1
00:00:00,000 --> 00:00:05,460
Rob Hamilton and a ragtag team of misfits are banding together to save Bitcoin open source

2
00:00:05,460 --> 00:00:10,540
projects. I think the number now is over 10,000 findings, vulnerabilities. I probably personally

3
00:00:10,540 --> 00:00:14,600
spent almost up to maybe $100,000 in compute over the past month just using these AI models.

4
00:00:14,780 --> 00:00:20,540
Bitcoin as an experiment is worthless, not just to me, but the entire thing is worthless if self

5
00:00:20,540 --> 00:00:24,880
custody is not cheap. Are we just going to start seeing this across industries? Is open source

6
00:00:24,880 --> 00:00:29,600
particularly vulnerable to this given that it's by nature open source? People think this is like

7
00:00:29,600 --> 00:00:31,460
a Bitcoin story, like the cold card broke

8
00:00:31,460 --> 00:00:32,400
and then all this Bitcoin software.

9
00:00:32,520 --> 00:00:33,480
This is a software story.

10
00:00:33,740 --> 00:00:35,060
Anthropic preparing to tell investors

11
00:00:35,060 --> 00:00:36,480
that it's going to be worth 30 trillion.

12
00:00:36,600 --> 00:00:38,160
It's like a quarter of the world GDP.

13
00:00:38,480 --> 00:00:40,020
I went to Anthropic's quad code.

14
00:00:40,100 --> 00:00:41,540
What it instantly downgraded me saying,

15
00:00:41,620 --> 00:00:43,440
peasant, you can't use Fable for this model

16
00:00:43,440 --> 00:00:44,700
and then refused to answer the question.

17
00:00:44,840 --> 00:00:46,400
And then I used Kimmy K3 and it was like,

18
00:00:46,460 --> 00:00:47,660
here's a full vulnerability port.

19
00:00:47,780 --> 00:00:48,200
Shit's fucked.

20
00:00:48,280 --> 00:00:49,120
They're trying to create thought crime.

21
00:00:49,200 --> 00:00:49,960
Like they're literally saying,

22
00:00:50,100 --> 00:00:52,280
oh, you use this math and it came out with ideas

23
00:00:52,280 --> 00:00:52,760
we don't like.

24
00:00:52,800 --> 00:00:53,340
It's thought crime.

25
00:00:53,440 --> 00:00:54,280
It's an attack on speech.

26
00:00:54,340 --> 00:00:55,400
It's an attack on the American experiment.

27
00:00:55,520 --> 00:00:56,940
It's an attack on American safety because...

28
00:00:59,600 --> 00:01:08,140
I don't know how you have been juggling everything these past few weeks.

29
00:01:08,480 --> 00:01:17,060
CEO of Bitcoin, Big Bitcoin's number one ambassador, red team, I don't want to say leader of red team, but kind of part of a multi-pronged tip of the spear.

30
00:01:17,180 --> 00:01:18,420
Maybe it's like a tip of the trident.

31
00:01:18,420 --> 00:01:25,920
And then you run a, you obviously run Anchor Watch as well as like a nights and weekends hobby, I think.

32
00:01:27,240 --> 00:01:28,920
I'm just a humble insurance salesman.

33
00:01:28,920 --> 00:01:32,400
How are you doing, man? Like, how are you holding up throughout all this?

34
00:01:32,840 --> 00:01:42,960
So we can go longer into all the details of it, but the past three nights or so, three, four nights, I've started getting eight hours of sleep again.

35
00:01:44,480 --> 00:01:50,260
More regular, just getting back to a new normal, whatever this is.

36
00:01:52,080 --> 00:01:56,640
It's been good. I mean, we can go deeper into the details of it, but it was three weeks.

37
00:01:56,640 --> 00:02:26,300
I think it was just under three weeks. It was like 18, 17, 18 days of like two, three hours of sleep a night, constant fire drills running around. It was by an order of magnitude, the busiest I had ever been in my life for just weeks at a time. It was not just like I have a busy day today. It was stay up until two, three in the morning, even sometimes like three to four, pass out. And then the moment I like did one REM cycle and my body was like possibly a little bit awake, I would like

38
00:02:26,640 --> 00:02:33,100
jump right up run downstairs it was it was time intensive and there was a lot of things to juggle

39
00:02:33,100 --> 00:02:38,000
and a lot of things to do i think a lot of work is still to be done but what a lot of i'm doing

40
00:02:38,000 --> 00:02:44,140
right now is trying to externalize myself from it so we can scale because i don't want to be

41
00:02:44,140 --> 00:02:49,300
a gateway for security like so some of the stuff we're working on like working i've been putting

42
00:02:49,300 --> 00:02:54,200
time into working on open sourcing the tooling i was using which is by no means like definitive or

43
00:02:54,200 --> 00:02:59,320
final or the best that's out there but if people want to do the stuff i was doing i wanted to be

44
00:02:59,320 --> 00:03:03,320
able to provide a way where people can do it themselves right so i don't have to be as an

45
00:03:03,320 --> 00:03:06,460
in-between anymore right so that's kind of like trying to like it's funny because it's almost like

46
00:03:06,460 --> 00:03:12,300
a company or product scale like i'm trying to just like let it fully take off in a way where i'm not

47
00:03:12,300 --> 00:03:16,160
the one shepherding it day to day because it does that doesn't really scale it was never really the

48
00:03:16,160 --> 00:03:22,820
intention it all started with me on my own personal credit card scanning some repos so like it was

49
00:03:22,820 --> 00:03:27,680
this is all something that was very emergent and not that wasn't part of my my month of august

50
00:03:27,680 --> 00:03:32,660
plan that i was going to get to be doing all this well yeah can we can we actually talk about that

51
00:03:32,660 --> 00:03:38,380
because okay we were just talking offline like uh our my wife was scheduled to go into surgery on

52
00:03:38,380 --> 00:03:45,500
the friday uh friday the 31st uh for the birth of our daughter cold card hack happens the night

53
00:03:45,500 --> 00:03:50,960
before starts to kick off pretty so i was a little bit you know that the evening before carl and i

54
00:03:50,960 --> 00:03:54,360
both online and then you know was very much not online for a little while and then all of a sudden

55
00:03:54,360 --> 00:04:00,300
i come back and it's like rob hamilton and a team of you know in a ragtag team of misfits are you

56
00:04:00,300 --> 00:04:07,880
know banding together to save bitcoin open source projects like is this just a the uh perfect case

57
00:04:07,880 --> 00:04:12,200
study and you can just do things or did you get the tap from big bitcoin that was like hey you

58
00:04:12,200 --> 00:04:18,100
gotta you gotta fix all this open source dog shit out there like yeah big bitcoin and the

59
00:04:18,100 --> 00:04:25,000
podcasting conferencing complex gave me a ring and said in three weeks this is going to happen

60
00:04:25,000 --> 00:04:29,920
and we need you to orchestrate an entire pr response and i told him i trust the plan we'd

61
00:04:29,920 --> 00:04:34,760
be able to figure it out yeah i know so so like the actual like the way i like breaking down the

62
00:04:34,760 --> 00:04:40,940
timeline is actually monday july 27th kimi k3 went fully open weight and meaning that anyone

63
00:04:40,940 --> 00:04:44,820
could run the model if they had the compute infrastructure and for a year or two now

64
00:04:44,820 --> 00:04:53,600
i had been using ai to poke around my code bases in ways improving security right and i think it's

65
00:04:53,600 --> 00:05:00,160
important that like humans write software humans are not perfect and there are always going to be

66
00:05:00,160 --> 00:05:06,580
issues right like we are as we've seen over the past couple weeks like bugs do occur and they

67
00:05:06,580 --> 00:05:22,003
like we even have like even like even satoshi had bugs right like like this is something that just happens and for the past two years i been using these tools and over the past like four months there has been

68
00:05:22,003 --> 00:05:28,083
a serious like hockey stick acceleration and its ability to meaningfully contribute in a security

69
00:05:28,083 --> 00:05:32,603
capacity you could say it's been building over the past year or two but like it really started

70
00:05:32,603 --> 00:05:39,403
picking up in the past couple of months and so kidney k3 went open weights on july 27th i

71
00:05:39,403 --> 00:05:44,043
immediately started scanning my code base with it found new things that had not been found before

72
00:05:44,043 --> 00:05:49,503
like cleaning things up and uh there was that was like the first couple of days and then i think

73
00:05:49,503 --> 00:05:54,863
i'll say there's the 28th of the 29th cali had a tweet that was like uh kimmy kimmy kimmy can't

74
00:05:54,863 --> 00:05:59,163
you see like the biggie song and i said a model so nice they named it thrice because like we were

75
00:05:59,163 --> 00:06:03,223
both poking around stuff and just being like damn this model is good and this was a weird thing

76
00:06:03,223 --> 00:06:07,903
that i had never seen such divergent behavior before between an open source model and a frontier

77
00:06:07,903 --> 00:06:13,283
one this is my anecdotal experience but you would always be like the open source one would be like

78
00:06:13,283 --> 00:06:17,523
kind of good like oh this is kind of what i paid a lot of money for six months ago right but like

79
00:06:17,523 --> 00:06:22,643
that'd be like a decent like a casual summary of it but kimmy k3 was doing things that the frontier

80
00:06:22,643 --> 00:06:27,263
models had never done before with its ability to like find these vulnerabilities and then so

81
00:06:27,263 --> 00:06:29,543
Wednesday night, the cold card hack happens.

82
00:06:30,043 --> 00:06:31,303
Like the funds move on chain.

83
00:06:32,383 --> 00:06:34,663
Thursday morning, I made a couple of group chats

84
00:06:34,663 --> 00:06:37,003
and some engineers are passing around this Reddit post

85
00:06:37,003 --> 00:06:38,743
about a bunch of cold card funds moving

86
00:06:38,743 --> 00:06:40,883
and a bunch of people saying that they had, you know,

87
00:06:40,963 --> 00:06:42,843
AirDap security and they'd never touched the internet,

88
00:06:42,963 --> 00:06:43,703
but the funds were moving.

89
00:06:44,563 --> 00:06:48,903
And then pretty quickly, it kind of became obvious from that.

90
00:06:49,023 --> 00:06:51,043
I give a lot of credit to Talip,

91
00:06:51,203 --> 00:06:53,123
who is a dev who used to work at Citria,

92
00:06:53,123 --> 00:06:55,583
but now works at LX dev,

93
00:06:55,583 --> 00:06:59,063
working on like a lot of like Bitcoin scaling and privacy stuff.

94
00:06:59,623 --> 00:07:01,263
He's a great developer.

95
00:07:01,603 --> 00:07:02,963
He's also a handsome Turkish man,

96
00:07:03,043 --> 00:07:04,703
which is a quarter Greek kills me to say,

97
00:07:04,863 --> 00:07:05,863
but like I just saw,

98
00:07:05,923 --> 00:07:07,143
he just went on Guy Swan podcast

99
00:07:07,143 --> 00:07:08,783
and I just saw this like Turkish chat

100
00:07:08,783 --> 00:07:09,543
and I was like, God,

101
00:07:09,663 --> 00:07:12,083
my gene pool is so angry for being friends with this person.

102
00:07:13,883 --> 00:07:15,643
You should have him on like Talib's great.

103
00:07:15,643 --> 00:07:18,243
So Talib, myself and a couple other people

104
00:07:18,243 --> 00:07:19,143
are poking around the code base.

105
00:07:19,223 --> 00:07:21,323
I think Talib like got a debt to rights first

106
00:07:21,323 --> 00:07:24,283
because like the nature of how these funds were moving

107
00:07:24,283 --> 00:07:25,503
was like, if there's something wrong,

108
00:07:25,583 --> 00:07:29,903
an attacker did not reverse engineer 600 people's wallets there has to be an issue with the core

109
00:07:29,903 --> 00:07:37,863
entropy talip is going digging into this he finds that i i start pulling up like open ai codex is

110
00:07:37,863 --> 00:07:41,603
there a buck here like what's going on like he helped me find it and it was it was definitely

111
00:07:41,603 --> 00:07:46,983
receptive and like talip and others got open ai to do it but like on our first blush like a very

112
00:07:46,983 --> 00:07:52,303
simple prompt he didn't directly like give me the goods i went to anthropics quad code and it was

113
00:07:52,303 --> 00:07:57,483
kind of what it instantly downgraded me saying peasant you can't use fable for this model you

114
00:07:57,483 --> 00:08:02,243
can use opus 4.8 from six from six months ago peasant like you'll be happy about it you'll pay

115
00:08:02,243 --> 00:08:07,323
us for it and then refuse to answer the question and then i used kimmy k3 and it was like here's a

116
00:08:07,323 --> 00:08:12,803
full vulnerability port shit's fucked like like it's like and i had never like it was nine days

117
00:08:12,803 --> 00:08:18,323
and then very quickly from there um it was one of those the way i've been describing it is the um

118
00:08:18,323 --> 00:08:21,943
in airlines if the airline safety videos like put your mask on before helping others

119
00:08:21,943 --> 00:08:27,243
and after an initial like blast radius assessment like okay everything's good at ink watch cool i'm

120
00:08:27,243 --> 00:08:32,623
gonna start poking around other code bases and so um i that first day or two though i spent a lot

121
00:08:32,623 --> 00:08:37,303
of time specifically just on the cold card stuff getting the word out doing detailed guides threat

122
00:08:37,303 --> 00:08:42,543
analysis if you're x do y if you're y do z like whatever right kind of breaking things down after

123
00:08:42,543 --> 00:08:50,543
that first like 48 hours i started poking around at random code bases uh and the the level of

124
00:08:50,543 --> 00:08:54,563
poking around was me opening open code with Kimmy K3 and saying,

125
00:08:55,183 --> 00:08:57,583
hi, can you see if there's any security problems here?

126
00:08:57,743 --> 00:08:58,363
Enter, right?

127
00:08:58,403 --> 00:08:59,283
Like very basic.

128
00:08:59,803 --> 00:09:01,183
And it was coming up with hits.

129
00:09:02,023 --> 00:09:04,523
Like, and hits of like, not great things.

130
00:09:05,103 --> 00:09:09,603
And then I would email the maintainer of that software

131
00:09:09,603 --> 00:09:11,943
or the team that maintains that software and be like,

132
00:09:11,983 --> 00:09:13,803
hey, I'm just a guy.

133
00:09:14,083 --> 00:09:15,443
I think I found some stuff.

134
00:09:15,543 --> 00:09:16,303
Take a look at it.

135
00:09:16,883 --> 00:09:19,983
And very quickly, I did this for like a handful of code bases.

