1
00:00:00,000 --> 00:00:09,140
All right. What's up, what's up, what's up, Freedom Tech. Welcome to Freedom Tech weekend.

2
00:00:09,320 --> 00:00:18,160
My name is Marks. I am here every Friday at 1530 UTC. Sometimes I'm right on time. Sometimes I'm

3
00:00:18,160 --> 00:00:22,720
a little late, a few minutes late, just as I continue to get things figured out here.

4
00:00:22,900 --> 00:00:28,460
I'm also running a company. I run OpenSecret, which I happen to be wearing an OpenSecret shirt

5
00:00:28,460 --> 00:00:34,280
today. Uh, just a coincidence, but, um, open secret, we make maple AI, which is actually

6
00:00:34,280 --> 00:00:41,380
going to be the focus of today is maple AI. Um, I have not wanted to do an actual episode on maple

7
00:00:41,380 --> 00:00:47,540
just because maple is part of every show as a tool. It's on screen being used to help run the

8
00:00:47,540 --> 00:00:52,780
show every day. I don't need these headphones on anymore. Let's take these off here. Pardon me.

9
00:00:52,780 --> 00:00:57,820
I'm just chatting a little bit anyways right now as people join the stream.

10
00:00:59,100 --> 00:01:05,900
Okay, and checking. It's working. All right, great. So yeah, I have one. I've been avoiding

11
00:01:05,900 --> 00:01:12,560
doing an episode specifically about Maple because I am a Freedom Tech developer as well. Me and my

12
00:01:12,560 --> 00:01:19,020
co-founder, we build the most private cloud AI that you can find out there on the market right

13
00:01:19,020 --> 00:01:27,380
now. And we are trying to enable individuals, businesses, activists, journalists, organizations

14
00:01:27,380 --> 00:01:32,020
around the world that are trying to fight against oppression in their countries. We're trying to

15
00:01:32,020 --> 00:01:38,420
enable and build these tools that people can use to have some semblance of privacy and sovereignty

16
00:01:38,420 --> 00:01:45,040
as they start to embrace AI and use AI to help make them more productive. So that's what we're

17
00:01:45,040 --> 00:01:48,340
going to actually talk about today, but we're going to talk about some other AI tools as well.

18
00:01:48,340 --> 00:01:55,420
It's not just going to be Maple. I want to show off. There are a couple of local AI tools that

19
00:01:55,420 --> 00:02:00,320
I like to use. There are so many other ones that you can just keep playing with them, like just

20
00:02:00,320 --> 00:02:06,280
try out new tools and see how they go. But I'll show you a couple just to help get you going down

21
00:02:06,280 --> 00:02:11,420
the path if you aren't already going down that path on your own. All right. We are working live

22
00:02:11,420 --> 00:02:16,040
on YouTube. That's great. We have a few people already in here. Welcome. I'm going to open up

23
00:02:16,040 --> 00:02:20,060
the chat as well. Feel free to drop in the chat. If you have any AI tools that you like using,

24
00:02:20,860 --> 00:02:25,060
I would love to know about them and we can even show them off here a little bit as well.

25
00:02:25,920 --> 00:02:32,640
And then if you also have just like concerns about privacy with AI or any thoughts on the topic,

26
00:02:32,860 --> 00:02:39,440
drop them in, drop questions, drop concerns, and we can have a chat about them. So that is great.

27
00:02:39,440 --> 00:02:45,740
Welcome everybody. I like to give just a few minutes for people to join and kind of get into

28
00:02:45,740 --> 00:02:52,540
get into the room if you will. I'm going to make sure that this is live on X.

29
00:02:53,320 --> 00:02:58,000
So let me go over here and just click on that. I got the little live circle around my avatar.

30
00:02:58,380 --> 00:03:04,800
That's great. Sweet. It is streaming and then I'm going to go retweet this.

31
00:03:06,380 --> 00:03:11,160
Normally I don't have Maple retweet my Freedom Tech Weekend stuff but since we are talking about

32
00:03:11,160 --> 00:03:21,900
maple today. I'm going to retweet it somehow. Let's see. Come in here.

33
00:03:21,900 --> 00:03:38,460
here we go quote okay come join to learn about privacy

34
00:03:38,460 --> 00:03:44,700
let's see chat gpt data retention problems

35
00:03:44,700 --> 00:03:56,060
and how you can use private AI tools like Maple

36
00:03:56,060 --> 00:04:02,540
to protect yourself.

37
00:04:04,680 --> 00:04:05,600
Sweet.

38
00:04:05,600 --> 00:04:18,040
okay you are the faithful you are here right now already in the stream so welcome we've got 11

39
00:04:18,040 --> 00:04:23,720
people now welcome everybody um just quick recap drop things in the chat we're going to be talking

40
00:04:23,720 --> 00:04:29,100
about the chat gpt data retention problems and kind of break down the story and go through all

41
00:04:29,100 --> 00:04:33,040
the drama that has been breaking this week it's actually been brewing for a few weeks

42
00:04:33,040 --> 00:04:39,380
really for a long time, but really coming out in the last few weeks. And then we're going to look

43
00:04:39,380 --> 00:04:43,460
at some solutions. What are some local AI solutions that we can use? And then what are

44
00:04:43,460 --> 00:04:49,940
some hosted solutions that we can use? So I know that a lot of people who watch this are also

45
00:04:49,940 --> 00:04:53,720
freedom tech fanatics. So I'm sure that you've already figured out some AI tools that work for

46
00:04:53,720 --> 00:05:01,220
you. So drop them in the chat and we will check them out and talk about them. I'm going to make

47
00:05:01,220 --> 00:05:05,460
sure that I have zap stream open so I can see the zap stream chat. It does not come into my tool

48
00:05:05,460 --> 00:05:09,520
like everything else does. So I just don't want, I don't want to miss any of the chats over there.

49
00:05:09,900 --> 00:05:16,580
Great. Okay. I got zap.stream set up. Let me move this over here and we are going to start sharing

50
00:05:16,580 --> 00:05:26,140
screen and jump right in. Okay. So hit the screen share here. We're going to share my home server

51
00:05:26,140 --> 00:05:32,540
screen that I've got right there. Last week I kept getting booted off. Hopefully it doesn't

52
00:05:32,540 --> 00:05:39,700
happen this time. I don't like this window setup. That is dumb. So let's go like this,

53
00:05:40,700 --> 00:05:52,060
like that, and then change the size. Perfect. Okay. Looking good. Like I should.

54
00:05:52,060 --> 00:05:58,600
went and took a walk around the neighborhood. Okay. So here we are. Let's go in. What is all

55
00:05:58,600 --> 00:06:05,640
this drama about? So we have ChatGPT, right? A lot of people use them. They are the biggest,

56
00:06:05,780 --> 00:06:11,940
most well-known AI company out there. There are others that compete with them. Claude being

57
00:06:11,940 --> 00:06:16,660
probably the second biggest one and then Grok, right? So in my mind, those are the big three.

58
00:06:16,660 --> 00:06:23,400
ChatGPT, Claude and Grok. Each one of them are hosting the servers. They're hosting your data.

59
00:06:23,580 --> 00:06:28,260
They're hosting the models. So everything that you do and say goes through their servers

60
00:06:28,260 --> 00:06:33,840
and is processed by them. All three of those companies train their own models.

61
00:06:34,540 --> 00:06:40,780
And so they need data to train off of, oh, I guess I, there's meta also, right? Facebook,

62
00:06:40,980 --> 00:06:44,740
let's not leave them out because they are huge and massive as well. But they're going down a

63
00:06:44,740 --> 00:06:50,300
slightly different path of the open source route, but they also gather data and host data. They have

64
00:06:50,300 --> 00:06:55,160
all of the Facebook and Instagram and WhatsApp data. So they have lots of stuff that they train

65
00:06:55,160 --> 00:07:00,640
on as well. And Mark Zuckerberg has been very open about this. I was in the room when they had

66
00:07:00,640 --> 00:07:07,920
LamaCon back in April and got invited out there because we build on top of Lama as one of our

67
00:07:07,920 --> 00:07:13,540
options. And so I was in the room when Mark basically told us all that we have all this

68
00:07:13,540 --> 00:07:17,380
data from Facebook, Instagram, and WhatsApp, and we're not going to give it to anybody else.

69
00:07:17,380 --> 00:07:22,100
And we're just going to train the heck out of it. Um, so that's what you're doing when you

70
00:07:22,100 --> 00:07:27,680
use those platforms, your training models, all of these platforms, all four of these,

71
00:07:27,720 --> 00:07:33,640
you're training, helping them to train their models. Um, and yeah, I'll get to that in a

72
00:07:33,640 --> 00:07:40,100
little bit. So there's really a couple of ways to look at, look at companies, right. Um,

73
00:07:40,100 --> 00:07:46,900
is that you have the big four that are training these big beefy models and grabbing as much data

74
00:07:46,900 --> 00:07:51,920
as they can. And then you have all these other AI companies that are building on top of those tools.

75
00:07:52,220 --> 00:07:56,140
A lot of people call those wrappers, right? We call these wrappers in the space where you're

76
00:07:56,140 --> 00:08:05,860
basically wrapping a user interface around these big services and these services expose APIs so

77
00:08:05,860 --> 00:08:10,840
that your app doesn't have to train its own models. You don't have to run your own models.

78
00:08:10,940 --> 00:08:18,200
You can just call through to chat GPT or Claude and use their stuff. So a big example of this is

79
00:08:18,200 --> 00:08:25,960
cursor, right? Maybe you've heard of cursor. It's one of the big vibe, vibe coding tools. If you're

80
00:08:25,960 --> 00:08:32,060
not familiar with vibe coding, it basically means using AI to write code. And so a lot of people

81
00:08:32,060 --> 00:08:37,820
download cursor and they run it on their computer but cursor actually calls through to the clod

82
00:08:38,460 --> 00:08:47,820
api and uses that then you either pay clod for a premium pro subscription in order to keep using

83
00:08:47,820 --> 00:08:53,260
it because you hit a cap on your free tier or you can switch cursor into a local mode where it uses

84
00:08:53,260 --> 00:08:59,340
a local ai that is not as powerful right so that's one example of a wrapper okay there's a lot of

85
00:08:59,340 --> 00:09:06,320
of these. So what's, uh, what's the big, um, the drama here? Uh, we'll get to that in just one

86
00:09:06,320 --> 00:09:11,800
second. Sorry. I want to show you one more thing. And that is, let's see, can I get to my channel?

