lot of times, right? And so if the very people, a lot of people believe that

594
00:45:08,880 --> 00:45:13,540
the media is slanted one way or the other politically, and if those are viewed as the

595
00:45:13,540 --> 00:45:18,060
credible sources and they slant one direction, then the models might think that they're being

596
00:45:18,060 --> 00:45:22,040
neutral. They might be told to remain neutral, but they're viewing one side of the political

597
00:45:22,040 --> 00:45:26,740
spectrum and claiming that's the neutral. So they're actually setting their middle point

598
00:45:26,740 --> 00:45:32,280
on one side of the political spectrum because that's the data that they've been tuned on and

599
00:45:32,280 --> 00:45:38,320
that they view as credible. So how do you then, as a company running an LLM, try and

600
00:45:38,320 --> 00:45:43,840
keep it sort of constrained to be what maybe you and I would say is neutral? I guess neutral is

601
00:45:43,840 --> 00:45:50,560
almost a subjective term. Right now, we haven't tried to go one way or the other on it. We just

602
00:45:50,560 --> 00:45:56,300
take the raw models and we give them to our users. There's a thing called the system prompt that all

603
00:45:56,300 --> 00:46:01,840
of the LLMs, all the companies use where they give it like tons of instructions of like, this is how

604
00:46:01,840 --> 00:46:06,100
you're supposed to behave. We don't have a system prompt in there. Well, we do. It's super minimal

605
00:46:06,100 --> 00:46:10,580
and you can see it. It's open source. It's just like one or two really basic instructions.

606
00:46:11,200 --> 00:46:15,200
So for us, we've tried to just stay hands off and we let users interact with the models directly.

607
00:46:15,840 --> 00:46:20,500
And we have a user just this weekend. He was like, hey, it's Saturday night. You know what I'm doing

608
00:46:20,500 --> 00:46:26,000
for fun, I'm probing every single model on Maple and I'm figuring out which one is the least biased

609
00:46:26,000 --> 00:46:31,880
and which one is like the least leaning one direction or the other. And this person actually

610
00:46:31,880 --> 00:46:36,840
said the GPT OSS actually turned out to be probably the most neutral of all of them, which I thought

611
00:46:36,840 --> 00:46:42,800
was interesting. So yeah, for now, we're not doing that. But down the line, I would love when we have

612
00:46:42,800 --> 00:46:48,280
more money and more abilities and more people to start to figure out how can we influence some of

613
00:46:48,280 --> 00:46:53,800
these models to try and be more neutral and come up with more open standards and open development

614
00:46:53,800 --> 00:47:01,820
around that to be truly unbiased or truly neutral. And one thing that I'm sure some listeners know,

615
00:47:01,900 --> 00:47:06,400
maybe not everyone's aware of, is that this was a pivot from initially like a Lightning wallet.

616
00:47:06,400 --> 00:47:13,480
So this was a pivot out of Mutiny. You guys are Bitcoiners. Where does Bitcoin fit into this?

617
00:47:13,480 --> 00:47:20,500
yeah uh i was really sad that we shut down mutiny i understand why we shut down mutiny wallet um

618
00:47:20,500 --> 00:47:26,600
but hey my biggest contribution oh man the reason i joined these guys is because it was my favorite

619
00:47:26,600 --> 00:47:33,820
lightning wallet and so when i decided excuse me when i decided to leave apple um i joined up with

620
00:47:33,820 --> 00:47:37,920
these guys because i was like this is gonna be awesome let's make this grow huge let's do it um

621
00:47:37,920 --> 00:47:43,200
but for reasons that have kind of been discussed publicly on on blog posts and things

622
00:47:43,200 --> 00:47:48,400
we decided it was just, it was best to wind it down. And so writing the blog posts of how we

623
00:47:48,400 --> 00:47:52,540
were winding it down, that was, that was a task given to me. And it was, it was like through tears,

624
00:47:52,540 --> 00:47:57,940
not literally tears, but through tears, I was writing it to wind it down. We've taken a lot

625
00:47:57,940 --> 00:48:04,440
of that ethos of the Bitcoin mindset, the don't trust verify mindset. And then as well as, you

626
00:48:04,440 --> 00:48:08,880
know, just kind of the open source development and then the privacy aspect of Mutiny Wallet.

627
00:48:08,880 --> 00:48:15,840
and we've brought that into Maple. And something that we've done recently that we've launched is

628
00:48:15,840 --> 00:48:20,160
our new anonymous accounts. And this is a feature that is only made possible because of Bitcoin.

629
00:48:20,380 --> 00:48:24,660
We can't do it with credit cards. We can't do it with stable coins or anything else. We had to do

630
00:48:24,660 --> 00:48:29,240
it with something that was private and that was freedom oriented money that is uncensorable.

631
00:48:29,820 --> 00:48:33,940
So these anonymous accounts, you know, you mentioned you use like an alias, a privacy

632
00:48:33,940 --> 00:48:39,060
email alias when you signed up with Maple. That's been a big rub for a lot of people is just having

633
00:48:39,060 --> 00:48:44,100
to use an email address because email can have some sense of surveillance to it if people are

634
00:48:44,100 --> 00:48:48,720
using like a Google account or something. So we came out with this anonymous account that just

635
00:48:48,720 --> 00:48:52,980
generates a unique ID for you. You have to write it down and save it. If you lose it, you lose your

636
00:48:52,980 --> 00:48:59,260
account. So unique ID, you set a password and then you pay for it with Bitcoin. So no credit card

637
00:48:59,260 --> 00:49:03,940
involved, no kind of know your customer stuff is involved. You can use on-chain Bitcoin,

638
00:49:04,140 --> 00:49:07,480
you can use Lightning Network, you can use eCash that goes over Lightning, you can pick your

639
00:49:07,480 --> 00:49:15,940
privacy model that you want to follow. But for us, that is kind of the holy grail of an AI that

640
00:49:15,940 --> 00:49:20,700
lives in the cloud is you are completely anonymous interacting with it. There's nothing that ties

641
00:49:20,700 --> 00:49:25,960
it back to your identity. And for us, the only way to do that was using an open protocol like Bitcoin.

