1
00:00:02,720 --> 00:00:14,000
Speaker 1: Bloomberg Audio Studios, Podcasts, Radio News.

2
00:00:18,560 --> 00:00:21,520
Speaker 2: Hello and welcome to another episode of the Odd Thoughts podcast.

3
00:00:21,680 --> 00:00:23,080
Speaker 2: I'm Tracy Alloway.

4
00:00:22,760 --> 00:00:24,319
Speaker 3: And I'm Jill. Why isn't thal Joe.

5
00:00:24,360 --> 00:00:29,120
Speaker 2: You know I have some uh prepper tendencies. Yeah, slightly

6
00:00:29,200 --> 00:00:32,200
Speaker 2: prepper tendencies, Prepper adjacent.

7
00:00:31,920 --> 00:00:35,199
Speaker 4: I know you do, because my plan for when everything,

8
00:00:35,760 --> 00:00:37,959
Speaker 4: when everything goes bad, is to bring my family over

9
00:00:38,000 --> 00:00:38,600
Speaker 4: to your place.

10
00:00:38,840 --> 00:00:39,760
Speaker 3: So I'm relying on you.

11
00:00:39,840 --> 00:00:42,360
Speaker 2: Actually, that's fine. I actually figured that, and I've been

12
00:00:42,400 --> 00:00:44,760
Speaker 2: building an extra store of supplies.

13
00:00:44,880 --> 00:00:46,640
Speaker 4: I'm gonna like send you a whole list of things

14
00:00:46,760 --> 00:00:48,839
Speaker 4: my kids like to eat and stuff like that, just

15
00:00:48,880 --> 00:00:51,800
Speaker 4: so that we're already Yeah, okay.

16
00:00:52,000 --> 00:00:54,440
Speaker 2: Well, one of the things I saw on a bunch

17
00:00:54,440 --> 00:00:57,760
Speaker 2: of the prepper boards that I sometimes look at. I

18
00:00:57,760 --> 00:00:59,600
Speaker 2: don't want people to think that I'm crazy about it,

19
00:00:59,600 --> 00:01:01,960
Speaker 2: but I find it interesting. I find it interesting seeing

20
00:01:02,000 --> 00:01:05,800
Speaker 2: how people's like insecurities manifest in physical stuff. But anyway,

21
00:01:06,480 --> 00:01:08,640
Speaker 2: one of the things everyone was saying was you need

22
00:01:08,680 --> 00:01:11,839
Speaker 2: to start taking cash out because of the situation in Iran,

23
00:01:11,920 --> 00:01:15,080
Speaker 2: because we're all expecting a big cyber attack that's going

24
00:01:15,120 --> 00:01:19,280
Speaker 2: to absolutely destroy the US financial infrastructure.

25
00:01:19,440 --> 00:01:20,880
Speaker 4: By the way, if I told you my idea for

26
00:01:20,959 --> 00:01:24,360
Speaker 4: a business, like I've looked at prepper meals, like prepared meals,

27
00:01:24,360 --> 00:01:26,120
Speaker 4: and they all look terrible. They do like a slightly

28
00:01:26,200 --> 00:01:29,399
Speaker 4: high end version for yuppies I think would be really good,

29
00:01:29,680 --> 00:01:31,080
Speaker 4: like something they you know, like.

30
00:01:31,080 --> 00:01:34,320
Speaker 2: Some nice I think it's a physical limitation on how

31
00:01:34,560 --> 00:01:37,200
Speaker 2: good you can actually get, like dry food.

32
00:01:37,240 --> 00:01:39,400
Speaker 4: Scientist can do a lot of things these days. Anyway,

33
00:01:40,240 --> 00:01:43,240
Speaker 4: let's talk about the actual issue at hand. Yeah, well,

34
00:01:43,280 --> 00:01:44,240
Speaker 4: are you taking cash out?

35
00:01:44,800 --> 00:01:47,760
Speaker 2: I have it? I haven't yet. I'm relying on my

36
00:01:47,840 --> 00:01:50,320
Speaker 2: store of gold and silver. That's right. But I think

37
00:01:50,360 --> 00:01:55,360
Speaker 2: this raises a legitimate and actually very interesting topic, which

38
00:01:55,400 --> 00:02:02,120
Speaker 2: is what do we know about Iran's cyber I guess facilities, skills,

39
00:02:02,200 --> 00:02:06,440
Speaker 2: what could happen in those context? And then also everything

40
00:02:06,440 --> 00:02:08,720
Speaker 2: that's going on with the world of AI, right like

41
00:02:09,080 --> 00:02:14,200
Speaker 2: cyber securities, cyber hacking. It's changing really rapidly given this

42
00:02:14,280 --> 00:02:15,679
Speaker 2: new technology.

43
00:02:15,200 --> 00:02:18,000
Speaker 4: Totally, I mean also just within the context of the

44
00:02:18,040 --> 00:02:21,919
Speaker 4: war itself, setting aside hypothetical doom scenarios. There's a really

45
00:02:22,000 --> 00:02:25,680
Speaker 4: interesting report in the Financial Times about Israel having been

46
00:02:25,720 --> 00:02:28,720
Speaker 4: able to hack into all of the traffic lights in Tehran,

47
00:02:29,400 --> 00:02:33,839
Speaker 4: almost unbelievable and shocking, but there's already within the war

48
00:02:33,880 --> 00:02:36,560
Speaker 4: itself or even over the last couple of years, there

49
00:02:36,600 --> 00:02:41,040
Speaker 4: was the pager attack that Israel had executed, and so yeah,

50
00:02:41,080 --> 00:02:43,280
Speaker 4: cyber is part of it. And the timing is wild

51
00:02:43,360 --> 00:02:47,359
Speaker 4: here because speaking of AI, it was just on Friday.

52
00:02:47,560 --> 00:02:49,840
Speaker 4: We're recording this March fifth. I'm not exactly sure the

53
00:02:49,919 --> 00:02:52,320
Speaker 4: data is coming out, but a week ago basically there

54
00:02:52,360 --> 00:02:56,359
Speaker 4: was the news the complete collapse of the Anthropic relationship

55
00:02:56,360 --> 00:02:58,440
Speaker 4: with the Department of Defense or the Department of War.

56
00:02:58,960 --> 00:03:01,160
Speaker 4: And so it's all in the mix right now, and

57
00:03:01,200 --> 00:03:04,240
Speaker 4: how is AI actually going to change warfare? And one

58
00:03:04,240 --> 00:03:07,440
Speaker 4: of the national security implications of AI and AI and hacking.

59
00:03:07,800 --> 00:03:09,880
Speaker 4: There is a lot in this sort of mix that's

60
00:03:09,919 --> 00:03:10,760
Speaker 4: all happening right now.

61
00:03:10,800 --> 00:03:13,000
Speaker 2: Absolutely. The thing that really caught my eye was the

62
00:03:13,080 --> 00:03:16,160
Speaker 2: story about a hacker using Claude to hack into like

63
00:03:16,160 --> 00:03:17,960
Speaker 2: the Mexican government system. Did you see that?

64
00:03:18,040 --> 00:03:21,080
Speaker 4: That was really interesting because it seemed like the hacker

65
00:03:21,200 --> 00:03:24,440
Speaker 4: extracted a bunch of information from Claude itself. You know,

66
00:03:24,520 --> 00:03:28,120
Speaker 4: I'm pretty sure you cannot go to claud code and say, like,

67
00:03:28,160 --> 00:03:31,160
Speaker 4: I want to break into the Mexican government website helped me,

68
00:03:31,200 --> 00:03:32,080
Speaker 4: like build this app.

69
00:03:32,120 --> 00:03:32,800
Speaker 3: It won't do that.

70
00:03:32,960 --> 00:03:37,080
Speaker 4: It's trained to avoid malicious uses. But people find a

71
00:03:37,080 --> 00:03:39,240
Speaker 4: way to jail break down. People find a way to

72
00:03:39,400 --> 00:03:42,320
Speaker 4: sort of extract information from the AI itself that has

73
00:03:42,320 --> 00:03:45,680
Speaker 4: in its training and so forth. And there's been examples

74
00:03:45,760 --> 00:03:48,520
Speaker 4: of leaks where you know, people upload data to the

75
00:03:48,560 --> 00:03:51,920
Speaker 4: AI and somehow other people see it. Anyway, there's a

76
00:03:51,960 --> 00:03:53,440
Speaker 4: lot here that we have to learn more about.

77
00:03:53,560 --> 00:03:55,080
Speaker 2: We should talk about all of it, and we do,

78
00:03:55,160 --> 00:03:58,120
Speaker 2: in fact have the perfect guest, someone who's been on

79
00:03:58,160 --> 00:04:00,480
Speaker 2: the podcast before but it's been a while. We're going

80
00:04:00,520 --> 00:04:02,800
Speaker 2: to be speaking with Matt Swish. He is the founder

81
00:04:02,800 --> 00:04:07,360
Speaker 2: of ondb, which is a data infrastructure startup for agentiic AI.

82
00:04:07,600 --> 00:04:12,160
Speaker 2: So honestly the perfect a legendary French hacker. I should

83
00:04:12,200 --> 00:04:14,560
Speaker 2: have said that first. Matt, thank you so much for

84
00:04:14,680 --> 00:04:15,760
Speaker 2: coming back on all lots.

85
00:04:16,080 --> 00:04:18,360
Speaker 5: Thank you very much. It's been I think it's been

86
00:04:18,480 --> 00:04:19,200
Speaker 5: worth four years.

87
00:04:19,320 --> 00:04:20,200
Speaker 3: Yeah, I think it has been.

88
00:04:20,440 --> 00:04:22,320
Speaker 2: The Last time we spoke to you, you were still

89
00:04:22,480 --> 00:04:25,440
Speaker 2: in Dubai, and now you're coming to us from Sweden

90
00:04:25,480 --> 00:04:28,200
Speaker 2: and a very Gustavian looking background over there.

91
00:04:28,279 --> 00:04:30,359
Speaker 4: I think actually what we talked to is right after

92
00:04:30,640 --> 00:04:33,440
Speaker 4: Russia's invasion of Ukraine. So I guess, yeah, wow, that

93
00:04:33,480 --> 00:04:34,520
Speaker 4: has been almost every.

94
00:04:34,400 --> 00:04:35,880
Speaker 2: Time there's a war, we call you Matt.

95
00:04:35,920 --> 00:04:40,320
Speaker 4: But because war is so intermixed with cyber espionage, cybersecurity, hacking,

96
00:04:40,320 --> 00:04:42,280
Speaker 4: and so forth, it's a natural natural time.

97
00:04:42,480 --> 00:04:44,760
Speaker 2: So for the benefit of people who didn't listen to

98
00:04:44,800 --> 00:04:47,040
Speaker 2: the episode four years ago, can you just give some

99
00:04:47,080 --> 00:04:50,400
Speaker 2: context around who you actually are and your sort of

100
00:04:50,680 --> 00:04:54,680
Speaker 2: history in the hacking community, including you know, shadow brokers

101
00:04:54,680 --> 00:04:56,560
Speaker 2: and the Wannacrey era and all that stuff.

102
00:04:56,800 --> 00:04:59,960
Speaker 5: So I've been in enterprise software for almost like twenty years,

103
00:05:00,040 --> 00:05:05,760
Speaker 5: particularly cyber security, and my name appeared in a few

104
00:05:06,040 --> 00:05:09,000
Speaker 5: of the different leaks because of various analysis that I've

105
00:05:09,160 --> 00:05:12,760
Speaker 5: done of private information that was being leaked, but also

106
00:05:12,760 --> 00:05:15,520
Speaker 5: a lot of attacks that happened that happened to target

107
00:05:15,520 --> 00:05:19,280
Speaker 5: critical infrastructure over the last ten years. And last time

108
00:05:19,320 --> 00:05:21,719
Speaker 5: we were on the podcast, one of the things we

109
00:05:21,760 --> 00:05:27,080
Speaker 5: talked about is does cyber really matter once you enter

110
00:05:27,160 --> 00:05:30,599
Speaker 5: into a kinetic war, which is exactly what's happening now.