87
00:09:15,380 --> 00:09:20,160
Hey, I got on the first try. Sweet. And look, we're streaming. So I just spoke last week at

88
00:09:20,160 --> 00:09:25,000
the Oslo freedom forum, basically on this topic. I didn't realize, I know that some of these

89
00:09:25,000 --> 00:09:29,700
articles were published, but I didn't realize at the time that all of this was going to explode

90
00:09:29,700 --> 00:09:33,920
that that happened with chat GPT and their data retention. I already knew that this stuff was

91
00:09:33,920 --> 00:09:38,860
going on and the writings on the wall, like their incentive is to capture as much data as you can.

92
00:09:39,000 --> 00:09:44,140
I recommend people go check this out. I normally don't like shill my own speeches and talks and

93
00:09:44,140 --> 00:09:50,160
stuff, but, um, gave a talk a lot about, um, freedom of speech, but then freedom of thought

94
00:09:50,160 --> 00:09:56,240
and how we have fought for centuries over freedom of speech. And that, that battle is still going on

95
00:09:56,240 --> 00:10:00,700
that we have not won the freedom of speech, but I mean, we, we obviously have a lot of freedoms.

96
00:10:00,700 --> 00:10:04,500
I'm speaking to you today. I don't give a shit about which country you're in. Like, I don't,

97
00:10:04,560 --> 00:10:08,420
I don't care. You can come from any country you want to, you can speak any language you want to.

98
00:10:08,520 --> 00:10:14,940
AI can help translate this for you, whatever. But, um, I I'm speaking to you and I hope that you can

99
00:10:14,940 --> 00:10:18,960
consume this content. I hope that you have a voice where you can speak. The new battle that I see is

100
00:10:18,960 --> 00:10:25,040
over freedom of thought. And that is, we have all these AI products now that are gathering our data

101
00:10:25,040 --> 00:10:31,260
and then are helping us to think, right? And so I could just type into, you know, to Maple and ask

102
00:10:31,260 --> 00:10:39,860
it to do something, right? Help me come up with a business plan for a lemonade stand. Okay.

103
00:10:40,800 --> 00:10:46,720
This is going to help me think, right? Now, Maple is not recording this. It's all encrypted. It's

104
00:10:46,720 --> 00:10:51,400
private and we'll get to that. Right. But this is what I mean. So we're, we're asking it to help us

105
00:10:51,400 --> 00:10:56,080
think we still have to read this. We still have to think critically ourselves, but with these other

106
00:10:56,080 --> 00:11:03,420
platforms, Chad GPT, Claude, Grock, they are turning on these long-term memory things where

107
00:11:03,420 --> 00:11:10,080
they will start to remember stuff for us forever intentionally. And then as we talk to the AI,

108
00:11:10,080 --> 00:11:17,060
we are going to become more dependent on it to kind of be a digital brain for us. So we have our

109
00:11:17,060 --> 00:11:22,560
meat space, our physical brain, and then we'll have this digital brain. And what's kind of this

110
00:11:22,560 --> 00:11:31,520
more crazy, futuristic, dystopian thought is that they are the ones hosting our brain for us,

111
00:11:32,300 --> 00:11:38,140
hosting our long-term memory. And they actually have the ability to alter that memory in any way

112
00:11:38,140 --> 00:11:42,400
that they want to. Let me jump over spoilers. I'm just giving you spoilers all over the place here.

113
00:11:42,620 --> 00:11:46,260
So let's jump over to censorship, right? There are multiple layers where they can censor us.

114
00:11:47,540 --> 00:11:54,200
And I'm focusing right now on memory where they could go in and they could delete stuff from our

115
00:11:54,200 --> 00:11:58,600
memory. Like if we are, if we're just having a thought about something that's very controversial,

116
00:11:58,600 --> 00:12:05,680
as we do, every human has thoughts that are really dark, scary, negative, bad. We would never share

117
00:12:05,680 --> 00:12:11,100
with anybody because that's how our brain processes the real world. We take input and we,

118
00:12:11,220 --> 00:12:18,520
um, we have to like game play out some of these scenarios to understand what is right and what is

119
00:12:18,520 --> 00:12:26,520
wrong. And, um, you know, 99.9% of these thoughts, these negative, or these kind of like immoral

120
00:12:26,520 --> 00:12:30,860
thoughts we have, we don't act upon right now. Some people act upon more of them than others.

121
00:12:30,860 --> 00:12:35,560
and those people usually end up in prison or worse.

122
00:12:37,000 --> 00:12:40,300
But us normal, happy, contributing members of society,

123
00:12:40,440 --> 00:12:41,280
we still have these thoughts.

124
00:12:41,500 --> 00:12:43,140
So as we start to chat with AI,

125
00:12:43,320 --> 00:12:45,340
we need to be in a safe, trusted space

126
00:12:45,340 --> 00:12:47,460
where we can debate about these things

127
00:12:47,460 --> 00:12:50,000
and really talk to an AI and really push it on it.

128
00:12:50,600 --> 00:12:54,420
And these things are gonna end up in our long-term memory.

129
00:12:55,540 --> 00:12:57,200
And as they end up in our long-term memory,

130
00:12:57,620 --> 00:13:00,020
are these platforms going to censor us?

131
00:13:00,020 --> 00:13:03,280
Are they going to remove these things from our memory or even worse?

132
00:13:03,280 --> 00:13:06,100
Are they going to add stuff into our memory that we aren't aware of?

133
00:13:06,600 --> 00:13:07,940
We have zero insight into this.

134
00:13:07,940 --> 00:13:10,280
We don't know what is being done to that.

135
00:13:10,580 --> 00:13:16,920
And so as we depend on this thinking, oh yes, this is our digital memory for us.

136
00:13:17,240 --> 00:13:18,380
And then we start asking questions.

137
00:13:18,380 --> 00:13:20,940
We're going to expect things that we actually thought about to be in there.

138
00:13:21,360 --> 00:13:25,820
And if they're not in there or if they've been altered, that is going to change the outcome

139
00:13:25,820 --> 00:13:34,500
of our AI chats and basically effectively change our thought process. And we start to lose our

140
00:13:34,500 --> 00:13:39,380
freedom of thought. Okay. So that is the backdrop here. I just wanted to set a backdrop for you

141
00:13:39,380 --> 00:13:46,660
on this entire conversation and where I see this. I call this censorship, subconscious censorship,

142
00:13:46,660 --> 00:13:52,340
because they are going to be, they want us to be dependent on them as a digital brain.

143
00:13:52,340 --> 00:13:57,340
and then they are going to be the keepers of what our brain has inside of it.

144
00:13:58,640 --> 00:14:01,000
So just want that to kind of sink in for a second,

145
00:14:01,140 --> 00:14:05,180
because that is where this whole thing starts to come into play.

146
00:14:05,900 --> 00:14:07,420
So I would recommend you go check that out.

147
00:14:07,500 --> 00:14:12,000
I think my talk will actually be published on the Oslo Freedom Forum channel as well.

148
00:14:12,060 --> 00:14:14,460
They're starting to chop those up, but I went ahead and chopped it up myself

149
00:14:14,460 --> 00:14:16,000
because I wanted to get it out there as well.

150
00:14:16,000 --> 00:14:30,100
So yesterday, this account on X, they've got 66,000 followers, posted a screenshot of a court order and an article about a court order for OpenAI.

151
00:14:30,360 --> 00:14:34,600
So OpenAI has been embroiled in a lawsuit here with the New York Times.

152
00:14:34,600 --> 00:14:47,640
New York Times is suing them for copyright infringement, saying that OpenAI is hoovering up, taking up all of the data, all the written works of the world, including New York Times, all of their copyrighted work.

153
00:14:48,560 --> 00:14:50,340
Think what you will about New York Times.

154
00:14:50,480 --> 00:14:58,120
I have my opinions about the New York Times and the validity and accuracy of a lot of their content.

155
00:14:58,480 --> 00:14:58,960
But whatever.

156
00:14:59,620 --> 00:15:01,600
That's not exactly here or there.

157
00:15:01,600 --> 00:15:04,180
They are taking all of this.

158
00:15:04,180 --> 00:15:15,304
And then when you start to chat with AI right Like when I just had this in here is this considered copyrighted content Am I reproducing the New York times

159
00:15:15,464 --> 00:15:19,904
So if I were to read a page from the New York times, a digital page,

160
00:15:19,944 --> 00:15:21,264
nobody reads the paper anymore.

161
00:15:21,264 --> 00:15:22,624
I kind of miss it, but nobody does.

162
00:15:23,024 --> 00:15:26,784
If I were to, um, take an article from the New York times and then, you know,

163
00:15:26,784 --> 00:15:29,864
say, Hey, write me, no, not using AI.

164
00:15:30,004 --> 00:15:32,504
If I just took it and just like typed it out word for word, but

165
00:15:32,504 --> 00:15:37,724
changed a few words here and there and then published it, I would be running afoul of copyright.

166
00:15:38,364 --> 00:15:43,624
Okay. And I don't know the exact threshold. I'm not a lawyer, but I don't know the exact threshold

167
00:15:43,624 --> 00:15:48,964
of how much I need to alter in order to be fair use, that kind of stuff. But there is some point

168
00:15:48,964 --> 00:15:57,784
where I have to do that. And maybe I should learn that. Let's ask how much of a piece of content

169
00:15:57,784 --> 00:16:14,224
do I need to alter in order to fall under the fair use copyright protection? Let's ask Maple,

170
00:16:14,824 --> 00:16:19,884
because I don't actually know the answer to that. But I would be running into problems.