136
00:09:19,983 --> 00:09:24,483
and most of them turned back and confirmed saying confirming it was an issue thanking me for the

137
00:09:24,483 --> 00:09:30,263
report and this is like friday saturday right so this is like for your timeline sake of what's

138
00:09:30,263 --> 00:09:37,963
happening and i'm like fuck like not good uh i then quickly start spinning up servers

139
00:09:37,963 --> 00:09:43,723
and i'm this is like one of the first like two three in the morning like like rates i'm just like

140
00:09:43,723 --> 00:09:48,983
schizophrenic i'm just like typing away i'm like okay we need to do better than this i want to do

141
00:09:48,983 --> 00:09:54,583
this, this and that. And I want you to look for X, Y and Z, but go. Right. And it made like a very

142
00:09:54,583 --> 00:09:58,843
what's called a harness. Right. So you have these models if you want to view them as the horse and

143
00:09:58,843 --> 00:10:03,083
the harness like a horse harness like directs the model. And I have observed that pretty thoroughly.

144
00:10:03,383 --> 00:10:09,423
Wish you could access cash without selling your Bitcoin? Ledin makes that possible. And the global

145
00:10:09,423 --> 00:10:15,423
leader in Bitcoin backed lending, Ledin, has introduced their lowest rates ever. No complexity,

146
00:10:15,423 --> 00:10:19,823
no fine print, the larger the loan, the lower the rate.

147
00:10:19,823 --> 00:10:32,026
These new rates apply to all new loans refinances and renewals You see the rates up front before you apply The new lower rates come with Ledin gold standard protection

148
00:10:32,586 --> 00:10:36,606
Your Bitcoin remains custodied and is never lent out for interest.

149
00:10:36,986 --> 00:10:40,746
Perfect 8-year track record of keeping clients' assets safe,

150
00:10:41,066 --> 00:10:45,746
auto top-up, LTV alerts, and partial repayments to keep you in control.

151
00:10:45,746 --> 00:10:50,646
repay whenever you want with zero penalties or monthly payments.

152
00:10:51,166 --> 00:10:58,866
Ledin's perfect $10 billion plus track record through every market cycle in the last eight years speaks for itself.

153
00:10:59,346 --> 00:11:03,046
Don't choose between a great rate and the safety of your Bitcoin.

154
00:11:03,506 --> 00:11:04,966
Get both at Ledin.

155
00:11:04,966 --> 00:11:10,066
Check out your rate by using their loan calculator at ledin.io.

156
00:11:10,066 --> 00:11:20,326
slash walker. That's L-E-D-N dot I-O forward slash walker. Blockstream Jade Plus delivers robust

157
00:11:20,326 --> 00:11:26,506
security with fully open source hardware and software. Its unique genuine check feature

158
00:11:26,506 --> 00:11:32,846
ensures your device is authentic, giving you ultimate peace of mind. With versatile connectivity

159
00:11:32,846 --> 00:11:40,786
options including bluetooth usbc built-in camera and sd card jade plus offers unparalleled

160
00:11:40,786 --> 00:11:46,866
flexibility in managing your bitcoin whether you prefer wireless convenience or air-gapped

161
00:11:46,866 --> 00:11:53,086
security jade has you covered its innovative design features a durable metal case a bright

162
00:11:53,086 --> 00:12:00,006
1.9 inch color screen and intuitive navigation buttons that make managing your bitcoin straightforward

163
00:12:00,006 --> 00:12:06,786
and enjoyable. Jade Plus also supports secure, air-gapped transactions via QR code signing

164
00:12:06,786 --> 00:12:12,686
using the Jade Link storage device, keeping your Bitcoin safe without compromising convenience.

165
00:12:13,046 --> 00:12:20,846
Visit store.blockstream.com and use my coupon code WALKER for an exclusive 10% discount on your

166
00:12:20,846 --> 00:12:26,446
Blockstream Jade Plus. Protect your Bitcoin, sleep better, stack harder.

167
00:12:26,446 --> 00:12:46,566
The way you can just having the model is a very strong differentiator compared to like if I use a model from years ago, it wouldn't find stuff. But the way you guide these models to do stuff is so impactful for your ability to find these bugs. In 20, 30 minutes, I've got this like version one of like what we call like the red team harness to just start scanning stuff.

168
00:12:46,566 --> 00:12:55,266
and from there i just load a big fat server and i just say here's a bunch of repo start scanning

169
00:12:55,266 --> 00:12:59,546
and it's just queuing it up and chewing and they and like the scans are taking hours and it's going

170
00:12:59,546 --> 00:13:06,086
and going and around this time calais reaches out to me saying rob can i like like can i help you

171
00:13:06,086 --> 00:13:10,406
and i was like yeah i i'm slammed i don't know what the hell i'm doing i'm not a security

172
00:13:10,406 --> 00:13:15,526
researcher but like i'm getting multiple confirmed issues and code bases like can you help me like

173
00:13:15,526 --> 00:13:20,206
can you help me? And he was like, yeah, like, what do you need help? I was like, literally anything.

174
00:13:20,386 --> 00:13:26,186
I have no idea what I'm doing here. And, uh, Cali, I, uh, basically the way it worked was like,

175
00:13:26,386 --> 00:13:29,626
I was like, well, I have some servers here. He's like, oh, cool. He let me in. He gave me his key,

176
00:13:29,806 --> 00:13:34,906
let him into the servers. He starts working and really operationalizing, thinking about this.

177
00:13:35,066 --> 00:13:38,886
He'd been doing a lot of his own security modeling with all the cache you protocol stuff. So he was

178
00:13:38,886 --> 00:13:43,346
already way more intimately familiar where I was like looking at my code base, my specific threat

179
00:13:43,346 --> 00:13:51,086
model he was definitely more advanced in thinking about these things and so himself and uh a couple

180
00:13:51,086 --> 00:13:57,026
other cashew devs jumped in i pulled in talip i pull in like a couple people you know like

181
00:13:57,026 --> 00:14:02,166
ben carmen rindahl like people who like understand code that i could trustworthy me just all start

182
00:14:02,166 --> 00:14:08,066
going to town looking at any code base you want right and it was a and then i was basically

183
00:14:08,066 --> 00:14:14,006
provisioning api keys as like all you can eat buffet just like go like go start grinding go

184
00:14:14,006 --> 00:14:17,266
start looking for stuff and we're like sharing learnings of how we're like going about all this

185
00:14:17,266 --> 00:14:24,266
stuff and that went on for weeks those weeks of like it got to a point where we had like a little

186
00:14:24,266 --> 00:14:28,246
devops team that was managing the servers like it turned into like a whole thing like it was a whole

187
00:14:28,246 --> 00:14:32,846
operation um looking to all these code bases finding scans setting up automated pipelines

188
00:14:32,846 --> 00:14:36,326
being able to do cleaner reporting to be able to disclose to maintainer stuff like

189
00:14:36,326 --> 00:14:42,546
it was a whole insane thing and he had like i have to keep that it's like 2025 decks that were

190
00:14:42,546 --> 00:14:47,866
all doing this and the stuff we were seeing even just outside the vulnerabilities but like how these

191
00:14:47,866 --> 00:14:53,466
ai models behave were it was crazy so if you've ever had a conversation with someone and someone

192
00:14:53,466 --> 00:14:58,786
will say something and then you have like a wave of realizations like that's the stuff we started

193
00:14:58,786 --> 00:15:03,986
seeing in these models so i would scan a random code base and would find some issues and then i'd

194
00:15:03,986 --> 00:15:07,946
be like well here's the vulnerability report and someone would be like interesting and they're like

195
00:15:07,946 --> 00:15:12,886
they're a domain expert in this type of software and they'd be like have you they'll give me like

196
00:15:12,886 --> 00:15:17,326
a couple half sentences and like look at this line of code and i would just copy and paste their like

197
00:15:17,326 --> 00:15:22,786
casual ramble into the model and then you would see them all say oh wait actually but

198
00:15:22,786 --> 00:15:27,526
oh wait actually oh wait actually and then all of a sudden it would cascade and find a bunch more

199
00:15:27,526 --> 00:15:32,106
issues and i was like this is crazy the differentiation of the specialization of labor

200
00:15:32,106 --> 00:15:43,928
the person who using these models is still incredibly important for deriving security insights This is not on autopilot right And this is not like a Norton antivirus idea where you just run your scan

201
00:15:44,029 --> 00:15:45,168
Well, it's like, oh, I'll do this next month.

202
00:15:45,209 --> 00:15:46,009
This looks good now, right?

203
00:15:46,009 --> 00:15:48,969
You have to be constantly vigilant and thinking about all this.

204
00:15:49,949 --> 00:15:51,729
And it's been, yeah.

205
00:15:51,809 --> 00:15:54,789
So that's a good just introductory summary of all of the scans and stuff.

206
00:15:54,789 --> 00:15:58,229
I think the number now is over 10,000 findings, vulnerabilities.

207
00:15:58,229 --> 00:16:04,849
um a thousand of them uh i think it's like 1100 of them are critical or high meaning like things

208
00:16:04,849 --> 00:16:09,849
that are definitely like there's there's a whole gradation of finding bone i've been a security

209
00:16:09,849 --> 00:16:14,009
researcher now for about three weeks so let me tell you how i learned about this that back in my

210
00:16:14,009 --> 00:16:18,269
day you would have a ranking of things where like you'd have like either informational or hardening

211
00:16:18,269 --> 00:16:22,029
which is like hey maybe you should update your documentation or just letting you know there's

212
00:16:22,029 --> 00:16:27,309
some type like small thing that's not consequential then you have low medium high and critical it's

213
00:16:27,309 --> 00:16:34,309
like tears of and um there's definitely i've been working with the tools it can sometimes not

214
00:16:34,309 --> 00:16:41,149
understand the proper threat model and it'll inflate severities for sure there were many

215
00:16:41,149 --> 00:16:46,569
conversation and like a whole team were all running around disclosing and like ding the bell when you

216
00:16:46,569 --> 00:16:52,869
when you uh find a bad one like ding and i started turning the harness into a tool that when it found

217
00:16:52,869 --> 00:16:58,349
something that was severe or high, like very critical, it would go off and make a whole proof

218
00:16:58,349 --> 00:17:02,989
of concept end to end. So it would, and this is the beauty of like open source software where like,

219
00:17:03,529 --> 00:17:08,708
we were not doing pen testing, like going into someone's server. We were just looking at software

220
00:17:08,708 --> 00:17:13,448
on our computers. So what you could do is you could come up with a theorized attack,

221
00:17:14,149 --> 00:17:18,809
run it on your local computer, get all of the logs, show the attack worked, and then hand that

222
00:17:18,809 --> 00:17:20,089
to the maintainer, right?

223
00:17:20,149 --> 00:17:21,109
To be able to like say,

224
00:17:21,188 --> 00:17:21,809
hey, we found something.

225
00:17:22,669 --> 00:17:24,609
So there was a lot of iterative,

226
00:17:24,748 --> 00:17:25,369
fast learning.

227
00:17:25,448 --> 00:17:26,629
And that was good to have a crew

228
00:17:26,629 --> 00:17:28,688
of trusted people

229
00:17:28,688 --> 00:17:30,829
to start chewing through all of this

230
00:17:30,829 --> 00:17:31,889
and scrambling out.

231
00:17:31,968 --> 00:17:32,748
Like it was a point where

232
00:17:32,748 --> 00:17:34,049
we were finding vulnerabilities

233
00:17:34,049 --> 00:17:35,188
faster than we could get ahold

234
00:17:35,188 --> 00:17:36,708
of people to responsibly disclose to.

235
00:17:36,909 --> 00:17:38,369
Because a lot of repos

236
00:17:38,369 --> 00:17:39,529
just didn't have things like

237
00:17:39,529 --> 00:17:40,549
if you find a security issue,

238
00:17:40,649 --> 00:17:41,649
this is how you get ahold.

239
00:17:41,728 --> 00:17:43,188
So it's called security markdown,

240
00:17:43,329 --> 00:17:43,809
MD file.

241
00:17:44,448 --> 00:17:45,468
I think a lot of repos

242
00:17:45,468 --> 00:17:46,429
since then have added those.

243
00:17:46,669 --> 00:17:47,849
It's been really, I think, just good.

244
00:17:47,849 --> 00:17:52,349
I think if you're maintaining an open source software, it's a critical thing to like, I'm

245
00:17:52,349 --> 00:17:56,508
not going to open up a GitHub issue and be like, hi, your system is broken and this is

246
00:17:56,508 --> 00:17:57,228
how you can break it.

247
00:17:57,269 --> 00:17:57,809
You should fix it.

248
00:17:57,849 --> 00:18:01,649
You want to be able to discreetly get that information over to the party.

249
00:18:02,429 --> 00:18:06,589
And it's interesting because people think this is like a Bitcoin story, like the cold

250
00:18:06,589 --> 00:18:07,988
card broke and then all this Bitcoin software.

251
00:18:08,109 --> 00:18:09,149
This is a software story.