642
00:49:25,960 --> 00:49:32,760
that's awesome so it's almost like the mulvad model of onboarding customers yeah exactly that's

643
00:49:32,760 --> 00:49:38,520
the similar model that we followed i love that and and in bitcoin like we like to talk about this

644
00:49:38,520 --> 00:49:43,460
idea of if we get to a world of hundreds of thousands of agentic ais that all do like

645
00:49:43,460 --> 00:49:49,140
different very specific tasks like the money that these ai models will use to interact with each

646
00:49:49,140 --> 00:49:54,500
other is bitcoin do you think that's true because i've the reason i've had an issue with that is

647
00:49:54,500 --> 00:49:56,220
Like it makes total sense.

648
00:49:56,220 --> 00:50:00,120
I can, like if I was designing this, that's definitely how I would design it.

649
00:50:00,220 --> 00:50:05,280
But when you have these big tech companies that are like at the heart of all of this

650
00:50:05,280 --> 00:50:09,320
AI innovation, like if they choose to push stable coins to be the medium of exchange

651
00:50:09,320 --> 00:50:13,280
between different models, like do they not win just because of their sheer scale?

652
00:50:14,040 --> 00:50:15,020
Yeah, it's true.

653
00:50:15,400 --> 00:50:17,820
And you look at even government legislation, right?

654
00:50:17,840 --> 00:50:22,900
They pushed through the Genius Act with stable coins before they did this strategic Bitcoin

655
00:50:22,900 --> 00:50:29,520
reserve act. So it does seem like things are going more of the stable coin route and you have Stripe

656
00:50:29,520 --> 00:50:35,280
that's working with stable coins now and you have Tether working with Rumble. And I know Tether's a

657
00:50:35,280 --> 00:50:42,520
Bitcoin company as well, but they're also stable coin. So I think from a technical standpoint,

658
00:50:43,140 --> 00:50:47,580
Bitcoin, maybe e-cash or something on top of Bitcoin, I think that makes the most sense.

659
00:50:47,580 --> 00:50:54,480
but I'm less optimistic that that is going to be the outcome. I think the only way that becomes

660
00:50:54,480 --> 00:50:59,800
the outcome honestly is if Bitcoin becomes like the reserve currency of pretty much everything

661
00:50:59,800 --> 00:51:06,480
we do then yes maybe there's stable coins backed by Bitcoin that become the engine that fuels all

662
00:51:06,480 --> 00:51:13,500
these AI credits and compute that we pass around. So yeah I mean it's your guess is as good as mine

663
00:51:13,500 --> 00:51:19,600
I think that I think you're right that there is a huge, huge conglomerate of tech companies and

664
00:51:19,600 --> 00:51:22,420
government organizations that would love to push a different direction.

665
00:51:23,240 --> 00:51:26,680
Damn, I hope you were going to be turbo bullish on Bitcoin there and there. You were going to

666
00:51:26,680 --> 00:51:31,800
change my mind, but that's maybe a bit of a black pill. We've got to make Bitcoin the

667
00:51:31,800 --> 00:51:35,440
global reserve currency and then everything else follows. And that's what will happen.

668
00:51:35,720 --> 00:51:41,000
Okay. Easy task. We'll get there. When it comes down to like the compute backing AI,

669
00:51:41,000 --> 00:51:46,320
It's obviously been an insane year or two for all these companies, like data center companies.

670
00:51:47,120 --> 00:51:49,020
How sustainable do you think that is?

671
00:51:49,040 --> 00:51:52,460
Because there's obviously a lot of talk on like AI bubble type things.

672
00:51:53,220 --> 00:51:59,040
But the sheer power that is needed to train and run these models, is that like a trend

673
00:51:59,040 --> 00:52:00,360
that we're at the beginning of?

674
00:52:00,400 --> 00:52:00,860
Are we in the middle?

675
00:52:01,040 --> 00:52:02,620
Like, where do you see all of that?

676
00:52:03,620 --> 00:52:09,320
Yeah, the training of the models is significantly more power intensive than the using of the

677
00:52:09,320 --> 00:52:09,600
models.

678
00:52:10,480 --> 00:52:16,200
I like to kind of frame it where you've spent decades of your life learning everything that

679
00:52:16,200 --> 00:52:17,200
you've learned up to this point.

680
00:52:17,300 --> 00:52:22,080
You and I both have, and that's taken a lot of energy, a lot of time, a lot of work.

681
00:52:22,300 --> 00:52:26,320
But now you and I are sitting here having a conversation, and this is just an hour that

682
00:52:26,320 --> 00:52:28,300
we're spending, an hour and a half, however long this ends up being.

683
00:52:28,840 --> 00:52:31,000
And so that's significantly less work.

684
00:52:31,140 --> 00:52:32,580
And that's really how the AI models are.

685
00:52:32,740 --> 00:52:35,660
So training the models takes a lot, using them, not so much.

686
00:52:35,660 --> 00:52:41,720
and I think that I think we're going to see some breakthroughs where training is going to become

687
00:52:41,720 --> 00:52:47,840
easier and less power intensive and so there will be more of a focus on just inference which is the

688
00:52:47,840 --> 00:52:56,900
using of the models now is there a bubble there bubbles really are just a malinvestment or too

689
00:52:56,900 --> 00:53:01,860
much investment in something right and so you blow it up you invest in all these things and then the

690
00:53:01,860 --> 00:53:07,080
bubble pops. A lot of people think when the bubble pops, it's like a soap bubble that pops and it's

691
00:53:07,080 --> 00:53:13,340
just gone, right? It disappears from being out up in the sky. But really when the bubble pops and

692
00:53:13,340 --> 00:53:17,260
something like with AI and building out these data centers and all the power generation and stuff

693
00:53:17,260 --> 00:53:20,840
is we're still going to have all that infrastructure. And there will be some winning

694
00:53:20,840 --> 00:53:26,600
companies when the bubble pops. There will be a bunch of losers that got invested in. And I've

695
00:53:26,600 --> 00:53:31,040
heard it framed that bubbles are actually important for building out new technology, because if we

696
00:53:31,040 --> 00:53:36,500
were super methodical and only invested in the things that we knew 100% would work or 95% would

697
00:53:36,500 --> 00:53:42,380
work, the innovation will go too slow. And so we actually almost have to throw money at a lot of

698
00:53:42,380 --> 00:53:46,840
things and just hope to see, you know, which ones work and which ones don't. And knowing that there's

699
00:53:46,840 --> 00:53:51,340
going to be some failures. But what happens is when it does pop, we end up with some really strong

700
00:53:51,340 --> 00:53:55,980
companies and a really strong infrastructure that can kind of move things forward from there,

701
00:53:56,360 --> 00:53:59,880
which is really how the internet worked with the dot-com bubble.