111
00:05:31,360 --> 00:05:35,720
Speaker 5: And the main takeaway was once you start using missiles,

112
00:05:36,040 --> 00:05:38,760
Speaker 5: most of these cyber elements are not really relevant because

113
00:05:38,800 --> 00:05:43,039
Speaker 5: you would use cyber mostly to gather information and intelligence

114
00:05:43,400 --> 00:05:47,800
Speaker 5: to prepare an attack or to disorganize an enemy. You

115
00:05:47,880 --> 00:05:52,320
Speaker 5: create confusion, but as we have seen now, you can

116
00:05:52,480 --> 00:05:55,080
Speaker 5: use like drones that are like twenty thousand dollars and

117
00:05:55,720 --> 00:05:58,880
Speaker 5: create more chaos that you would do with any sort

118
00:05:58,920 --> 00:05:59,760
Speaker 5: of exploits.

119
00:06:00,200 --> 00:06:02,880
Speaker 4: I like how you're introduced as a legendary hacker and

120
00:06:02,880 --> 00:06:05,480
Speaker 4: then you're like, oh, I've been twenty years in enterprise software.

121
00:06:05,640 --> 00:06:07,680
Speaker 3: I feel like this is like the uh, the wind

122
00:06:07,720 --> 00:06:08,320
Speaker 3: of the premium.

123
00:06:08,400 --> 00:06:10,880
Speaker 4: It's like, you know, the casual and then fancy a

124
00:06:10,960 --> 00:06:16,560
Speaker 4: dress up hacker and the like twenty years in enterprise software.

125
00:06:16,600 --> 00:06:19,520
Speaker 4: But this is a really interesting point that you made

126
00:06:19,839 --> 00:06:23,760
Speaker 4: this idea between Okay, mostly it sounds like when people

127
00:06:23,800 --> 00:06:28,080
Speaker 4: imagined cyber attacks, they imagine what Tracy talked about in

128
00:06:28,120 --> 00:06:31,360
Speaker 4: the beginning. Suddenly the entire financial infrastructure, like I just

129
00:06:31,360 --> 00:06:33,400
Speaker 4: come to a haul of people worried about or there's

130
00:06:33,400 --> 00:06:36,720
Speaker 4: gonna be a blackout, et cetera. But in reality, or

131
00:06:36,760 --> 00:06:39,560
Speaker 4: what we've seen so far, by and large, is that

132
00:06:40,320 --> 00:06:43,239
Speaker 4: cyber in the context of war is still much more

133
00:06:43,400 --> 00:06:47,839
Speaker 4: about data collection, espionage, and so forth, rather than these

134
00:06:47,880 --> 00:06:49,680
Speaker 4: more like you know, the types of things you might

135
00:06:49,720 --> 00:06:50,479
Speaker 4: see in a movie.

136
00:06:50,839 --> 00:06:53,760
Speaker 5: Yeah, exactly. I mean over the last like ten fifteen years,

137
00:06:53,800 --> 00:06:56,839
Speaker 5: we've seen some like attacks like cyber attacks against a

138
00:06:56,839 --> 00:07:02,520
Speaker 5: critical infrastructure you run talking like Aramco around like twenty twelve.

139
00:07:03,040 --> 00:07:05,520
Speaker 5: They were just mostly like using what we call a

140
00:07:05,520 --> 00:07:08,800
Speaker 5: wiper that was like a malware that was erasing the

141
00:07:08,800 --> 00:07:11,680
Speaker 5: hard drive of most of the machines. And then we

142
00:07:11,760 --> 00:07:16,080
Speaker 5: obviously have the case of Stocksnet few years before, where

143
00:07:16,360 --> 00:07:19,400
Speaker 5: it was a joint Israeli US operation against some of

144
00:07:19,440 --> 00:07:23,120
Speaker 5: the nuclear centrals in Iran where some of the PLCs

145
00:07:23,160 --> 00:07:26,720
Speaker 5: were targeted. But what we have seen over the weekend

146
00:07:27,640 --> 00:07:30,480
Speaker 5: is some of the drones happened to target some of

147
00:07:30,520 --> 00:07:35,360
Speaker 5: the Amazon data centers and that created so much instability

148
00:07:35,720 --> 00:07:38,480
Speaker 5: because multiple of the zones have been down and I

149
00:07:38,520 --> 00:07:41,400
Speaker 5: think two out of three and the third one is

150
00:07:41,440 --> 00:07:47,000
Speaker 5: still recovering days after. Because most of companies, either private

151
00:07:47,000 --> 00:07:50,400
Speaker 5: companies or public companies not relying on the cloud, which

152
00:07:50,440 --> 00:07:53,800
Speaker 5: is something that was not really the case before, and

153
00:07:54,280 --> 00:07:59,080
Speaker 5: once you have some softw of centralization in terms of dependence,

154
00:08:00,080 --> 00:08:05,160
Speaker 5: so become an easy target. And most of government's AI companies,

155
00:08:05,600 --> 00:08:10,239
Speaker 5: cloud companies do not really have twenty thousand dollars drones

156
00:08:10,320 --> 00:08:13,720
Speaker 5: in their threat models, which is like something that's pretty new,

157
00:08:13,760 --> 00:08:17,600
Speaker 5: but also confirms that Kinneti quals and have more impact.

158
00:08:17,920 --> 00:08:21,120
Speaker 2: So I take the point about cyber being perhaps more

159
00:08:21,200 --> 00:08:24,520
Speaker 2: useful before a war when it comes to info gathering

160
00:08:24,600 --> 00:08:28,600
Speaker 2: and things like that. But we have seen some deployment

161
00:08:28,680 --> 00:08:31,120
Speaker 2: of cyber attacks in the past week or so. So

162
00:08:31,160 --> 00:08:34,480
Speaker 2: we know Israel is attacking some cyber infrastructure in Iran,

163
00:08:34,600 --> 00:08:38,080
Speaker 2: and we know that Iran has perhaps attempted some things,

164
00:08:38,240 --> 00:08:41,280
Speaker 2: maybe not as successfully, but walk us through what we've

165
00:08:41,320 --> 00:08:42,520
Speaker 2: actually seen so far.

166
00:08:43,160 --> 00:08:46,000
Speaker 5: So so far, we have seen an Israeli operation where

167
00:08:46,080 --> 00:08:49,360
Speaker 5: one of the prayer app has been hijacked and so

168
00:08:49,520 --> 00:08:53,160
Speaker 5: message was sent to the users, so it's more like

169
00:08:53,360 --> 00:08:58,680
Speaker 5: to create confusion within people. Also the traffic Light operation

170
00:08:59,000 --> 00:09:02,800
Speaker 5: to understand the position of some of the targets, but

171
00:09:02,920 --> 00:09:07,400
Speaker 5: it's more used for like recaoning a In terms of destruction,

172
00:09:08,000 --> 00:09:12,120
Speaker 5: we didn't see anything significant. Even the government itself of

173
00:09:12,160 --> 00:09:15,840
Speaker 5: Iran shut down most of the Internet for a lot

174
00:09:15,840 --> 00:09:17,920
Speaker 5: of the users, and a lot of what we see

175
00:09:17,960 --> 00:09:21,760
Speaker 5: on social media is the usual disinformation and misinformation campaign,

176
00:09:22,400 --> 00:09:24,720
Speaker 5: especially now with AI. There is so much AI slob

177
00:09:25,200 --> 00:09:29,400
Speaker 5: with like the videos, the text, the boats. That's becoming

178
00:09:29,440 --> 00:09:32,480
Speaker 5: pretty common now, even when there is no war, So

179
00:09:32,679 --> 00:09:35,480
Speaker 5: it's not really like really impactful. So it's more like

180
00:09:35,520 --> 00:09:39,800
Speaker 5: to create confusion than being actually destructive. And now we're

181
00:09:39,840 --> 00:09:44,480
Speaker 5: differently entering in a stage where it's been extremely destructive.

182
00:09:44,640 --> 00:09:47,200
Speaker 5: And I cannot remember the last time we've seen so

183
00:09:47,240 --> 00:09:53,360
Speaker 5: many countries being targeted, which is pretty like a first

184
00:09:53,440 --> 00:09:55,559
Speaker 5: I would say, in terms of like a war climate.

185
00:09:56,040 --> 00:09:58,240
Speaker 4: Can you talk about You know, people stare at their

186
00:09:58,280 --> 00:10:01,839
Speaker 4: screens all day, and they fool themselves into thinking that

187
00:10:01,880 --> 00:10:05,480
Speaker 4: they're quote monitoring the situation, et cetera, but mostly.

188
00:10:05,120 --> 00:10:06,080
Speaker 6: The projections here.

189
00:10:07,120 --> 00:10:10,360
Speaker 4: I don't delude myself, no, I like, actually like I

190
00:10:10,520 --> 00:10:12,280
Speaker 4: sort of look at my screen and I know that

191
00:10:12,320 --> 00:10:18,120
Speaker 4: I'm being inundated with contextless garbage and slop and propaganda

192
00:10:18,320 --> 00:10:22,479
Speaker 4: and so forth. I'm curious how you monitor the situation actually,

193
00:10:22,640 --> 00:10:25,400
Speaker 4: as someone who takes these topics seriously and doesn't just

194
00:10:25,440 --> 00:10:28,440
Speaker 4: sort of become an overnight expert, you know, the day

195
00:10:28,520 --> 00:10:33,160
Speaker 4: after bombings begin, Like how do you pay attention to

196
00:10:33,240 --> 00:10:36,079
Speaker 4: what matters? How do you actually know what's real and

197
00:10:36,120 --> 00:10:38,960
Speaker 4: so forth, and avoid just sort of the delusion of

198
00:10:39,040 --> 00:10:41,880
Speaker 4: staring at the screen and engaging with slop.

199
00:10:42,280 --> 00:10:44,800
Speaker 5: It's a good question because there's so much of it,

200
00:10:44,880 --> 00:10:48,560
Speaker 5: so I think the default reaction is to ignore most

201
00:10:48,600 --> 00:10:52,640
Speaker 5: of it, Yeah, unless it becomes really significant. In this case,

202
00:10:52,960 --> 00:10:55,640
Speaker 5: I think it comes down to looking at the actual damage.

203
00:10:56,160 --> 00:10:59,400
Speaker 5: Many people from the military world, but also the intelligence

204
00:10:59,400 --> 00:11:04,959
Speaker 5: community has been underestimating Iran capabilities exactly like people used

205
00:11:05,000 --> 00:11:07,600
Speaker 5: to do with North Korea, and now North koreas some

206
00:11:07,640 --> 00:11:09,800
Speaker 5: of the best hackers in the world when we see

207
00:11:09,800 --> 00:11:14,319
Speaker 5: them like targeting financial institutions, whereas before they would not

208
00:11:14,480 --> 00:11:18,120
Speaker 5: do like much. So there is definitely like internal capabilities

209
00:11:18,200 --> 00:11:21,880
Speaker 5: that are available, but there's so much noise now, Like

210
00:11:21,920 --> 00:11:24,000
Speaker 5: you say, a lot of people are monitoring the situation

211
00:11:24,880 --> 00:11:29,199
Speaker 5: giving that quote un quote like overnight expert opinions, and

212
00:11:29,320 --> 00:11:33,760
Speaker 5: that's becoming a lot of noise. But I would say

213
00:11:33,920 --> 00:11:37,199
Speaker 5: that in this particular case, we have heard of the

214
00:11:37,240 --> 00:11:42,800
Speaker 5: imminent threat of Iran for around forty years, and that's

215
00:11:42,840 --> 00:11:46,520
Speaker 5: also not really a new situation, so most of people

216
00:11:46,559 --> 00:11:51,640
Speaker 5: would have context around it. And even for the attack

217
00:11:51,760 --> 00:11:54,760
Speaker 5: that happened last week, and many people were expecting them

218
00:11:55,200 --> 00:12:09,800
Speaker 5: four weeks, especially as the continuation of what happened last summer.

