The Nomadic AI Developer with Aaron Erickson の全文書き起こし

that's still a problem. 59 00:04:44.839 --> 00:04:47.600 Well no, I mean, so what did I do? I ran 60 00:04:47.639 --> 00:04:54.040 an instance of Linux as a DNS server to lie to the replay TV to 61 00:04:54.879 --> 00:05:00.680 run my loader of metadata. So that I had replay D be populated with 62 00:05:00.680 --> 00:05:02.319 the right stuff because it's just some digital recorder. But you know, this 63 00:05:02.399 --> 00:05:06.959 is the joys of being a programmer for yourself. Yep, totally. Yeah. 64 00:05:08.000 --> 00:05:10.959 Well ah that doesn't work. Let me write in a little after that. 65 00:05:11.040 --> 00:05:12.759 I'll just make a thing with the stuff, and then we're done. 66 00:05:12.800 --> 00:05:15.959 Make a thing with this stuff and you're done. I you know, I'll 67 00:05:15.959 --> 00:05:18.560 tell you another story just as an aside before we get into this, because 68 00:05:18.600 --> 00:05:20.920 of course, now that we've moved up to the coast, I'm moving like 69 00:05:20.959 --> 00:05:26.160 all of my services. So I met my new optometrists this week, okay, 70 00:05:26.240 --> 00:05:30.360 and I'm chatting with the front desk lady that the the lady actually runs 71 00:05:30.399 --> 00:05:36.519 the show, and we end up in a conversation about her husband's hearing problems 72 00:05:38.319 --> 00:05:41.639 and how he's on the list for a calcular implant, which you know takes 73 00:05:41.639 --> 00:05:43.720 a little time to get a certain don stort. It's like, that's this 74 00:05:43.839 --> 00:05:46.199 is not hearing aid bad, this is you cannot hear. You can't hear. 75 00:05:46.319 --> 00:05:51.759 And I showed her live transcribe on Android because she just didn't know. 76 00:05:53.360 --> 00:05:56.480 Wow, right, it's like right now she's like, I'm frustrated trying to 77 00:05:56.480 --> 00:05:58.240 talk to him because I have to shout at him, and it says, 78 00:05:58.240 --> 00:06:01.439 well, that automatically changes your mood, and you know you're frustrated. And 79 00:06:01.480 --> 00:06:03.879 I literally just fired it up on the phone and put it in front of 80 00:06:03.879 --> 00:06:08.600 her and we start kept talking, like look down, everything we're saying is 81 00:06:08.639 --> 00:06:11.800 there, Like, let me put us on your phone when you go home 82 00:06:11.839 --> 00:06:15.079 tonight. Just put it in front of him and start talking and see if 83 00:06:15.079 --> 00:06:17.759 you don't figure it out right away. That's so cool. So I got 84 00:06:17.800 --> 00:06:21.560 a call from her yesterday. It's like you've changed everything. That is awesome. 85 00:06:21.639 --> 00:06:26.279 I'm also really aware like cochlear implants are only so good, Like they're 86 00:06:26.319 --> 00:06:30.680 expecting hations for what an implant's going to be. Are you know me? 87 00:06:30.879 --> 00:06:33.600 You see a path ahead of someone where it's going to be months of sadness. 88 00:06:33.680 --> 00:06:36.360 It's like, look, there are other ways to address this. Yeah, 89 00:06:36.360 --> 00:06:40.680 Like if there's anything we've got going for us living in this technology world, 90 00:06:40.759 --> 00:06:44.199 it's take that knowledge to folks who don't have it. You're so right, 91 00:06:44.279 --> 00:06:46.800 she owned the phone, she didn't know the product existed, and it's 92 00:06:47.079 --> 00:06:50.879 free. Yeah, amazing, that's a great story, Richard. Yeah, 93 00:06:51.120 --> 00:06:55.160 well, anyway, that's our better note framework who's talking to us today. 94 00:06:55.199 --> 00:06:58.319 Mister Campbell grabbed the comment of the show eighteen ninety two. The one we 95 00:06:58.319 --> 00:07:01.480 did with Michelle Duke once was Michelle Mannering but then you know, got married, 96 00:07:01.560 --> 00:07:03.959 changed your name, all that good stuff. Who works for Gethub and 97 00:07:03.959 --> 00:07:06.759 we talked about gethub copilot, which we'd done before with her, so it 98 00:07:06.800 --> 00:07:12.040 was fun to get an update on that. And Joshua Hillarup, who's been 99 00:07:12.079 --> 00:07:15.199 on the show as well, you know guests and regular commenter, said, 100 00:07:15.279 --> 00:07:17.399 I'm