171
00:16:19,884 --> 00:16:26,144
And so where it comes with AI is if I ask AI, write me an article in the voice of the, you know,

172
00:16:26,144 --> 00:16:32,564
that I was, say I was publishing to the New York Times and this, this author, this journalist,

173
00:16:33,884 --> 00:16:40,044
write me an article like that, right? It starts to look very similar to the New York Times,

174
00:16:40,044 --> 00:16:45,564
but I am not actually copying the New York Times. So where are these lines? All right, let's look.

175
00:16:46,444 --> 00:16:50,064
Fair use is a doctrine that permits limited use of copyright material without obtaining permission

176
00:16:50,064 --> 00:16:54,204
from the copyright holder. To determine whether your use of copyright material falls under fair

177
00:16:54,204 --> 00:16:59,884
use courts, consider the four factors, following four factors, purpose and character of the use.

178
00:16:59,884 --> 00:17:04,524
Is your use for commercial or non-commercial purposes? Is it transformative, meaning you're

179
00:17:04,524 --> 00:17:10,044
using the material in a new or innovative way, or is it merely reproductive? Number two, the nature

180
00:17:10,044 --> 00:17:14,244
of the copyrighted work. Is the original work a factual or fictional piece? Is it published or

181
00:17:14,244 --> 00:17:21,104
unpublished? Three, amount and substantiality of the portion used. How much of the original work

182
00:17:21,104 --> 00:17:26,524
are you using and is the portion use significant to the original work? Number four, effect of the

183
00:17:26,524 --> 00:17:33,504
use on the market for the original work. Does your use harm or help the market for the original work?

184
00:17:34,164 --> 00:17:42,844
So pretty vague. Let's see if it gets into, let's see, short answer.

185
00:17:42,844 --> 00:17:50,184
how much can I use of the original?

186
00:17:54,344 --> 00:17:57,784
I think it's in there, but let's just ask this instead.

187
00:17:59,604 --> 00:18:04,984
There's no specific percentage, but a rough guideline is one to two sentences or 10 to 20%

188
00:18:04,984 --> 00:18:09,604
of the original work for images, 10 to 20% for music, five to 10 seconds for video, 10 to 30

189
00:18:09,604 --> 00:18:15,344
seconds of a clip. Okay. So those are just kind of general guidelines. So New York times is

190
00:18:15,344 --> 00:18:22,344
complaining that chat GPT is effectively using much more than that. And that you can reproduce

191
00:18:22,344 --> 00:18:26,064
the New York times. If you wanted to, we could create our own website. I'm not going to encourage

192
00:18:26,064 --> 00:18:30,104
you to do this, but you could create a website that spoofs the New York times regenerates all

193
00:18:30,104 --> 00:18:34,444
their articles, but maybe you feel like they aren't factual enough. So you're going to make

194
00:18:34,444 --> 00:18:41,404
them more factual who cares whatever that's the that's the gist of this lawsuit so a court order

195
00:18:41,404 --> 00:18:50,464
came down saying open ai now has to start retaining all of this user data so that it can be examined

196
00:18:50,464 --> 00:19:00,404
and looked at and i this is where i'm not 100 clear but i think it's because they want to

197
00:19:00,404 --> 00:19:06,484
effectively audit later on for lawsuit purposes to see if copyright infringement is is being

198
00:19:06,484 --> 00:19:13,084
breached right um so this caused a lot of drama um

199
00:19:13,084 --> 00:19:20,784
this article is from six days ago deleted or not court order and open ai silence leave users in the

200
00:19:20,784 --> 00:19:27,304
dark about how chat gpt user data is handled and then ars technica wrote an article two days ago

201
00:19:27,304 --> 00:19:33,784
openai slams court order to save all chat gpt logs included including deleted chats right so a lot of

202
00:19:33,784 --> 00:19:39,544
uh really long articles here but the the gist of it is that there are three things which article had

203
00:19:39,544 --> 00:19:46,744
it um sorry one of them had just like a nice bullet point list and i'm trying to remember which one it

204
00:19:46,744 --> 00:19:51,704
was let me search the page

205
00:19:57,304 --> 00:20:03,984
Okay. Well, what it came down to the three that I've seen discussed the most are deleted chats.

206
00:20:04,144 --> 00:20:07,584
So you as a user go in and delete a chat that you don't want to have stored anymore.

207
00:20:09,524 --> 00:20:16,424
Those now have to be retained. Temporary chats. This is, this is like private window on your

208
00:20:16,424 --> 00:20:22,104
browser, right? Incognito mode on your browser. But with AI, you say, don't remember this chat.

209
00:20:22,184 --> 00:20:26,504
I just want to talk about it right now. And as soon as I close this chat, you can forget about it.

210
00:20:26,504 --> 00:20:34,204
okay so that's the other one they have to retain those now and then the third one is api usage

211
00:20:34,204 --> 00:20:40,624
this is massive the api usage part because there are all these apps out there that use the chat

212
00:20:40,624 --> 00:20:44,144
gpt you know remember at the very beginning of the stream we talked about the wrappers

213
00:20:44,144 --> 00:20:50,484
right not the little waynes little waynes right here no i'm not talking about that kind of wrapper

214
00:20:50,484 --> 00:20:55,784
i'm talking about these uh wrapping paper right these apps out there that are ai apps that just

215
00:20:55,784 --> 00:21:02,804
call through to open AI, Claude, I don't think Grok has an API, um, meta, right?

216
00:21:02,924 --> 00:21:06,344
Some of these other ones, these hosted ones, the hosted deep seek meta and deep

217
00:21:06,344 --> 00:21:08,644
seek are different because they have open source versions.

218
00:21:08,864 --> 00:21:13,964
We are using the open source versions of, um, both llama, which is meta and deep

219
00:21:13,964 --> 00:21:17,344
seek, but, uh, they have hosted ones as well.

220
00:21:17,344 --> 00:21:19,904
And when you use the hosted ones, you are giving them your data.

221
00:21:20,324 --> 00:21:21,164
Just remember that.

222
00:21:21,984 --> 00:21:22,364
Okay.

223
00:21:23,184 --> 00:21:30,864
So these rappers use these APIs and a lot of these tools, um, I've met some of them

224
00:21:30,864 --> 00:21:37,604
when I was out at LamaCon, I spoke with a founder who is doing a, um, it's something

225
00:21:37,604 --> 00:21:40,084
that's kind of adjacent to the medical community with therapy.

226
00:21:40,864 --> 00:21:43,444
And he, he's like, oh, it's cool.

227
00:21:43,444 --> 00:21:48,084
We, we have a zero data retention policy with open AI ZDR.

228
00:21:48,084 --> 00:21:52,784
Those are all the rage right now with AI is zero data retention policies.

229
00:21:53,004 --> 00:21:56,124
So he felt like he was safe because they're not keeping his data.

230
00:21:56,584 --> 00:21:58,984
I'm sorry, but it's all bullshit.

231
00:21:59,584 --> 00:22:06,564
It's just company agreements saying we will not use your data, wink, wink, but it can change

232
00:22:06,564 --> 00:22:07,604
on a whim right here.

233
00:22:07,924 --> 00:22:09,104
This court has ordered them.

234
00:22:09,404 --> 00:22:13,804
Now we can debate whether or not the court should even be allowed to do this.

235
00:22:14,244 --> 00:22:16,684
I don't know that they necessarily should.

236
00:22:16,684 --> 00:22:21,884
I still want to learn more about this. This seems like potentially a big overreach for the courts

237
00:22:21,884 --> 00:22:31,884
to impinge, impinge, to infringe on user privacy. They're just going in now and reaching into user

238
00:22:31,884 --> 00:22:36,344
data and forcing a company to do something that it doesn't want to do, right? So there's a whole

239
00:22:36,344 --> 00:22:40,764
another debate here to be had on whether they can do this or not. However, they are doing it. So

240
00:22:40,764 --> 00:22:48,584
let's just talk about this for the moment. Um, they, uh, let me back up. I lost my train of

241
00:22:48,584 --> 00:22:54,204
thought, but, um, they have to keep all of our data now. Yeah. So the zero data retention policies

242
00:22:54,204 --> 00:22:59,584
can go out the window with a court order. Right. And even that you still don't know if they're

243
00:22:59,584 --> 00:23:04,344
actually not retaining your data. They're just signing an agreement saying we will not retain

244
00:23:04,344 --> 00:23:08,984
your information. And so if you have a therapy app that talks about very sensitive things

245
00:23:08,984 --> 00:23:16,384
and has the names and personal private information of underage children who are dealing with some

246
00:23:16,384 --> 00:23:23,104
really sad situations that you're putting into here, you hope that they're not retaining it.

247
00:23:23,204 --> 00:23:28,124
You hope that they are not, that it's not going to get leaked out in a data breach at some point.

248
00:23:28,124 --> 00:23:32,924
You hope that that information is not being trained. Models are not being trained on top

249
00:23:32,924 --> 00:23:37,804
of that information and that in a future instance, when somebody asks ChatGPT about

250
00:23:37,804 --> 00:23:45,264
a sensitive topic, somehow the name of that child and their very specific situation

251
00:23:45,264 --> 00:23:53,404
is used in an AI chat. You hope that it's not. It shouldn't, but there's nothing really

252
00:23:53,404 --> 00:23:58,224
preventing it from happening, right? So the more that you give this very personal information to

253
00:23:58,224 --> 00:24:07,484
these models, the more that you are joining this public library of data that the entire world can

254
00:24:07,484 --> 00:24:13,884
access. And you have no control over whether or not the AI uses the real names or the fake names,

255
00:24:14,324 --> 00:24:19,344
uses real life situations, or says, these are hypothetical situations that just happen to be

256
00:24:19,344 --> 00:24:25,984
the exact same situation that you typed into ChatGPT six months ago, three years ago. Who knows,

257
00:24:26,084 --> 00:24:31,984
right? I hope you're starting to see the big data privacy problem that we have on our hands here

258
00:24:31,984 --> 00:24:37,044
with all of this. And this was already a problem before this court order. The fact that

259
00:24:37,044 --> 00:24:48,244
that open AI, Claude, Grok, and Meta have all this data means that they can do what they will with it.