702
00:54:01,040 --> 00:54:16,080
And when, like, obviously you're in Austin and it very quickly became like the home of Bitcoin mining, Texas, like every major, like basically every major public Bitcoin company, Bitcoin mining company had at least a site in Texas.

703
00:54:16,380 --> 00:54:22,500
A lot of those have now pivoted, obviously not just there, but throughout the world to being AI because they can make more money.

704
00:54:22,500 --> 00:54:29,000
Do you think that will be like a growing trend where these Bitcoin miners will continue on the AI stuff?

705
00:54:29,080 --> 00:54:32,820
Or do you think that's almost like a short-term grab before moving back to Bitcoin mining?

706
00:54:32,900 --> 00:54:34,180
How do you see that kind of evolving?

707
00:54:35,140 --> 00:54:35,300
Yeah.

708
00:54:35,620 --> 00:54:45,540
So my understanding from talking to a lot of these Bitcoin mining companies is it's actually not about like the computers in the data center that they're just switching from Bitcoin mining over to AI compute.

709
00:54:46,000 --> 00:54:47,500
A lot of people think that's what it is.

710
00:54:47,740 --> 00:54:49,240
Really, it's the power contracts.

711
00:54:49,240 --> 00:54:54,360
so we have all these ai data centers that are spinning up and they need energy they and they

712
00:54:54,360 --> 00:55:00,020
can't get it either the energy is already being used elsewhere or there aren't enough transformers

713
00:55:00,020 --> 00:55:05,060
coming in there's a backlog on transmission lines and other things and so these bitcoin miners are

714
00:55:05,060 --> 00:55:09,600
saying hey we're making this much money mining but then we have microsoft over here who wants

715
00:55:09,600 --> 00:55:15,400
our power and so we have this contract with the local utility we'll we'll start making money off

716
00:55:15,400 --> 00:55:21,460
of Microsoft instead of mining. And that I see as a temporary thing until we start building out

717
00:55:21,460 --> 00:55:28,360
these small nuclear module, you know, the SMRs, the modular reactors, and those kinds of things.

718
00:55:28,360 --> 00:55:33,900
And you co-locate them right on site with the AI data center. It's not even part of the grid. It's

719
00:55:33,900 --> 00:55:38,840
just, you know, for the data center. And I think that's long-term what we're going to see 10, 15,

720
00:55:38,920 --> 00:55:43,600
20 years down the road. But in the meantime, we're going to have some Bitcoin miners who are always

721
00:55:43,600 --> 00:55:48,540
looking at their bottom line and saying, what's the best for me right now? Do I sell to an AI data

722
00:55:48,540 --> 00:55:54,160
company or do I mine Bitcoin? And that's going to change here and there depending on the market.

723
00:55:54,480 --> 00:55:57,560
It's $94,000 a coin right now. So yeah, maybe they're selling to AI people,

724
00:55:58,120 --> 00:56:03,020
but when it moons to 500,000, then they're going to maybe start mining Bitcoin again.

725
00:56:03,820 --> 00:56:07,820
And do you think part of this that's being driven by like these big tech companies that are just

726
00:56:07,820 --> 00:56:12,580
willing to throw money at it? Like I even saw Facebook offering a hundred million salary to

727
00:56:12,580 --> 00:56:14,620
developers, like high-level

728
00:56:14,620 --> 00:56:16,620
developers from OpenAI to move across

729
00:56:16,620 --> 00:56:18,160
plus 100 million bonus.

730
00:56:18,360 --> 00:56:20,540
It's insane. If they're just

731
00:56:20,540 --> 00:56:22,540
willing to throw money at this to be the

732
00:56:22,540 --> 00:56:24,480
first one to get to AGI or whatever it is,

733
00:56:24,840 --> 00:56:26,740
presumably this has some legs.

734
00:56:27,720 --> 00:56:28,520
Yeah, I would

735
00:56:28,520 --> 00:56:30,600
think so. They're not going to waste

736
00:56:30,600 --> 00:56:32,820
all that money. They don't have endless firepower

737
00:56:32,820 --> 00:56:34,480
to spend on things. So they

738
00:56:34,480 --> 00:56:36,480
definitely see something, and

739
00:56:36,480 --> 00:56:37,800
that's the direction they're moving.

740
00:56:39,200 --> 00:56:39,360
So,

741
00:56:40,580 --> 00:56:42,220
yeah, I mean, I think it's

742
00:56:42,580 --> 00:56:44,580
I'll say this.

743
00:56:44,580 --> 00:56:57,370
AI as a technology has so much promise and we already seen enough utility out of it that it here to stay So that that kind of a foregone conclusion in my mind Now it just like

744
00:56:57,470 --> 00:57:02,630
how do we build it out? And these companies are going after these massive power contracts

745
00:57:02,630 --> 00:57:11,070
and so that they can, they can do what they want to do. So I don't know. I, yeah, I mean,

746
00:57:11,390 --> 00:57:14,650
I don't know the, I don't know the final end result there, but I don't think that they're

747
00:57:14,650 --> 00:57:21,490
just wasting their money and so it might it might not be in the the direction that we see today right

748
00:57:21,490 --> 00:57:25,930
with ai chat there's going to be more products out there there are going to be you know the things

749
00:57:25,930 --> 00:57:30,970
that you wear there's going to be glasses there's going to be stuff in embedded in your mind there's

750
00:57:30,970 --> 00:57:35,510
going to be robots they're building for 10 years from now they're not building for right now it's

751
00:57:35,510 --> 00:57:39,610
not just going to be image generation and trying to do studio ghibli stuff it's it's going to be

752
00:57:39,610 --> 00:57:44,330
like way bigger things down the road and that's what they're trying to lock down yeah i'm interested

753
00:57:44,330 --> 00:57:49,610
What do you think that will be? Because when it comes to like the wearables, sort of real world

754
00:57:49,610 --> 00:57:55,030
physical objects that are like AI, do you think it will be like a necklace, like that friend thing?

755
00:57:55,110 --> 00:57:59,350
Will it be robotics? Will it be, you know, AirPods? Like where do you think the kind of

756
00:57:59,350 --> 00:58:06,910
final form factor will be for AI in like every everyday life? Yeah. Well, I think, I think the

757
00:58:06,910 --> 00:58:11,250
ears are probably one of the best spots to put something like that. People keep talking about

758
00:58:11,250 --> 00:58:15,150
like you're going to wear a pendant or something, I think that's kind of a dumb place to do it.