with Richard on the future of AI. I can't say what will happen 101 00:07:17.480 --> 00:07:20.759 hundreds of years from now in this space, but I think we before we 102 00:07:20.800 --> 00:07:25.319 would get a business person giving all the requirements to an ass generative program, 103 00:07:25.399 --> 00:07:30.959 we would get AI largely replacing that business person is dead good line. I 104 00:07:30.000 --> 00:07:34.720 mean, let's be clear, how much time do you spend getting business requirements 105 00:07:34.160 --> 00:07:39.800 from a domain expert like It takes a while, so that's not a trivial 106 00:07:39.839 --> 00:07:42.879 thing. But yeah, there's many changes coming in space. And I know 107 00:07:42.920 --> 00:07:46.000 we're going to talk to Aaron today and he's certainly working with companies that have 108 00:07:46.079 --> 00:07:50.680 a role to play in generative AI, so that I thought I'd grab that 109 00:07:50.680 --> 00:07:57.040 little story and think, you know, you know, don't don't take the 110 00:07:57.120 --> 00:08:01.480 obvious answers here. Every time we have a new automated automation tool in our 111 00:08:01.519 --> 00:08:07.519 industry or in any industry, that industry evolves and generally results in more work, 112 00:08:07.720 --> 00:08:13.279 not less. One, Joshua, I know you already have a copy 113 00:08:13.279 --> 00:08:16.480 of music code By, but I wrote the comment anyway because I really appreciate 114 00:08:16.519 --> 00:08:18.160 your insights and you can reach out to me anytime if you like it. 115 00:08:18.160 --> 00:08:20.079 If you'd like a copy of music code By, I write a colm on 116 00:08:20.120 --> 00:08:24.000 the website at don at Rocks dot com or on the facebooks. We publish 117 00:08:24.000

--> 00:08:26.319 every show there, and if you comment there and ever reading the show, 118 00:08:26.319 --> 00:08:28.120 I'll send you a copy of music Coba and you can follow us on Twitter 119 00:08:28.240 --> 00:08:31.720 or x or y or z or whatever the hell it is. Today. 120 00:08:31.759 --> 00:08:37.559 But the cool kids are hanging out at mastadon. I'm Carl Franklin at techhub 121 00:08:37.559 --> 00:08:41.720 dot social, and I'm Rich Campbell at Macedon dot social. Aaron Ericson the 122 00:08:41.840 --> 00:08:46.840 last time he was here was two thousand and nine. He leads a team 123 00:08:46.919 --> 00:08:52.120 building autonomous agents at Nvidio maybe a little company might have heard of. As 124 00:08:52.120 --> 00:08:56.720 I said, his first book, Nomadic Developer, dropped in two thousand and 125 00:08:56.759 --> 00:08:58.879 nine, and that's the last time we talked to him. And since then, 126 00:09:00.399 --> 00:09:05.039 he spent nine years at thought Works, moved to San Francisco, built 127 00:09:05.120 --> 00:09:09.559 internal developer platforms at Salesforce, became a VPE at New Relic, then lost 128 00:09:09.559 --> 00:09:13.279 his mind and decided to become a startup founder, where he attempt to San 129 00:09:13.320 --> 00:09:18.840 Francisco, where he attempted to I just want to know where mushrooms involved. 130 00:09:18.159 --> 00:09:24.639 Uh. Yeah, So he attempted to build a company there at the startup 131 00:09:24.639 --> 00:09:30.600 that used AI to automate your company reorg. Wow. After that startup didn't 132 00:09:30.639 --> 00:09:35.159 work, he joined Nvidia, where he now builds AI agents that optimized the 133 00:09:35.240 --> 00:09:41.600 usage of GPU resources across the GPU fleet for Nvidia. So basically he's a 134 00:09:41.639 --> 00:09:45.039 slacker. Yeah, just later on, Just dude, no wonder we haven't 135 00:09:45.039 --> 00:09:46.759 talked to him in fifteen years. Didn't have anything, he wasn't doing anything. 136 00:09:46.840 --> 00:09:52.679 Oh my god, what kind of bio is that? How are you 137 00:09:52.720 --> 00:09:56.039 doing? Aaron? And welcome back? I feel fantastic. I mean, 138 00:09:56.159 --> 00:10:01.039 you know, what kind of life charmed life? Do you live? Where 139 00:10:01.039 --> 00:10:03.240 you fail at a startup, and they say, hey, why don't you 140 00:10:03.360 --> 00:10:09.960 just do even crazier stuff with AI than automating reorcs, which already sounds, 141 00:10:11.240 --> 00:10:15.159 I don't know, a little bit audacious. You could say, yeah, 142 00:10:15.200 --> 00:10:18.080 that's a nice word. Yeah, I don't know, dude. That seems 143 00:10:18.159 --> 00:10:24.240 like we already have a problem with too much remoteness as it is, right, 144 00:10:24.320 --> 00:10:28.480 Like nobody wants to be laid off by text message. It's just the 145 00:10:28.519 --> 00:10:31.600 phrase that gives me shells. I'm sure we could humanize it it certainly. 