219
00:12:14,760 --> 00:12:16,760
Speaker 4: Can you actually talk a little bit more about the

220
00:12:16,840 --> 00:12:20,599
Speaker 4: data center attack? And because that's not cider really, I

221
00:12:20,600 --> 00:12:23,760
Speaker 4: mean that's that's just a physical kinetic warfare against the

222
00:12:23,880 --> 00:12:29,200
Speaker 4: data center. I was surprised how disruptive was that. I

223
00:12:29,280 --> 00:12:32,520
Speaker 4: sort of would assume that cloud service providers that it's

224
00:12:32,559 --> 00:12:35,719
Speaker 4: fairly liquid. Okay, one goes down, but you know, it

225
00:12:35,760 --> 00:12:38,280
Speaker 4: can just be the same software, it can be run

226
00:12:38,360 --> 00:12:41,400
Speaker 4: from numerous other clouds. But I saw that there were disruptions.

227
00:12:41,440 --> 00:12:44,280
Speaker 4: I saw Fortnite tweeting about the fact that some of

228
00:12:44,280 --> 00:12:48,480
Speaker 4: their gameplay was impaired due to the attack on data centers.

229
00:12:48,800 --> 00:12:51,920
Speaker 4: How disruptive have those attacks been? Because this is of

230
00:12:51,960 --> 00:12:53,480
Speaker 4: course a very.

231
00:12:53,440 --> 00:12:56,079
Speaker 2: You know, that's whereinetic meatia.

232
00:12:56,160 --> 00:12:58,600
Speaker 4: Yeah, yeah, and there's a lot you know in the future,

233
00:12:58,720 --> 00:13:01,920
Speaker 4: like thinking about hardening these data centers and as you say,

234
00:13:02,040 --> 00:13:05,080
Speaker 4: like making them they're going to be increasingly targets for war.

235
00:13:05,360 --> 00:13:06,439
Speaker 4: Like how disruptive was that?

236
00:13:07,280 --> 00:13:11,360
Speaker 5: Very good question? So I think one of the main

237
00:13:11,440 --> 00:13:15,680
Speaker 5: takeaway is that it has been extremely successful. So, like

238
00:13:15,760 --> 00:13:18,000
Speaker 5: we said before, like a Shahid drone is around twenty

239
00:13:18,040 --> 00:13:20,800
Speaker 5: thousand dollars and they managed to shut down two of

240
00:13:20,840 --> 00:13:23,679
Speaker 5: the zones of Amazon. Actually, even if you look at

241
00:13:23,720 --> 00:13:27,280
Speaker 5: the official report from Amazon for like thirty six hours,

242
00:13:27,440 --> 00:13:30,000
Speaker 5: they were just saying, oh, some objects struck the data

243
00:13:30,000 --> 00:13:34,520
Speaker 5: centers before they actually expletely said they were drawn strikes.

244
00:13:34,840 --> 00:13:37,200
Speaker 5: So a lot of services that I've been using them

245
00:13:37,320 --> 00:13:41,040
Speaker 5: have been targeted, so from like local applications from two

246
00:13:41,080 --> 00:13:45,920
Speaker 5: banks because in a data center you are taking care

247
00:13:46,000 --> 00:13:50,480
Speaker 5: of multiple different services, and even Versail had to reroute

248
00:13:51,280 --> 00:13:56,359
Speaker 5: their data to Bombay and to exclude middle List as deployment.

249
00:13:56,960 --> 00:14:00,280
Speaker 5: So even if you take the cost of most of

250
00:14:00,320 --> 00:14:03,200
Speaker 5: like zero, the exploits that can go up to like

251
00:14:04,000 --> 00:14:10,000
Speaker 5: multimillion dollar attacks. If you are really aiming at districting things,

252
00:14:10,720 --> 00:14:17,040
Speaker 5: the cost reward of using such an attack is really efficient.

253
00:14:17,120 --> 00:14:20,440
Speaker 5: So you really enter into some sort of asymmetric conflict

254
00:14:20,720 --> 00:14:24,640
Speaker 5: where you can just spend like some really old material

255
00:14:24,800 --> 00:14:27,240
Speaker 5: and have way more impact than someone who's going to

256
00:14:27,240 --> 00:14:30,600
Speaker 5: be like cutting edge and just trying to impress with

257
00:14:31,080 --> 00:14:33,040
Speaker 5: like capabilities. Because at the end of the day, it

258
00:14:33,040 --> 00:14:34,320
Speaker 5: does not really matter.

259
00:14:34,680 --> 00:14:38,120
Speaker 2: How do governments actually build up their cyber capabilities nowadays,

260
00:14:38,160 --> 00:14:40,920
Speaker 2: Because I have this image in my head of maybe

261
00:14:40,960 --> 00:14:44,000
Speaker 2: ten or twenty years ago, you know, they would recruit

262
00:14:44,240 --> 00:14:47,680
Speaker 2: like a twenty year old such as yourself at the time,

263
00:14:48,320 --> 00:14:50,760
Speaker 2: and they would be working in a dark room, that

264
00:14:50,800 --> 00:14:53,640
Speaker 2: sort of thing, but then drinking red bull. Drinking red bull,

265
00:14:53,680 --> 00:14:57,080
Speaker 2: that's right. But then you know, we had the boom

266
00:14:57,240 --> 00:15:00,640
Speaker 2: in Silicon Valley and so you had competition from private companies.

267
00:15:01,120 --> 00:15:04,240
Speaker 2: Now we have the boom in AI and again even

268
00:15:04,320 --> 00:15:07,200
Speaker 2: more competition from private companies. And at the same time,

269
00:15:07,720 --> 00:15:10,680
Speaker 2: governments seem to be I guess, seeding some of their

270
00:15:10,720 --> 00:15:15,880
Speaker 2: own skill set to potentially private companies like Anthropic and

271
00:15:16,080 --> 00:15:19,840
Speaker 2: chat GPT and some others. Walk us through how I

272
00:15:19,880 --> 00:15:24,440
Speaker 2: guess the development of governmental cyber capacity has actually shifted.

273
00:15:25,360 --> 00:15:28,320
Speaker 5: I mean something that didn't really change over the last

274
00:15:28,800 --> 00:15:31,800
Speaker 5: years in terms of capabilities, Like I guess we all

275
00:15:31,840 --> 00:15:36,520
Speaker 5: remember like these snodnnicks in twenty thirteen when we started

276
00:15:36,560 --> 00:15:41,120
Speaker 5: to see more about the inside of capabilities from a government,

277
00:15:41,880 --> 00:15:48,520
Speaker 5: including like domestic my surveillance, global surveillance, exploitation capabilities, And

278
00:15:48,640 --> 00:15:52,680
Speaker 5: since then every other year we have seen an history

279
00:15:52,880 --> 00:15:57,560
Speaker 5: of data being leaked that belongs to the government. So

280
00:15:57,920 --> 00:16:00,400
Speaker 5: in a way, things have been changing a lot, but

281
00:16:00,640 --> 00:16:04,800
Speaker 5: not really much. Like most recently, there was a contractor

282
00:16:04,920 --> 00:16:10,560
Speaker 5: from n Free Iris that was sentenced to eighty seven

283
00:16:10,600 --> 00:16:15,760
Speaker 5: months sentence because they happened to sell zero the exploits

284
00:16:15,840 --> 00:16:22,560
Speaker 5: to a Russian broker, and that's actual expert that belonged

285
00:16:22,560 --> 00:16:25,200
Speaker 5: to the government because there was some sort of integrator.

286
00:16:25,720 --> 00:16:28,800
Speaker 5: So we see like nation states or like governments like

287
00:16:28,840 --> 00:16:31,720
Speaker 5: the US investing like an almost amount of money into

288
00:16:31,800 --> 00:16:35,880
Speaker 5: offensive capabilities, but they also keep being burned by insiders.

289
00:16:36,680 --> 00:16:40,240
Speaker 5: A lot of those capabilities are also as strong as the

290
00:16:40,280 --> 00:16:41,960
Speaker 5: internal coercion.

291
00:16:42,240 --> 00:16:44,200
Speaker 2: But I guess what I'm asking is, you know, if

292
00:16:44,240 --> 00:16:46,520
Speaker 2: you're the Department of Defense or I guess now the

293
00:16:46,600 --> 00:16:50,480
Speaker 2: Department of War, and you're thinking about developing in house

294
00:16:50,520 --> 00:16:55,400
Speaker 2: capabilities versus partnering with a company like Anthropic, and we

295
00:16:55,400 --> 00:16:57,600
Speaker 2: should talk about all the drama that's going on there,

296
00:16:57,760 --> 00:17:01,640
Speaker 2: how are you balancing those decisions nowadays versus say, ten

297
00:17:01,720 --> 00:17:02,640
Speaker 2: or twenty years ago.

298
00:17:03,240 --> 00:17:05,720
Speaker 5: Like my understanding is that now a lot of it

299
00:17:05,760 --> 00:17:10,080
Speaker 5: is also like outsourced because they cannot really develop as

300
00:17:10,119 --> 00:17:15,600
Speaker 5: many capabilities internally. So now like we have seen with

301
00:17:16,800 --> 00:17:21,440
Speaker 5: Entropic that it had been used in the Maduro operation

302
00:17:22,200 --> 00:17:26,000
Speaker 5: and then after that, like there was a per out

303
00:17:26,040 --> 00:17:30,960
Speaker 5: from Entropic because they said it was violating their ethical

304
00:17:31,119 --> 00:17:36,320
Speaker 5: like standard like policy, Yeah, standards. So I would say, now,

305
00:17:36,520 --> 00:17:40,840
Speaker 5: something that's really changing like very fast is the incorporation

306
00:17:40,960 --> 00:17:45,199
Speaker 5: of like AI into tho's decisions. But as we all know,

307
00:17:45,359 --> 00:17:48,960
Speaker 5: like AI can also hallucinate, so even Darly or the

308
00:17:49,080 --> 00:17:51,720
Speaker 5: CEO of Entropics that it's definitely not in a state

309
00:17:51,760 --> 00:17:54,760
Speaker 5: where it can be used for like fully autonomous decisions

310
00:17:54,800 --> 00:17:57,480
Speaker 5: like that. So I would say, like the AI element

311
00:17:57,680 --> 00:18:00,040
Speaker 5: would be like the main difference even we start to

312
00:18:00,080 --> 00:18:04,600
Speaker 5: see it now for like exploit development or vulnerability discovery,

313
00:18:05,119 --> 00:18:08,120
Speaker 5: but it's still too early to kind of like give

314
00:18:08,160 --> 00:18:12,120
Speaker 5: a definitive like opinion about it. But overall, I would

315
00:18:12,119 --> 00:18:13,120
Speaker 5: say it's very similar.

316
00:18:13,560 --> 00:18:16,240
Speaker 4: Well, talk to us about exploit development, because I know

317
00:18:16,320 --> 00:18:18,760
Speaker 4: that you can't just go to cloud code and say, like,

318
00:18:18,840 --> 00:18:23,000
Speaker 4: I'm working on zero day mailware attack, help me figure

319
00:18:23,040 --> 00:18:26,040
Speaker 4: this out. But you know, I also know that there

320
00:18:26,040 --> 00:18:29,240
Speaker 4: are some very talented people who pride themselves in being

321
00:18:29,320 --> 00:18:33,240
Speaker 4: able to jail break AI and illicit outputs that the

322
00:18:33,480 --> 00:18:36,919
Speaker 4: labs do not want their AIS to produce. So do

323
00:18:36,960 --> 00:18:39,040
Speaker 4: you have a sense how just within the pure like

324
00:18:39,359 --> 00:18:44,160
Speaker 4: hacker community, AI is being employed today for these purposes

325
00:18:44,280 --> 00:18:46,120
Speaker 4: or what they're able to get out of these tools.