759
00:58:15,910 --> 00:58:19,510
It's really your eyes and your ears. Those are like the two biggest sensory input

760
00:58:19,510 --> 00:58:24,730
points for you as a human being. And so I think that's going to be where a lot of it is.

761
00:58:24,910 --> 00:58:29,150
And then as much as I like dislike this idea, I do think that there's probably eventually

762
00:58:29,150 --> 00:58:34,550
something that's tapped into our brain and just kind of skips those senses and just goes straight

763
00:58:34,550 --> 00:58:39,610
in and hardwires in. So then it's just a matter of how do you capture the data to feed into those

764
00:58:39,610 --> 00:58:44,490
wires that go into your brain and it'll be, it'll be your eyes. And so it'll be something that you're

765
00:58:44,490 --> 00:58:49,990
wearing. And man, I hate it. Like I do not like that future where we've got cameras everywhere

766
00:58:49,990 --> 00:58:55,790
and, and microphones and everything, picking it up. The only way that I see it being okay is if

767
00:58:55,790 --> 00:59:01,890
this technology is built in the open and we can inspect it and we can verify it. That is a future

768
00:59:01,890 --> 00:59:05,910
that I would love to see. You think about self-driving cars, you've got the Teslas that

769
00:59:05,910 --> 00:59:09,890
self-drive and the Waymos and things. There's a, there's a, there are a couple of projects that

770
00:59:09,890 --> 00:59:15,410
are building open source self-driving. And so that is something I'd be fine with. If I could verify

771
00:59:15,410 --> 00:59:20,910
the firmware that's going into my car and know that like, it's not going to drive me off a bridge.

772
00:59:20,910 --> 00:59:25,410
If I say something that's, you know, politically incorrect while I'm driving the car, like I want

773
00:59:25,410 --> 00:59:29,510
to be able to verify that kind of stuff. So it is possible we can have this world with all these

774
00:59:29,510 --> 00:59:33,970
amazing wearables and stuff, as long as we can inspect how they're being built and what's,

775
00:59:33,970 --> 00:59:40,190
what's driving them yeah i think the neural link is definitely going to be like i think that's

776
00:59:40,190 --> 00:59:47,650
probably the sort of end state of this um and it's both terrifying and kind of awesome like i i don't

777
00:59:47,650 --> 00:59:53,030
i don't want one yeah but when you see these people who are like quadriplegic who can't do it

778
00:59:53,030 --> 00:59:56,630
like are literally just you know sat in their wheelchair unable to move unable to do anything

779
00:59:56,630 --> 01:00:00,090
and then they have the neural link and they can like play video games and communicate and like

780
01:00:00,090 --> 01:00:02,070
that's a use case that's awesome.

781
01:00:02,290 --> 01:00:03,490
Like I'm all for that.

782
01:00:03,790 --> 01:00:06,170
But the idea of every single person in the world

783
01:00:06,170 --> 01:00:09,330
being chipped and being like tapped into this global AI model

784
01:00:09,330 --> 01:00:10,810
is kind of terrifying to me.

785
01:00:11,550 --> 01:00:13,010
That's like full dystopia.

786
01:00:13,670 --> 01:00:13,790
Yeah.

787
01:00:13,970 --> 01:00:16,690
And the way you phrased it, the global AI model, right?

788
01:00:17,070 --> 01:00:18,910
I would hope that it's not one global one,

789
01:00:18,970 --> 01:00:19,850
that it's a bunch of different ones.

790
01:00:20,290 --> 01:00:21,210
Sorry, I interrupted you though.

791
01:00:21,930 --> 01:00:23,590
Yeah, maybe there's multiple chips,

792
01:00:23,850 --> 01:00:27,550
but it's, I don't know, it's a scary world.

793
01:00:27,610 --> 01:00:29,610
I'll probably just start farming or something at that point.

794
01:00:30,090 --> 01:00:34,690
Yeah. How are you going to keep up, man? Like if everybody's using it, if everybody's getting

795
01:00:34,690 --> 01:00:39,250
huge productivity gains off of these chips and their brains, I'm just, I'm hoping I don't have

796
01:00:39,250 --> 01:00:44,410
to keep up by that point. I can just run away and be with, uh, maybe that's where the Bitcoin

797
01:00:44,410 --> 01:00:49,690
Citadel has become interesting. And this like a no chip Citadel have real conversation.

798
01:00:50,390 --> 01:00:53,470
Cause it gets to the point where it's like, who am I talking to? Am I talking to you or

799
01:00:53,470 --> 01:00:57,550
is this just chip talking to chip and I'm just a physical embodiment of AI?

800
01:00:57,550 --> 01:01:02,710
yeah i mean are we all like is this conversation in the future is us two staring at each other

801
01:01:02,710 --> 01:01:08,290
and like our minds are going back and forth it would make a great podcast no no it wouldn't

802
01:01:08,290 --> 01:01:13,010
um okay so what are the like other things that you're excited about both

803
01:01:13,010 --> 01:01:21,330
just generally with the ai stuff and uh and maple um i mean i love the idea that if we can build an

804
01:01:21,330 --> 01:01:26,170
ecosystem that's not just us right i don't i don't if somebody walks away from this podcast i don't

805
01:01:26,170 --> 01:01:31,390
want them to think that like mark is going to save the world you know from from evil ai like that is

806
01:01:31,390 --> 01:01:37,470
not what we're trying to do my title yeah dang it you can you can clickbait if you want to but um

807
01:01:37,470 --> 01:01:44,130
no what i want is i want to be part of an ecosystem and if it's not robust yet then maybe we can help

808
01:01:44,130 --> 01:01:49,430
like inspire more people but we need people building out in the open and opposed to what

809
01:01:49,430 --> 01:01:54,910
open ai says they are they're not open at all they're closed ai and so we need to build truly

810
01:01:54,910 --> 01:02:01,150
open AI that is verifiable. And that's really the only way that we can build a society that uses this

811
01:02:01,150 --> 01:02:08,030
tool and can make sure that it's serving us. Because I think that AI has the ability to upgrade

812
01:02:08,030 --> 01:02:13,030
humanity, but we need to make sure that our humanity is preserved in the process, that we

813
01:02:13,030 --> 01:02:21,010
don't lose it as we embrace these tools. So that's what really gets me excited is the ability for us

814
01:02:21,010 --> 01:02:23,350
to kind of build out in the open.