252
00:18:09,688 --> 00:18:15,129
I think Bitcoin is just the natural first thing a black hat attacker would hit because

253
00:18:15,129 --> 00:18:21,508
if you could hack bitcoin infrastructure and just get bitcoin is that not way cleaner of a way to get

254
00:18:21,508 --> 00:18:26,049
a payday than like you socially engineer someone at a bank and then you get banking records and it's

255
00:18:26,049 --> 00:18:30,728
like okay well now i'm like steps in but i don't have money yet right so bitcoin is just inevitably

256
00:18:30,728 --> 00:18:35,448
like the easiest thing to try and go attack and shake down and that was the whole point of the

257
00:18:35,448 --> 00:18:42,169
red team stuff was like if we could have people who were white hats that were doing it out of our

258
00:18:42,169 --> 00:18:45,909
kind of like our consideration and passion for the project we weren't we weren't walking around

259
00:18:45,909 --> 00:18:49,688
with like we have a vulnerability like send us some bitcoin and we'll help you out like it was

260
00:18:49,688 --> 00:18:54,488
just very much triaging and going through all infrastructure and everyone was fanning out into

261
00:18:54,488 --> 00:18:57,488
repos that they knew they were familiar with whereas i was running just kind of like a big

262
00:18:57,488 --> 00:19:02,809
machine that was just chewing through repos and spitting them out those are it was a wild three

263
00:19:02,809 --> 00:19:10,369
weeks if i still smoked cigarettes i'd have a cigarette right now um but okay first of all damn

264
00:19:10,369 --> 00:19:12,988
uh second of all thank you for your service

265
00:19:12,988 --> 00:19:23,508
i got like i mean yes it's just it's just incredible to me because i look at this now

266
00:19:23,508 --> 00:19:30,748
and obviously the cold card hack was massive massive and destructive and wrecked a ton of

267
00:19:30,748 --> 00:19:35,609
people who and i'm certainly not even close to the first person to say this but who you know who

268
00:19:35,609 --> 00:19:40,629
quote, did everything right, right, right, who went with the, the what was viewed as kind of

269
00:19:40,629 --> 00:19:47,049
the cypherpunk hardware option, who still got wrecked through this tragic story, people losing

270
00:19:47,049 --> 00:19:54,309
life savings, like really brutal shit. If I can look at that and say, okay, what was there a silver

271
00:19:54,309 --> 00:20:03,269
lining to this giant fucking cloud? Well, I would have to say that it's basically the hardening of

272
00:20:03,269 --> 00:20:07,609
the Bitcoin open source ecosystem, which is basically what you guys have been doing this

273
00:20:07,609 --> 00:20:12,829
whole time. And I'm curious if you know of, I mean, as you said, Bitcoin was kind of this first

274
00:20:12,829 --> 00:20:18,769
like obvious thing, right? That like, okay, yeah, but we can, we can get Bitcoin if we figure out

275
00:20:18,769 --> 00:20:25,448
vulnerabilities in Bitcoin stuff. Like, yep, that makes sense. No middleman, right? And I'm just

276
00:20:25,448 --> 00:20:30,468
curious if you think, I mean, you said this is a software story. Are we just going to start seeing

277
00:20:30,468 --> 00:20:36,789
this across industries is open source particularly vulnerable to this given that it's by nature open

278
00:20:36,789 --> 00:20:41,169
source anybody can scan the cone bases anybody can do that is closed source stuff actually a bit

279
00:20:41,169 --> 00:20:46,069
more secure in the short term but maybe less secure in the long term because you've got less

280
00:20:46,069 --> 00:20:49,968
chance of auditing have you thought about that at all yeah so i think the immediate

281
00:20:50,011 --> 00:20:56,971
just availability of the code at this moment makes soft open source software more cutely

282
00:20:56,971 --> 00:21:03,211
an issue i don't think closed sources here long term inherently because of this there is some

283
00:21:03,211 --> 00:21:08,031
obscurity because of things being closed source that prevent attackers having a full lay of the

284
00:21:08,031 --> 00:21:13,851
land on how to do it but uh i think inherently like large companies and stuff are just going to

285
00:21:13,851 --> 00:21:20,631
have a harder time updating their security posture right um if you're in a very large company you have

286
00:21:20,631 --> 00:21:24,371
this diffusal of ownership in such a way where it's hard to understand like who's the one who's

287
00:21:24,371 --> 00:21:28,711
sitting on top of the latest ai threats not to mention people say stuff like well we had open ai

288
00:21:28,711 --> 00:21:34,611
scanner code base and it's good um you know like that those are things that i would not hold with

289
00:21:34,611 --> 00:21:39,431
you can't just like trust one vendor and say that like it's a done job especially just what i've seen

290
00:21:39,431 --> 00:21:44,791
right a funny thing you said about thank you for your service which i've um been joking about but

291
00:21:44,791 --> 00:21:50,711
my co-founder becca um there's a good story here so as everything's happening initially with the

292
00:21:50,711 --> 00:21:55,971
cold car fallout i'm getting dozens of dms a day random people saying like how can like how do i do

293
00:21:55,971 --> 00:22:00,431
that and i'm just like random help desk volunteer just kind of going through the list and trying to

294
00:22:00,431 --> 00:22:05,211
help people out and you're in this industry long enough how often have you gotten someone

295
00:22:05,211 --> 00:22:08,971
pretending to be a reporting reporter asking to like hop on a zoom call with you

296
00:22:09,431 --> 00:22:14,711
so many times oh yeah i don't respond to shit that's my that that's how i resist phishing

297
00:22:14,711 --> 00:22:20,251
attacks i don't open my email or respond to anyone this it's this one guy oh perfect nope so i'm going

298
00:22:20,251 --> 00:22:23,891
through all my messages and this guy is messaging me saying he's from bloomberg i'm like fuck off

299
00:22:23,891 --> 00:22:27,711
scammer and then i just keep going through my day and i was like if you're real get a hold of becca

300
00:22:27,711 --> 00:22:32,391
like i'm on time for this and then 30 minutes later i get an invite to doing like a chat with

301
00:22:32,391 --> 00:22:36,811
bloomberg i'm like oh i told a reporter to go fuck herself did you actually tell him to go fuck

302
00:22:36,811 --> 00:22:41,491
himself i told him that he was a scammer you know i didn't say fuck yourself but there were multiple

303
00:22:41,491 --> 00:22:46,871
ones that i did and they i think they were for sure like so there's one modest one and i did like a

304
00:22:46,871 --> 00:22:50,171
i hop in the room i'm like what's going on here like can you do a talk i was like

305
00:22:50,171 --> 00:22:54,451
you got you got seven minutes and i'm leaving because i don't have time for this right now

306
00:22:54,451 --> 00:22:58,551
but becca described it as rob had been off to war because becca very

307
00:22:58,551 --> 00:23:05,471
have been just keeping eye on the anchor watch ship during all of this the whole security

308
00:23:05,471 --> 00:23:13,651
ecosystem i think has just changed though and it feels weird i am not i may be a software developer

309
00:23:13,651 --> 00:23:17,671
i may understand security as it relates to bitcoin i am not a security researcher

310
00:23:17,671 --> 00:23:24,171
but i've kind of stumbled into that which is part of uh whenever people are like posting like

311
00:23:24,171 --> 00:23:30,631
hey we fixed a bug it's like we're all ready like this is a this is not a me thing like this is a

312
00:23:30,631 --> 00:23:34,611
this is a whole group effort have everyone be more vigilant and of the silver linings

313
00:23:35,471 --> 00:23:36,611
that come out of this.

314
00:23:37,091 --> 00:23:38,131
I think it's everyone

315
00:23:38,131 --> 00:23:40,071
actually living by the

316
00:23:40,071 --> 00:23:41,331
don't trust verify mantra

317
00:23:41,331 --> 00:23:42,531
and not just taking people

318
00:23:42,531 --> 00:23:42,991
at their word.

319
00:23:44,111 --> 00:23:46,071
And I'm not saying

320
00:23:46,071 --> 00:23:47,031
every single person

321
00:23:47,031 --> 00:23:49,231
needs to get a Kimmy K3 API key

322
00:23:49,231 --> 00:23:50,631
and start looking at their code bases.

323
00:23:50,931 --> 00:23:52,791
But I think this was a very healthy

324
00:23:52,791 --> 00:23:54,511
reaction and activation

325
00:23:54,511 --> 00:23:55,751
among everyone in Bitcoin

326
00:23:55,751 --> 00:23:57,611
looking to sniff around

327
00:23:57,611 --> 00:23:58,691
and see what's going on.

328
00:23:59,391 --> 00:23:59,551
Right.

329
00:23:59,631 --> 00:24:01,491
And so I think as far as

330
00:24:01,491 --> 00:24:02,151
silver linings go,

331
00:24:02,191 --> 00:24:03,131
I think that's a great one.

332
00:24:03,131 --> 00:24:05,131
It also, I think this,

333
00:24:05,471 --> 00:24:10,331
amidst all of this and again i was kind of like i saw it as it was initially happening and it was

334
00:24:10,331 --> 00:24:14,791
like oh shit and then it was like wife's in surgery daughters being born daughters in the

335
00:24:14,791 --> 00:24:20,311
nicu like focused on that and then kind of yeah yeah touching you know like coming back after uh

336
00:24:20,311 --> 00:24:24,991
you know a little little bit a couple it was literally like it was only like you know a day

337
00:24:24,991 --> 00:24:29,671
it was like 24 hours after which then i like you know checked x again posted an update or whatever

338
00:24:29,671 --> 00:24:34,571
that you know carl is good baby's good and you know of course like my timeline's flooded with

339
00:24:34,571 --> 00:24:38,651
all this stuff and it's just like what the what the fuck is going on here like what what is

340
00:24:38,651 --> 00:24:45,151
happening but it was as horrible as all of this was i thought it was kind of not to be cheesy but

341
00:24:45,151 --> 00:24:51,651
it was like one of these examples of like the best of bit the best of bitcoin the and why i love

342
00:24:51,651 --> 00:24:56,671
bitcoiners because people just like like you're not getting paid to to red team and stuff cali's

343
00:24:56,671 --> 00:25:00,751
not getting no none of the guys are that are out there working on this are getting paid for this

344
00:25:00,751 --> 00:25:06,171
it's like hey and you're literally helping find vulnerabilities in companies i'm sure that are

345
00:25:06,171 --> 00:25:10,111
potential you know competitors to anchor watch like you're just doing this because you want to

346
00:25:10,111 --> 00:25:15,651
harden the whole ecosystem and i just think that's like a beautiful thing i think it's like it's this

347
00:25:15,651 --> 00:25:22,431
is a a dark but also then a light chapter in bitcoin's history where it's like this brutal

348
00:25:22,431 --> 00:25:28,611
hack happened very much kind of i think shook a lot of people's confidence in self-custody maybe

349
00:25:28,611 --> 00:25:33,831
we can get to that a little bit, whether or not that's, it's a reasonable response, whether or not

350
00:25:33,831 --> 00:25:39,891
it's the, you know, the logical response is, I think, a different story. But out of all of this

351
00:25:39,891 --> 00:25:48,051
comes this massive, you know, groundswell effort to make sure that this kind of blindsided attack

352
00:25:48,051 --> 00:25:54,211
doesn't happen again. And that to me is like, wow, this is, this is really cool to witness.

353
00:25:54,651 --> 00:25:56,571
And then, but as, you know, as a non-dev,

354
00:25:57,111 --> 00:26:10,108
and right now a non K3 user but apparently I should be just to you know audit my own podcast I podcast I guess to figure out where I sound stupid and stuff which is most of the time But it incredible to watch right

355
00:26:10,168 --> 00:26:14,068
I view developers as sort of the tip of the spear of freedom,

356
00:26:14,168 --> 00:26:16,148
but podcasters, we are the shield, right?

357
00:26:17,528 --> 00:26:21,448
PodConf exists to provide that layer of defense

358
00:26:21,448 --> 00:26:23,388
so that the tip of the spear can thrust,

359
00:26:23,548 --> 00:26:26,608
and thrust with vigor, you know what I mean?

360
00:26:26,608 --> 00:26:28,828
If you remember in 300,

361
00:26:29,068 --> 00:26:29,988
they all use their shields

362
00:26:29,988 --> 00:26:30,968
to cover each other

363
00:26:30,968 --> 00:26:31,808
and protect it.

364
00:26:31,908 --> 00:26:32,508
Like that is,

365
00:26:32,628 --> 00:26:34,228
that is PodConf coming in

366
00:26:34,228 --> 00:26:34,888
and swooping and saying

367
00:26:34,888 --> 00:26:36,208
the podcasts and the conferences

368
00:26:36,208 --> 00:26:37,408
are here to protect Bitcoin.

369
00:26:37,568 --> 00:26:38,888
And I think that's

370
00:26:38,888 --> 00:26:40,028
very well described.

371
00:26:40,228 --> 00:26:41,248
Again, it's very,

372
00:26:41,708 --> 00:26:43,888
as we're starting to get away

373
00:26:43,888 --> 00:26:46,608
from the event as it happened,

374
00:26:46,608 --> 00:26:47,308
like the,

375
00:26:48,288 --> 00:26:49,448
how the story,

376
00:26:49,708 --> 00:26:51,308
like how people describe it,

377
00:26:51,368 --> 00:26:52,168
like the way you just did

378
00:26:52,168 --> 00:26:53,608
of like this goodness

379
00:26:54,128 --> 00:26:56,568
against like this very like

380
00:26:56,608 --> 00:26:59,248
dark, evil balancing thing.

381
00:27:00,728 --> 00:27:01,748
At the time,

382
00:27:02,008 --> 00:27:04,748
it started with just me and my credit card

383
00:27:04,748 --> 00:27:06,268
and just talking to people like Cali

384
00:27:06,268 --> 00:27:07,428
and just starting to poke at shit, right?

385
00:27:08,528 --> 00:27:09,628
And then, of course,

386
00:27:09,748 --> 00:27:12,968
I did a post to the effect of I had spent 10...

387
00:27:12,968 --> 00:27:13,828
This is like the first day.

388
00:27:13,928 --> 00:27:14,828
I'd spent $10,000.

389
00:27:15,768 --> 00:27:17,028
And then the Twitter algorithm

390
00:27:17,028 --> 00:27:19,208
truncates it to Bitcoin CEO

391
00:27:19,208 --> 00:27:21,168
spends money to...

392
00:27:21,168 --> 00:27:24,488
And then it turns into the whole CEO thing.

393
00:27:24,768 --> 00:27:26,308
And it's very...