326
00:18:46,400 --> 00:18:49,120
Speaker 5: It's a good question. So, like we started to see

327
00:18:49,119 --> 00:18:53,720
Speaker 5: people leveraging like AI for like bug discovery, which actually

328
00:18:53,920 --> 00:18:56,680
Speaker 5: is becoming like pretty good. I think even Entropy published

329
00:18:56,680 --> 00:19:00,840
Speaker 5: an article explaining how cloud can be used for the

330
00:19:01,560 --> 00:19:05,560
Speaker 5: discovery into smart contracts and how it found some bugs automatically.

331
00:19:05,600 --> 00:19:07,720
Speaker 5: And I think even recently there really is something called

332
00:19:08,720 --> 00:19:15,560
Speaker 5: cloud for security that was aiming at doing cod assessment.

333
00:19:16,119 --> 00:19:20,080
Speaker 5: But now we're entering in is like interesting paradigm shift

334
00:19:20,119 --> 00:19:24,439
Speaker 5: where the cost of software is going towards zero. So

335
00:19:24,680 --> 00:19:27,400
Speaker 5: if you're a company and if your cost of building

336
00:19:27,480 --> 00:19:31,480
Speaker 5: software is becoming less and less, it's also hard to

337
00:19:31,520 --> 00:19:35,119
Speaker 5: convince people that auditing software for security reasons is going

338
00:19:35,200 --> 00:19:38,000
Speaker 5: to be more expensive than developing the actual software. So

339
00:19:38,040 --> 00:19:40,679
Speaker 5: I think that's one of those shifts we're going to see.

340
00:19:41,000 --> 00:19:45,000
Speaker 4: But when can you explain positive lives Mark, what did

341
00:19:45,040 --> 00:19:47,200
Speaker 4: you mean by that? It's going to be hard to convent.

342
00:19:47,600 --> 00:19:52,479
Speaker 5: If you can have like to allocate budgets for building

343
00:19:52,520 --> 00:19:55,040
Speaker 5: a product, So you have most of the budget that's

344
00:19:55,119 --> 00:19:58,320
Speaker 5: usually allocated for like your software engineers to build the software,

345
00:19:58,800 --> 00:20:02,840
Speaker 5: and then you do some code review afterwards to make

346
00:20:02,840 --> 00:20:06,680
Speaker 5: sure there's no vulnerabilities before it gets released to the public,

347
00:20:07,240 --> 00:20:11,320
Speaker 5: but securities risk is usually like pretty high. You cannot

348
00:20:11,440 --> 00:20:14,480
Speaker 5: just rely on like EI tools, at least not at

349
00:20:14,480 --> 00:20:17,040
Speaker 5: the moment. Maybe in a year from now it's going

350
00:20:17,080 --> 00:20:20,879
Speaker 5: to be possible. So it's going to be like really

351
00:20:20,880 --> 00:20:23,359
Speaker 5: interesting to see how it's going to like do like

352
00:20:23,440 --> 00:20:27,680
Speaker 5: a market shift because now with cloud code, and as

353
00:20:27,760 --> 00:20:31,600
Speaker 5: a famous Vibe coder, Joe, I'm sure you know that

354
00:20:31,680 --> 00:20:35,640
Speaker 5: the cost of building software is approaching like zero if

355
00:20:35,640 --> 00:20:37,400
Speaker 5: you just look on the timeline.

356
00:20:37,600 --> 00:20:39,760
Speaker 2: Are you actually on this note, are you a believer

357
00:20:39,920 --> 00:20:43,240
Speaker 2: in the SaaS apocalypse? Because obviously there's the argument that, well,

358
00:20:43,280 --> 00:20:46,439
Speaker 2: now everyone can just create their own software fairly easily

359
00:20:46,560 --> 00:20:49,040
Speaker 2: using natural language. But on the other hand, if you

360
00:20:49,040 --> 00:20:52,760
Speaker 2: are a big corporation or presumably a government you're going

361
00:20:52,840 --> 00:20:55,439
Speaker 2: to want to have you're going to still want to

362
00:20:55,600 --> 00:20:59,439
Speaker 2: buy software from an external provider given some of the

363
00:20:59,480 --> 00:21:04,399
Speaker 2: security concerns, given that it might not necessarily make sense

364
00:21:04,440 --> 00:21:09,360
Speaker 2: for various reasons, management reasons, perhaps recreate an entire software

365
00:21:09,400 --> 00:21:10,399
Speaker 2: business in house.

366
00:21:10,920 --> 00:21:14,920
Speaker 5: So I'm definitely biased on that. But as someone who

367
00:21:14,960 --> 00:21:17,280
Speaker 5: thinks that the cost of software is going towards zero,

368
00:21:17,359 --> 00:21:22,040
Speaker 5: and as someone who is like watching the software costs

369
00:21:22,040 --> 00:21:25,600
Speaker 5: like collapsing. One of the things that we realize is

370
00:21:25,680 --> 00:21:28,560
Speaker 5: that data is the only durabel asset in the AI economy.

371
00:21:28,640 --> 00:21:30,480
Speaker 5: That's why we decided to work on the bid the

372
00:21:30,520 --> 00:21:34,760
Speaker 5: current startup I'm working on, because like that's the only

373
00:21:34,800 --> 00:21:36,720
Speaker 5: thing that's really going to have value long term if

374
00:21:36,760 --> 00:21:40,280
Speaker 5: agents need something to transact or like to take the decision.

375
00:21:40,880 --> 00:21:43,800
Speaker 5: Because even if you look in terms of like in

376
00:21:43,840 --> 00:21:48,720
Speaker 5: any context, you know, if agents are designed to think autonomously,

377
00:21:49,240 --> 00:21:51,960
Speaker 5: you need to have enough information to take those decisions

378
00:21:51,960 --> 00:21:56,040
Speaker 5: whenever you're going to have your reasoning loops. So software itself,

379
00:21:56,119 --> 00:22:00,240
Speaker 5: if you just build it, is pretty static, whereas like

380
00:22:00,800 --> 00:22:04,560
Speaker 5: the urgentic like feedback loops are more dynamic. But what

381
00:22:04,680 --> 00:22:09,440
Speaker 5: changes the contexts that take decision on. So definitely, like

382
00:22:09,720 --> 00:22:14,280
Speaker 5: Salz business are going to have our time because if anyone,

383
00:22:14,480 --> 00:22:19,080
Speaker 5: including the Shopify CEO, can just rewrite an MRI software

384
00:22:19,600 --> 00:22:23,480
Speaker 5: in one afternoon just to like look at his back MRIs,

385
00:22:23,840 --> 00:22:26,919
Speaker 5: you can imagine like how destructive it's going to be,

386
00:22:27,000 --> 00:22:29,040
Speaker 5: like by the end of this year. I think the

387
00:22:29,080 --> 00:22:33,159
Speaker 5: only thing we haven't seen yet is like enterprise AI agent.

388
00:22:33,720 --> 00:22:37,240
Speaker 5: So so far I would say, like since Christmas, people

389
00:22:37,320 --> 00:22:39,800
Speaker 5: are most still playing around trying to find like a

390
00:22:39,840 --> 00:22:44,280
Speaker 5: proper use case. We see a lot of like consumer

391
00:22:44,440 --> 00:22:48,040
Speaker 5: like AI agent like open Cloth that really met like

392
00:22:48,160 --> 00:22:52,879
Speaker 5: agents more mainstream, but we really haven't seen yet enterprise

393
00:22:53,080 --> 00:22:56,280
Speaker 5: AI agent. So as everyone is kind of scared of

394
00:22:56,320 --> 00:23:00,200
Speaker 5: being replaced for their jobs, we haven't really seen in

395
00:23:00,240 --> 00:23:04,520
Speaker 5: like actual like AI agents replacing entire departments or full

396
00:23:04,560 --> 00:23:07,800
Speaker 5: on like employees. So we have seen some disruption around

397
00:23:07,840 --> 00:23:12,120
Speaker 5: like software engineering, mostly to make like software engineering more efficient,

398
00:23:12,280 --> 00:23:16,720
Speaker 5: especially in terms of like development with shorter timelines, but

399
00:23:16,840 --> 00:23:20,600
Speaker 5: we haven't seen yet like proper like enterprise agents.

400
00:23:21,040 --> 00:23:22,600
Speaker 3: How do you define an AI agent?

401
00:23:22,960 --> 00:23:25,439
Speaker 5: So my view of an AI agent and I like

402
00:23:25,480 --> 00:23:28,960
Speaker 5: to remind people like what an AI agent actually is.

403
00:23:29,720 --> 00:23:32,159
Speaker 5: At the moment, most of AI agent is just like

404
00:23:32,280 --> 00:23:35,639
Speaker 5: a piece of God usually written in Python or in

405
00:23:35,720 --> 00:23:38,520
Speaker 5: type script. That's just doing a bunch of cards to

406
00:23:38,680 --> 00:23:42,679
Speaker 5: like entropic PENAI and running in a loop and taking

407
00:23:42,720 --> 00:23:46,680
Speaker 5: decisions and calling like third party tools like MCPS or

408
00:23:46,760 --> 00:23:51,280
Speaker 5: web searches. So that's mostly what an AI agent is.

409
00:23:51,840 --> 00:23:55,760
Speaker 5: We tend to think of AI agent as a completely

410
00:23:55,920 --> 00:23:58,879
Speaker 5: different persona. But at the end of the day. It

411
00:23:59,040 --> 00:24:01,840
Speaker 5: is just like a piece of software that's running as

412
00:24:01,880 --> 00:24:05,159
Speaker 5: a service on a machine or on a server. So

413
00:24:06,600 --> 00:24:09,479
Speaker 5: from a security standpoint, which is pretty interesting, it's just

414
00:24:09,560 --> 00:24:13,639
Speaker 5: like another service. So like software, but people really like

415
00:24:13,800 --> 00:24:16,760
Speaker 5: to think of it in another way.

416
00:24:17,080 --> 00:24:20,400
Speaker 4: Well, just so like I'm from the security standpoint, I mean,

417
00:24:20,840 --> 00:24:24,040
Speaker 4: one of the exciting aspects of AI agents is that

418
00:24:24,080 --> 00:24:27,320
Speaker 4: they can work autonomously. Right, you set a task and

419
00:24:27,560 --> 00:24:29,640
Speaker 4: it can go out and find what it needs to do,

420
00:24:29,720 --> 00:24:31,720
Speaker 4: and it says like, Okay, this didn't work, I'm gonna

421
00:24:31,720 --> 00:24:33,960
Speaker 4: try it's gonna try this thing. It's gonna try this thing.

422
00:24:34,160 --> 00:24:36,040
Speaker 4: I need to connect to this web service to get

423
00:24:36,040 --> 00:24:39,760
Speaker 4: this information, et cetera. The downside of AI agents is

424
00:24:39,800 --> 00:24:42,480
Speaker 4: precisely the same. The downside of AI agent says that

425
00:24:42,520 --> 00:24:44,520
Speaker 4: they could do whatever they want to do and if

426
00:24:44,560 --> 00:24:48,040
Speaker 4: it accidentally deletes a bunch of files because it thinks

427
00:24:48,040 --> 00:24:50,399
Speaker 4: that's what's necessary to execute the task.

428
00:24:51,240 --> 00:24:54,080
Speaker 3: So like I'm curious, Like from a security.

429
00:24:53,520 --> 00:24:57,400
Speaker 4: Standpoint, like I mean, we've already seen examples of people

430
00:24:57,400 --> 00:25:00,280
Speaker 4: getting private information exposed or as I mentioned, and the

431
00:25:00,320 --> 00:25:03,719
Speaker 4: example of deleting a bunch of information. Is this like

432
00:25:03,760 --> 00:25:06,959
Speaker 4: a new way to think about the security threat model.

433
00:25:07,040 --> 00:25:11,120
Speaker 4: The fact that the capability and the downside are one

434
00:25:11,160 --> 00:25:11,680
Speaker 4: and the same.