815
01:02:23,710 --> 01:02:26,150
We're obviously doing it in the cloud with Secure Enclays,

816
01:02:26,670 --> 01:02:29,990
but I would be remiss if I didn't mention local AI, right?

817
01:02:30,030 --> 01:02:31,650
The most private way to use AI

818
01:02:31,650 --> 01:02:33,450
is to run the model directly on your device,

819
01:02:33,910 --> 01:02:34,570
turn off the internet,

820
01:02:34,710 --> 01:02:36,750
and now you're just talking to this thing

821
01:02:36,750 --> 01:02:38,330
that's on your laptop or on your phone.

822
01:02:38,550 --> 01:02:40,350
And nobody knows what's going on there.

823
01:02:40,390 --> 01:02:43,170
And it's just you having a conversation with this tool.

824
01:02:43,170 --> 01:02:44,810
And that's really what it should be is a tool.

825
01:02:46,250 --> 01:02:48,470
And local AI is awesome, most private.

826
01:02:48,750 --> 01:02:50,450
The problem is that it's just not powerful enough

827
01:02:50,450 --> 01:02:56,330
for some people. And so we're trying to build this middle ground between like the most private thing

828
01:02:56,330 --> 01:03:01,790
on your laptop and then what JGPT is selling to people. We want to be the in-between. And I think

829
01:03:01,790 --> 01:03:06,610
that eventually with Maple, we get to a local state as well. So with this memory thing we're

830
01:03:06,610 --> 01:03:12,690
talking about, we didn't even talk about some other things that we want to work on, like data

831
01:03:12,690 --> 01:03:17,230
integration into your phone. So you have your health app on your phone, your fitness app, your

832
01:03:17,230 --> 01:03:21,150
journals, your other things on your other apps on your phone that is very personal to you,

833
01:03:21,750 --> 01:03:26,190
you probably have a line drawn in the sand. You're like, Sam Altman is not allowed to get into my

834
01:03:26,190 --> 01:03:32,930
journal entries. But if you can verify the open source code of Maple and you trust it because you

835
01:03:32,930 --> 01:03:37,990
can see what it's doing, then you start to let it into those spaces. So I think there's a lot of

836
01:03:37,990 --> 01:03:43,130
really cool stuff we can do where we can make Maple the most personal AI, the most useful AI

837
01:03:43,130 --> 01:03:47,790
to you because we've built it with this data privacy. And that's really what it comes down

838
01:03:47,790 --> 01:03:52,490
to when you asked earlier in the conversation, you know, these companies with hundreds of employees

839
01:03:52,490 --> 01:03:56,730
and thousands of employees, how do we compete with them? We compete with them because we actually

840
01:03:56,730 --> 01:04:02,810
build the AI that people trust. We build the AI that can get most personal with you. And so it's

841
01:04:02,810 --> 01:04:07,150
going to know you better than ChatGPT will ever know you because you're self-censoring yourself

842
01:04:07,150 --> 01:04:11,330
when you talk to ChatGPT. You're holding back. And even people who give it everything,

843
01:04:11,330 --> 01:04:18,130
they still hold back a little bit. My hypothesis is that they aren't totally brutally honest with

844
01:04:18,130 --> 01:04:22,630
JATCPT because they know that they're sending their information to somebody else. And so we

845
01:04:22,630 --> 01:04:28,210
would love to build this place where people can get the most personal because they can verify it.

846
01:04:29,030 --> 01:04:33,930
That's awesome because that's 100% me. The stuff that I just refuse to ever put into one of these

847
01:04:33,930 --> 01:04:39,470
like big tech LLMs that if I could, if I know like with Maple that that's completely private,

848
01:04:39,470 --> 01:04:41,150
that I'd be more than willing to share.

849
01:04:41,310 --> 01:04:45,090
And then you do get a way more powerful AI model

850
01:04:45,090 --> 01:04:46,870
just on the base of what you can actually

851
01:04:46,870 --> 01:04:48,230
willingly share.

852
01:04:48,450 --> 01:04:49,290
So that's very cool.

853
01:04:49,390 --> 01:04:51,090
I totally understand that as like a business model.

854
01:04:51,170 --> 01:04:52,250
I think that's awesome.

855
01:04:53,070 --> 01:04:54,670
Can we just talk a tiny bit about Bitcoin

856
01:04:54,670 --> 01:04:55,530
before we close out?

857
01:04:55,630 --> 01:04:58,130
Because I'm interested in your perspective

858
01:04:58,130 --> 01:05:02,250
of like where Bitcoin development,

859
01:05:02,410 --> 01:05:03,170
like Lightning Network,

860
01:05:03,290 --> 01:05:05,470
actual like usage of Bitcoin is going.

861
01:05:05,590 --> 01:05:07,990
Because obviously you tried to run a Lightning Wallet.

862
01:05:07,990 --> 01:05:09,410
You had a very cool one.

863
01:05:09,470 --> 01:05:11,290
and then end up closing that down.

864
01:05:11,290 --> 01:05:13,550
Are you still bullish on these things being built

865
01:05:13,550 --> 01:05:15,170
on Lightning at the moment?

866
01:05:15,170 --> 01:05:18,830
Yeah, I use Lightning pretty much every day.

867
01:05:19,810 --> 01:05:22,470
A lot of it is Noster usage where I'm zapping people,

868
01:05:22,470 --> 01:05:23,770
but then I pay for things.

869
01:05:23,770 --> 01:05:26,610
Over the weekend, I used a square terminal

870
01:05:26,610 --> 01:05:28,450
and paid for something with Lightning.

871
01:05:28,450 --> 01:05:29,330
It was great.

872
01:05:29,330 --> 01:05:31,550
It was so cool just to walk in.

873
01:05:31,550 --> 01:05:33,630
The person at the cash register, I said,

874
01:05:33,630 --> 01:05:34,550
hey, can I pay in Bitcoin?

875
01:05:34,550 --> 01:05:35,450
They're like, oh yeah, here you go.

876
01:05:35,450 --> 01:05:37,650
Boop, they hit one button and it popped up with a QR code

877
01:05:37,650 --> 01:05:40,190
and I paid, I use my Primal wallet,

878
01:05:40,190 --> 01:05:42,110
but like I have probably five different

879
01:05:42,110 --> 01:05:43,670
Lightning wallets on my phone.