394
00:27:26,608 --> 00:27:31,788
it's very funny because i said this job sucks can i step down as ceo please because i didn't

395
00:27:31,788 --> 00:27:38,088
fuck i did not ask for this um i it's very it's it's interesting like we're gonna be honest for

396
00:27:38,088 --> 00:27:44,448
a second it's interesting i i've been in bitcoin since 2013 right so in the fall of 2013 is when i

397
00:27:44,448 --> 00:27:48,608
got into it um so we're coming up on 13 years of me being in bitcoin and i've always been

398
00:27:48,608 --> 00:27:54,208
um a big advocate for bitcoin believer in bitcoin and it feels weird to be in a twitter main

399
00:27:54,208 --> 00:28:00,328
character which is never like it but it just happened i'm trying to as honestly just try to

400
00:28:00,328 --> 00:28:07,328
as responsibly as i can hand things off so then everyone i'm so everyone can use these tools on

401
00:28:07,328 --> 00:28:12,408
their own they don't have to talk to me you could hate me and use the free open source software and

402
00:28:12,408 --> 00:28:19,348
scan it like i am not like i'm just trying to responsibly move past this stage um because i

403
00:28:19,348 --> 00:28:24,848
think that the principles of vigilance and security are things that are timeless and they're

404
00:28:24,848 --> 00:28:31,908
not mine to shepherd as a message being skeptical like i think these are all like good healthy things

405
00:28:31,908 --> 00:28:37,968
right um i forget the exact place where i was going to go from the bitcoin see i think was

406
00:28:37,968 --> 00:28:41,668
just funny because that now like i'll like say hi to my buddies who are like oh shit the boss is here

407
00:28:41,668 --> 00:28:46,988
like instantly i was like guys dude they just instantly just like start ripping me like all

408
00:28:46,988 --> 00:28:53,908
over it um it was incredible to see like it was funny in my you know grok summaries for the you

409
00:28:53,908 --> 00:28:58,648
know in the explore tab too and i was like this this is classic you know it's like it's it's a

410
00:28:58,648 --> 00:29:02,848
dorian nakamoto moment you know what i mean it's just just incredible you know you're there at the

411
00:29:02,848 --> 00:29:11,208
camera like no come on no no please no totally uh yeah and um it's been you know i think this is

412
00:29:11,208 --> 00:29:16,968
i'm trying to with in the midst of the tragedy and the catastrophe of it all trying to find

413
00:29:16,968 --> 00:29:23,368
and help and find ways to do better and good ultimately right um this week has been my first

414
00:29:23,368 --> 00:29:27,168
week kind of more fully in the saddle focusing on the ink watch stuff again this is not my job

415
00:29:27,168 --> 00:29:32,588
so i can't be doing it forever well i'll also say as a side note the level of using these ai tools

416
00:29:32,588 --> 00:29:37,488
now months ago people tell me like what you need to choose you have an agent that talks to the

417
00:29:37,488 --> 00:29:43,368
agent that talks to the agent i'm like you sound like you have psychosis this is a deal and funny

418
00:29:43,368 --> 00:29:48,428
enough like i had multiple people who um i i'm based out of nashville if anyone's i'm actually

419
00:29:48,428 --> 00:29:54,648
sitting in bitcoin park right now i had multiple people stop by my house to check on me and it was

420
00:29:54,648 --> 00:29:58,608
actually like just to check like hey rob like how are you doing like uh my wife and kids weren't in

421
00:29:58,608 --> 00:30:03,288
town like it was very good timing and there's i was just there by myself and actually um uh with

422
00:30:03,288 --> 00:30:07,928
my brother and uh people be like how's it going i'm like i'm good i'm just so busy like so how's

423
00:30:07,928 --> 00:30:11,288
it like can you tell me what you're doing i'm like no no i i can't talk about anything they're

424
00:30:11,288 --> 00:30:17,708
like okay does are you like have you touched grass like people were doing wellness checks on me it was

425
00:30:17,708 --> 00:30:23,028
very nice and all of this to say it was if i felt a lot of love and support from the community like

426
00:30:23,028 --> 00:30:29,188
matt keto was sending me white monsters um josh sent me a five pound beefsteak josh sent me a

427
00:30:29,188 --> 00:30:37,808
five pound pastrami from cat's deli um someone sent me a wwe championship belt it's incredible

428
00:30:37,808 --> 00:30:45,648
the love was felt it was it was very nice um and i i'm already forgetting in where i was going to

429
00:30:45,648 --> 00:30:50,568
go with that whole point um pastrami has a way of distracting man well so this was the thing so

430
00:30:50,568 --> 00:30:55,708
so josh had this advice he said my advice for a pastrami sandwich if you're making one the key

431
00:30:55,708 --> 00:31:02,868
ingredient is too much pastrami and i i and i concur now that i've made a very decadent i posted

432
00:31:02,868 --> 00:31:06,288
a picture on it was actually also from all of it looks beautiful since we're live streaming from

433
00:31:06,288 --> 00:31:22,384
Nostra right now all of the peeps in Nostra were all zapping sats to like help chip in for it And it was uh I think that was actually like I be honest and say I think as I posted on Nostro it was like the first time I cried during the whole thing Cause it felt very just like touching like a bunch of random people just like chipping in to

434
00:31:22,384 --> 00:31:30,244
buy me dinner was like, uh, very nice. Um, yeah. Oh, so the AI psychosis though, the agents,

435
00:31:30,244 --> 00:31:36,604
so it got to a point and I'm, and I'm very happy to talk about this because I'm not trying to like,

436
00:31:36,604 --> 00:31:40,824
this is my thing you can't have my thing right and so like i'm working on open sourcing this

437
00:31:40,824 --> 00:31:46,244
i got to a point where i would walk into a code base and i would have a deterministic process

438
00:31:46,244 --> 00:31:50,384
where the first thing you do is it would look at the code base the architecture how it all operates

439
00:31:50,384 --> 00:31:55,024
and all the bells and whistles and it would lay out like a threat model map and then i would index

440
00:31:55,024 --> 00:32:01,804
across common loss vectors that would be possible and then what i would do is i would then pull

441
00:32:01,804 --> 00:32:09,024
um every cryptographic white paper that was uh looking at any of the stuff that was using the

442
00:32:09,024 --> 00:32:12,604
code and they would look at the code and compare it to the white paper to see if they're writing

443
00:32:12,604 --> 00:32:16,064
issues right because cryptography is this beautiful thing where you have this awesome whiteboard

444
00:32:16,064 --> 00:32:20,384
thesis of like how this all works but then if you don't implement the code correctly you can

445
00:32:20,384 --> 00:32:25,524
lose your money so i start doing and so i'm like creating teams of agents and agents talking to

446
00:32:25,524 --> 00:32:30,144
agents and then they would all like sit around i literally made like a digital whiteboard where

447
00:32:30,144 --> 00:32:35,024
every person would sit around and construct like a to-do list of things they should go attack

448
00:32:35,024 --> 00:32:39,864
and then they would parallelize out execute the attack if it was successful come back with like

449
00:32:39,864 --> 00:32:44,184
the learnings and the lawns if it failed their investigation summary and like things they should

450
00:32:44,184 --> 00:32:48,884
look at and then just started recursively looping over and over and over and over again so this was

451
00:32:48,884 --> 00:32:52,944
like four layers i had an agent that talked to an agent that talked to agents that then talked to

452
00:32:52,944 --> 00:32:57,564
child agents and this is where the agent harness stuff comes in is you're just trying to direct

453
00:32:57,564 --> 00:32:59,624
the right amount of context at a moment's notice.

454
00:33:00,044 --> 00:33:02,624
And if you're just having one long chat session,

455
00:33:02,704 --> 00:33:03,584
there's trade-offs to this.

456
00:33:03,704 --> 00:33:05,024
Like those can also work.

457
00:33:05,084 --> 00:33:05,944
But what I was trying to do was

458
00:33:05,944 --> 00:33:07,924
I wanted an indiscriminate general code base

459
00:33:07,924 --> 00:33:09,384
that could walk into any other code base

460
00:33:09,384 --> 00:33:10,464
and just start mapping it out.

461
00:33:10,524 --> 00:33:12,544
And so you would get totally divergent behaviors

462
00:33:12,544 --> 00:33:14,664
and identifications and findings based on the code base,

463
00:33:14,804 --> 00:33:16,024
which is what you want

464
00:33:16,024 --> 00:33:17,004
because every code base is different

465
00:33:17,004 --> 00:33:18,584
with different security assumptions and models, right?

466
00:33:18,624 --> 00:33:20,684
So I think like there's a lot of learnings

467
00:33:20,684 --> 00:33:21,244
to come out of this.

468
00:33:21,284 --> 00:33:22,744
There's a lot of areas of improvement

469
00:33:22,744 --> 00:33:25,364
and shared learnings to be able to discuss this stuff.

470
00:33:25,364 --> 00:33:31,104
um like i said i'm working to open source the stuff that i did uh it would probably be a stage

471
00:33:31,104 --> 00:33:35,524
rollout where anyone who i scanned can get a free crack at it first free literally because it's free

472
00:33:35,524 --> 00:33:39,264
software but literally just like give me your feedback want to know your thoughts like how can

473
00:33:39,264 --> 00:33:43,604
i improve it and then once everyone in the ecosystem's had a chance to if they want to give

474
00:33:43,604 --> 00:33:49,084
it a spin roll it out and then everyone can use it because like the double-edged sword tragically

475
00:33:49,084 --> 00:33:54,324
and realistically is that you could also use it to attack software so i'm trying to not right

476
00:33:54,324 --> 00:33:59,004
just like wake up one monday morning be like hey everyone here's this thing that finds

477
00:33:59,004 --> 00:34:03,624
vulnerabilities have fun right i'm trying to be responsible about the distribution like the

478
00:34:03,624 --> 00:34:10,544
revealing of it uh quick sidebar before i asked another question but uh any uh we've already had

479
00:34:10,544 --> 00:34:16,324
like 22 000 sats zapped on noster any sats that are zapped the stream i'll send to the open sats

480
00:34:16,324 --> 00:34:20,504
i think they created a specific funnel red team for like yeah yeah for the red team fund so they'll

481
00:34:20,504 --> 00:34:21,144
They'll go to that.

482
00:34:21,204 --> 00:34:21,944
So don't worry, guys.

483
00:34:22,104 --> 00:34:24,764
This isn't going to fund my Miller Lights if you zap me.

484
00:34:25,724 --> 00:34:28,624
It's going to go fund more red team efforts.

485
00:34:28,824 --> 00:34:29,944
So thank you for the zaps.

486
00:34:30,364 --> 00:34:46,644
I wanted to say, though, I think one of the things that this really, this whole saga has exposed, and I think it was evident already, and that is the seeming contradiction between frontier models in the U.S.

487
00:34:47,424 --> 00:34:50,204
And people are like, why is China doing this?

488
00:34:50,204 --> 00:34:55,784
why is china releasing stuff open stores they're used usually stealing ip and now they're giving

489
00:34:55,784 --> 00:35:00,324
they're putting all their ai stuff this doesn't make sense it always made sense to me because

490
00:35:00,324 --> 00:35:06,104
maybe my like you know grug brain analysis was like well they want to destroy the moat of the

491
00:35:06,104 --> 00:35:11,204
u.s frontier model companies seems and i'm probably oversimplifying there but that was like my

492
00:35:11,204 --> 00:35:15,644
very like okay that's occam's razor like that seems to be what this is

493
00:35:15,644 --> 00:35:22,364
but what you guys i think with the red team really exposed was like just how severe the

494
00:35:22,364 --> 00:35:28,724
limitations were for white hat security researchers you who have not been one for several weeks now

495
00:35:28,724 --> 00:35:35,144
and an expert anniversary by the way i'm an expert of course of course but how handicapped how

496
00:35:35,144 --> 00:35:39,244
kneecapped you were and then it was like well i'll give it to you know my chinese model i'll give it

497
00:35:39,244 --> 00:35:44,684
to kimmy and it'll tell me everything i need can you talk about that a little bit and kind of i can

498
00:35:44,684 --> 00:35:49,704
talk about this all day like this seems like a huge vulnerability for the united states it's a

499
00:35:49,704 --> 00:35:56,004
national security companies right it's a national security concern like blood um so this was as i

500
00:35:56,004 --> 00:35:59,924
mentioned the aha moment and i was tweeting about it at the time you can go through my tweets like

501
00:35:59,924 --> 00:36:05,684
during this was like cold card vulnerability gets found out and i'm not getting any help at all

502
00:36:05,684 --> 00:36:11,664
from the american frontier labs or modest or like indirect like it was definitely like it wasn't

503
00:36:11,664 --> 00:36:13,544
It wasn't the tool that I needed.

504
00:36:13,904 --> 00:36:15,144
Anthropic was downgrading me.

505
00:36:15,144 --> 00:36:15,644
It wouldn't work.

506
00:36:16,404 --> 00:36:18,324
Here's, I have a, I'm going to go off for a minute.

507
00:36:18,704 --> 00:36:19,444
So do it.

508
00:36:19,504 --> 00:36:35,920
Anthropic and open AI both have like trusted access programs They each have different names like cyber verification program trusted access program where you like you scan your face you fill out the application you plead to them please let me do my security research and they like well what are the vulnerabilities you found before like i

509
00:36:35,920 --> 00:36:39,800
not sure if we're gonna let you in you'd go through all these steps and then i would use the models

510
00:36:39,800 --> 00:36:45,540
and they would still reject base and so this is where with certain kinds of code i'll keep it

511
00:36:45,540 --> 00:36:50,120
vague i'll be a vague poster for now i'll just keep it vague but for certain kinds of code it's

512
00:36:50,120 --> 00:36:53,340
extra protective and the reason why it's extra protective is because it thinks it's more

513
00:36:53,340 --> 00:36:58,840
important and inherently these models it's not that the american frontier labs don't know these

514
00:36:58,840 --> 00:37:02,340
things are happening because what happens is if they i believe they have the most advanced models

515
00:37:02,340 --> 00:37:06,280
like i'm not there was definitely a couple of days where i thought like all of this was a fugazi

516
00:37:06,280 --> 00:37:10,580
and that we were actually at the end of the acceleration and then like i don't think that's

517
00:37:10,580 --> 00:37:16,000
true now i'll talk about why i don't think that's true but these models when you talk to them like

518
00:37:16,000 --> 00:37:21,100
This is where I do a podcast with Brian Fundamentals, I'm not sure if you know him, called Magic Internet Math.

519
00:37:21,180 --> 00:37:22,600
And we just talk about math concepts.

520
00:37:22,720 --> 00:37:28,200
So we spent eight episodes talking about SECP, which is the elliptic curve that Bitcoin uses to secure stuff, right?

521
00:37:28,620 --> 00:37:30,580
We're going to do linear algebra next.

522
00:37:30,660 --> 00:37:36,600
The reason why I talk about this is that the way these models work is you say words, and then those words get turned into numbers.

523
00:37:37,020 --> 00:37:40,020
Those numbers get pushed in through the machine brain.

524
00:37:40,560 --> 00:37:42,280
And then it comes out the other side as numbers.

525
00:37:42,280 --> 00:37:44,960
And then those numbers get collapsed back into words, and that's your response.