260
00:24:49,044 --> 00:24:54,784
And hackers can get in and steal the data. And then there's a middle ground of not even hackers,

261
00:24:55,444 --> 00:25:03,464
but rogue employees, employees who have been bought off to sell this data, to grab this data,

262
00:25:03,464 --> 00:25:16,024
to expose this data, right? I don't know if you follow the big news last month, but Coinbase

263
00:25:16,024 --> 00:25:29,064
had a large breach of user data. What happened here is they leaked a bunch of

264
00:25:29,064 --> 00:25:32,744
of government documents for people.

265
00:25:32,844 --> 00:25:33,944
So when you sign up for Coinbase,

266
00:25:34,024 --> 00:25:37,164
if you are wanting to buy Bitcoin or cryptocurrencies,

267
00:25:37,504 --> 00:25:40,484
like fart coin that they just added, which is so dumb.

268
00:25:41,384 --> 00:25:43,784
But let's say that you are using this platform

269
00:25:43,784 --> 00:25:45,904
to buy Bitcoin and cryptocurrencies.

270
00:25:46,804 --> 00:25:49,124
You have to upload your government ID.

271
00:25:49,344 --> 00:25:50,524
So that could be a driver's license.

272
00:25:50,524 --> 00:25:52,804
That could be a passport or some other government ID.

273
00:25:53,104 --> 00:25:54,764
And they say that they're going to keep it safe.

274
00:25:54,804 --> 00:25:55,884
And I believe them.

275
00:25:55,884 --> 00:26:00,024
They do their best to keep these under lock and key as much as possible.

276
00:26:00,604 --> 00:26:05,944
But due to government regulations, they are required to collect this and hold on to this data, right?

277
00:26:06,324 --> 00:26:07,104
That's another debate.

278
00:26:07,504 --> 00:26:10,264
Is this even something the government should be requiring of them?

279
00:26:10,444 --> 00:26:11,504
I don't think it should be.

280
00:26:11,924 --> 00:26:21,444
I 100% disagree with these laws that require all of these companies to collect our private information in the name of anti-money laundering.

281
00:26:21,444 --> 00:26:27,764
I think it's a fake made up law and rule that should go away.

282
00:26:28,824 --> 00:26:31,164
But right now we have to live within these rules.

283
00:26:31,524 --> 00:26:32,944
So Coinbase is collecting this information.

284
00:26:33,304 --> 00:26:41,424
They had employees in their company that were working with outside third parties to slowly

285
00:26:41,424 --> 00:26:47,384
find ways into the different parts of the company so that they could gain elevated access

286
00:26:47,384 --> 00:26:54,264
and eventually steal a bunch of these government documents and transaction records and a whole

287
00:26:54,264 --> 00:26:59,464
bunch of information, right? You can go read up on this. This is not a Bitcoin cryptocurrency show,

288
00:26:59,684 --> 00:27:06,144
but it's a very real thing. So anytime you give data over to these companies, it is sitting there

289
00:27:06,144 --> 00:27:11,424
as a treasure trove for internal employees or hackers on the outside, or both to collude together

290
00:27:11,424 --> 00:27:13,944
to make this information public.

291
00:27:14,684 --> 00:27:17,844
Okay, what does Chachapiti store?

292
00:27:18,304 --> 00:27:20,424
This is an article from January this year,

293
00:27:20,804 --> 00:27:22,544
just kind of, it looks like it's probably something

294
00:27:22,544 --> 00:27:23,684
they do at the beginning of the year

295
00:27:23,684 --> 00:27:26,164
to rehash what's being stored on the servers.

296
00:27:26,604 --> 00:27:27,904
I think this article is really great.

297
00:27:28,264 --> 00:27:30,204
This website is nightfall.ai.

298
00:27:30,604 --> 00:27:33,944
It could very well be generated, a generated article.

299
00:27:34,124 --> 00:27:35,704
I don't know, but it seems helpful.

300
00:27:36,184 --> 00:27:37,704
You do need to go verify everything, right?

301
00:27:38,164 --> 00:27:40,404
You've always had to verify everything on the internet,

302
00:27:40,404 --> 00:27:45,564
but now that things are generated with AI, you need to verify them even more. But they collect

303
00:27:45,564 --> 00:27:51,844
all sorts of things about you. I'm going to come down here. Why do they do it? They fine-tune their

304
00:27:51,844 --> 00:27:59,284
models. They have human review. Humans are reading your data. Okay. Despite these practices,

305
00:27:59,484 --> 00:28:04,804
OpenAI explicitly states it does not use data for marketing or advertising campaigns or selling

306
00:28:04,804 --> 00:28:10,784
information to third parties without consent. Without consent, right? Well, you most likely

307
00:28:10,784 --> 00:28:15,744
gave consent in that really long end user license agreement they didn't read and that you just

308
00:28:15,744 --> 00:28:19,324
checked that box because you tried to sign in without checking it. And then it was like, hold

309
00:28:19,324 --> 00:28:22,604
on. And a little red thing popped up that said, you need to check this box in order to continue.

310
00:28:22,904 --> 00:28:27,304
You did it and you continued. I don't know that for a fact, but this is usually where they hide

311
00:28:27,304 --> 00:28:35,344
that consent is just in the fine print. Okay. What stays and what gets deleted? Deletion timelines,

312
00:28:35,504 --> 00:28:40,644
user initiated deletion. By the way, this is, like I said, January, this is before the court order.

313
00:28:41,224 --> 00:28:45,724
They delete a chat from the interface 30 days later. So if you delete a chat in the interface,

314
00:28:45,804 --> 00:28:50,504
it's going to hang around for 30 days, presumably so you can like undo it if you want to, but also

315
00:28:50,504 --> 00:28:56,324
so that they can train on it if they want to. Account termination, full account deletion erases

316
00:28:56,324 --> 00:29:01,284
profile data but may retain anonymized interactions for model training unless opted out so if you

317
00:29:01,284 --> 00:29:08,644
didn't go into the settings i don't remember but i'm sure that they're on by default where you're

318
00:29:08,644 --> 00:29:18,244
going to donate your data to be trained excuse me then it's going to make it into their training

319
00:29:18,244 --> 00:29:23,924
practices and then operator exceptions data from openai's operator agent including screenshots and

320
00:29:23,924 --> 00:29:29,964
browsing activity persists for 90 days post-deletion for abuse monitoring. I didn't read this bullet

321
00:29:29,964 --> 00:29:36,984
point earlier. Okay, what is that? Operator is this thing where you can tell it to control your

322
00:29:36,984 --> 00:29:42,644
web browser. I'm smiling right now. If you're on the audio, you can't see me smiling. But like,

323
00:29:42,684 --> 00:29:48,504
this is some dark dystopian stuff that they are building that is really cool and is going to

324
00:29:48,504 --> 00:29:55,844
continue to happen. I think these are powerful tools that we need to have privacy underpinning

325
00:29:55,844 --> 00:30:02,964
them, not the way that they're being done right now on Claude and on ChatGPT. It is taking

326
00:30:02,964 --> 00:30:10,444
screenshots of your screen and your browsing activity. And then even after you delete it,

327
00:30:10,444 --> 00:30:29,148
it going to stay on their servers for 90 days nine zero And they say it for abuse monitoring right That usually means child stuff right It could be other things Maybe it means abuse of their system I don know But a lot of these

328
00:30:29,148 --> 00:30:34,768
privacy sacrifices, right? These compromises we make for our privacy are in the name of things

329
00:30:34,768 --> 00:30:41,388
like terrorism or child abuse. You name it. They pick some really scary things to throw out there

330
00:30:41,388 --> 00:30:46,248
and say, we need you to sacrifice. We need you to turn over your data. We need you to

331
00:30:46,248 --> 00:30:51,328
open up your doors. You're not a criminal, are you? You're not hiding and harboring terrorists

332
00:30:51,328 --> 00:30:56,368
and child abusers. So you have nothing to hide. So just go ahead and give us all your stuff,

333
00:30:56,368 --> 00:31:04,108
right? No, that's not the way this works. We are selective about our privacy. We selectively reveal

334
00:31:04,108 --> 00:31:10,248
what we think is needed at that moment with the interaction, with the transaction that we're

335
00:31:10,248 --> 00:31:14,208
happening with someone. Okay. The relationship we have with somebody, the business transaction we

336
00:31:14,208 --> 00:31:21,148
have with somebody, we need to be in power of revealing that. It is not let us collect everything.

337
00:31:21,148 --> 00:31:25,608
And then even after you tell us to delete it, we're going to hang onto it for 90 more days

338
00:31:25,608 --> 00:31:31,908
in the name of something scary, but really they can do all sorts of other stuff, right? That's

339
00:31:31,908 --> 00:31:36,728
just the reason, but then they can train on it. They can sell it. They can give it to the government.

340
00:31:36,728 --> 00:31:46,368
they can do all sorts of things. Okay. Europe has the GDPR stuff, and there's some shortfalls with

341
00:31:46,368 --> 00:31:55,228
that. Insufficient anonymization. Let's see. 2024 EU audit found that 63% of ChatGPT user data

342
00:31:55,228 --> 00:32:02,088
contained personally identifiable information, PII. If you don't know this term, learn this, PII.