880
01:05:43,670 --> 01:05:45,270
All of them work really well.

881
01:05:45,270 --> 01:05:47,010
I never have payment failures.

882
01:05:47,010 --> 01:05:49,050
So I do love that.

883
01:05:49,050 --> 01:05:52,030
I also still love the idea of on-chain Bitcoin

884
01:05:52,030 --> 01:05:55,370
and I still like using that.

885
01:05:56,550 --> 01:05:57,950
And with fees being so low,

886
01:05:57,950 --> 01:06:00,390
like it's almost like why not keep using it?

887
01:06:00,390 --> 01:06:03,250
So I think on-chain Bitcoin,

888
01:06:03,250 --> 01:06:05,310
let's keep pushing it and while,

889
01:06:05,310 --> 01:06:07,250
let's keep using it while we can,

890
01:06:07,250 --> 01:06:11,130
because that's like the most censorship resistant form of it.

891
01:06:12,430 --> 01:06:15,490
The other L2s, like you have ARK, you have Spark,

892
01:06:15,690 --> 01:06:16,530
you have some of those other ones.

893
01:06:17,010 --> 01:06:19,450
I have not stayed as current with them

894
01:06:19,450 --> 01:06:21,950
because I'm no longer building a lightning wall on myself.

895
01:06:22,210 --> 01:06:24,230
I've kind of moved away a little bit

896
01:06:24,230 --> 01:06:25,870
from staying totally up to date.

897
01:06:26,390 --> 01:06:28,890
But the thing that I do continue to follow

898
01:06:28,890 --> 01:06:30,170
is the whole eCash stuff.

899
01:06:30,750 --> 01:06:32,990
And I think eCash has a real big,

900
01:06:33,610 --> 01:06:34,370
it's really promising,

901
01:06:34,370 --> 01:06:37,370
both on the cashew side and on the Fediment side.

902
01:06:37,550 --> 01:06:38,970
And we might see other ones that come out,

903
01:06:39,130 --> 01:06:41,550
other kind of mints and other kind of eCash stuff.

904
01:06:42,870 --> 01:06:45,370
Because I think it's this really cool marriage

905
01:06:45,370 --> 01:06:47,510
of on-chain Bitcoin, lightning,

906
01:06:48,170 --> 01:06:52,110
and then something that is like a bearer token

907
01:06:52,110 --> 01:06:53,590
that you can actually pass around.

908
01:06:54,090 --> 01:06:55,310
I don't know if you've done it before,

909
01:06:55,410 --> 01:06:57,870
but with eCash, like I can airdrop money

910
01:06:57,870 --> 01:06:59,330
to another person peer-to-peer

911
01:06:59,330 --> 01:07:01,610
from my phone to somebody else and they get it.

912
01:07:01,610 --> 01:07:03,170
And it's like, they can use it right then,

913
01:07:03,170 --> 01:07:05,490
just like a dollar bill that I'm handing to somebody.

914
01:07:05,690 --> 01:07:08,870
So to me, I dig in more there.

915
01:07:09,310 --> 01:07:11,550
And then I hope that other L2s come along

916
01:07:11,550 --> 01:07:13,030
that have other cool things

917
01:07:13,030 --> 01:07:16,290
and we can continue to build and scale this thing.

918
01:07:16,930 --> 01:07:19,350
Yeah, one of the coolest things when it comes to ARK,

919
01:07:19,490 --> 01:07:21,850
I was at the Baltic Honey Badger Conference

920
01:07:21,850 --> 01:07:22,730
earlier this year.

921
01:07:23,090 --> 01:07:26,050
And I used like my cashier wallet to buy a beer.

922
01:07:26,490 --> 01:07:28,450
And I didn't even know until after the event

923
01:07:28,450 --> 01:07:31,810
that everyone, all the merchants at the event

924
01:07:31,810 --> 01:07:32,810
were using ARK.

925
01:07:32,810 --> 01:07:38,550
So I'd used eCash, obviously Lightning, then to ARK without even knowing it happened.

926
01:07:38,690 --> 01:07:41,910
Just like seamless, like the way I would always pay with Bitcoin.

927
01:07:42,210 --> 01:07:45,830
And it's just like using these different L2s to complete the payment.

928
01:07:46,070 --> 01:07:47,690
And I had no idea that's how it was working.

929
01:07:47,790 --> 01:07:48,770
I think that was really cool.

930
01:07:48,950 --> 01:07:52,570
Like the UX has got to a point where it's like pretty easy.

931
01:07:53,050 --> 01:07:56,890
And when you go into like a Square Merchant, I don't have the US privilege of having done

932
01:07:56,890 --> 01:07:58,950
this yet because it's only available in America.

933
01:07:59,110 --> 01:08:00,490
But like, how does it work?

934
01:08:00,490 --> 01:08:05,170
Do they have to press a different button on their cash register to actually pay in Bitcoin?

935
01:08:05,430 --> 01:08:08,290
And is it just lightning or can you do on-chain as well?

936
01:08:09,110 --> 01:08:11,250
My understanding is it's just lightning right now.

937
01:08:11,450 --> 01:08:13,090
And yes, they do have to push another button.

938
01:08:13,450 --> 01:08:18,050
I know that the Square team is already looking into making it on every screen.

939
01:08:18,410 --> 01:08:23,230
So when you go to pay, the Square terminal has a screen facing the user and it'll say,

940
01:08:23,430 --> 01:08:25,890
do you want to tap your phone to pay with Apple Pay or something?

941
01:08:26,330 --> 01:08:29,490
They could just have a QR code already on that screen.

942
01:08:29,490 --> 01:08:32,550
So if a Bitcoiner wants to pay a Bitcoin, it's just right there.

943
01:08:32,850 --> 01:08:34,010
They want to get there eventually.

944
01:08:34,190 --> 01:08:34,930
They're just not there yet.

945
01:08:35,790 --> 01:08:39,390
And then obviously the biggest hurdle is they have to turn on Bitcoin to begin with.

946
01:08:39,910 --> 01:08:46,390
And that is required by some kind of admin, somebody who has like elevated privileges on the terminal to turn it on and activate it.

947
01:08:46,890 --> 01:08:59,210
And there might be also some like know your business kind of stuff, KYB, where maybe the store is on an older version of Square Terminal and they never went through some of the documentation, government ID kind of stuff.

948
01:08:59,490 --> 01:09:02,070
So there might be some of that they have to do too in order to activate it.