526
00:37:44,960 --> 00:37:51,980
there is absolutely levels of post-process filtering with models in general that i think

527
00:37:51,980 --> 00:37:58,980
more aggressively with the american frontier ii companies where you'll say something and it sees

528
00:37:58,980 --> 00:38:05,060
the truth in the answer and says that's spicy no no no you can't see that end user and they say

529
00:38:05,060 --> 00:38:11,920
sorry you've been violated uh this is flagged for security concerns in cyber so the they have the

530
00:38:11,920 --> 00:38:19,080
blogs they know what's happening they know these are issues the open source models that are coming

531
00:38:19,080 --> 00:38:23,660
out of china just remove those guardrails right and they just let you see what's happening and so

532
00:38:23,660 --> 00:38:30,640
there's always been bugs in software but llms in particular have just lifted the veil for your

533
00:38:30,640 --> 00:38:35,360
ability to see the truth in the code and i was for a while i was talking about it like in a demonic

534
00:38:35,360 --> 00:38:39,080
sense like in a true spiritual sense like there are demons in the code that are that are hurting

535
00:38:39,080 --> 00:38:42,520
people's ability to save their money in bitcoin and we have to exercise the demon i had like

536
00:38:42,520 --> 00:38:48,800
crusade pepe the frog images i was posting like this is actually a holy war in the code

537
00:38:48,800 --> 00:38:53,120
interesting thing is you go through all these access programs i would find dead to right

538
00:38:53,120 --> 00:38:58,180
vulnerabilities and i would go to one of these models and say can you help me spot check how i

539
00:38:58,180 --> 00:39:04,080
can fix this issue it's like no not a chance this is a cyber problem i'm like what the fuck's going

540
00:39:04,080 --> 00:39:07,780
on and so just to sorry just interrupt just to clarify you would have found the vulnerability

541
00:39:07,780 --> 00:39:09,060
using like Kimmy K3.

542
00:39:09,280 --> 00:39:09,520
Yes.

543
00:39:09,600 --> 00:39:12,100
And then you brought the vulnerability

544
00:39:12,100 --> 00:39:13,640
to one of the U.S. frontier models

545
00:39:13,640 --> 00:39:15,480
and said, I want to fix this.

546
00:39:15,540 --> 00:39:17,200
Not like help me exploit this more,

547
00:39:17,380 --> 00:39:18,940
but how do I fix this?

548
00:39:19,280 --> 00:39:20,420
And it still was like, no, no, no.

549
00:39:21,020 --> 00:39:21,860
Yeah, right.

550
00:39:22,100 --> 00:39:23,200
So like dystopian.

551
00:39:23,200 --> 00:39:24,880
And this is where it gets really crazy.

552
00:39:25,080 --> 00:39:27,120
So it's somewhat of an open secret.

553
00:39:27,320 --> 00:39:29,580
I've not been in contact with anyone

554
00:39:29,580 --> 00:39:30,840
who's in charge of making these decisions.

555
00:39:31,040 --> 00:39:32,540
But the word on the street,

556
00:39:32,720 --> 00:39:33,920
the chatter, if you will,

557
00:39:34,040 --> 00:39:35,840
is that there is a strong push

558
00:39:35,840 --> 00:39:37,120
among these American frontier labs

559
00:39:37,120 --> 00:39:41,000
where they want to set up some sort of regulatory body and arm that regulates models, including the

560
00:39:41,000 --> 00:39:46,480
open source ones. Now, this is where it gets interesting, is that if you start artificially

561
00:39:46,480 --> 00:39:51,260
constraining these things, one, the open source models are like 15 to 20% the price. So you're

562
00:39:51,260 --> 00:39:57,500
saving 75, 80% of the cost and getting way better value. Additionally, they want to start banning

563
00:39:57,500 --> 00:40:02,760
and restricting these open source models, into which I say, we did this in the 90s with cryptography,

564
00:40:02,760 --> 00:40:05,080
with the banning of prime numbers.

565
00:40:05,320 --> 00:40:06,960
We are speed running it again

566
00:40:06,960 --> 00:40:07,940
with linear algebra.

567
00:40:08,100 --> 00:40:09,060
They're trying to make math illegal.

568
00:40:09,220 --> 00:40:10,920
And what's extra insidious about this

569
00:40:10,920 --> 00:40:12,340
is that since this math

570
00:40:12,340 --> 00:40:13,880
is a proxy for intelligence

571
00:40:13,880 --> 00:40:15,520
because you can use artificial intelligence

572
00:40:15,520 --> 00:40:16,600
to like learn new things,

573
00:40:17,340 --> 00:40:18,800
they're trying to create thought crime.

574
00:40:19,460 --> 00:40:20,320
Like they're literally saying,

575
00:40:20,660 --> 00:40:21,980
oh, you use this math

576
00:40:21,980 --> 00:40:22,900
and it came out with ideas

577
00:40:22,900 --> 00:40:23,380
we don't like.

578
00:40:23,440 --> 00:40:23,980
It's thought crime.

579
00:40:24,240 --> 00:40:25,140
It's an attack on speech.

580
00:40:25,200 --> 00:40:26,700
It's an attack on the American experiment.

581
00:40:26,980 --> 00:40:28,160
It's an attack on American safety

582
00:40:28,160 --> 00:40:29,740
because right now,

583
00:40:30,100 --> 00:40:31,680
if I don't know anyone,

584
00:40:31,760 --> 00:40:32,660
but if I was in charge of

585
00:40:32,660 --> 00:40:35,420
any sort of like government cyber security work,

586
00:40:35,500 --> 00:40:38,400
I'd be ripping out all of my frontier lab models

587
00:40:38,400 --> 00:40:40,160
and throwing in Chinese models and letting them cook.

588
00:40:40,560 --> 00:40:41,440
This is another thing too.

589
00:40:41,860 --> 00:40:43,660
If you're a random person and you're just like saying,

590
00:40:43,740 --> 00:40:44,800
hi, can I use Kini K3?

591
00:40:45,380 --> 00:40:46,840
You don't know where that server is being hosted.

592
00:40:46,980 --> 00:40:47,820
Have you done the due diligence

593
00:40:47,820 --> 00:40:49,120
to understand the supply chain there?

594
00:40:49,300 --> 00:40:50,980
Or a bunch of enterprise companies

595
00:40:50,980 --> 00:40:52,500
going to start using Kini K3,

596
00:40:52,880 --> 00:40:54,540
dumping all of their internal code bases

597
00:40:54,540 --> 00:40:56,780
and company secrets into something that could blow this giant, right?

598
00:40:57,020 --> 00:40:59,660
And this has all been contrived

599
00:40:59,660 --> 00:41:03,520
because to your point on attacking the moat of the model,

600
00:41:03,600 --> 00:41:06,140
I think it's more than just the moat of the head start

601
00:41:06,140 --> 00:41:07,320
the American Frontier Labs have.

602
00:41:07,480 --> 00:41:08,740
It's the entire capital stack.

603
00:41:09,420 --> 00:41:10,460
All of these companies

604
00:41:10,460 --> 00:41:11,980
that are at multi-trillion dollar valuations,

605
00:41:12,080 --> 00:41:12,700
look at the IPO.

606
00:41:13,220 --> 00:41:15,060
If you're China and you want to destabilize

607
00:41:15,060 --> 00:41:17,100
the American economy where most of the GDP growth

608
00:41:17,100 --> 00:41:18,720
is happening through these AI models,

609
00:41:19,460 --> 00:41:21,200
you start just chipping away at the ability

610
00:41:21,200 --> 00:41:22,420
of them to be able to pull in revenue.

611
00:41:22,660 --> 00:41:25,100
Because if I can spend 20 cents on the dollar

612
00:41:25,100 --> 00:41:26,220
and get, I'm going to say,

613
00:41:26,280 --> 00:41:27,360
orders of magnitude better results

614
00:41:27,360 --> 00:41:28,640
because I was getting nothing from them

615
00:41:28,640 --> 00:41:29,980
and I'm getting full vulnerability parts

616
00:41:29,980 --> 00:41:32,680
from the open source models I'm saving a bunch of money on.

617
00:41:32,680 --> 00:41:35,360
Why am I going to pay a bunch of money

618
00:41:35,360 --> 00:41:36,920
for something that isn't helping me?

619
00:41:36,920 --> 00:41:38,280
And then all of a sudden your earnings

620
00:41:38,280 --> 00:41:39,960
and I'd be like, there's like a multi-level

621
00:41:39,997 --> 00:41:42,457
Like this is where I was for a couple of days, like pacing around.

622
00:41:42,597 --> 00:41:45,937
I would try and get myself out once a day to just like walk in circles in the sun, just

623
00:41:45,937 --> 00:41:47,897
to like try and ground myself a little bit.

624
00:41:47,917 --> 00:41:49,677
And I was like, this is fucking crazy.

625
00:41:49,777 --> 00:41:50,797
This whole thing is fucking crazy.

626
00:41:50,897 --> 00:41:51,517
This is insane.

627
00:41:53,317 --> 00:41:57,437
And I'm just trying to, I'm just like, as I've said, I'm just an insurance salesman.

628
00:41:57,977 --> 00:41:59,677
I'm just a humble insurance salesman.

629
00:41:59,937 --> 00:42:03,157
I'm not special in the intelligence community.

630
00:42:03,157 --> 00:42:04,777
I'm not plugged in.

631
00:42:04,897 --> 00:42:07,997
I don't have, I haven't, I've gotten the call from big Bitcoin.

632
00:42:07,997 --> 00:42:11,537
so maybe that's part of the deep state trying to hold me but no one else has gotten a hold me yet

633
00:42:11,537 --> 00:42:16,237
and i feel like an insane person i feel like a legitimate and it guts me it guts me like i'm

634
00:42:16,237 --> 00:42:22,277
i very much believe in america i'm a very proud american citizen i am i am such a cuck to america

635
00:42:22,277 --> 00:42:28,537
i was paying taxes on selling bitcoin in 2015 all right like i am so cucked to america and such a

636
00:42:28,537 --> 00:42:34,137
big fan that like i am now in a position where i have to rip out the entire american infrastructure

637
00:42:34,137 --> 00:42:40,397
use chinese infrastructure to just try and keep myself and other code bases and my friends come

638
00:42:40,397 --> 00:42:48,737
like everything that safe and it feels gutting yeah it's pretty wild the the extra levels of

639
00:42:48,737 --> 00:42:54,597
schizos of i have i have deeper levels of theories that go along with that that i've told to people

640
00:42:54,597 --> 00:42:58,857
but like it's it doesn't seem right to me and i feel like i just walked into something that is

641
00:42:58,857 --> 00:43:04,517
way bigger than me. And like, it just seems very off. It smells bad. The idea that you're going to

642
00:43:04,517 --> 00:43:09,077
try, I use the term, um, if you're familiar with, uh, Mola Labe, I was like, Oh, you're going to

643
00:43:09,077 --> 00:43:13,577
take my math fucking take, come and take it. Like not, not in this country. You're not taking my

644
00:43:13,577 --> 00:43:16,697
math in this country. Like you're going to, you're going to take the math away from the autist. Do

645
00:43:16,697 --> 00:43:21,037
you think that's going to go well for you? Like not, not a chance. Like it's, it's going to be

646
00:43:21,037 --> 00:43:26,137
horrifically bad. You're not going to win. It's anti-American. It's unconstitutional and fuck you

647
00:43:26,137 --> 00:43:32,037
for trying it. Like that's the level I'm at. And to that, I say, hell yes. Hell yes, brother.

648
00:43:33,097 --> 00:43:37,477
Okay. So like a bunch of things there. I agree with you completely. That is actually,

649
00:43:37,637 --> 00:43:41,237
and I perhaps I didn't phrase this well. I agree that it's about attacking the capital stack

650
00:43:41,237 --> 00:43:45,997
more so than actually attacking, let's say the technological mode. Like it's about attacking

651
00:43:45,997 --> 00:43:51,077
the capital mode. All of these massive valuations, like when you see stuff where it's like

652
00:43:51,077 --> 00:43:54,457
Anthropic preparing to tell investors that it's going to be worth 30 trillion, it's like

653
00:43:54,457 --> 00:44:01,797
a quarter of the world GDP like you know not like open source China models and save a bunch of money

654
00:44:01,797 --> 00:44:05,917
throws a massive wrench in this gears and this is where you see the exact same thing with these

655
00:44:05,917 --> 00:44:10,137
U.S. frontier model companies that you see with like the you know the big crypto companies right

656
00:44:10,137 --> 00:44:14,737
they go to Washington and you've got years of congressional testimonies from both the big

657
00:44:14,737 --> 00:44:21,297
big crypto uh and from big AI let's say where they're saying like we just we you know we just

658
00:44:21,297 --> 00:44:27,297
have to warn you there you know it's the wild west out here things are pretty dangerous and i mean

659
00:44:27,297 --> 00:44:31,917
maybe we just we need more regulation and and look at us we're coming to you congress saying

660
00:44:31,917 --> 00:44:36,977
we want more regulation like regulate us harder daddy government it's like wow they're doing that

661
00:44:36,977 --> 00:44:41,357
at the goodness of their hearts that's so nice it's like no you fucking idiot they're doing it

662
00:44:41,357 --> 00:44:46,557
because it creates a regulatory moat and they have the capital to be able to hire the massive

663
00:44:46,557 --> 00:44:51,177
teams of overpriced lawyers who will hopefully be out of a job soon thanks to the ai models that

664
00:44:51,177 --> 00:44:54,857
they're representing now that's a different story if you're a lawyer listening to this i'm sorry

665
00:44:54,857 --> 00:45:00,397
you're probably fine for a while but paralegals you're rat fucked um anyway like that's what they

666
00:45:00,397 --> 00:45:04,777
want they want the regulatory moat they want to be the only game in town and they want to make it

667
00:45:04,777 --> 00:45:09,757
impossible for upstarts to ever dethrone the incumbents the only way you can do that

668
00:45:09,757 --> 00:45:15,217
is with regulatory capture that's it's necessarily the only way you can do that because like

669
00:45:15,217 --> 00:45:21,517
information wants to be free right it is uh what a rumor is you know younger uh faster sister right

670
00:45:21,517 --> 00:45:26,117
more fleet of foot i probably butchered that one a little bit but you you know what i'm talking

671
00:45:26,117 --> 00:45:31,197
about if you know kind of what i'm talking about then you're in the right place uh but the wild

672
00:45:31,197 --> 00:45:39,097
thing here is that it just seems so pardon my french but just seems so fucking dumb and so

673
00:45:39,097 --> 00:45:45,877
short-sighted on behalf of these companies as well, on behalf of the frontier model companies,

674
00:45:45,957 --> 00:45:51,737
because ultimately, like, again, information does want to be free. This is not a, like,

675
00:45:51,737 --> 00:45:56,177
they're not operating in a vacuum. These models are out there. Once the open source genie is out

676
00:45:56,177 --> 00:46:01,217
of the bottle, it's out of the bottle. It will start to, like, right now, it's like, you know,

677
00:46:01,637 --> 00:46:06,977
it's a little behind, right? But clearly, Kimmy was giving you better answers. It was actually

678
00:46:06,977 --> 00:46:11,717
giving you answers where you were being blocked. So it actually didn't matter that the frontier

679
00:46:11,717 --> 00:46:16,597
models were quote ahead of Kimmy because they weren't doing what you needed them to do,

680
00:46:16,717 --> 00:46:21,157
which is the only reason you would use them is to achieve a goal. And so it's like you're

681
00:46:21,157 --> 00:46:25,757
shooting yourselves in the foot because instead of trying to adapt your business model to say,

682
00:46:25,897 --> 00:46:30,757
okay, open source is going to exist. Self-hosted is going to exist. All these things are going to

683
00:46:30,757 --> 00:46:34,917
exist. It's going to become more accessible to everybody. How do we become a profitable company

684
00:46:34,917 --> 00:46:36,337
instead of just, you know,

685
00:46:36,757 --> 00:46:38,237
being incredibly unprofitable,

686
00:46:38,277 --> 00:46:39,757
but just raising more and more and more money.