343
00:32:02,088 --> 00:32:06,368
Okay. You're going to, you're going to start seeing it everywhere now that you've heard it,

344
00:32:06,508 --> 00:32:15,768
but it basically means really personal information that is about you. Um, this is huge. Okay. So 63%

345
00:32:15,768 --> 00:32:23,668
of it contained PII with only 22% of users aware that opt out settings existed, right? Most users,

346
00:32:23,668 --> 00:32:27,948
when they open up a piece of software, when they download an app, when they sign up for something

347
00:32:27,948 --> 00:32:34,068
online, they go with the default settings. Okay. The, the software developer and the vibe coder,

348
00:32:34,168 --> 00:32:39,288
whoever is behind building that piece of software, they put their opinions in there. They set the

349
00:32:39,288 --> 00:32:43,148
defaults and a majority of the users are just going to go with the defaults and never change

350
00:32:43,148 --> 00:32:47,828
them. Even if the settings are right in their face, they trust that the, that the developer

351
00:32:47,828 --> 00:32:53,548
knows what to do. And, uh, with reason, right? The developer has put a lot of effort into this,

352
00:32:53,548 --> 00:32:58,248
trying to maximize user experience because they want users to be happy with the product and find

353
00:32:58,248 --> 00:33:04,208
value in the product. And so they set the defaults to be the quote best thing for the users. Well,

354
00:33:04,828 --> 00:33:10,108
there are other defaults that are set that maybe are not the best for the users, but are the best

355
00:33:10,108 --> 00:33:17,248
for the company. And that are things that those are things like data retention. So I'm sorry,

356
00:33:17,248 --> 00:33:21,848
this is going long. Hell, we've already been going, we're over 30 minutes. Dang. Okay.

357
00:33:21,848 --> 00:33:28,288
I haven't even gotten into Maple and local AI. Might have to make this a two-part thing, or

358
00:33:28,288 --> 00:33:32,988
I'm going to keep going as long as I can, but we'll just keep on going as long as we have people.

359
00:33:33,048 --> 00:33:40,148
We've got 51 people here. Welcome. Welcome to the stream. We are going over the ChatGPT data

360
00:33:40,148 --> 00:33:45,028
privacy problems that exist, and then we were going to talk about some AI things that you can do

361
00:33:45,028 --> 00:33:49,748
for yourself to have privacy with AI. Okay, so there are all sorts of opt-out things. They bring

362
00:33:49,748 --> 00:33:55,408
up temporary chats that, um, that won't record it. Activate in temporary chat mode prevents

363
00:33:55,408 --> 00:34:00,408
interactions from appearing in history and delete some after 30 days, though they may still train

364
00:34:00,408 --> 00:34:05,968
models. If the opt out is inactive. I didn't even know this part. I thought temporary chat legit

365
00:34:05,968 --> 00:34:13,228
meant that it wasn't saved on the servers. It was saved for 30 days even before this court order.

366
00:34:13,348 --> 00:34:18,228
Okay. So sorry, but if you ever had a temporary chat where you were like, I'm incognito now,

367
00:34:18,228 --> 00:34:24,068
I'm going to say all of these really weird, crazy things that I don't want anybody to ever know about

368
00:34:24,068 --> 00:34:29,808
me. Sorry, those were already being saved for 30 days and they were training the models off of it

369
00:34:29,808 --> 00:34:35,528
unless you opted out of that. Okay. Now the court order is having them retain them indefinitely

370
00:34:35,528 --> 00:34:44,288
beyond the 30 days. Enterprise users can have end-to-end encryption for all interactions. So

371
00:34:44,288 --> 00:34:50,308
if you are a large corporation and want to pay a ton of money to open AI, then you can encrypt

372
00:34:50,308 --> 00:34:56,268
your stuff end to end. I don't know the specifics about that, but there is still some trust involved

373
00:34:56,268 --> 00:35:04,548
that open AI is how they're running the servers, but this is likely pretty good. You just have to

374
00:35:04,548 --> 00:35:11,608
pay a lot of money for it. And then you can also set custom retention policies. If you are a rich

375
00:35:11,608 --> 00:35:15,968
person, right? A wealthy company. There was something else in here that I read through

376
00:35:15,968 --> 00:35:22,268
earlier that I don't remember what I wanted to call out, but it had to do with enterprise users.

377
00:35:22,888 --> 00:35:29,388
Free, let me search for the word free. That might be my keyword to go off of. Okay. Yes.

378
00:35:30,068 --> 00:35:35,308
Indefinite storage for non-enterprise users. A pivotal 2024 policy change removed the ability

379
00:35:35,308 --> 00:35:41,308
for free and plus users to disable chat history, meaning, and then this is all bolded, meaning all

380
00:35:41,308 --> 00:35:47,408
prompts and interactions are retained indefinitely unless manually deleted enterprise and team

381
00:35:47,408 --> 00:35:53,868
subscribers retain this opt-out capability with data purged after 30 days so sorry if you're a

382
00:35:53,868 --> 00:36:00,228
free user or even just a pro user your shit is on their servers and it was being retained

383
00:36:00,228 --> 00:36:06,308
indefinitely already before the courts ordered them to retain even more information okay i hope

384
00:36:06,308 --> 00:36:12,788
with sinking in. There are more articles that talk about this. Yesterday was a very busy day

385
00:36:12,788 --> 00:36:19,468
on X and on Noster for me because this was going all over the place. This guy right here is the

386
00:36:19,468 --> 00:36:23,928
co-founder and CEO of Hugging Face, which is one of the biggest platforms for sharing AI models.

387
00:36:25,168 --> 00:36:28,608
He was talking about, you know, who's working on AI encryption and privacy.

388
00:36:29,088 --> 00:36:35,028
It's a very important topic in my opinion. I would love it if people jumped into this thread

389
00:36:35,028 --> 00:36:41,588
on X and posted TriMaple AI because we're working on it. I'd love for him to know more about it.

390
00:36:41,988 --> 00:36:46,628
Obviously, I mentioned it and we got a few others in there on their own who did it and asked them to.

391
00:36:46,788 --> 00:36:52,488
Really grateful for that. But he's not the only one. There are multiple big accounts on X asking

392
00:36:52,488 --> 00:36:56,948
this question now. What are we doing about data privacy? I'm sorry, but like, why was nobody

393
00:36:56,948 --> 00:37:01,648
talking about this before? Like this, to me, this seemed obvious, which is why we started this

394
00:37:01,648 --> 00:37:10,348
company. Um, and so I don't want to say I told you so, but like, come on. Okay. What can we do

395
00:37:10,348 --> 00:37:15,848
about all this enough about the drama? You can go read up more on it. What can we do about this?

396
00:37:16,628 --> 00:37:20,448
We're running low. I don't want to run a long time. We have all the time that we want, but,

397
00:37:20,448 --> 00:37:27,088
um, I try to keep these to half hour because I post this on the podcast stream and then we have

398
00:37:27,088 --> 00:37:34,828
a YouTube playlist and then obviously it's on X and zap.stream. So I want to keep these relatively

399
00:37:34,828 --> 00:37:40,508
the same so that they're digestible. So I'll quickly go into some of the things that we can do.

400
00:37:40,508 --> 00:37:48,308
What can we do? One of the local LLMs that I like to use, local LLM things, is called LM Studio.

401
00:37:49,708 --> 00:37:55,288
Okay, so you have options. You can run all of your, these are the two options really, right?

402
00:37:57,088 --> 00:38:00,588
You can run all of your AI locally, right?

403
00:38:00,688 --> 00:38:03,348
And what that means is you run it on your laptop.

404
00:38:03,568 --> 00:38:05,408
You run it on your phone, right?

405
00:38:05,428 --> 00:38:06,028
I've got a phone.

406
00:38:07,868 --> 00:38:09,388
Or you get like a little home server.

407
00:38:09,908 --> 00:38:11,508
There is a server called Start9.

408
00:38:12,388 --> 00:38:14,568
We've brought them up on previous episodes, right?

409
00:38:15,068 --> 00:38:17,208
You can get something like Start9, Umbral,

410
00:38:17,808 --> 00:38:21,208
or you can get like an actual legit server

411
00:38:21,208 --> 00:38:23,328
that has like a really nice GPU inside of it.

412
00:38:23,388 --> 00:38:25,408
That's really more what you need, but it's more expensive.

413
00:38:25,408 --> 00:38:30,308
and you can just have a computer running in your home that's plugged in all the time, 24-7 running.

414
00:38:30,628 --> 00:38:35,168
And then you set up the internet connection so that when you're away from the home,

415
00:38:35,248 --> 00:38:37,788
your phone can talk to that home server and do your AI for you.

416
00:38:38,168 --> 00:38:40,908
This is the most private AI that you can get.

417
00:38:41,768 --> 00:38:43,648
Local, running on your own box.

418
00:38:44,268 --> 00:38:48,788
You could even make it so that you can have the internet turned off while you're using the AI

419
00:38:48,788 --> 00:38:51,428
and know that your stuff is not going anywhere.

420
00:38:51,428 --> 00:38:53,388
You can see exactly what's being stored.

421
00:38:53,728 --> 00:38:55,208
Okay, this is awesome.

422
00:38:55,408 --> 00:38:57,588
and it used to be difficult to do this.

423
00:38:57,688 --> 00:39:00,028
It's getting easier every day because more tools are coming out.

424
00:39:00,528 --> 00:39:03,148
LM Studio is one of those really easy to use tools.

425
00:39:03,708 --> 00:39:05,688
Okay, what it lets you do is you download it,

426
00:39:05,748 --> 00:39:09,728
and it's available on like all the major platforms, I believe.

427
00:39:10,528 --> 00:39:12,068
Let me get to the download page somewhere.

428
00:39:13,708 --> 00:39:14,268
Downloads.

429
00:39:20,268 --> 00:39:21,268
Download LM Studio.

430
00:39:22,808 --> 00:39:25,028
As you can see, they have Mac, Windows, Linux.