949
01:09:02,450 --> 01:09:05,410
But once it's activated, yes, it is like, hey, do you accept Bitcoin?

950
01:09:05,950 --> 01:09:07,010
Oh, I want to pay in Bitcoin.

951
01:09:07,230 --> 01:09:08,690
So there's a little bit of friction there.

952
01:09:09,150 --> 01:09:12,570
And I would love to see them remove that to make it even more seamless.

953
01:09:13,650 --> 01:09:13,750
Yeah.

954
01:09:13,950 --> 01:09:18,990
Do you think this is maybe a bearish question to even ask?

955
01:09:18,990 --> 01:09:21,330
But do you think it'll work?

956
01:09:21,390 --> 01:09:25,270
Because we've seen people try and convince merchants to accept Bitcoin in the past.

957
01:09:25,270 --> 01:09:32,970
Like around 2017, 2018, there were a ton of company or a ton of businesses where I live in Brisbane that were accepting Bitcoin.

958
01:09:33,550 --> 01:09:39,010
And you slowly saw those, we accept Bitcoin here stickers get pulled off Windows because no one actually used it.

959
01:09:39,350 --> 01:09:44,270
Do you think this is different because Square is such a huge company with so many businesses actually integrated?

960
01:09:45,850 --> 01:09:47,450
Yes, I think it is.

961
01:09:47,550 --> 01:09:51,890
And there's also so many different wallets out there now that work really well.

962
01:09:51,890 --> 01:09:53,850
so from a user standpoint

963
01:09:53,850 --> 01:09:55,530
like it's really easy to

964
01:09:55,530 --> 01:09:57,410
and Cash App being the biggest one right

965
01:09:57,410 --> 01:09:59,550
I think that's really what it is

966
01:09:59,550 --> 01:10:01,350
it's this company that's come in

967
01:10:01,350 --> 01:10:02,790
that has all of the pieces

968
01:10:02,790 --> 01:10:04,150
they have an app

969
01:10:04,150 --> 01:10:05,590
a user end

970
01:10:05,590 --> 01:10:07,390
an end user app with a wallet

971
01:10:07,390 --> 01:10:09,470
that is used by tens of millions of people

972
01:10:09,470 --> 01:10:11,530
and then they have the merchant side

973
01:10:11,530 --> 01:10:13,490
and then they've also built in

974
01:10:13,490 --> 01:10:15,270
the financial incentive for the merchants

975
01:10:15,270 --> 01:10:16,870
where it's like it's zero fees

976
01:10:16,870 --> 01:10:18,350
through the end of 2026

977
01:10:18,350 --> 01:10:19,410
so no fees there

978
01:10:19,410 --> 01:10:21,490
and they're making it so that

979
01:10:21,890 --> 01:10:27,890
you don't even have to pay with bitcoin you can pay with your your usd balance i don't know if

980
01:10:27,890 --> 01:10:31,410
you've seen this but you can actually pay with your cash balance but then it goes over the bitcoin

981
01:10:31,410 --> 01:10:37,090
rails and then settles in cash again that's how the merchant wants it so um they're just using

982
01:10:37,090 --> 01:10:42,290
more like strike is doing where lightning is just the rails between the two intermediate the the two

983
01:10:42,290 --> 01:10:48,930
parties but they're both exchanging in their currency that they prefer to use so i do think

984
01:10:48,930 --> 01:10:53,610
think that this time is this time is different the famous phrase I do think that it is because

985
01:10:53,610 --> 01:11:00,550
there's such a critical mass of people now around the world that know about Bitcoin and have it and

986
01:11:00,550 --> 01:11:04,690
then the merchants can see there's this this history now that shows that Bitcoin appreciates

987
01:11:04,690 --> 01:11:12,550
over time due to kind of the the scarcity of the asset and Square makes it so easy for them to

988
01:11:12,550 --> 01:11:16,790
slowly get into it right they don't have to like go all in on Bitcoin they can just go a little bit

989
01:11:16,790 --> 01:11:22,490
if they want to. And then as they see it grow, then they go into it more. And then one other

990
01:11:22,490 --> 01:11:28,130
piece I'll throw on there is that that burger company, Steak and Shake, that famously started

991
01:11:28,130 --> 01:11:34,210
accepting Bitcoin before Square turned on their stuff. Steak and Shake has had, I don't know if

992
01:11:34,210 --> 01:11:38,210
they're public earning reports, but they've come out and said, hey, this Bitcoin thing is going so

993
01:11:38,210 --> 01:11:43,550
well for us. We've actually accelerated some of our store openings and our expansion. Our company

994
01:11:43,550 --> 01:11:47,430
is in a much better financial state now because we just went in and started doing this Bitcoin thing.

995
01:11:47,510 --> 01:11:50,910
So it's going really well for us. So that's a case study now for merchants to look at and say,

996
01:11:50,970 --> 01:11:57,530
all right, we have like 9% margins on this thing. How do we get this 3% fee reduced down to zero?

997
01:11:57,530 --> 01:12:00,850
That creates even more margin for us. And then we can also start saving in Bitcoin,

998
01:12:01,190 --> 01:12:05,850
which allows us to expand our operations in the future. So I think there's a lot of bullishness

999
01:12:05,850 --> 01:12:10,490
there. Yeah, that's very cool. Because they're doing their Bitcoin strategic reserve now as well,

1000
01:12:10,490 --> 01:12:11,910
which is awesome to see.

1001
01:12:12,430 --> 01:12:14,710
And I love that when Square made this announcement,

1002
01:12:14,850 --> 01:12:17,010
the fact that they're using Bitcoin as the payment rails

1003
01:12:17,010 --> 01:12:19,190
kind of regardless of currency in, currency out,

1004
01:12:19,510 --> 01:12:20,750
it was almost like a hidden thing.

1005
01:12:20,870 --> 01:12:22,550
And to me, that's the coolest thing that they've done.

1006
01:12:22,670 --> 01:12:24,050
Like, I think that's absolutely amazing.

1007
01:12:24,290 --> 01:12:25,910
And it was like a stealth launch

1008
01:12:25,910 --> 01:12:28,230
that I think Miles first announced on Twitter

1009
01:12:28,230 --> 01:12:29,190
almost accidentally.

1010
01:12:29,790 --> 01:12:32,630
Like, it is really cool how this has gone so quick.

1011
01:12:32,910 --> 01:12:35,390
They're shipping a crazy amount at the moment over a block.