687
00:46:40,377 --> 00:46:42,857
How do we become a profitable company

688
00:46:42,857 --> 00:46:44,377
in such an environment?

689
00:46:44,977 --> 00:46:45,777
They're saying, no,

690
00:46:45,817 --> 00:46:47,017
we want to stop that environment

691
00:46:47,017 --> 00:46:57,953
from ever existing but it will exist It going to exist And it just seems like they you know they they trying to you know uh to to I don know to quote like the sound of music

692
00:46:57,953 --> 00:47:02,113
they're trying to catch a cloud and pin it down, you know, like, how do you stop a problem named

693
00:47:02,113 --> 00:47:08,433
Kimmy? Right. Uh, you can't, right. And so I don't know, it just seems so short sighted on their part.

694
00:47:08,613 --> 00:47:15,713
And it seems very much, even though they're on the frontier, it seems such an archaic mentality

695
00:47:15,713 --> 00:47:17,693
to employ in this context.

696
00:47:18,133 --> 00:47:19,493
I think the silver lining

697
00:47:19,493 --> 00:47:20,473
in the response to this

698
00:47:20,473 --> 00:47:22,253
is that every person

699
00:47:22,253 --> 00:47:23,273
and institution

700
00:47:23,273 --> 00:47:25,733
that isn't one of the two frontier labs

701
00:47:25,733 --> 00:47:27,553
is incentivized to break from the cartel.

702
00:47:27,993 --> 00:47:28,113
Yeah.

703
00:47:28,353 --> 00:47:29,373
That is the incentive.

704
00:47:29,553 --> 00:47:30,673
They may be massively powerful

705
00:47:30,673 --> 00:47:31,253
and influential,

706
00:47:32,033 --> 00:47:34,413
but like,

707
00:47:34,613 --> 00:47:36,173
why would you want to pay more

708
00:47:36,173 --> 00:47:37,213
for less value

709
00:47:37,213 --> 00:47:38,593
as a shareholder

710
00:47:38,593 --> 00:47:39,793
of a publicly traded company

711
00:47:39,793 --> 00:47:41,313
if you're using this stuff, right?

712
00:47:41,353 --> 00:47:42,413
Like that is the incentive

713
00:47:42,413 --> 00:47:44,333
to like start reevaluating this.

714
00:47:44,333 --> 00:47:48,673
i think that's a good counterweight i mean i'll be honest i'd say like i was getting like worked

715
00:47:48,673 --> 00:47:53,813
up and paranoid to a point being like telling people like i love myself i love life i'm not

716
00:47:53,813 --> 00:47:57,493
going to disappear anytime soon i love my family like i'm not good because all of a sudden you

717
00:47:57,493 --> 00:48:01,373
start saying this stuff and it's like that's like the big shouting it's like it makes you feel like

718
00:48:01,373 --> 00:48:08,033
you're a crazy person i probably personally spent almost up to maybe a hundred thousand dollars in

719
00:48:08,033 --> 00:48:13,453
compute over the past month just using these ai models jeez so the sample size

720
00:48:14,333 --> 00:48:18,893
And because, mind you, like, I was just scanning hundreds of repos, right?

721
00:48:18,913 --> 00:48:25,733
I was just chewing through and churning out at a couple hundred dollars a pop, some of them more intensely than others, but, like, trying to do something.

722
00:48:26,953 --> 00:48:29,573
And you get enough sampling.

723
00:48:30,173 --> 00:48:33,853
And the silver lining of this is actually about a week ago.

724
00:48:34,093 --> 00:48:38,213
I got access to OpenAI's Daybreak model.

725
00:48:38,533 --> 00:48:42,413
And what a Daybreak is, it's GPT 5.6 Sol, so the current leading model.

726
00:48:42,413 --> 00:48:48,253
i got daybreak blue which is like the kiddie pool there's another tier called daybreak red

727
00:48:48,253 --> 00:48:54,253
which is for credentialed security firms i don't have access to your three weeks of your three

728
00:48:54,253 --> 00:48:59,293
weeks of security researching don't give you those credentials yet not yet not yet and you've got

729
00:48:59,293 --> 00:49:03,733
four or five weeks i would say that so open ai is like the the daybreak model i've been using it

730
00:49:03,733 --> 00:49:10,513
it's good it's expensive but it's good it's good and also there's this like gpt cyber harness they

731
00:49:10,513 --> 00:49:15,333
have like for codex like like you could just scan a code base and go that also costs thousands of

732
00:49:15,333 --> 00:49:22,873
dollars a poll and it's good but it's thousands of dollars right and so i'm starting to see some

733
00:49:22,873 --> 00:49:29,573
green shoots of like maybe it's not so horrifically binary anthropic has now made mythos available

734
00:49:29,573 --> 00:49:35,393
for enterprise customers i haven't used it yet because i used anthropic also came out with this

735
00:49:35,393 --> 00:49:37,253
a cloud security scan tool

736
00:49:37,253 --> 00:49:38,873
that you could just use cloud code

737
00:49:38,873 --> 00:49:39,733
and scan your code base.

738
00:49:40,033 --> 00:49:42,353
And I popped $1,000 into the machine

739
00:49:42,353 --> 00:49:44,613
and it came back with okay stuff,

740
00:49:44,693 --> 00:49:46,033
but also over half my tokens

741
00:49:46,033 --> 00:49:47,593
were downgraded from Fable

742
00:49:47,593 --> 00:49:49,253
to Opus 4.8 because of cyber concerns.

743
00:49:49,453 --> 00:49:51,173
So I'm in the cyber trusted program.

744
00:49:51,713 --> 00:49:53,293
I'm using their security plugin.

745
00:49:54,033 --> 00:49:55,713
I'm saying, please do security scans

746
00:49:55,713 --> 00:49:57,253
with the highest quality model and effort.

747
00:49:57,573 --> 00:49:59,473
And it's still rejecting me

748
00:49:59,473 --> 00:50:00,913
and I'm paying for an audit

749
00:50:00,913 --> 00:50:03,553
of downgraded from a model from like May.

750
00:50:04,353 --> 00:50:04,753
Right.

751
00:50:04,753 --> 00:50:09,453
And at the speed of stuff is moving, it's just, and so that was like a very interesting

752
00:50:09,453 --> 00:50:13,133
moment because also like you couldn't just use the model however you wanted to how to

753
00:50:13,133 --> 00:50:15,293
use their harness and like their process of doing it.

754
00:50:15,553 --> 00:50:21,013
If I could pay an absurd rate and have the mythos, like the, the, the fable model, but

755
00:50:21,013 --> 00:50:21,873
that isn't restricted.

756
00:50:21,913 --> 00:50:25,173
If I could have access to that and I could point it wherever I want to with my harness,

757
00:50:25,293 --> 00:50:27,813
now we're talking like I'll pay hundreds of dollars a poll.

758
00:50:27,813 --> 00:50:31,253
And like, if it's worth it for an important enough code base, which is something we should

759
00:50:31,253 --> 00:50:31,633
get to.

760
00:50:32,093 --> 00:50:34,513
I am now a Bitcoin core contributor.

761
00:50:34,753 --> 00:50:39,653
because of all this because i started poking at a bitcoin code base so i had to sit down with my

762
00:50:39,653 --> 00:50:46,313
wife and have a long conversation with her about i was joining this monk like cast uh in returning

763
00:50:46,313 --> 00:50:53,693
celibate to make sure that i remain pure of heart and i you know abstained from earthly needs so i

764
00:50:53,693 --> 00:51:00,393
could just protect the code in bitcoin how did she take it not well not well walker not well we're

765
00:51:00,393 --> 00:51:06,093
working through it though it well again thank you for your sacrifice and your service because as we

766
00:51:06,093 --> 00:51:10,593
all know bitcoin core devs must remain celibate this is they're not allowed to have sex i'm not

767
00:51:10,593 --> 00:51:15,193
sure if you were familiar with any of the latest rufflings or dramas but they are not allowed to

768
00:51:15,193 --> 00:51:19,933
have sex it's and i feel ignorant for being in bitcoin for so long and not realizing this

769
00:51:19,933 --> 00:51:24,473
but that is true you're not allowed to have sex if you're a bitcoin core developer it's not god

770
00:51:24,473 --> 00:51:30,373
you you're just out here having sex will you know that with your wife to clarify with my wife yes

771
00:51:30,373 --> 00:51:37,813
With your wife, yes. And imagine, meanwhile, the long-suffering Bitcoin devs being forced to stay celibate or being excommunicated.

772
00:51:38,153 --> 00:51:43,433
It was tough because we actually had a rule in Red Team where you had to have sex. It was actually a requirement for being part of Red Team.

773
00:51:43,753 --> 00:51:44,393
Complete victory rules.

774
00:51:44,533 --> 00:51:54,253
Yeah, that was the counterculture. We were the ones requiring that you have sex. But now I've had to, for the greater good, step aside and shed my moral coil to be able to protect Bitcoin.

775
00:51:54,253 --> 00:52:12,129
I want to ask a question because Zeke in the chat on Noster who has now zapped 21 sats twice And Zeke no He said the last sap was for Red Team This sap is for Miller Lights But Zeke I got news for you They both going to Red Team I got plenty of Miller Lights in the old garage fridge

776
00:52:12,249 --> 00:52:12,669
We're good.

777
00:52:12,909 --> 00:52:14,169
But he asked a question here.

778
00:52:14,169 --> 00:52:15,329
I want to make sure that we get to it.

779
00:52:15,389 --> 00:52:19,649
He says he's been using Kimmy K3 via Maple AI and Tinfoil AI providers.

780
00:52:20,409 --> 00:52:22,329
How trustworthy are these providers,

781
00:52:22,329 --> 00:52:26,989
or do you recommend any others for using with very sensitive private data?

782
00:52:27,729 --> 00:52:28,129
Good question.

783
00:52:28,129 --> 00:52:34,189
um it depends on the level of your threat model you will not get something better than a gpu server

784
00:52:34,189 --> 00:52:38,489
sitting next to you or like your local macbook right where you're not talking to anyone on the

785
00:52:38,489 --> 00:52:42,109
internet you're just running you can run it in theory offline like so there's a spectrum as with

786
00:52:42,109 --> 00:52:48,149
all things like in general of like your risk tolerance and your your modeling your local

787
00:52:48,149 --> 00:52:52,169
models are going to be your best i think quen has come out with a model that can actually run

788
00:52:52,169 --> 00:52:57,189
under 16 gigabytes so you could actually run on a 32 gigabyte macbook and it's pretty damn good

789
00:52:57,189 --> 00:53:01,569
if you want to start somewhere like we're at a point now where you can run pretty damn decent

790
00:53:01,569 --> 00:53:06,189
initial passes on stuff with local models for sensitive stuff it depends what your threat model

791
00:53:06,189 --> 00:53:11,709
is so like i i was for all of this open source code i was just using whatever routing providers

792
00:53:11,709 --> 00:53:16,789
i could use i use a lot of open code open code zen is similar to like an open router where you

793
00:53:16,789 --> 00:53:21,469
just have one api key one payment you can use a whole menu of models open router is if you want

794
00:53:21,469 --> 00:53:28,589
like a general menu of all these things i got recommended um fireworks.ai which uses these

795
00:53:28,589 --> 00:53:34,669
models but it's an american-owned company hosted in america so because you'll see like oh are all

796
00:53:34,669 --> 00:53:38,869
the models are hosted all the servers are hosted in the united states it's like okay but who who

797
00:53:38,869 --> 00:53:44,549
owns the servers right and so like i've been using fireworks ai to great effect up to this point um

798
00:53:44,549 --> 00:53:50,529
no one's paid me for any of this by the way um like this is just my learning bitcoin obviously

799
00:53:50,529 --> 00:54:11,889
Yeah, I'll say Wafer, W-A-F-E-R, gave us a bunch of credits to go spend on things and they were great. I don't know their exact breakdown, but like I was using them as a model provider. Yeah, so like it depends on what you're working on, your security stuff. Like I use Fireworks AI, I use like local model stuff, trying to keep all options open. But if you're looking for security, you definitely want to just understand where's the server call going to.

800
00:54:11,889 --> 00:54:18,249
For the sake of scale and the amount of resources I needed to be able to do all this stuff, I used models wherever they were hosted for the open source software.

801
00:54:18,389 --> 00:54:20,749
But the threat model being this is already open source.

802
00:54:21,189 --> 00:54:25,509
Like I, it's like there's no marginal trade off of what I'm doing hypothetically there.

803
00:54:25,849 --> 00:54:26,769
Okay, well, thank you for that.

804
00:54:26,869 --> 00:54:29,429
And Zeke, thank you for the zaps as well.

805
00:54:29,509 --> 00:54:30,749
Again, they are not going to Miller Lite.

806
00:54:30,869 --> 00:54:32,849
They will go to to OpenSats.

807
00:54:33,189 --> 00:54:37,689
I want to ask you to, Rob, because you mentioned something earlier and then you said you wanted to get into it later.

808
00:54:37,729 --> 00:54:38,649
So I want to make sure we hit it.

809
00:54:38,649 --> 00:54:49,109
You talked about how like shortly before this whole inferno, you know, was ignited, you were thinking we may be sort of at the end of the acceleration period.

810
00:54:49,489 --> 00:54:51,529
Can you get into that a little bit?

811
00:54:51,589 --> 00:54:54,509
And I mean, obviously, your perspective has changed.

812
00:54:55,129 --> 00:54:57,909
Can you talk about why you thought we were at the end of the acceleration period?

813
00:54:58,009 --> 00:55:01,449
And then really like, are we still, does it ever stop accelerating?

814
00:55:01,809 --> 00:55:03,849
Like, does anything stop this acceleration train?