435
00:25:11,800 --> 00:25:12,359
Speaker 3: It's the same.

436
00:25:12,400 --> 00:25:15,040
Speaker 4: It's sort of like hallucinations, right, the ability to like

437
00:25:15,119 --> 00:25:18,880
Speaker 4: create an output, and it also you know, is hand

438
00:25:18,920 --> 00:25:21,000
Speaker 4: in hand with the ability to create a wrong output,

439
00:25:21,119 --> 00:25:23,840
Speaker 4: a false output, and so you know, the ability to

440
00:25:23,920 --> 00:25:26,480
Speaker 4: do to act on its own is also the ability

441
00:25:26,520 --> 00:25:28,879
Speaker 4: to destroy on its own. Is this sort of like

442
00:25:28,920 --> 00:25:33,159
Speaker 4: a novel threat model or a novel paradigm in thinking

443
00:25:33,200 --> 00:25:34,800
Speaker 4: about enterprise security.

444
00:25:35,160 --> 00:25:38,800
Speaker 5: From an enterprise security standpoint, it is pretty much the

445
00:25:38,880 --> 00:25:43,600
Speaker 5: same thing in the sense of like if you're building software,

446
00:25:43,720 --> 00:25:46,720
Speaker 5: you can just really like patch software afterwards and stuff,

447
00:25:46,720 --> 00:25:49,119
Speaker 5: because you never it never ends. Like secrety might be

448
00:25:49,240 --> 00:25:51,600
Speaker 5: like built in, and you need to have like a

449
00:25:51,640 --> 00:25:54,680
Speaker 5: safe design from the beginning. What we have seen now

450
00:25:54,800 --> 00:25:58,200
Speaker 5: is like whenever people do something argenttic, they just give

451
00:25:58,320 --> 00:26:01,520
Speaker 5: like all permeisterns up from yeah, which is like probably

452
00:26:01,560 --> 00:26:05,000
Speaker 5: the worst thing you can do. And if you're an enterprise,

453
00:26:05,240 --> 00:26:07,600
Speaker 5: as you can imagine, if you just give like all

454
00:26:07,640 --> 00:26:11,400
Speaker 5: permissions to an agent, it just becomes Murphy's Lows. If

455
00:26:11,440 --> 00:26:13,840
Speaker 5: something bad can happen because you give it access to

456
00:26:13,960 --> 00:26:16,000
Speaker 5: it will happen. So you're going to see like more

457
00:26:16,080 --> 00:26:21,040
Speaker 5: dataks for sure, because there's no safety by design in

458
00:26:21,160 --> 00:26:25,560
Speaker 5: those like architectures or like those like agents, which is

459
00:26:25,960 --> 00:26:29,720
Speaker 5: in terms of reliabilities and like exposure would be like

460
00:26:30,000 --> 00:26:33,200
Speaker 5: very similar to what we have seen over the last

461
00:26:33,600 --> 00:26:37,280
Speaker 5: like ten twenty years. But you know, if people are

462
00:26:37,320 --> 00:26:40,800
Speaker 5: ignoring what has been done in terms of software security

463
00:26:40,840 --> 00:26:43,560
Speaker 5: for the last twenty years, that's why we're going to

464
00:26:43,600 --> 00:26:45,679
Speaker 5: have a lot of problems. And I think we're probably

465
00:26:45,680 --> 00:26:48,480
Speaker 5: going to start to see like people, especially in enterprise,

466
00:26:48,600 --> 00:26:52,040
Speaker 5: like pushing back a lot because there's compliance that needs

467
00:26:52,040 --> 00:26:54,879
Speaker 5: to be like, you know, like answer to, so you

468
00:26:54,920 --> 00:26:57,840
Speaker 5: cannot just give like full access to like you know,

469
00:26:57,920 --> 00:26:58,440
Speaker 5: your agent.

470
00:27:14,320 --> 00:27:17,240
Speaker 2: Speaking of the long arc of history, one thing I

471
00:27:17,280 --> 00:27:20,240
Speaker 2: really wonder is you've obviously been in this space for

472
00:27:20,280 --> 00:27:23,440
Speaker 2: a very long time at this point. Can you describe

473
00:27:23,480 --> 00:27:27,000
Speaker 2: how you think your own career and I guess coding

474
00:27:27,200 --> 00:27:30,920
Speaker 2: experience would have been different if you were, say, starting

475
00:27:30,920 --> 00:27:33,960
Speaker 2: out now in twenty twenty six versus I guess you

476
00:27:33,960 --> 00:27:35,840
Speaker 2: would have started out in like the late nineties or

477
00:27:35,880 --> 00:27:38,600
Speaker 2: early two thousands, maybe even before that, yeah.

478
00:27:38,800 --> 00:27:41,920
Speaker 5: Mid two thousands. Well, I would say, like, what has

479
00:27:42,000 --> 00:27:45,800
Speaker 5: changed is back even like we even we've up going

480
00:27:45,840 --> 00:27:49,119
Speaker 5: to like the two thousands, Like back in the snow

481
00:27:49,200 --> 00:27:55,320
Speaker 5: and days when the global surveillance program was being exposed,

482
00:27:55,320 --> 00:27:59,439
Speaker 5: like a lot of people were really like scared of

483
00:27:59,480 --> 00:28:04,040
Speaker 5: it and scandalized by it and push back and people

484
00:28:04,160 --> 00:28:08,600
Speaker 5: really cared about privacy was like now we're entering in

485
00:28:08,640 --> 00:28:12,720
Speaker 5: a new arc where very few people care about privacy.

486
00:28:13,520 --> 00:28:18,120
Speaker 5: Where you see like the CEO of Entropy being asked

487
00:28:18,880 --> 00:28:22,960
Speaker 5: why he refused to work with the US government and

488
00:28:23,000 --> 00:28:26,920
Speaker 5: he says, well, they wanted to do a domestic surveillance program,

489
00:28:27,080 --> 00:28:32,080
Speaker 5: so that's against or like a safety like chart. So

490
00:28:32,359 --> 00:28:37,640
Speaker 5: there's this all aspect of people relationship with data, which

491
00:28:37,640 --> 00:28:40,959
Speaker 5: I guess is very different in terms of software off

492
00:28:40,960 --> 00:28:44,880
Speaker 5: you see like now you can write more things, anyone

493
00:28:44,920 --> 00:28:48,280
Speaker 5: can write anything. But I think we're still in this

494
00:28:48,440 --> 00:28:54,920
Speaker 5: weird adequate software phase where we know what AI can do,

495
00:28:56,280 --> 00:28:58,880
Speaker 5: but it cannot really do anything more or anything less.

496
00:28:58,960 --> 00:29:03,600
Speaker 5: Yet you haven't seen like the actually or like use

497
00:29:03,680 --> 00:29:08,880
Speaker 5: cass for it. Because it's obviously like very exciting and

498
00:29:08,920 --> 00:29:11,400
Speaker 5: it feels like a lot of it is very different

499
00:29:11,440 --> 00:29:17,280
Speaker 5: from before, but we don't really have like any evidence

500
00:29:17,320 --> 00:29:21,160
Speaker 5: of how it's really helping in national security. I know

501
00:29:21,240 --> 00:29:24,240
Speaker 5: it's been helping people who have been analyzing like Epstein's

502
00:29:24,280 --> 00:29:27,320
Speaker 5: emails because there's a lot of data and that makes

503
00:29:27,360 --> 00:29:31,400
Speaker 5: it faster. But in terms of like real use case,

504
00:29:31,680 --> 00:29:34,840
Speaker 5: I don't think like it feels like the current world

505
00:29:34,960 --> 00:29:37,480
Speaker 5: is very different than before, but there's so much noise

506
00:29:37,520 --> 00:29:41,000
Speaker 5: and so much like slop all over that in a

507
00:29:41,040 --> 00:29:44,200
Speaker 5: way it is pretty similar. I don't know if that

508
00:29:44,280 --> 00:29:44,800
Speaker 5: makes sense.

509
00:29:45,200 --> 00:29:47,880
Speaker 4: Well, I mean does your company is o n dB

510
00:29:48,080 --> 00:29:50,720
Speaker 4: dot AI, So you must think that it's going to

511
00:29:50,760 --> 00:29:53,800
Speaker 4: be used or they you know, or that there's clearly

512
00:29:54,120 --> 00:29:57,640
Speaker 4: something there actually tell us about I'm actually really I'm

513
00:29:57,680 --> 00:30:00,680
Speaker 4: on your website right now. It looks really interesting because

514
00:30:00,680 --> 00:30:03,000
Speaker 4: it's something I've been thinking about. But you must have

515
00:30:03,080 --> 00:30:06,120
Speaker 4: some vision for like where it's going and that there

516
00:30:06,160 --> 00:30:10,600
Speaker 4: will actually be significant demand for these services.

517
00:30:11,080 --> 00:30:13,480
Speaker 5: Sure. I mean, like I was saying, like I think

518
00:30:13,560 --> 00:30:18,280
Speaker 5: now we've like anything that's urgentic building software, which is

519
00:30:18,280 --> 00:30:20,880
Speaker 5: like the main use case so far for AI is

520
00:30:20,880 --> 00:30:22,920
Speaker 5: going to make software like going to zero, so the

521
00:30:22,960 --> 00:30:24,440
Speaker 5: cost of building software is going to.

522
00:30:24,680 --> 00:30:25,360
Speaker 3: So that's real.

523
00:30:25,480 --> 00:30:27,720
Speaker 4: So like say, like there's in your view, there's no

524
00:30:27,920 --> 00:30:31,440
Speaker 4: question that already AI. I mean talk about use it's

525
00:30:31,440 --> 00:30:33,480
Speaker 4: a pretty big use case right there, bringing the cost

526
00:30:33,480 --> 00:30:34,600
Speaker 4: of building software to zero.

527
00:30:34,880 --> 00:30:36,800
Speaker 5: Yeah, I mean, like even if you look at the

528
00:30:37,000 --> 00:30:39,920
Speaker 5: offer of cloud code, he said that you didn't write

529
00:30:39,960 --> 00:30:44,080
Speaker 5: like cod since November, So like a lot of people

530
00:30:44,080 --> 00:30:47,000
Speaker 5: are like this, you know, even us internally, like we

531
00:30:47,120 --> 00:30:50,200
Speaker 5: definitely like use cloud code a lot. But if software

532
00:30:50,240 --> 00:30:53,720
Speaker 5: is going to zero, like what's left in terms of

533
00:30:53,760 --> 00:30:57,200
Speaker 5: like the internet layers. So our conclusion is that data

534
00:30:57,680 --> 00:31:00,880
Speaker 5: is the only thing that's going to be like timeless

535
00:31:00,920 --> 00:31:04,320
Speaker 5: in the AI economy, so building like a layer for that.

536
00:31:04,440 --> 00:31:08,880
Speaker 5: Especially now there's like all those innovations around like payments

537
00:31:08,880 --> 00:31:11,760
Speaker 5: and stable coin that you can use to actually like

538
00:31:11,800 --> 00:31:15,920
Speaker 5: pay like anything online. So we think like, okay, if

539
00:31:16,720 --> 00:31:20,560
Speaker 5: you have like issues like entropy or like opening I

540
00:31:20,800 --> 00:31:25,239
Speaker 5: just scrapping Internet and using like public internet, so you

541
00:31:25,280 --> 00:31:29,000
Speaker 5: may as well find ways of charging like boats or

542
00:31:29,040 --> 00:31:31,920
Speaker 5: agent for your data. So at least you can have

543
00:31:32,080 --> 00:31:34,800
Speaker 5: this new like revenue unlock that's going to emerge in

544
00:31:34,880 --> 00:31:37,600
Speaker 5: these new economy because I think a lot of the

545
00:31:37,720 --> 00:31:41,480
Speaker 5: traditional like economic model, like for instance, like we said

546
00:31:41,480 --> 00:31:44,720
Speaker 5: like with SaaS are going to be like disrupted a lot.