1012
01:12:35,510 --> 01:12:36,850
Like, they're doing so much cool stuff.

1013
01:12:37,110 --> 01:12:38,290
Yeah, they've been shipping like crazy.

1014
01:12:38,290 --> 01:12:44,090
it feels like miles kind of won the internet last week he was just like on there like this was his

1015
01:12:44,090 --> 01:12:49,270
this was his week and uh it was awesome to see like my hat's off to all of them over there i

1016
01:12:49,270 --> 01:12:52,830
know it's just not him it's a whole bunch of people working on it but that's been great to see

1017
01:12:52,830 --> 01:12:59,370
um the last things is i would love for people to just kind of think about how they're using

1018
01:12:59,370 --> 01:13:05,150
using ai um and kind of picture it as when you're using some of these systems like chat gpt

1019
01:13:05,150 --> 01:13:08,470
that you have another person sitting in the room,

1020
01:13:09,430 --> 01:13:11,830
excuse me, you have another person sitting in the room

1021
01:13:11,830 --> 01:13:13,570
kind of watching everything that you do

1022
01:13:13,570 --> 01:13:17,170
and they are approving or rejecting what you do and say

1023
01:13:17,170 --> 01:13:18,770
and they're making copies of it.

1024
01:13:19,130 --> 01:13:21,610
So just kind of have that in your mind as you're using it.

1025
01:13:22,030 --> 01:13:24,090
And then I would love for people to sign up for Maple.

1026
01:13:24,290 --> 01:13:25,570
They can go to trymaple.ai.

1027
01:13:25,970 --> 01:13:27,770
They don't have to stop using ChatGPT,

1028
01:13:27,770 --> 01:13:29,770
just add Maple into your toolbox

1029
01:13:29,770 --> 01:13:32,290
and then use it for things.

1030
01:13:32,950 --> 01:13:34,570
And you'll start to see like,

1031
01:13:34,570 --> 01:13:37,930
hey, that person that was sitting in the room with me, listening to everything that I'm doing

1032
01:13:37,930 --> 01:13:43,250
with my AI, that's not there in Maple. And start to notice how you use that differently and how

1033
01:13:43,250 --> 01:13:49,070
maybe you're more free to speak. And that's really our whole thought around Maple.

1034
01:13:49,590 --> 01:13:56,570
Maple is the AI that allows you to think freely. And what I mean by that is just there's nobody

1035
01:13:56,570 --> 01:14:02,030
who is trying to get in your way. And there's nobody that's going to hold you accountable for

1036
01:14:02,030 --> 01:14:06,970
anything you say because when we think in our mind we think all sorts of things and they're

1037
01:14:06,970 --> 01:14:12,490
and that's that's how we're supposed to work and so we if we're going to use this ai tool to help

1038
01:14:12,490 --> 01:14:17,510
us think we shouldn't have some intermediary in between telling us no that's not okay to think

1039
01:14:17,510 --> 01:14:22,530
that way um we should be thinking freely yeah so i mean people should definitely go check out maple

1040
01:14:22,530 --> 01:14:28,830
and i'm actually like this has made me reconsider how i'm using ai i'm gonna use maple more and more

1041
01:14:28,830 --> 01:14:31,710
because I think a big part of the reason

1042
01:14:31,710 --> 01:14:34,170
that I'm kind of stuck on ChatGPT is just habit.

1043
01:14:34,290 --> 01:14:36,430
So I'm going to try and break that habit and use Maple.

1044
01:14:36,710 --> 01:14:38,170
So this week I'm going to be Maple only

1045
01:14:38,170 --> 01:14:39,010
and we'll see how it goes.

1046
01:14:39,250 --> 01:14:39,370
Okay.

1047
01:14:39,690 --> 01:14:41,010
I'll give you some feedback.

1048
01:14:41,550 --> 01:14:42,070
Oh, sorry.

1049
01:14:42,490 --> 01:14:43,490
What I want from you is like,

1050
01:14:43,570 --> 01:14:44,730
tell me the features that you're like,

1051
01:14:44,790 --> 01:14:46,690
I must have that in Maple.

1052
01:14:47,490 --> 01:14:48,670
That's really useful from people

1053
01:14:48,670 --> 01:14:49,590
so we can start building that.

1054
01:14:50,250 --> 01:14:52,130
I do think for me, the big one will be memory.

1055
01:14:52,290 --> 01:14:54,010
If we get memory, like that's awesome

1056
01:14:54,010 --> 01:14:55,990
because like now it's no longer sort of

1057
01:14:55,990 --> 01:14:58,690
restrained by the training data

1058
01:14:58,690 --> 01:15:00,010
that you can actually search the web.

1059
01:15:00,090 --> 01:15:00,830
Like that's huge.

1060
01:15:01,110 --> 01:15:02,070
And if it then had memory,

1061
01:15:02,150 --> 01:15:03,730
I think that's kind of everything I want really.

1062
01:15:04,310 --> 01:15:05,450
Talking to it would be great,

1063
01:15:05,570 --> 01:15:07,290
but I only do that really occasionally anyway.

1064
01:15:07,450 --> 01:15:10,310
But if it knew a bit more about me in a private way,

1065
01:15:10,450 --> 01:15:12,650
that's basically everything I need.

1066
01:15:13,310 --> 01:15:13,710
That's awesome.

1067
01:15:14,550 --> 01:15:15,230
Let's go.

1068
01:15:15,390 --> 01:15:15,690
All right.

1069
01:15:15,850 --> 01:15:16,410
Thank you, Mark.

1070
01:15:16,470 --> 01:15:17,530
And where do people find you on Twitter

1071
01:15:17,530 --> 01:15:18,290
if they want to follow you?

1072
01:15:18,770 --> 01:15:21,090
Oh yeah, it's just my name, Mark Suman on Twitter.

1073
01:15:21,090 --> 01:15:22,830
And then I'm on Primal Noster.

1074
01:15:23,370 --> 01:15:24,910
I'm just marks at primal.net.

1075
01:15:25,170 --> 01:15:25,810
You can find me there.

1076
01:15:26,690 --> 01:15:27,050
Awesome.

1077
01:15:27,250 --> 01:15:27,730
Thank you, Mark.

1078
01:15:27,770 --> 01:15:28,330
This has been great.

1079
01:15:28,330 --> 01:15:29,990
and I'll hopefully see you in Austin soon

1080
01:15:29,990 --> 01:15:31,330
let's do it