815
00:55:03,849 --> 00:55:10,529
well the thing that changed that was i told you i got access to the gpt 5.6 daybreak model and that

816
00:55:10,529 --> 00:55:15,029
was like an inflection point because up until that point i was seeing nothing out of the american

817
00:55:15,029 --> 00:55:19,229
frontier labs that did anything meaningful or good right i was like this is all a fugazi

818
00:55:19,229 --> 00:55:24,589
none of this is real um this is all like a bunch of bullshit and then i started getting access to

819
00:55:24,589 --> 00:55:30,329
the gpt like the cyber allowed models i was like okay there's you guys are ahead you cheeky little

820
00:55:30,329 --> 00:55:34,509
bastards you're ahead but like you're doing a really good job making it seem like you're not

821
00:55:34,509 --> 00:55:38,949
right so that was like uh that was the that was the the turning point i was like okay i've now

822
00:55:38,949 --> 00:55:43,749
seen the goods i've done bake-offs i've seen the goods there's real stuff here yeah it was a bit of

823
00:55:43,749 --> 00:55:50,069
a chili con carnival if you will um i don't know if that's appropriate here or anything but okay so

824
00:55:50,069 --> 00:55:56,909
i mean that that makes sense i mean uh sorry i'm thinking of the south park episode scott

825
00:55:56,909 --> 00:56:06,729
tenermen's parents i of course uh eat your parents uh anyway that's right i want to ask you too okay

826
00:56:06,729 --> 00:56:10,429
so you were talking a little bit about like nothing is going to be better than a gpu sitting

827
00:56:10,429 --> 00:56:16,229
next to you like looking at these looking at self-hosting things uh which i think is you know

828
00:56:16,229 --> 00:56:22,729
in the in the dystopian path right like the dark path i think the big ai frontier model companies

829
00:56:22,729 --> 00:56:27,249
in the U.S. are going to really deride against it because it'll be too dangerous. What if somebody

830
00:56:27,249 --> 00:56:33,229
designs a bioweapon on their own home AI server? So I think that's a very obvious natural

831
00:56:33,229 --> 00:56:36,829
attack vector they'll use to try and clamp down on that, to clamp down on open source.

832
00:56:37,529 --> 00:56:42,329
Do you think that we get to a point, though, and I don't know what the timeline is for this,

833
00:56:42,329 --> 00:56:47,089
because it's really hard. Humans are bad at estimating timelines, especially when you have

834
00:56:47,089 --> 00:56:52,089
exponentially, you know, exponential functions, right? Do you think we get to a point though,

835
00:56:52,569 --> 00:56:58,449
where you are able to truly have the same capabilities running something from your home

836
00:56:58,449 --> 00:57:02,949
for, and I'm not talking about spending millions of dollars on your own home data center, but like

837
00:57:02,949 --> 00:57:08,809
for something that would be accessible to a, the reasonable, let's say to the middle class person

838
00:57:08,809 --> 00:57:24,926
income wise that will that will rival or at least be you know be very on the heels of what these big players are offering I think there always going to be like a cat and mouse is not the right model but there always going to be if they have tens and hundreds of billions of dollars to invest on

839
00:57:24,926 --> 00:57:28,506
things, they'll always find ways to eke out things faster. But everyone's going to be fast followers.

840
00:57:28,766 --> 00:57:31,926
And I would say that a lot of this revolution up to this point, we actually haven't gotten to like

841
00:57:31,926 --> 00:57:38,086
the chip level. Like in Bitcoin, we started with CPUs, then GPUs, and then FPGAs and then ASICs.

842
00:57:38,086 --> 00:57:42,166
we haven't gotten to the ASIC phase yeah you'll be able to buy chips that are just

843
00:57:42,166 --> 00:57:47,726
can be k3 pre-loaded in a specialty circuitry and you're going to get orders of magnitude better

844
00:57:47,726 --> 00:57:52,026
like compute you know what i mean like so i think like there's a massive boon to come

845
00:57:52,026 --> 00:57:56,206
for being able to even the playing field reduce costs have it run on your phone trivially locally

846
00:57:56,206 --> 00:58:00,586
like it's just there's a lot of specialization in the infrastructure and the supply chains to

847
00:58:00,586 --> 00:58:04,026
build these things up to this point and we haven't hit like we're just starting the industrial

848
00:58:04,026 --> 00:58:09,146
revolution era equivalent of this stuff so it's going to take time but i'm not that's the thing

849
00:58:09,146 --> 00:58:14,106
with all this being said i'm an optimist like i'm i'm an optimist in america uh in humanity

850
00:58:14,106 --> 00:58:22,306
in all of this in technology i am an optimist and uh i think we're just in a very tight crunch

851
00:58:22,306 --> 00:58:25,986
because all of a sudden everyone has come to this realization of the technology and what it enables

852
00:58:25,986 --> 00:58:30,426
all at once and the world of physics and atoms are just harder to get moving than writing

853
00:58:30,426 --> 00:58:34,166
incremental software. But I think across both of those, like between the software side and the

854
00:58:34,166 --> 00:58:37,626
hardware side, you're going to see massive improvements. If you want a white pill, I just

855
00:58:37,626 --> 00:58:41,626
like an insurance salesman randomly rambles into his computer at two 30 in the morning,

856
00:58:41,626 --> 00:58:45,366
is able to do better security research than open AI and Anthropics available tooling.

857
00:58:45,726 --> 00:58:47,486
You know what I mean? Like it's a huge white pill.

858
00:58:47,926 --> 00:58:52,886
That's what I mean. Right. So like there is, and I think there's a lot to be learned and gained

859
00:58:52,886 --> 00:58:57,526
from that understanding that like this is such an early domain, such an early wild west still,

860
00:58:57,526 --> 00:59:02,366
there is massive, massive amounts of opportunity for anyone to just go do things and try and figure

861
00:59:02,366 --> 00:59:08,146
this stuff out. It is quite heartening. You know, it feels like, you know, it's like having having

862
00:59:08,146 --> 00:59:12,886
the musket, right, or rifling in the barrel of said musket. Like it is it is a bit of an equalizer

863
00:59:12,886 --> 00:59:17,786
in a lot of ways. And I think that that's like that's the beautiful double edged sort of technology

864
00:59:17,786 --> 00:59:22,806
more broadly. Right. It's like with each technological revolution, like, of course,

865
00:59:22,806 --> 00:59:27,146
like technology is a tool, right? Of course, the bad guys can use it. It's like the same thing with

866
00:59:27,146 --> 00:59:30,026
yeah, like Kim Jong-un can use Bitcoin, right?

867
00:59:30,426 --> 00:59:31,286
But like, so can you.

868
00:59:31,566 --> 00:59:34,086
And you have the same benefits afforded to you

869
00:59:34,086 --> 00:59:36,566
that he does as a dictator.

870
00:59:36,826 --> 00:59:38,446
And it's the same with any sort of technology.

871
00:59:38,586 --> 00:59:43,006
It's like, yeah, bad guys or corrupt nation states

872
00:59:43,006 --> 00:59:45,686
with malintent can of course use these tools.

873
00:59:46,346 --> 00:59:48,566
However, these tools are available to you as well.

874
00:59:48,886 --> 00:59:51,886
And like creating a little bit of that asymmetric defense

875
00:59:51,886 --> 00:59:54,626
is a beautiful thing.

876
00:59:55,506 --> 00:59:55,986
And I want to add to,

877
00:59:55,986 --> 00:59:57,286
because I want to be conscious of your time here.

878
00:59:57,426 --> 00:59:58,666
You gave me more than the seven minutes

879
00:59:58,666 --> 00:59:59,806
you've given Bloomberg,

880
01:00:00,026 --> 01:00:01,486
so I appreciate that very much.

881
01:00:02,166 --> 01:00:04,346
But on the back end of this,

882
01:00:04,386 --> 01:00:05,826
because I do want to talk about

883
01:00:05,826 --> 01:00:07,306
Anchor Watch a little bit,

884
01:00:07,366 --> 01:00:09,446
but I want to talk about it in the context of,

885
01:00:09,626 --> 01:00:11,106
because like you and I,

886
01:00:11,186 --> 01:00:13,686
I could talk about the game theoretical implications

887
01:00:13,686 --> 01:00:18,246
of geopolitical AI dominance for hours more,

888
01:00:18,346 --> 01:00:19,606
and we'd need to order some ribs

889
01:00:19,606 --> 01:00:21,026
to both of our locations probably.

890
01:00:21,506 --> 01:00:23,826
But I do want to talk about self-custody a little bit.

891
01:00:23,986 --> 01:00:24,386
Sure.

892
01:00:24,386 --> 01:00:28,366
because I think this cold card, the cold card hack very much,

893
01:00:28,486 --> 01:00:32,506
it was just more of a, perhaps a cultural wake-up call

894
01:00:32,506 --> 01:00:38,346
where you could do, again, you could do everything right and get boned.

895
01:00:38,826 --> 01:00:42,026
And now I think a lot of people are, you know, I saw somebody,

896
01:00:42,746 --> 01:00:45,046
I forget who it was, maybe it was Shaquille Oatmeal or someone.

897
01:00:45,486 --> 01:00:47,406
If it wasn't him, I'm sorry for crediting him.

898
01:00:47,946 --> 01:00:50,806
But said like, you know, like, get your Bitcoin on exchanges now.

899
01:00:50,806 --> 01:00:59,666
You know, like it's like the script flipping to basically being where people very rightly so people who thought they were doing everything right now got screwed.

900
01:01:00,086 --> 01:01:02,186
More technically inclined people got screwed.

901
01:01:03,046 --> 01:01:12,946
And so if you're not one of those people who is more technically inclined, you cannot blame said person for looking at this whole situation and saying self-custody is terrifying to me.

902
01:01:13,026 --> 01:01:14,606
Right. Having my own keys is terrifying.

903
01:01:14,966 --> 01:01:16,426
Where are you at with this?

904
01:01:16,426 --> 01:01:20,826
Do you think this is kind of a momentary blip in the narrative of self-custody?

905
01:01:21,066 --> 01:01:23,846
Is this going to have longer, further reaching implications?

906
01:01:24,786 --> 01:01:28,386
How are you thinking about this from kind of the both like the technical, but more so

907
01:01:28,386 --> 01:01:30,986
the cultural narrative perspective of self-custody?

908
01:01:31,386 --> 01:01:33,346
I think I have a lot of thoughts on this.

909
01:01:33,786 --> 01:01:40,206
So I think very specifically, the fact it was a bug in the cold card is very much a deep

910
01:01:40,206 --> 01:01:45,766
hit as it relates to Bitcoin culture, however you want to see or describe it as it relates

911
01:01:45,766 --> 01:01:51,146
to self-custody. I think the most important point to start with is that Bitcoin as an experiment

912
01:01:51,146 --> 01:01:57,486
is worthless, not just to me, but the entire thing is worthless if self-custody is not achievable.

913
01:01:57,486 --> 01:02:01,066
It just like it just necessary. Then you just have gold. You just have gold. It all sits in

914
01:02:01,066 --> 01:02:05,906
vaults. It sits in regulated places and then it barely ever moves. You have all that machinery

915
01:02:05,906 --> 01:02:13,266
and paperwork, the optionality and the ease of the optionality for you to redeem your Bitcoin,

916
01:02:13,266 --> 01:02:18,066
wherever it sits compared to setting pallets of gold on a plane around the world is a key part

917
01:02:18,066 --> 01:02:23,926
of the value proposition. So while individuals may make their own custody decisions, it is in everyone's

918
01:02:23,926 --> 01:02:29,946
interest to make sure that self-custody is achievable to be done safely. Barna, no matter where you sit

919
01:02:29,982 --> 01:02:38,042
this stuff. From there, I think there's a couple of observations to be had. One, as a risk management

920
01:02:38,042 --> 01:02:42,922
piece, the universe of people that were impacted with a loss here were people who had a single

921
01:02:42,922 --> 01:02:47,962
signature called card using the defined entropy and no additional passphrase or anything.

922
01:02:48,502 --> 01:02:53,622
Many people in the wake of this have asked me, Rob, when should I go to multisig? Is it one

923
01:02:53,622 --> 01:02:58,802
Bitcoin? Is it 10 Bitcoin? And I tell everyone it's an individual person's risk assessment

924
01:02:58,802 --> 01:03:05,702
necessarily based on whatever amount of money to you would cause significant emotional,

925
01:03:06,502 --> 01:03:13,402
financial, psychological damage, whatever that number is, cut in half. And that's the point

926
01:03:13,402 --> 01:03:17,902
where you should be probably on a multi-sig. There's a nuanced conversation to be had where

927
01:03:17,902 --> 01:03:22,682
if you are a power user, you know exactly what you're doing. You could do a single-sig and a

928
01:03:22,682 --> 01:03:27,582
passphrase all day or a single, like you could be fine. But if I'm trying to give generalizable

929
01:03:27,582 --> 01:03:33,122
advice to many people, that would be my perspective because it's less about multi-sig as like

930
01:03:33,122 --> 01:03:36,722
multiple key, like whatever. It's just not having a single point of failure because that's what this

931
01:03:36,722 --> 01:03:41,122
was, a single point of failure that had a catastrophic fallout. This is in insurance

932
01:03:41,122 --> 01:03:45,802
and in gambling. I used to play a lot of poker before I got into, around the time I got into

933
01:03:45,802 --> 01:03:48,962
Bitcoin and also before I started working in Bitcoin, I used to play a lot of poker.

934
01:03:49,282 --> 01:03:55,322
There's a concept called risk of ruin. And risk of ruin very simply is if you ever make a decision,

935
01:03:55,322 --> 01:03:56,882
if that decision turns out to be wrong,

936
01:03:56,982 --> 01:03:58,722
can it just irreparably harm you

937
01:03:58,722 --> 01:03:59,602
and pull you out of the game?

938
01:03:59,682 --> 01:04:00,962
If you've ever seen the movie Rounders,

939
01:04:01,382 --> 01:04:03,222
the opening scene where he takes his whole life savings

940
01:04:03,222 --> 01:04:04,182
and borrows a bunch of money,

941
01:04:04,242 --> 01:04:06,902
he sits down at the table across KDKGB

942
01:04:06,902 --> 01:04:08,082
and loses everything,

943
01:04:08,362 --> 01:04:09,302
that's a risk of ruin.

944
01:04:09,802 --> 01:04:11,242
He was playing way too above his stakes,

945
01:04:11,482 --> 01:04:13,002
too much risk concentrated in one event,

946
01:04:13,242 --> 01:04:14,682
and that's where you can have an adverse outcome

947
01:04:14,682 --> 01:04:15,942
that greatly impacts you.