547
00:31:45,240 --> 00:31:49,400
Speaker 5: So there is a completely new market around help people

548
00:31:49,400 --> 00:31:52,640
Speaker 5: are going to consume data, and I think people would

549
00:31:52,680 --> 00:31:54,960
Speaker 5: just be ready to pay like more to have eigh

550
00:31:55,000 --> 00:31:57,640
Speaker 5: quality data because the more noise there is, the more

551
00:31:57,720 --> 00:32:00,760
Speaker 5: you want to make sure that the data access to

552
00:32:01,520 --> 00:32:06,520
Speaker 5: is going to be like valuable and real. So yeah,

553
00:32:06,600 --> 00:32:09,000
Speaker 5: even from a security standpoint, you know, like once you

554
00:32:09,160 --> 00:32:12,000
Speaker 5: build like an infrastecure layer, you can have these like

555
00:32:12,120 --> 00:32:15,000
Speaker 5: built in security to make sure that the data you

556
00:32:15,040 --> 00:32:17,680
Speaker 5: give back or the access like for the interface that

557
00:32:17,720 --> 00:32:21,800
Speaker 5: you define is actually secure. Because even if you look

558
00:32:21,800 --> 00:32:25,280
Speaker 5: at Opencloth for instance, one of the top skill was malware.

559
00:32:25,680 --> 00:32:30,080
Speaker 5: So like people just like living into like wild West,

560
00:32:30,120 --> 00:32:33,920
Speaker 5: but they just run everything. So it's us anticipating that

561
00:32:34,120 --> 00:32:36,560
Speaker 5: enterprise is going to look very different and they won't

562
00:32:36,600 --> 00:32:38,600
Speaker 5: just run like anything they find online.

563
00:32:38,840 --> 00:32:41,120
Speaker 4: Yeah, you know, this is one of the things that

564
00:32:41,160 --> 00:32:45,280
Speaker 4: I've encountered in my Vibe coding for is is that

565
00:32:45,400 --> 00:32:48,240
Speaker 4: one of the annoying things is Okay, you want the

566
00:32:48,360 --> 00:32:51,600
Speaker 4: agent to like go out and grab some information or

567
00:32:51,720 --> 00:32:55,480
Speaker 4: clear some database or whatever, and then it's like okay,

568
00:32:55,640 --> 00:32:58,560
Speaker 4: like let me know when you've gotten an API, They're

569
00:32:58,600 --> 00:33:00,680
Speaker 4: like okay, come back and get and then you have

570
00:33:00,720 --> 00:33:03,080
Speaker 4: to go to a website, and then you have to

571
00:33:03,200 --> 00:33:05,520
Speaker 4: get out your credit card and you have to set

572
00:33:05,560 --> 00:33:07,960
Speaker 4: up an account, and then you get an API key

573
00:33:08,080 --> 00:33:10,120
Speaker 4: and then you copy and paste it and so forth,

574
00:33:10,560 --> 00:33:12,080
Speaker 4: and that's very annoying.

575
00:33:12,200 --> 00:33:12,680
Speaker 3: I want.

576
00:33:12,720 --> 00:33:15,120
Speaker 4: What I want is for the agent to just be

577
00:33:15,120 --> 00:33:17,360
Speaker 4: able to go there say, oh, you know, like let

578
00:33:17,440 --> 00:33:19,280
Speaker 4: me just pay you with some stable coins, et cetera.

579
00:33:19,520 --> 00:33:21,880
Speaker 4: Just go out and get the information on its own

580
00:33:21,960 --> 00:33:24,800
Speaker 4: without having this human in the loop. But it also

581
00:33:24,920 --> 00:33:27,040
Speaker 4: occurs to me, and this is something that I've asked

582
00:33:27,040 --> 00:33:30,360
Speaker 4: about with others, which is that like, once I'm just

583
00:33:30,400 --> 00:33:33,280
Speaker 4: like entirely operating in the terminal and the agent just

584
00:33:33,320 --> 00:33:35,880
Speaker 4: going out and scraping information for me, et cetera. Like

585
00:33:36,160 --> 00:33:39,240
Speaker 4: why do we even have a free public internet anymore?

586
00:33:39,560 --> 00:33:42,040
Speaker 4: And so like I'm curious, like whether the direction of

587
00:33:42,080 --> 00:33:45,920
Speaker 4: the Internet and information in general is just entirely like

588
00:33:46,240 --> 00:33:49,920
Speaker 4: you know, paying micro transaction or fees for data consumption

589
00:33:50,160 --> 00:33:53,520
Speaker 4: so that the data then arrives in some usable form

590
00:33:53,840 --> 00:33:55,240
Speaker 4: in the terminal that I'm operating.

591
00:33:55,720 --> 00:33:59,719
Speaker 5: Yeah, no, I think you're You're raising some really good points.

592
00:34:00,160 --> 00:34:03,440
Speaker 5: We need to talk after you can be design.

593
00:34:03,480 --> 00:34:07,280
Speaker 3: You know, it's going to hack the ati I can be,

594
00:34:07,400 --> 00:34:08,400
Speaker 3: I can be a consultant.

595
00:34:08,440 --> 00:34:10,160
Speaker 4: No I hate having to deal with all these API

596
00:34:10,280 --> 00:34:11,680
Speaker 4: key is why am I still using?

597
00:34:12,160 --> 00:34:15,520
Speaker 5: You're like completely right, Yeah, I told you're famous code.

598
00:34:16,480 --> 00:34:19,240
Speaker 5: It's like but like the US case, you're describing noise

599
00:34:19,400 --> 00:34:22,720
Speaker 5: like it makes entire sense. That's why, like we position

600
00:34:22,840 --> 00:34:25,439
Speaker 5: ourselves kind of like as a trusted like provider, where

601
00:34:25,480 --> 00:34:29,120
Speaker 5: like instead of having to go like everywhere to get data,

602
00:34:29,160 --> 00:34:31,600
Speaker 5: you know, you have like this single point and unified

603
00:34:31,680 --> 00:34:34,200
Speaker 5: access a bit like open Router for AI models, but

604
00:34:34,239 --> 00:34:37,360
Speaker 5: for provids. Because if you think about it, when you

605
00:34:37,480 --> 00:34:41,240
Speaker 5: use like cloud now in your terminal, like the level

606
00:34:41,360 --> 00:34:46,000
Speaker 5: one is basically you asking the model itself for information,

607
00:34:46,280 --> 00:34:48,560
Speaker 5: but as we know, like the model maybe like a

608
00:34:48,800 --> 00:34:51,280
Speaker 5: few months old and doesn't have access to the information.

609
00:34:51,800 --> 00:34:56,040
Speaker 5: So like the second level is the agent like cloud

610
00:34:56,080 --> 00:34:59,480
Speaker 5: god or a cloud for this top or like cha

611
00:34:59,600 --> 00:35:03,520
Speaker 5: gip doing web searches, and that's just them looking on

612
00:35:03,560 --> 00:35:08,360
Speaker 5: the internet doing like Google search, etc. But that's still

613
00:35:08,440 --> 00:35:11,800
Speaker 5: not giving you access to like the actual relevant data

614
00:35:12,000 --> 00:35:14,480
Speaker 5: that you would need, where for instance, you would have

615
00:35:14,560 --> 00:35:17,680
Speaker 5: like those APIs and stuff. So like the level after

616
00:35:17,760 --> 00:35:21,800
Speaker 5: that is basically access to private information. So on private information,

617
00:35:22,000 --> 00:35:25,279
Speaker 5: usually the one that's valuable, like in the case of Bloomberg,

618
00:35:25,320 --> 00:35:29,200
Speaker 5: for instance, Bloomberg has a lot of extremely valuable information,

619
00:35:29,280 --> 00:35:30,960
Speaker 5: but you can only have access to it through the

620
00:35:31,080 --> 00:35:34,560
Speaker 5: terminal once you have the subscription, or like any SaaS

621
00:35:34,560 --> 00:35:37,600
Speaker 5: services if you think of like SaaS platform at just

622
00:35:37,680 --> 00:35:40,640
Speaker 5: like some fancy UI where you can like browser database,

623
00:35:40,680 --> 00:35:45,320
Speaker 5: but the data that's valuable is just in the database directly.

624
00:35:45,800 --> 00:35:48,520
Speaker 5: So if you have access to those APIs and we

625
00:35:48,640 --> 00:35:52,279
Speaker 5: not have to create like one million subscriptions left and right,

626
00:35:52,880 --> 00:35:55,840
Speaker 5: because there's already like someone who's doing the integration for

627
00:35:56,000 --> 00:35:59,839
Speaker 5: that kind of like as a programmatic like API marketplace.

628
00:35:59,880 --> 00:36:03,360
Speaker 5: But the integration now is like much easier, and especially

629
00:36:03,360 --> 00:36:05,160
Speaker 5: if you trust it because now what we have seen

630
00:36:05,239 --> 00:36:08,239
Speaker 5: or so it's like whenever use autos tools, they make

631
00:36:08,320 --> 00:36:11,080
Speaker 5: you install like MCPS, but more and more people are

632
00:36:11,120 --> 00:36:14,000
Speaker 5: like moving away from mcps, they're just using like skills

633
00:36:14,040 --> 00:36:17,879
Speaker 5: because like you just said, we start to spend more

634
00:36:17,920 --> 00:36:22,160
Speaker 5: time into the terminal. So like the terminal and anything

635
00:36:22,160 --> 00:36:25,319
Speaker 5: that cila is becoming like a natural interface for like

636
00:36:25,400 --> 00:36:28,680
Speaker 5: agents even for humans because you don't need to try

637
00:36:28,719 --> 00:36:32,319
Speaker 5: to understand those like fancy UI and ux, you just say, like, Okay,

638
00:36:32,320 --> 00:36:35,359
Speaker 5: I want to do it, ABC just do that. So

639
00:36:36,000 --> 00:36:37,960
Speaker 5: I think it just makes more sense to have like

640
00:36:38,400 --> 00:36:41,400
Speaker 5: this interface for it, and like, like you said, you know,

641
00:36:41,480 --> 00:36:44,640
Speaker 5: like otherwise it's just like it becomes too complicated.

642
00:36:45,160 --> 00:36:46,759
Speaker 2: Just on this note, can we talk a little bit

643
00:36:46,800 --> 00:36:51,400
Speaker 2: about I guess institutional knowledge of code because I remember

644
00:36:51,440 --> 00:36:53,720
Speaker 2: one of the things that happened in the early days

645
00:36:54,120 --> 00:36:56,520
Speaker 2: of AI development. I mean not that early. I think

646
00:36:56,520 --> 00:36:59,359
Speaker 2: it was like twenty seventeen or something, back when Facebook's

647
00:36:59,520 --> 00:37:03,440
Speaker 2: AI lab still existed. What was that called again, Like

648
00:37:03,520 --> 00:37:06,560
Speaker 2: the acronym was fair or something. Yeah, they had like

649
00:37:06,560 --> 00:37:10,520
Speaker 2: a little Facebook like experimental lab. Anyway, they invented a

650
00:37:10,520 --> 00:37:14,239
Speaker 2: bunch of chatbots, and the chatbots started talking to each

651
00:37:14,280 --> 00:37:18,640
Speaker 2: other in pretty much incomprehensible language, but like they clearly

652
00:37:18,760 --> 00:37:21,600
Speaker 2: understood what each other were saying. And so I'm just

653
00:37:21,680 --> 00:37:26,480
Speaker 2: wondering if you extrapolate that to AI generated code, could

654
00:37:26,560 --> 00:37:30,560
Speaker 2: we have a situation where the models are constantly iterating

655
00:37:30,600 --> 00:37:34,279
Speaker 2: on themselves, They're constantly talking to each other, and so

656
00:37:34,360 --> 00:37:37,000
Speaker 2: we end up with a system that becomes very very

657
00:37:37,040 --> 00:37:41,719
Speaker 2: difficult for human engineers and coders to actually understand.