146 00:10:31.879 --> 00:10:35.000 The biggest problem I have with reorcs is often people don't go through all of 147 00:10:35.039 --> 00:10:39.759 the steps to take care of folks properly. And so if I had a 148 00:10:39.840 --> 00:10:43.679 digital checklist that's making sure I'm doing the right things, maybe I can actually 149 00:10:43.720 --> 00:10:48.159 do it better. But that's a tough elevator pitch, it is. Yeah, 150 00:10:48.679 --> 00:10:56.159 the funny was so funny enough. You know where we landed with this. 151 00:10:56.799 --> 00:11:00.559 So we're working on a problem which was I think a little bit more 152 00:11:00.639 --> 00:11:05.159 virtuous when we started, which was how do we help engineering leaders reorganize our 153 00:11:05.159 --> 00:11:09.440 teams in a better way, in a way that kind of has multiple displaynering 154 00:11:09.480 --> 00:11:13.639 teams. I think it started out that way, and then, as what 155 00:11:13.720 --> 00:11:16.600 happened to a lot of startups in twenty twenty three or late twenty one two, 156 00:11:16.360 --> 00:11:20.200 chech gt comes out and everybody's like, well, how the hell do 157 00:11:20.279 --> 00:11:24.360 I raise another round? And how do I like it if I don't have 158 00:11:24.440 --> 00:11:28.200 this? Yeah, how do I plant dot ai at the end of whatever 159 00:11:28.240 --> 00:11:31.840 I'm doing and make it moderately relevant? And so our dumb idea was, 160 00:11:31.879 --> 00:11:35.720 well, we already have this engine that allows you to kind of do your 161 00:11:35.039 --> 00:11:39.080 change your organization in a draft mode and then have your friends approve it or 162 00:11:39.080 --> 00:11:43.120 in your org and then go and do it. Why don't we just put 163 00:11:43.159 --> 00:11:46.320 a chech tee wrapper around it and say we're going to write that reorger email 164 00:11:46.320 --> 00:11:50.600 that everybody hates writing It already sounds robotic. I mean, who has ever 165 00:11:50.639 --> 00:11:54.519 gotten an email from HR that didn't look like it came from chet gipt in 166 00:11:54.519 --> 00:12:00.559 the first place. It certainly infused with HR legal ees, yes and double 167 00:12:00.600 --> 00:12:03.480 speak. I mean, it's to me, it's the classical case. I 168 00:12:03.519 --> 00:12:05.639 mean, there's a lot of things you do not want Jenner Todai to do, 169 00:12:05.759 --> 00:12:09.279 you know. I look at the music that you know, Carl that 170 00:12:09.320 --> 00:12:11.159 you're into, and I look at a lot of really creative pursuits, and 171 00:12:11.159 --> 00:12:15.120 I think creativity is going to be you know, despite all the noise, 172 00:12:15.200 --> 00:12:18.159 right, it's going to be one of the last bestiges where jenner Ai is 173 00:12:18.159 --> 00:12:22.279 really useful robotic things from HR. I think it is extraordinarily useful in that 174 00:12:22.320 --> 00:12:26.399 because at least we can get rid of the pretense that anybody really cares about 175 00:12:26.440 --> 00:12:31.120 that

communication, right, because mostly they care about it for risk management. 176 00:12:31.159 --> 00:12:33.879 They don't like, how do we make it sound as anodyne as possible? 