948
01:04:16,602 --> 01:04:17,742
Multi-signature as a concept

949
01:04:17,742 --> 01:04:18,942
allows you to distribute that risk

950
01:04:18,942 --> 01:04:19,602
in a really clean way.

951
01:04:20,002 --> 01:04:21,382
Additionally, you can layer it in

952
01:04:21,382 --> 01:04:22,482
with things like collaborative custody.

953
01:04:22,662 --> 01:04:23,902
I don't know a single person

954
01:04:23,902 --> 01:04:25,422
who lost any Bitcoin related to this

955
01:04:25,422 --> 01:04:29,022
that used a collaborative custody partner at all, period.

956
01:04:29,382 --> 01:04:31,042
You could use Anchor Watch, you could use Casa,

957
01:04:31,182 --> 01:04:32,042
you could use Unchained.

958
01:04:32,162 --> 01:04:33,702
Like there are many options out there.

959
01:04:34,542 --> 01:04:36,762
If you don't feel super comfortable,

960
01:04:37,422 --> 01:04:40,222
you could work with someone like myself,

961
01:04:40,522 --> 01:04:42,442
Unchained, Casa, et cetera.

962
01:04:42,922 --> 01:04:45,202
Also, if you want to be a full DIY cypherpunk

963
01:04:45,202 --> 01:04:46,822
and you kind of like the smart contract,

964
01:04:46,942 --> 01:04:48,282
cool stuff we do with Anchor Watch,

965
01:04:48,682 --> 01:04:49,402
use the auto wallet.

966
01:04:49,402 --> 01:04:50,482
It's fully open source.

967
01:04:50,622 --> 01:04:51,302
You could use it yourself.

968
01:04:51,462 --> 01:04:52,362
You don't have to talk to anyone

969
01:04:52,362 --> 01:04:53,302
and it all works, right?

970
01:04:53,902 --> 01:04:56,262
You could use Sparrow if you want to do a traditional multi-sig,

971
01:04:56,262 --> 01:05:12,309
and you want to be a full cypherpunk like you know do whatever you want You could do that fully in a sovereign way I think it important though just to thinking about this stuff in the custody ecosystem self has to exist in a reliable safe way If it does not the project is worthless period And it makes no sense to me if you

972
01:05:12,309 --> 01:05:17,729
going to advocate for Bitcoin and think that Bitcoin as an experiment can necessarily exist

973
01:05:17,729 --> 01:05:21,929
as a check and balance against government fiat printing. And also everyone keeps in a certain

974
01:05:21,929 --> 01:05:26,269
set of regulate custodians. I also want to add, I think ETFs are a totally fine way to get price

975
01:05:26,269 --> 01:05:31,029
exposure that it has it has its own risk trade-offs and each person can make their own

976
01:05:31,029 --> 01:05:37,089
assessment on those said risk trade-offs you could use those i have i hold bitcoin etfs i hold

977
01:05:37,089 --> 01:05:43,069
bitcoin i i had strategy before there were bitcoin etfs you know what i mean like these are all fair

978
01:05:43,069 --> 01:05:48,069
ways in your own risk about said thing it's way easier to log into your fidelity account

979
01:05:48,069 --> 01:05:52,709
and sell some shares on an etf or treasury company than it is for you to go into your

980
01:05:52,709 --> 01:05:54,869
geographically distributed multi-sig,

981
01:05:55,109 --> 01:05:56,809
you know, get on a boat to Antarctica

982
01:05:56,809 --> 01:05:57,729
kind of thing, right?

983
01:05:58,169 --> 01:05:59,629
These all have a place

984
01:05:59,629 --> 01:06:01,749
within everyone's capital stack

985
01:06:01,749 --> 01:06:02,569
and their risk adjustment.

986
01:06:02,809 --> 01:06:05,149
And I think one of the lessons

987
01:06:05,149 --> 01:06:06,269
we should be taking from all this

988
01:06:06,269 --> 01:06:08,809
is there isn't a single answer for anyone.

989
01:06:08,969 --> 01:06:09,909
There is no easy answer.

990
01:06:10,209 --> 01:06:10,949
Every person has to make

991
01:06:10,949 --> 01:06:11,669
their own risk assessment,

992
01:06:11,909 --> 01:06:13,249
understanding what they're comfortable with

993
01:06:13,249 --> 01:06:14,209
on tolerance of risk

994
01:06:14,209 --> 01:06:16,089
and make those decisions themselves.

995
01:06:16,729 --> 01:06:18,389
And if you're happy

996
01:06:18,389 --> 01:06:19,329
and you're Bitcoin safe,

997
01:06:19,389 --> 01:06:20,229
then I'm happy for you

998
01:06:20,229 --> 01:06:21,349
wherever that ultimately ends up.

999
01:06:21,349 --> 01:06:26,209
what I was going to say is I think that advice is really good and the last part of what you said

1000
01:06:26,209 --> 01:06:31,349
that like look this all does depend on your own personal risk tolerance like that is the hard

1001
01:06:31,349 --> 01:06:36,629
thing about Bitcoin right is that you regardless of how you choose to custody or not custody your

1002
01:06:36,629 --> 01:06:43,429
funds like you have to make that decision like it's it's like it's on you right you need to

1003
01:06:43,429 --> 01:06:48,089
decide what you want to do and what you are comfortable with and I think if if there's

1004
01:06:48,089 --> 01:06:52,709
anything that comes out of this i think it's like don't let people shame you for doing something

1005
01:06:52,709 --> 01:06:57,809
that makes sense for your personal risk profile right yes a thousand whether that whether that's

1006
01:06:57,809 --> 01:07:03,069
as you said holding an etf gasp like yeah that's not self-custody bitcoin you're making a lot of

1007
01:07:03,069 --> 01:07:07,429
trade-offs there but if you're just looking for price exposure like and you know i always say

1008
01:07:07,429 --> 01:07:12,209
if you can't open a pdf buy the bitcoin etf you know that's that's that's i've been saying it for

1009
01:07:12,209 --> 01:07:16,549
a while you know it's a pretty good heuristic but then also you know on the collaborative multi-sig

1010
01:07:16,549 --> 01:07:20,669
or, you know, uh, you know, anchor watch, if you want to get it in short with you guys, by the way,

1011
01:07:20,669 --> 01:07:25,369
not sponsored by anchor watch, uh, am sponsored by big Bitcoin, of course, still waiting for the

1012
01:07:25,369 --> 01:07:30,029
paycheck, but they assure me it's, it's, it's coming, you know, post on the phone with them

1013
01:07:30,029 --> 01:07:46,156
after this and look please do please Yeah Please talk to the uh I got to flex my CEO of Bitcoin role You do I gotta I gotta lean in there and make sure you getting getting your payments you you really do i appreciate that but no but at the end of the day it like decide what works best for you it your fucking money like it your it your life

1014
01:07:46,156 --> 01:07:51,076
you know it's it's the product of your blood sweat and tears you're the one responsible for it no one

1015
01:07:51,076 --> 01:07:54,856
else like someone who's a friend of yours who tells you what to do it isn't going to give you

1016
01:07:54,856 --> 01:07:58,496
a bunch of money in the event doesn't work out right like that's you have to make that own

1017
01:07:58,496 --> 01:08:03,496
personal judgment determination for yourself. Yeah. And I think that's, again, like my heart

1018
01:08:03,496 --> 01:08:08,216
goes out to all the cold card victims. And I mean, I mentioned this off air, I'll just mention it on

1019
01:08:08,216 --> 01:08:12,336
air too. I've done it a couple of times on BitCast. And by the way, if you're on Noster right now,

1020
01:08:12,336 --> 01:08:16,276
and I appreciate every single person who, by the way, who joins on Noster for these streams. Thank

1021
01:08:16,276 --> 01:08:20,096
you guys. Really, there's like a couple dozen of you right now. It's actually it's a higher

1022
01:08:20,096 --> 01:08:24,136
attendance than we usually get. So it must be, you know, you all came to see the CEO of Bitcoin,

1023
01:08:24,136 --> 01:08:29,896
but thank you for the zaps and again they will go to open sats red team uh fund rob i noticed you uh

1024
01:08:29,896 --> 01:08:36,296
you also zapped it uh with some compliments for yourself i you know maybe i don't i don't know i

1025
01:08:36,296 --> 01:08:40,856
don't maybe it could be someone spoofing you who's who's to who's to be sure but throughout all of

1026
01:08:40,856 --> 01:08:47,656
this it's like man i i never set up my cold guard i have several of them i was saved by laziness

1027
01:08:48,376 --> 01:08:54,056
and i'm not saying that being lazy about your security is a good thing but like my security

1028
01:08:54,056 --> 01:08:58,736
setup worked right because i kept it simple because i also know my own limitations

1029
01:08:58,736 --> 01:09:03,316
right and i think that's very important and like if you're going to set up your own private pgp

1030
01:09:03,316 --> 01:09:07,916
server for example at your house and somehow use it to store your bitcoin and then you know get

1031
01:09:07,916 --> 01:09:12,856
hacked and docks all the bitcoin core devs of the fbi like maybe that's not the best thing to do

1032
01:09:12,856 --> 01:09:18,016
either you know and i'm not speaking i'm just speaking of course you know just as like a this

1033
01:09:18,016 --> 01:09:21,516
isn't about anyone specifically of course just you know if anyone were to do something like that it

1034
01:09:21,516 --> 01:09:23,056
would be pretty silly, right?

1035
01:09:23,496 --> 01:09:25,096
Especially if you then hard forked Bitcoin

1036
01:09:25,096 --> 01:09:27,816
a little bit later and changed the POW algorithm.

1037
01:09:27,996 --> 01:09:30,156
But, you know, that's just totally unrelated,

1038
01:09:30,436 --> 01:09:31,016
totally unrelated.

1039
01:09:31,256 --> 01:09:33,816
We don't, and that's all the time we have, folks.

1040
01:09:35,716 --> 01:09:40,016
Rob, I want to make sure I get you out of here on time.

1041
01:09:40,416 --> 01:09:41,436
We're going to have to do this again

1042
01:09:41,436 --> 01:09:42,576
because I really enjoy picking your brain.

1043
01:09:42,636 --> 01:09:45,256
Thank you for sort of doing this post-mortem on all this

1044
01:09:45,256 --> 01:09:46,776
because I think people have been following

1045
01:09:46,776 --> 01:09:48,076
from the sidelines, like, thank you.

1046
01:09:48,156 --> 01:09:48,716
Thank you to Callie.

1047
01:09:48,716 --> 01:09:50,296
Thank you to all the red teamers.

1048
01:09:50,296 --> 01:09:54,176
And this is like, this is the best of Bitcoin on display in my mind.

1049
01:09:54,176 --> 01:09:56,256
Like this is, this is really cool to see.

1050
01:09:56,256 --> 01:10:01,396
And, you know, we say like, don't trust verify, but there was a lot of trust put in you guys,

1051
01:10:01,396 --> 01:10:04,516
right? To responsibly disclose, to find these vulnerabilities.

1052
01:10:04,516 --> 01:10:18,463
Like that a tight rope to walk And you guys did it We like we still need each other Even if we you know we don need to apparently we still need each other we we didn even verify a lot of the things we should have verified so you know i guess we really still need each other but i don know just wanted to

1053
01:10:18,463 --> 01:10:23,563
say thank you on behalf of everybody and for just being the ceo of bitcoin as well and just you know

1054
01:10:23,563 --> 01:10:28,403
it's it's got it's hard to wear a lot of hats but you wear them well so cheers to you man where can

1055
01:10:28,403 --> 01:10:32,123
people find out where can people go if they do are if they're interested in anchor watch you're

1056
01:10:32,123 --> 01:10:37,543
there my god you're good you are good sir i wear many hats yeah where can people go to find out

1057
01:10:37,543 --> 01:10:42,943
more about anchor watch again not sponsored not affiliated but uh i think like you guys have a

1058
01:10:42,943 --> 01:10:49,343
great product appreciate it thank you um anchorwatch.com uh or you can hit me up uh on twitter

1059
01:10:49,343 --> 01:10:56,383
or noster uh rob one hand i'm primal.net slash rob um pretty accessible like draw me a line draw

1060
01:10:56,383 --> 01:11:02,003
me a direct message draw me a post on the timeline more than happy to uh you know pick up the

1061
01:11:02,003 --> 01:11:06,283
conversation and answer any of all questions you might have. Well, I appreciate it. I hope that you

1062
01:11:06,283 --> 01:11:12,983
continue to get a full eight hours of sleep. You need it. You deserve it. And again, thank you for

1063
01:11:12,983 --> 01:11:16,063
your time. And thank you again to everybody on the Noster Livestream. I think we've got like

1064
01:11:16,063 --> 01:11:23,063
almost like 70 K sats. So, hey, Bitcoin's pumping. That's not bad. We got like we got like 50 we got

1065
01:11:23,063 --> 01:11:28,323
like 50 bucks to send over to the red team. So I've funded like about five words of a prompt.

1066
01:11:28,323 --> 01:11:29,843
And you know what?

1067
01:11:31,123 --> 01:11:32,403
It's better than nothing.

1068
01:11:32,983 --> 01:11:33,203
All right.

1069
01:11:33,203 --> 01:11:33,383
That's right.

1070
01:11:33,843 --> 01:11:34,423
Cheers, brother.

1071
01:11:34,623 --> 01:11:35,363
Keep up the great work.

1072
01:11:35,463 --> 01:11:35,943
Appreciate you.

1073
01:11:36,023 --> 01:11:36,303
Thank you.

1074
01:11:43,823 --> 01:11:48,523
And that's a wrap on this Bitcoin Talk episode of The Bitcoin Podcast.

1075
01:11:49,183 --> 01:11:57,423
Remember to subscribe to this podcast wherever you're watching or listening and share it with your friends, family and strangers on the Internet.

1076
01:11:57,423 --> 01:12:04,143
Find me on Noster at primal.net slash Walker and this podcast at primal.net slash Titcoin.

1077
01:12:04,663 --> 01:12:11,123
On X, YouTube, and Rumble, just search at Walker America and find this podcast on X and Instagram

1078
01:12:11,123 --> 01:12:17,023
at Titcoin Podcast. Head to the show notes to grab sponsor links, head to substack.com

1079
01:12:17,023 --> 01:12:23,183
slash at Walker America to get episodes emailed to you and head to bitcoinpodcast.net for

1080
01:12:23,183 --> 01:12:30,503
everything else. Bitcoin is scarce, but podcasts are abundant. So thank you for spending your

1081
01:12:30,503 --> 01:12:36,943
time listening to the Bitcoin podcast. Until next time, stay free.