431
00:39:25,408 --> 00:39:30,208
different processors, different versions, that kind of stuff. Right. And then they're also open

432
00:39:30,208 --> 00:39:39,668
source, I believe. Trying to remember. Yes. Get up. There it is. Okay. So they're open source as

433
00:39:39,668 --> 00:39:44,548
well. So you can see the code behind this, what's powering this, right? So a very powerful tool,

434
00:39:44,548 --> 00:39:48,328
but what makes it so awesome is then it's just point and click to install these models.

435
00:39:49,028 --> 00:39:52,928
So they have all these models in their catalog and you can just be like, oh, DeepSeek R1. I heard

436
00:39:52,928 --> 00:39:56,548
everybody talking about this because it supposedly crashed the stock market for a few days,

437
00:39:56,628 --> 00:40:03,268
which was so dumb. But because having better AI out there just means we're going to be able to do

438
00:40:03,268 --> 00:40:09,308
more stuff with it, not less. So I thought that was silly. But you can just, when you're in the

439
00:40:09,308 --> 00:40:13,508
LM Studio app, you can just like click on these and install them. And it tells you roughly how

440
00:40:13,508 --> 00:40:17,308
much RAM, how much memory you need on your computer to be able to power them, that kind of stuff.

441
00:40:17,748 --> 00:40:22,728
Okay. Really easy point and click installation. So LM Studio is great. I recommend people

442
00:40:22,728 --> 00:40:26,828
download that and just try it out and play with it, right? What are the trade-offs? Well, you get

443
00:40:26,828 --> 00:40:32,988
super strong privacy, but you have an underpowered computer or your laptop does not have a high-end

444
00:40:32,988 --> 00:40:42,228
NVIDIA GPU inside of it that is going to run as fast as ChatGPG or Maple. You also, you have to

445
00:40:42,228 --> 00:40:48,288
kind of maintain it yourself. So you have to make sure you're updating the models or if you're going

446
00:40:48,288 --> 00:40:52,988
to do a home server, you have to make sure that it's always turned on, that it has security patches

447
00:40:52,988 --> 00:40:58,768
applied to it, that you have an internet connection to it when you're away from the house, poking

448
00:40:58,768 --> 00:41:04,408
through your own personal firewall, that kind of stuff. So there's cost and work involved with that.

449
00:41:05,408 --> 00:41:10,168
But you get great privacy. Okay. So we wanted to come up with something different

450
00:41:10,168 --> 00:41:15,728
that took the power of cloud compute, right? Let me sign out.

451
00:41:18,288 --> 00:41:24,428
actually no I'll just do a private window try maple.ai we want to take the privacy of local

452
00:41:24,428 --> 00:41:31,448
but bring it to the cloud because nvidia has these massive gpus called the h100s and they're

453
00:41:31,448 --> 00:41:36,168
always innovating and coming out with bigger stronger ones right that can run these big models

454
00:41:36,168 --> 00:41:42,408
with local ai with something like lm studio I don't know if you notice this but if I go to the

455
00:41:42,408 --> 00:41:48,928
model catalog. They have these numbers next to them, right? 3.8B, 14.7B. These are the number

456
00:41:48,928 --> 00:41:55,808
of parameters that can be run in a model. And the more parameters, the more RAM your device has to

457
00:41:55,808 --> 00:42:02,048
have just to be able to load the model. But then also it has to process it, the speed at which it

458
00:42:02,048 --> 00:42:12,068
can process it. So your devices can run these sizes. Well, Llama 3.3 that we run is 70 billion

459
00:42:12,068 --> 00:42:18,248
parameters, right? That doesn't run very well on an iPhone or on a Linux laptop,

460
00:42:18,828 --> 00:42:26,128
unless you install massive amounts of RAM. So in order to run the strongest, biggest models that

461
00:42:26,128 --> 00:42:32,328
are the fastest and the most accurate models, you need to have cloud compute level stuff.

462
00:42:32,788 --> 00:42:35,808
And those are H100s and anything better that comes along.

463
00:42:35,808 --> 00:42:46,848
so we set out to say what can we build that combines the two okay and that that's where

464
00:42:46,848 --> 00:42:51,428
maple was born maple uses something called secure enclaves or confidential computing

465
00:42:51,428 --> 00:42:58,608
to do all this and so what you get is um you get your phone here and it makes an encrypted

466
00:42:58,608 --> 00:43:03,128
connection over to our servers which is normal right but then it uses secure enclaves

467
00:43:03,128 --> 00:43:09,968
Secure enclaves are the things that are on your phone that secure your face, your thumbprint for ID.

468
00:43:10,828 --> 00:43:15,148
It secures your wallet when you use like Google Pay or Apple Pay, that kind of stuff.

469
00:43:15,248 --> 00:43:18,328
The secure enclave is making that all secure and safe.

470
00:43:18,928 --> 00:43:20,728
So these exist on cloud servers now.

471
00:43:21,688 --> 00:43:24,148
And so you can use secure enclaves in the cloud.

472
00:43:24,648 --> 00:43:26,028
AWS has a version.

473
00:43:27,028 --> 00:43:29,208
Microsoft Azure has a version.

474
00:43:30,248 --> 00:43:32,508
NVIDIA has one on their GPUs as well.

475
00:43:33,128 --> 00:43:34,728
And so they put these on there.

476
00:43:34,848 --> 00:43:40,308
And what it does is it creates this safe box where you can put code.

477
00:43:40,748 --> 00:43:43,028
Everybody can see the code that's running in the box,

478
00:43:43,028 --> 00:43:45,808
but they can't see the data that passes through it, right?

479
00:43:45,828 --> 00:43:49,088
So if you're in the kitchen making food and you're using a recipe,

480
00:43:49,868 --> 00:43:52,008
everybody can see the recipe that you're using,

481
00:43:52,168 --> 00:43:54,668
but they can't see what actual ingredients,

482
00:43:54,788 --> 00:43:57,368
what kind of flour you might have used or what kind of,

483
00:43:57,368 --> 00:43:59,668
if you put sugar in it, what kind of sugar you're using

484
00:43:59,668 --> 00:44:00,548
or what kind of eggs?

485
00:44:00,648 --> 00:44:03,828
Were they eggs that were out from free range chickens

486
00:44:03,828 --> 00:44:07,388
in your own property or were they bought at a store?

487
00:44:07,568 --> 00:44:08,148
They don't know.

488
00:44:08,508 --> 00:44:09,888
They just know the recipe that you use.

489
00:44:10,148 --> 00:44:11,368
That's what secure enclaves are.

490
00:44:12,208 --> 00:44:14,748
So you know the data that's being used,

491
00:44:15,248 --> 00:44:18,868
but you don't, sorry, you know the code that is being run,

492
00:44:18,948 --> 00:44:19,828
but you don't know the data.

493
00:44:20,248 --> 00:44:22,608
And this is big when, you know, we just a second ago,

494
00:44:22,628 --> 00:44:23,868
we talked about ChatGPT

495
00:44:23,868 --> 00:44:26,308
and how they have these agreements with enterprises

496
00:44:26,308 --> 00:44:29,508
and they say the zero data retention policy,

497
00:44:29,508 --> 00:44:30,748
They say, we're not going to keep your data.

498
00:44:30,848 --> 00:44:31,888
We're not going to store it, right?

499
00:44:32,588 --> 00:44:33,368
We promise.

500
00:44:34,028 --> 00:44:37,008
Well, you don't actually have a way to look at what code is running on the servers.

501
00:44:37,248 --> 00:44:41,168
You don't know if they actually inserted a thing because one of the governments came

502
00:44:41,168 --> 00:44:44,608
and knocked on their door and said, hey, there's this business over here.

503
00:44:45,208 --> 00:44:47,348
We want to have access to what they're doing.

504
00:44:47,808 --> 00:44:51,468
So will you please just, we know you have a zero data retention policy.

505
00:44:51,728 --> 00:44:55,128
If it ever comes up in courts, you know, you can fight it.

506
00:44:55,168 --> 00:44:57,428
You've got a lot of money or maybe we'll help you fight it, whatever.

507
00:44:57,428 --> 00:45:03,848
please just like insert a line of code that makes a copy of all of their data and puts it in a secret

508
00:45:03,848 --> 00:45:12,128
database that'll never be seen by anybody other than us that can happen um because you can't

509
00:45:12,128 --> 00:45:16,068
verify what's running on the servers so confidential computing gives you the ability to see what's

510
00:45:16,068 --> 00:45:22,308
running on those servers and that is through a process called attestation right big long a word

511
00:45:22,308 --> 00:45:34,872
but um you know live secure enclave verification this is on our website before we even sign in and you can come see that our servers are verified So I click on this this is my proof of security This gives me like a fingerprint of the code

512
00:45:34,872 --> 00:45:36,952
the build of code that's running on the server.

513
00:45:36,952 --> 00:45:38,752
And then you can actually look at all the servers

514
00:45:38,752 --> 00:45:43,132
that are in between me and Maple, right?

515
00:45:43,132 --> 00:45:45,552
Their AWS has their own servers.

516
00:45:45,552 --> 00:45:49,712
So if you have been around app development

517
00:45:49,712 --> 00:45:53,432
and internet security for any kind of time,

518
00:45:53,432 --> 00:45:54,972
This looks familiar to you.

519
00:45:54,972 --> 00:45:57,712
This is very similar to HTTPS,

520
00:45:57,712 --> 00:46:01,152
to secure socket layer TLS kind of stuff, right?

521
00:46:01,152 --> 00:46:04,172
So this little lock icon that every browser has now,

522
00:46:04,172 --> 00:46:06,652
we've all been conditioned to see this lock icon.

523
00:46:06,652 --> 00:46:09,812
If I click on this, I get to see the certificates

524
00:46:09,812 --> 00:46:13,472
so that I know that this domain, trymapl.ai,

525
00:46:13,472 --> 00:46:16,392
going over HTTPS is secure.

526
00:46:16,392 --> 00:46:21,392
It's secure because some company up on the stack of trust

527
00:46:21,392 --> 00:46:28,992
signed it and said, yes, this domain was registered by this person and they are the ones running it.