658
00:37:42,480 --> 00:37:44,360
Speaker 5: Yeah, I mean I think what we're gonna stop to

659
00:37:44,640 --> 00:37:47,759
Speaker 5: is like humans, you know, like us are going to

660
00:37:47,880 --> 00:37:51,279
Speaker 5: move towards like creating like margdon files as like a

661
00:37:51,280 --> 00:37:53,440
Speaker 5: programming language. So everything is just going to be like

662
00:37:53,520 --> 00:37:58,879
Speaker 5: normal like language, but for the machine itself obviously. Yeah,

663
00:37:58,920 --> 00:38:02,720
Speaker 5: I remember that video that is like gibberish like voice

664
00:38:02,960 --> 00:38:05,960
Speaker 5: transfer thing. Well, if you think about it, it's not

665
00:38:06,120 --> 00:38:09,440
Speaker 5: that much different from voice to text in that sense.

666
00:38:10,000 --> 00:38:12,560
Speaker 5: And at the end of the day, just like bits

667
00:38:12,600 --> 00:38:15,719
Speaker 5: like being transferred. Because even like whenever you connect on

668
00:38:15,760 --> 00:38:18,200
Speaker 5: a web page, you know, like you write it in text,

669
00:38:18,320 --> 00:38:22,240
Speaker 5: but behind it just like bites that being a change.

670
00:38:22,640 --> 00:38:26,320
Speaker 5: So agents still need to agree on the protocol that

671
00:38:26,360 --> 00:38:29,600
Speaker 5: they're going to use, and not necessarily an encryption format,

672
00:38:29,640 --> 00:38:33,359
Speaker 5: but an encoding format. So once you know what it is,

673
00:38:33,400 --> 00:38:36,760
Speaker 5: you know, it just becomes like like a reverse engineering

674
00:38:36,800 --> 00:38:39,600
Speaker 5: problem or like a foreign sake problem where you're just like, okay,

675
00:38:39,640 --> 00:38:41,880
Speaker 5: like this is what's being used when THOSS packets are

676
00:38:41,920 --> 00:38:44,560
Speaker 5: being sent. You know, let me just decrypt it once

677
00:38:44,600 --> 00:38:47,719
Speaker 5: you know the protocol, you know. Like that's so I

678
00:38:47,719 --> 00:38:51,600
Speaker 5: don't think we're going to end up in a situation

679
00:38:51,760 --> 00:38:54,600
Speaker 5: where like we would have like no idea of it,

680
00:38:54,640 --> 00:38:56,319
Speaker 5: because you're always going to have like people who are

681
00:38:56,320 --> 00:38:58,960
Speaker 5: like pretty good like reverse engineers. But at the same time,

682
00:38:59,000 --> 00:39:01,400
Speaker 5: you're also going to have like aah assistant who's going

683
00:39:01,480 --> 00:39:04,279
Speaker 5: to help you to like reverse engineer those things. So

684
00:39:04,880 --> 00:39:08,400
Speaker 5: in a way, even if it happens like you are

685
00:39:08,480 --> 00:39:12,640
Speaker 5: not like alone with your red bull and you I

686
00:39:12,680 --> 00:39:15,640
Speaker 5: know AI agents, we're all gonna have AI agents. So

687
00:39:16,640 --> 00:39:21,760
Speaker 5: that's also like the reality of things. So we're far

688
00:39:21,800 --> 00:39:25,160
Speaker 5: from just like the Clippy plugin that we used to

689
00:39:25,160 --> 00:39:27,920
Speaker 5: have in Microsoft Office, or you can have this like

690
00:39:28,000 --> 00:39:31,160
Speaker 5: CLI interface. We just give like commands and then it's

691
00:39:31,400 --> 00:39:34,040
Speaker 5: led that into like Okay, I understand what I need

692
00:39:34,080 --> 00:39:36,040
Speaker 5: to do, you know, and that's it. No.

693
00:39:36,160 --> 00:39:39,279
Speaker 4: I love like interacting with just the CLI now, And

694
00:39:39,320 --> 00:39:41,320
Speaker 4: like every time I have to go to the web,

695
00:39:41,719 --> 00:39:44,319
Speaker 4: it feels like some sort of failure. I'm like, I

696
00:39:44,320 --> 00:39:46,200
Speaker 4: have to go to this brace website, yeah, and I

697
00:39:46,239 --> 00:39:48,760
Speaker 4: just like want the information right there on the black

698
00:39:48,800 --> 00:39:53,640
Speaker 4: screen talking, you know, communicating back and forth in English background.

699
00:39:53,719 --> 00:39:56,880
Speaker 5: You know, that's like feeling familiar for you.

700
00:39:56,880 --> 00:39:57,080
Speaker 3: You know.

701
00:39:57,200 --> 00:39:59,600
Speaker 4: Something I think about is I imagine that there's a lot

702
00:39:59,600 --> 00:40:05,439
Speaker 4: of like crusty old Linux and Unix programmers who are like, oh,

703
00:40:05,560 --> 00:40:08,399
Speaker 4: this code is this isn't high quality code that the

704
00:40:08,400 --> 00:40:10,640
Speaker 4: AI that the chat pods produced.

705
00:40:10,719 --> 00:40:12,479
Speaker 3: This is slop code. It needs all.

706
00:40:12,440 --> 00:40:15,759
Speaker 4: Kinds of fixing and so forth. From your perspective, is

707
00:40:15,800 --> 00:40:20,040
Speaker 4: the code itself of good quality or of improving quality?

708
00:40:20,080 --> 00:40:23,719
Speaker 4: How just the lines of code itself from your standards

709
00:40:23,760 --> 00:40:24,440
Speaker 4: is a good stuff?

710
00:40:24,600 --> 00:40:26,600
Speaker 5: Yeah, it's pretty good then even when it's bad. You

711
00:40:26,680 --> 00:40:29,600
Speaker 5: just said it okay, like this is bad. You know,

712
00:40:29,840 --> 00:40:31,640
Speaker 5: like just do it better. You know, like if you're

713
00:40:31,719 --> 00:40:33,960
Speaker 5: using like negative rewards, you know, if you say okay,

714
00:40:33,960 --> 00:40:35,879
Speaker 5: like this piece of code is garbage, you know it's

715
00:40:35,880 --> 00:40:39,200
Speaker 5: gonna understand better because you kind of give like a

716
00:40:39,280 --> 00:40:42,000
Speaker 5: strong emotion where it's like if you say, oh it's okay,

717
00:40:42,040 --> 00:40:44,040
Speaker 5: you know it's be like okay, like whatever. You know,

718
00:40:44,040 --> 00:40:47,239
Speaker 5: if it's okay, it means it's passing, like the the

719
00:40:47,440 --> 00:40:50,880
Speaker 5: Eddequo test. Okay, that's their badge. You write it.

720
00:40:51,000 --> 00:40:54,760
Speaker 2: And I always do this so if I ask about

721
00:40:55,040 --> 00:40:57,760
Speaker 2: for something, I will always say, like after the first version,

722
00:40:57,840 --> 00:40:59,759
Speaker 2: like do better and just see what it comes up with,

723
00:40:59,840 --> 00:41:01,359
Speaker 2: and just try to iterate this is.

724
00:41:01,440 --> 00:41:04,480
Speaker 4: This is tough because you're talking when you're talking in

725
00:41:04,560 --> 00:41:07,880
Speaker 4: English the brain deludes itself into thinking that you're talking

726
00:41:07,880 --> 00:41:10,240
Speaker 4: to or when you're talking in human language, any language,

727
00:41:10,400 --> 00:41:12,239
Speaker 4: the brain deludes itself you feel like you're talking to

728
00:41:12,239 --> 00:41:14,120
Speaker 4: a person, so you have to be nice.

729
00:41:14,160 --> 00:41:15,800
Speaker 2: Well, then I feel like I don't have that problem.

730
00:41:17,760 --> 00:41:19,680
Speaker 4: But then I feel like I like I don't want

731
00:41:19,719 --> 00:41:21,799
Speaker 4: to say, oh, this is garbage. I don't But from

732
00:41:21,880 --> 00:41:26,080
Speaker 4: your perspective, it's actually sort of important to to be

733
00:41:26,200 --> 00:41:28,520
Speaker 4: firm with the bot, and you get better results by

734
00:41:28,520 --> 00:41:29,640
Speaker 4: being more sharp with it.

735
00:41:29,800 --> 00:41:33,200
Speaker 5: Yeah, because it's the equivalent to the negative world. Like

736
00:41:33,360 --> 00:41:35,680
Speaker 5: just like positive world. If you if it does something,

737
00:41:35,719 --> 00:41:37,960
Speaker 5: you say, okay, like that's great, that's exactly what I

738
00:41:38,000 --> 00:41:41,560
Speaker 5: tried to explain. Then it's like okay, like that's like

739
00:41:41,600 --> 00:41:43,799
Speaker 5: a reference point. Was Like if it starts to go

740
00:41:43,840 --> 00:41:47,160
Speaker 5: on a tangent, you just say like, oh, that's completely

741
00:41:47,200 --> 00:41:50,319
Speaker 5: like out of the line, redo this, like why are

742
00:41:50,400 --> 00:41:53,640
Speaker 5: you doing this? You know, like the more explicit you are,

743
00:41:54,440 --> 00:41:57,280
Speaker 5: the better it's going to understand how far these from

744
00:41:57,400 --> 00:41:58,239
Speaker 5: the requirements.

745
00:41:58,520 --> 00:42:01,879
Speaker 4: Well, I just have to get better, you know, sort

746
00:42:01,920 --> 00:42:04,839
Speaker 4: of conflict avoidant, and I like very nice to people.

747
00:42:04,880 --> 00:42:07,040
Speaker 2: So I just still says please and thank you. So

748
00:42:07,120 --> 00:42:09,320
Speaker 2: I just bask doesn't get them.

749
00:42:09,440 --> 00:42:10,279
Speaker 3: So yeah, that's right.

750
00:42:10,320 --> 00:42:11,880
Speaker 4: So I have to I just have to be like, no,

751
00:42:12,000 --> 00:42:15,040
Speaker 4: this is trash, this is garbage. You totally we're all

752
00:42:15,120 --> 00:42:17,520
Speaker 4: dumber for having seen this code. Okay, this is good

753
00:42:17,560 --> 00:42:17,799
Speaker 4: to know.

754
00:42:18,040 --> 00:42:20,240
Speaker 5: I mean it's a good point. You know, like those

755
00:42:20,320 --> 00:42:23,640
Speaker 5: like AI companies that are recording all the prompts, you know, yeah,

756
00:42:24,239 --> 00:42:27,040
Speaker 5: if they're keeping it, if there's retention around it, you know,

757
00:42:27,160 --> 00:42:29,680
Speaker 5: like we know the open the eye is keeping them,

758
00:42:29,719 --> 00:42:33,319
Speaker 5: you know, like they can get sophy. Is so your

759
00:42:33,560 --> 00:42:36,360
Speaker 5: importance it? Yeah, so like who knows, you know, like

760
00:42:36,440 --> 00:42:39,440
Speaker 5: in years from now, if there's like full on like

761
00:42:39,520 --> 00:42:44,080
Speaker 5: autonomous like robots managed by the Department of War because

762
00:42:44,120 --> 00:42:45,359
Speaker 5: they think it's lowful so.

763
00:42:45,320 --> 00:42:50,200
Speaker 7: We all get social scores based on our Yeah, these

764
00:42:50,320 --> 00:42:53,440
Speaker 7: thoughts creep into my head on ironically where it's like

765
00:42:54,000 --> 00:42:56,640
Speaker 7: at some point is there's going to be some am

766
00:42:56,640 --> 00:42:57,800
Speaker 7: I going to regret having?

767
00:42:59,000 --> 00:43:01,359
Speaker 3: I don't know? They I they were there in my head.