528
00:46:29,052 --> 00:46:34,852
They put the certificate on their servers. We agree that this is who they say they are. So that

529
00:46:34,852 --> 00:46:40,892
way some other company isn't pretending to be try maple.ai and spoofing and, but they don't have the

530
00:46:40,892 --> 00:46:48,952
certificate, right? Well, secure enclaves are this next evolution, right? We started with HTTP. I'm

531
00:46:48,952 --> 00:46:53,972
just going to highlight it here on the screen, right? The letters HTTP. This was the first version

532
00:46:53,972 --> 00:46:58,512
of the internet that was not secure. Everything just kind of went over the wire. And if you were

533
00:46:58,512 --> 00:47:02,512
able to clamp into that wire, metaphorically, you could see all the traffic going across.

534
00:47:02,752 --> 00:47:08,732
Well, then we started to log into bank accounts and buy things on eBay and buy plane tickets and

535
00:47:08,732 --> 00:47:13,132
other things. So then we said, let's make this secure. So we added an S on the end and called

536
00:47:13,132 --> 00:47:18,272
it HTTPS. And we used encryption to encrypt everything as it goes across the wire. So now

537
00:47:18,272 --> 00:47:23,072
if you tap into that wire and try to listen, it's just a bunch of garbled, a bunch of garbled mess,

538
00:47:23,152 --> 00:47:31,972
right? It's encrypted. Uh, we think that we are on the cusp of a new upgrade and that is HTTP SE.

539
00:47:31,972 --> 00:47:37,952
That's what I'm calling it. Uh, HTTP over secure enclaves. And these are pieces of software that

540
00:47:37,952 --> 00:47:42,632
are running in different enclaves. So we have maple running an enclave. We have open secret

541
00:47:42,632 --> 00:47:49,072
backend running in an enclave and watch. Okay. Sorry. Maple, I guess technical maple's running

542
00:47:49,072 --> 00:47:52,492
in your browser. It's a JavaScript app running in your browser that talks to the enclave

543
00:47:52,492 --> 00:47:59,152
of open secret in the backend, but you can have more enclaves. Other apps are running and their

544
00:47:59,152 --> 00:48:04,592
API is running an enclave and you can start to verify, not only verify that your stuff is

545
00:48:04,592 --> 00:48:09,172
encrypted in transit, but now you can verify what software is running on that server and that they

546
00:48:09,172 --> 00:48:14,852
are protecting your data, encrypting your data in the way that they say they are. So a promise that

547
00:48:14,852 --> 00:48:22,012
we make here with Maple is we make the software commitment to you that you have a private key

548
00:48:22,012 --> 00:48:27,652
that is going to privately encrypt all of your chats. So let me go to our GitHub.

549
00:48:30,412 --> 00:48:35,052
I know it's on here somewhere. Oh, there it is. Okay.

550
00:48:35,052 --> 00:48:42,192
so we're the company open secret we make maple and so you can come on here and you can say okay

551
00:48:42,192 --> 00:48:47,492
open secret this is the server software that is powering it and if you are a developer you can

552
00:48:47,492 --> 00:48:53,192
actually go in and look at our rust code or you can just simply feed this code to ai and say hey

553
00:48:53,192 --> 00:48:59,172
please help me understand this code base what are they what are they doing to you treat my public or

554
00:48:59,172 --> 00:49:04,832
to treat my data my user data are they saving and recording my user data are they doing it any way

555
00:49:04,832 --> 00:49:11,872
that is unencrypted or are they staying true to their encryption promises, right? So our encryption

556
00:49:11,872 --> 00:49:18,332
promise to our users is that when they sign up, right, when they go here on trymable.ai

557
00:49:18,332 --> 00:49:26,952
and they click log in and sign up, they get these options. It used to be, and it still is with a lot

558
00:49:26,952 --> 00:49:32,212
of privacy tools are kind of hard to use because they require you to generate your own private key

559
00:49:32,212 --> 00:49:38,172
and it ends up being this long string of text or like a group of 12 words or something that you now

560
00:49:38,172 --> 00:49:45,232
have to manage and maintain on your own. And then if you somehow lose that, right, let's say you lose

561
00:49:45,232 --> 00:49:51,692
your encryption key, all of your data is now rendered unreadable. And that's just how encryption

562
00:49:51,692 --> 00:49:57,132
works. If you're not familiar with like private public key encryption, it's effectively a one-way

563
00:49:57,132 --> 00:50:03,092
street. You have to have that private key that you encrypt and it goes one way. And the only way to

564
00:50:03,092 --> 00:50:07,672
decrypt that is with the private key. If you don't have the private key, mathematically, you cannot

565
00:50:07,672 --> 00:50:17,032
unscramble all the stuff until we have new computers. In theory, quantum computing might

566
00:50:17,032 --> 00:50:22,652
be able to break that encryption algorithm. We don't know. But for now, the whole world operates

567
00:50:22,652 --> 00:50:26,752
under this model that that encryption has not been broken. I will say it's possible that

568
00:50:26,752 --> 00:50:33,272
governments, defense departments maybe have already broken that encryption, but their incentive is

569
00:50:33,272 --> 00:50:36,952
actually to keep that quiet because they want to, if you've broken that encryption, you're just going

570
00:50:36,952 --> 00:50:42,472
to want to watch all of the world's traffic go by. So you are not incentivized to let everybody know

571
00:50:42,472 --> 00:50:46,472
that you've done that. So for now, we're operating on this model that this encryption is solid and

572
00:50:46,472 --> 00:50:53,532
working well. Okay. With that said, because we're using secure enclaves, we can take that scary user

573
00:50:53,532 --> 00:50:58,492
experience away from people and we can just give them something that they know and use already.

574
00:50:58,492 --> 00:51:02,712
You can sign in with email or you can sign in with some kind of OAuth provider, right?

575
00:51:03,472 --> 00:51:09,752
Eventually we want to allow other options like maybe log in with Noster or maybe you could log

576
00:51:09,752 --> 00:51:14,072
in with a private key. So if you do want to have an extra layer of security, maybe you generate

577
00:51:14,072 --> 00:51:17,272
your own private key and log in with that. These are things that we've kind of tossed around,

578
00:51:17,372 --> 00:51:22,132
right? Pass key, other things. For now, you just do this. And so you can just, boom, log in with

579
00:51:22,132 --> 00:51:26,672
one of them or sign up with one of them. And it's very easy. It looks like we have login here when

580
00:51:26,672 --> 00:51:34,512
it should say sign up. Um, or maybe not, I don't know. But what happens is once you do that,

581
00:51:34,752 --> 00:51:39,612
you get signed into Maple, like I'm signed in over here, right. And you get dumped into this page

582
00:51:39,612 --> 00:51:44,952
and a private key has been generated for you. So that long complicated key is generated for you

583
00:51:44,952 --> 00:51:52,112
in the enclave and then stored encrypted in our backend. So if I were to log into our database,

584
00:51:52,112 --> 00:51:56,772
right now, all I see are a bunch of encrypted rows that I cannot read. And I'll never be able

585
00:51:56,772 --> 00:52:03,772
to read unless I have your private key, right? Or unless I, uh, unless I push a malicious version

586
00:52:03,772 --> 00:52:09,932
of our code to the servers, and then you log in with your private key. Um, and that, that version

587
00:52:09,932 --> 00:52:15,972
of the code saves your private key down, but that version of the code is right here on GitHub.

588
00:52:16,792 --> 00:52:21,692
So if I were to push something new, it's going to show up on GitHub and your Maple client

589
00:52:21,692 --> 00:52:28,492
knows which version of the code it's talking to. And if I push something that the Maple client is

590
00:52:28,492 --> 00:52:34,892
not expecting, this will not connect. This green verified badge does not show up. It turns red with

591
00:52:34,892 --> 00:52:38,752
big red X. And when you try to do anything with Maple, it's going to say, sorry, I'm not going to

592
00:52:38,752 --> 00:52:44,312
let you do this because I am not talking to the correct software on the server. Something's been

593
00:52:44,312 --> 00:52:49,672
tampered with. You need to go check that out, right? So that's where this system of trust

594
00:52:49,672 --> 00:52:51,972
becomes an open system of trust,

595
00:52:52,052 --> 00:52:53,912
which is why we call our company OpenSecret

596
00:52:53,912 --> 00:52:56,732
because all of your secrets are protected by the fact

597
00:52:56,732 --> 00:52:58,412
that we are running open source code

598
00:52:58,412 --> 00:53:01,032
that runs in enclaves that can be verified by anybody.

599
00:53:01,972 --> 00:53:03,952
Now, down here in the instructions on GitHub,

600
00:53:04,272 --> 00:53:05,372
we have instructions for you

601
00:53:05,372 --> 00:53:06,992
to how to make your own build of the code.

602
00:53:07,192 --> 00:53:09,192
And then you can get that checksum

603
00:53:09,192 --> 00:53:11,832
that you can verify yourself.

604
00:53:11,952 --> 00:53:13,372
So let's go back here, right?

605
00:53:13,432 --> 00:53:16,352
This fingerprint, we explain to you in the instructions

606
00:53:16,352 --> 00:53:17,652
how to generate that fingerprint.

607
00:53:17,652 --> 00:53:19,952
and then you can compare the two yourself.

608
00:53:20,232 --> 00:53:22,312
The software is making that comparison for you already,

609
00:53:22,512 --> 00:53:24,472
but don't trust verify, just do it yourself.

610
00:53:25,112 --> 00:53:28,272
And you can make sure that you are operating

611
00:53:28,272 --> 00:53:30,152
under the privacy model that we told you

612
00:53:30,152 --> 00:53:31,972
we're gonna provide, right?

613
00:53:32,952 --> 00:53:34,912
So you can keep an eye on our GitHub

614
00:53:34,912 --> 00:53:38,352
and look at changes that come over the different commits.

615
00:53:39,492 --> 00:53:43,512
And then you can also watch the Maple thing as well

616
00:53:43,512 --> 00:53:45,052
because we make changes there.

617
00:53:45,692 --> 00:53:47,192
So we develop in the open.

618
00:53:47,192 --> 00:53:51,172
And people are able to see what's in there and see that we are not inserting back doors.

619
00:53:51,312 --> 00:53:53,472
We're not inserting things that log your stuff, right?

620
00:53:55,532 --> 00:53:59,512
And I think this is the way forward for cloud computing.

621
00:54:00,992 --> 00:54:05,252
Start9 has the opinion that the way forward is no cloud computing, but home servers.

622
00:54:05,992 --> 00:54:08,092
And I really, that resonates with me.

623
00:54:08,132 --> 00:54:09,112
And I think that's important.

624
00:54:09,212 --> 00:54:09,912
I think that's great.

625
00:54:10,052 --> 00:54:11,092
I run my own home servers.

626
00:54:11,212 --> 00:54:12,692
I know a lot of people run their own home servers.

627
00:54:13,092 --> 00:54:14,212
Let's do more of that.

628
00:54:14,212 --> 00:54:19,272
and let's build software that can be truly decentralized and can interact in a very easy way

629
00:54:19,272 --> 00:54:24,952
with other people using the trust and the privacy that we set on our own servers.

630
00:54:25,792 --> 00:54:31,972
We're not in that world yet, but they're building it. There is going to be a long transition period

631
00:54:31,972 --> 00:54:38,012
between right now where we are on these cloud systems like ChatGPT that capture all of our data,

632
00:54:38,012 --> 00:54:44,992
right this is the reality we live in right now and to get to home servers where everybody's

633
00:54:44,992 --> 00:54:48,752
running their own private ai is going to be a lot of work and effort to get there let's keep

634
00:54:48,752 --> 00:54:53,912
building it okay this is not in place of that but i think that there's a huge middle area for this

635
00:54:53,912 --> 00:54:58,352
transition where we need to bring people better privacy we need to protect people we need to

636
00:54:58,352 --> 00:55:06,512
build a safer world right now um we can't wait for that because people are getting screwed over

637
00:55:06,512 --> 00:55:12,652
right now. And so that's where Maple comes in. And then even in that world where we all have

638
00:55:12,652 --> 00:55:22,212
our own private servers, if secure enclaves in the long run get adopted heavily, then we might

639
00:55:22,212 --> 00:55:25,792
have this dual system where you have your own home server and secure enclaves in the cloud.

640
00:55:26,052 --> 00:55:29,432
The cloud is always going to have more powerful hardware than you have at home because

641
00:55:29,432 --> 00:55:35,452
businesses can afford bigger, beefier hardware and can share the cost and spread the cost over

642
00:55:35,452 --> 00:55:40,652
lots of users. So we might end up in this really cool world where you have a home server and you

643
00:55:40,652 --> 00:55:45,452
have cloud software that's running on secure enclaves that talk to each other. And I think

644
00:55:45,452 --> 00:55:49,832
that's an even cooler future where we combine the two. So I want to keep marching toward that

645
00:55:49,832 --> 00:55:53,692
direction. We're going to do this with Maple, start nine, and others are going to do it with

646
00:55:53,692 --> 00:55:59,612
home servers. And let's work together. And let's, let's build this brighter future that leaves

647
00:55:59,612 --> 00:56:06,072
behind just plain HTTPS leaves behind this trust where you can't see what's running on the servers.

648
00:56:06,752 --> 00:56:12,792
Okay. We are coming up on an hour. I'm going to call it here. Okay. We got up to, we have 77 people

649
00:56:12,792 --> 00:56:16,412
on the live stream. Welcome everybody. I really appreciate you tuning in. I've been doing this

650
00:56:16,412 --> 00:56:20,592
every week now for a little over two months. We're showing off different freedom tech tools.

651
00:56:20,752 --> 00:56:26,572
We're trying to give you something, one thing that you can do, one app you can play with this weekend

652
00:56:26,572 --> 00:56:33,152
to try and decouple yourself from the captured technology systems that are out there,

653
00:56:33,272 --> 00:56:38,632
the things like the chat GPT, right? So this weekend, my challenge to you is to play with

654
00:56:38,632 --> 00:56:44,572
one of these tools. Go sign up for Maple, right? You can get a free account. Let me go back. Okay.

655
00:56:45,312 --> 00:56:51,672
You can get a free account and it's super easy. Is this hard right here? Sign up with email,

656
00:56:51,852 --> 00:56:55,132
sign in with one of these. There are some privacy people that don't want to do that. That's fine.

657
00:56:55,132 --> 00:56:58,572
I just want you to know that over half of our users use a privacy email.

658
00:56:58,792 --> 00:57:04,112
We don't know who they are because it's just some temporary generated thing that is an

659
00:57:04,112 --> 00:57:08,132
alias that forwards to some other email somewhere, or it truly is a burner email account that

660
00:57:08,132 --> 00:57:09,192
they use, right?

661
00:57:09,592 --> 00:57:13,732
I don't recommend the burner ones necessarily unless you know how to operate with it because

662
00:57:13,732 --> 00:57:18,892
in the future, if you need to reset your password, you might run into some problems, but maybe

663
00:57:18,892 --> 00:57:19,732
that's fine for you.

664
00:57:19,792 --> 00:57:22,112
Maybe you just want to use this as a temporary account and get a new one.

665
00:57:22,112 --> 00:57:27,212
Everybody has their own risk model that they operate under, but this couldn't be easier.

666
00:57:27,412 --> 00:57:29,332
So just go to try maple.ai.

667
00:57:29,572 --> 00:57:34,272
That's the website or go into the Apple app store and download maple AI.

668
00:57:35,192 --> 00:57:37,252
Leave us five-star review while you're there.

669
00:57:37,352 --> 00:57:43,832
We would appreciate it, but download it and sign in this weekend and play around with a

670
00:57:43,832 --> 00:57:49,952
version of AI in the cloud that is similar to chat GPT, but is 100 is, is completely

671
00:57:49,952 --> 00:57:55,412
private and confidential and gives you protection of your privacy. We do not have as many features

672
00:57:55,412 --> 00:58:00,652
as ChatGPT. That is readily apparent as soon as you log in, right? But we're adding them constantly.

673
00:58:00,652 --> 00:58:05,112
Every day we're adding new things. We just added this last week. This lets you put in your own

674
00:58:05,112 --> 00:58:11,172
character, your own system prompt for your chat, right? We are adding DeepSeek. We are adding

675
00:58:11,172 --> 00:58:16,872
document upload. We are going to add long-term memory, excuse me, long-term memory storage in

676
00:58:16,872 --> 00:58:21,292
the future that is done correctly with privacy. We're going to add a web search. So you can do

677
00:58:21,292 --> 00:58:25,872
things like deep research where it goes on searches web for you with privacy. We want to bring all of

678
00:58:25,872 --> 00:58:31,472
these really important, um, productive and valuable tools, but do it in a way where you don't have to

679
00:58:31,472 --> 00:58:36,312
sacrifice your privacy and give things up. Okay. So that would be my recommendation. Go to Maple,

680
00:58:36,512 --> 00:58:41,352
but also play around with the local stuff. LM studio is great. One that I wanted to show off

681
00:58:41,352 --> 00:58:48,132
was whisper. Um, I'm on a Mac and so I use one called Mac whisper. This is how I do. Oh, I'm in

682
00:58:48,132 --> 00:58:54,572
private. This is how I do my transcripts from my weekly podcast that I do right. Mac whisper

683
00:58:54,572 --> 00:59:02,192
is when I'm done recording, I take my recording and I dump it into Mac whisper and it generates

684
00:59:02,192 --> 00:59:07,032
the transcript for me. It knows how to differentiate between different speakers. So if I have a guest

685
00:59:07,032 --> 00:59:11,612
on. It'll differentiate between those. And it's all local. You can have it go out to ChatGPT if

686
00:59:11,612 --> 00:59:15,492
you want to, but I don't. I just run it locally. It takes about 10 minutes to transcribe my episode.

687
00:59:15,892 --> 00:59:19,212
So I just walk away and do other stuff while it's doing that. And then I come back and it's done,

688
00:59:19,552 --> 00:59:25,732
right? Grab a local AI tool, sign up for try maple.ai or download the app in the app store.

689
00:59:26,092 --> 00:59:30,732
That's your challenge for this weekend. All right. Looks like the stream stayed running this whole

690
00:59:30,732 --> 00:59:36,472
time, which is dope. The screen share kept working the whole time. It doesn't always do that.

691
00:59:36,472 --> 00:59:42,112
I'm going to stop sharing the screen here now. Thanks everybody for joining today on this episode

692
00:59:42,112 --> 00:59:48,392
of Freedom Tech Weekend. We are here every Friday, 1530 UTC, and we're talking about a different tool

693
00:59:48,392 --> 00:59:56,552
that you can play with this weekend to decouple your life from captured proprietary systems that

694
00:59:56,552 --> 01:00:03,452
hang on to your data and use your data against your wishes. I would just like to say, hope you

695
01:00:03,452 --> 01:00:07,572
have a great weekend. And if you have any recommendations for tools that you want to look

696
01:00:07,572 --> 01:00:12,072
at, or if you say I'm using this thing and I depend on every single day, I don't know how to

697
01:00:12,072 --> 01:00:19,232
break away from it. Shoot me a message on Twitter on X. My username is marks underscore F TW for the

698
01:00:19,232 --> 01:00:25,952
win. Or I'm also on Noster marks at primal.net. Hit me up on one of those or leave a comment on

699
01:00:25,952 --> 01:00:31,672
the YouTube video on this X post as well. And let me know what you want us to look at in the future.

700
01:00:31,672 --> 01:00:35,432
and we will do that. All right. Have a great weekend. Peace.