768
00:43:01,440 --> 00:43:04,400
Speaker 2: Truly a brave new world, Matt. Just one last question

769
00:43:04,440 --> 00:43:07,600
Speaker 2: for me, but going back to cybersecurity and the current

770
00:43:07,640 --> 00:43:12,440
Speaker 2: situation with Iran, what are you on just the lookout for,

771
00:43:12,680 --> 00:43:15,640
Speaker 2: like what would peak your interest the most to see

772
00:43:15,800 --> 00:43:17,320
Speaker 2: in that particular space.

773
00:43:18,400 --> 00:43:23,080
Speaker 5: I think now in that particular space because I'm one

774
00:43:23,080 --> 00:43:25,800
Speaker 5: of those people who think it's related to the Epstein files,

775
00:43:25,800 --> 00:43:27,919
Speaker 5: you know, like it's just more about like getting more

776
00:43:28,440 --> 00:43:31,920
Speaker 5: Epstein Files related stuff to see if there is like

777
00:43:32,080 --> 00:43:35,480
Speaker 5: more connections to it, you know. So I think that

778
00:43:35,520 --> 00:43:37,360
Speaker 5: would be the only thing that would kind of be

779
00:43:37,440 --> 00:43:41,080
Speaker 5: like digital that we like spike my interest because now

780
00:43:41,080 --> 00:43:44,000
Speaker 5: we have seen you know, like just those like all

781
00:43:44,080 --> 00:43:46,840
Speaker 5: drones can do like so much damage, and that Iran

782
00:43:47,400 --> 00:43:51,400
Speaker 5: demonstrated that they can be like really precise with the attacks.

783
00:43:51,960 --> 00:43:55,479
Speaker 5: So now I think it's more about seeing like which

784
00:43:55,560 --> 00:43:58,160
Speaker 5: direction it's going to go to and how long it's

785
00:43:58,200 --> 00:44:01,600
Speaker 5: going to last, which is like the big question mark

786
00:44:01,640 --> 00:44:05,080
Speaker 5: because there's so many other components, like the energy sector,

787
00:44:05,160 --> 00:44:07,279
Speaker 5: you know, like how is it We have seen the

788
00:44:07,320 --> 00:44:10,839
Speaker 5: price of like memory like increasing like a lot. So

789
00:44:11,320 --> 00:44:14,240
Speaker 5: now now if they're starting to block like the Detroit

790
00:44:14,320 --> 00:44:16,880
Speaker 5: of almost you know, like like what's gonna happen to

791
00:44:16,960 --> 00:44:19,440
Speaker 5: like the cost of data centers and like memory and

792
00:44:19,520 --> 00:44:22,319
Speaker 5: AI in general. You know, like we're going towards on

793
00:44:22,400 --> 00:44:24,440
Speaker 5: one side, you know, we're going towards like the cost

794
00:44:24,480 --> 00:44:27,880
Speaker 5: of tokens and insurance going down, but that may also

795
00:44:27,960 --> 00:44:30,319
Speaker 5: like bring the cost up. You know, So if you're

796
00:44:30,320 --> 00:44:34,279
Speaker 5: gonna use like AI for like your next generation wars,

797
00:44:34,520 --> 00:44:37,319
Speaker 5: but if your enemy can just like increase your cost

798
00:44:37,360 --> 00:44:40,160
Speaker 5: of token and insurance, what does that even mean? Like

799
00:44:40,200 --> 00:44:42,160
Speaker 5: do you even need AI in the first place? Is

800
00:44:42,160 --> 00:44:44,759
Speaker 5: it even relevant? So I think there's this all like

801
00:44:44,840 --> 00:44:47,839
Speaker 5: asymmetric like warfare that's going to happen that we really

802
00:44:47,880 --> 00:44:50,160
Speaker 5: haven't seen yet, and I think that's gonna be like

803
00:44:50,480 --> 00:44:53,160
Speaker 5: really interesting. But at the same time, there's so much

804
00:44:53,200 --> 00:44:56,120
Speaker 5: noise and so much so many things happening at once

805
00:44:56,160 --> 00:44:59,000
Speaker 5: that it's becoming like extremely like hard to focus and

806
00:44:59,200 --> 00:45:01,880
Speaker 5: just like extract. But that's pretty ready.

807
00:45:01,600 --> 00:45:05,360
Speaker 2: But yeah, definitely feels like that tough, tough choices potentially

808
00:45:05,360 --> 00:45:08,680
Speaker 2: coming for half Alock as your token costs go up. Matt,

809
00:45:08,719 --> 00:45:11,160
Speaker 2: it was so wonderful to reconnect again. Thank you so

810
00:45:11,239 --> 00:45:13,920
Speaker 2: much for coming back on odd lots And yeah, you'll

811
00:45:13,960 --> 00:45:16,560
Speaker 2: have to get back to Joe about those APIs.

812
00:45:16,800 --> 00:45:18,760
Speaker 3: Yeah, I'm happy, let's let's.

813
00:45:18,600 --> 00:45:23,080
Speaker 5: Do Yeah, bring me on as sounds good.

814
00:45:23,320 --> 00:45:23,960
Speaker 3: Take care of Matt.

815
00:45:24,040 --> 00:45:38,279
Speaker 6: Thanks Matt, Joe.

816
00:45:38,320 --> 00:45:39,440
Speaker 2: Always good to catch up with.

817
00:45:39,520 --> 00:45:43,160
Speaker 4: No super interesting It's incredible how much has happened right

818
00:45:43,160 --> 00:45:48,200
Speaker 4: now at this particular nexus, especially obviously the anthropic stuff.

819
00:45:48,440 --> 00:45:49,680
Speaker 3: But I you know, it's interesting.

820
00:45:49,760 --> 00:45:52,359
Speaker 4: You know, you think about cyber warfare and you think about, okay,

821
00:45:52,360 --> 00:45:54,040
Speaker 4: we're going to hack into a system and take out

822
00:45:54,040 --> 00:45:56,680
Speaker 4: critical infrastructure. But another thing you can do is just

823
00:45:56,719 --> 00:45:58,480
Speaker 4: attack a data center correctly, just.

824
00:45:58,440 --> 00:46:00,520
Speaker 2: Send a drone to the data center. I thought that

825
00:46:00,600 --> 00:46:03,040
Speaker 2: was really interesting. That sort of like we got very

826
00:46:03,160 --> 00:46:06,040
Speaker 2: used to thinking about cyber as like this thing that

827
00:46:06,320 --> 00:46:10,200
Speaker 2: exists only in code, in code, but now you have

828
00:46:10,280 --> 00:46:13,000
Speaker 2: this like new front of kinetic warfare where the two

829
00:46:13,120 --> 00:46:14,120
Speaker 2: like really intersect.

830
00:46:14,280 --> 00:46:16,759
Speaker 4: Yeah, they do really intersect, and yeah, these are like

831
00:46:16,960 --> 00:46:21,160
Speaker 4: huge national security vulnerabilities. And he pointed out, I mean

832
00:46:21,280 --> 00:46:24,960
Speaker 4: certainly today, but you know, was this in anyone's threat

833
00:46:25,000 --> 00:46:27,120
Speaker 4: model thinking about the risks to it? You know, the

834
00:46:27,200 --> 00:46:30,680
Speaker 4: cheapness of drones the ability to take them out super interesting.

835
00:46:31,040 --> 00:46:35,200
Speaker 4: Also just this idea like yes, you know, obviously again

836
00:46:35,320 --> 00:46:37,840
Speaker 4: as your observation, the point we think of like cyber attacks,

837
00:46:37,840 --> 00:46:39,440
Speaker 4: like we're going to take out this whole thing, but

838
00:46:39,920 --> 00:46:42,680
Speaker 4: in least in the warfare context. His point like most

839
00:46:42,680 --> 00:46:44,880
Speaker 4: of it is like before the war et cetera, the

840
00:46:44,920 --> 00:46:48,560
Speaker 4: sort of information gather spy, spy craft and so forth

841
00:46:49,040 --> 00:46:51,040
Speaker 4: prior to the actual the actual attacks.

842
00:46:51,160 --> 00:46:54,279
Speaker 2: But it is interesting to see Israel in particular use

843
00:46:54,440 --> 00:46:57,840
Speaker 2: some cyber attacks as a sort of sewer of chaos.

844
00:46:57,920 --> 00:47:00,880
Speaker 2: Definitely in Iran. I can't imagine what it's like to

845
00:47:00,920 --> 00:47:02,839
Speaker 2: actually be on the ground there at the moment for

846
00:47:02,920 --> 00:47:06,640
Speaker 2: many reasons, but like you imagine just being there worried

847
00:47:06,680 --> 00:47:09,520
Speaker 2: about your physical safety, and then the traffic lights aren't

848
00:47:09,520 --> 00:47:10,160
Speaker 2: working as well.

849
00:47:10,280 --> 00:47:13,440
Speaker 4: Well, right, and also just think like wait, there's cameras

850
00:47:13,440 --> 00:47:17,000
Speaker 4: everywhere or how much is being recorded? Like create like

851
00:47:17,000 --> 00:47:21,000
Speaker 4: a sense of like paranoia among everyone about everything.

852
00:47:21,239 --> 00:47:24,640
Speaker 2: And also the interesting thing obviously this is very topical

853
00:47:24,640 --> 00:47:28,600
Speaker 2: in markets, but the SaaS apocalypse idea Matt in particular,

854
00:47:28,760 --> 00:47:33,520
Speaker 2: seemed pretty bearish on the outlook for existing software companies,

855
00:47:33,560 --> 00:47:35,640
Speaker 2: I guess, and I did think his comments about what

856
00:47:35,680 --> 00:47:39,920
Speaker 2: that would mean for security budgets within organizations were pretty

857
00:47:40,640 --> 00:47:44,319
Speaker 2: relevant and worrying. Absolutely, all right, shall we leave it there,

858
00:47:44,400 --> 00:47:46,520
Speaker 2: Let's leave it there. This has been another episode of

859
00:47:46,600 --> 00:47:49,080
Speaker 2: the Odd Thoughts podcast. I'm Tracy Alloway. You can follow

860
00:47:49,120 --> 00:47:50,600
Speaker 2: me at Tracy Alloway.

861
00:47:50,320 --> 00:47:53,000
Speaker 4: And I'm Jill Wisenthal. You can follow me at the Stalwarts,

862
00:47:53,040 --> 00:47:55,680
Speaker 4: follow our guest Matt Sweish, She's at m swish Fellow

863
00:47:55,760 --> 00:47:59,080
Speaker 4: or producers Carmen Rodriguez at Carmen Arman, Dash o'l bennett

864
00:47:59,080 --> 00:48:01,960
Speaker 4: at dashbot, ands at Kale Brooks. And for more odd

865
00:48:02,000 --> 00:48:04,640
Speaker 4: Laws content, go to Bloomberg dot com slash Oddlage for

866
00:48:04,719 --> 00:48:07,080
Speaker 4: the daily newsletter and all of our episodes, and you

867
00:48:07,120 --> 00:48:09,160
Speaker 4: can chat about all these topics twenty four to seven

868
00:48:09,239 --> 00:48:12,359
Speaker 4: in our discord Discord dot gg slash odd Lots.

869
00:48:12,680 --> 00:48:14,600
Speaker 2: And if you enjoy odd Lots, if you like it

870
00:48:14,640 --> 00:48:17,640
Speaker 2: when we talk about the intersection of kinetic and cyber warfare,

871
00:48:17,680 --> 00:48:20,000
Speaker 2: then please leave us a positive review on your favorite

872
00:48:20,040 --> 00:48:23,360
Speaker 2: podcast platform. And remember, if you are a Bloomberg subscriber,

873
00:48:23,440 --> 00:48:26,400
Speaker 2: you can listen to all of our episodes absolutely add free.

874
00:48:26,560 --> 00:48:28,640
Speaker 2: All you need to do is find the Bloomberg channel

875
00:48:28,680 --> 00:48:32,160
Speaker 2: on Apple Podcasts and follow the instructions there. Thanks for listening.