177 00:12:33.919 --> 00:12:37.960 That that's the whole point. So why not have can we soften the blow? 178 00:12:37.120 --> 00:12:41.399 Ye? Yeah, well and not expose ourselves to legal issues, right, 179 00:12:41.600 --> 00:12:45.279 you know, I mean that therein lies the bigger thing. Like I 180 00:12:45.320 --> 00:12:48.360 know, I know folks in HR who are remarkably compassionate people. They're just 181 00:12:48.399 --> 00:12:54.600 not allowed to be most of the time because you slam into labor laws and 182 00:12:54.600 --> 00:13:00.559 and anything you say can be used. You know, you're almost automatic living 183 00:13:00.600 --> 00:13:05.360 in Miranda Rights Land, just by talking about HR related issues. Yeah, 184 00:13:05.559 --> 00:13:09.080 I mean I used to do this thing at conferences where I'd say I'd write 185 00:13:09.120 --> 00:13:11.360 the memo, but then you could put at the end and write it in 186 00:13:11.399 --> 00:13:16.639 the form of a Homrian epic or sometimes a limerick. Yes, and you 187 00:13:16.720 --> 00:13:20.000 have been let go. But you know what, it's kind of dumb. 188 00:13:20.000 --> 00:13:22.519 But it was like I would really appreciate that if I was being let go 189 00:13:22.639 --> 00:13:26.080 to actually read it, you know, in Iambic pentaminter for example, here 190 00:13:26.159 --> 00:13:30.320 you go rhyming couples. You know, at least it would make me laugh. 191 00:13:33.960 --> 00:13:37.720 Man. I mean, it's kind of crazy that it's been fifteen years 192 00:13:37.720 --> 00:13:43.080 since Nomadic Developer. I mean, spending nine years at Thought Works, is 193 00:13:43.080 --> 00:13:46.559 that really nomadding You kind of put down roots, I mean, except for 194 00:13:46.600 --> 00:13:50.759 the fact that you're going from company to company to company and with a banner 195 00:13:50.799 --> 00:13:54.559 over your head. Yeah. Yeah, and my personal stuff, I mean 196 00:13:54.559 --> 00:13:58.919 I was I went from engineered a bunch of other roles, you know. 197 00:13:58.960 --> 00:14:01.440 I think it was like, hey, be a product manager, be you 198 00:14:01.519 --> 00:14:03.240 this, be you that, And you know, at the end of the 199 00:14:03.320 --> 00:14:07.039 day, it's like, decide what you want to do for a living. 200 00:14:07.159 --> 00:14:11.399 Maybe you are you an engineer? Are you not? I'm still thinking about 201 00:14:11.399 --> 00:14:13.440 it. I don't know. Yeah, yeah, I don't know what. 202 00:14:13.559 --> 00:14:16.480 I don't know what I want to be when I grow up. That same 203 00:14:16.559 --> 00:14:22.240 thing, but again, it's still is the nomadding mindset, the idea of 204 00:14:22.240 --> 00:14:24.519 what's always the next thing. And you know what most people just say was 205 00:14:24.559 --> 00:14:28.600 a managed, well managed career. You kind of put the nomad's been on 206 00:14:28.720 --> 00:14:31.840 at least that part of it. I think a lot of it for me, 207 00:14:33.440 --> 00:14:35.480 A lot of it, despite the randomness of it, was intentional. 208 00:14:37.320 --> 00:14:39.279 And it was intentional because I wanted to, you know, if you want 209 00:14:39.279 --> 00:14:41.720 to be, you know, somebody leading something. I think it's good to 210 00:14:41.759 --> 00:14:45.399 know a little bit of sales. It's good to know a little bit of 211 00:14:45.440 --> 00:14:48.279 product. It makes you a better engineer. If you know a little bit 212 00:14:48.320 --> 00:14:50.039 of product. You know a little bit about why, why are we building 213 00:14:50.039 --> 00:14:54.799 this? Who does it matter to have you done customer research? You know, 214 00:14:54.919 --> 00:14:58.080 you become a better engineer if you ride along totally right something. We 215 00:14:58.080 --> 00:15:03.879 did it strangely, we started rotating devs into Strangely, we were building network 216 00:15:03.919 --> 00:15:07.879 appliances, and so it turned out that the lynchpin of the whole company were 217 00:15:07.919 --> 00:15:11.720 our installers, our system integrators. Because you're touching, you're getting, you're 218 00:15:11.720 --> 00:15:15.120 putting stuff into people's networks. Everybody's network has the ugliness. This is the 219 00:15:15.200 --> 00:15:18.799 question of where you know where it is right and the Sis and they are 220 00:15:18.879 --> 00:15:22.360 all geniuses, like just brilliant, brilliant people inevitably had to fix something in 221 00:15:22.399 --> 00:15:26.480 a network to have the appliance work, like it's just not there was never 222 00:15:26.960 --> 00:15:31.840 not that. And when we started sticking devs with them, you do a 223 00:15:31.919 --> 00:15:37.759 six week rotation with the SIS as a dev just be on the call see 224 00:15:37.759 --> 00:15:41.000 some of the things. Every single time they came back, they're like, 225 00:15:41.440 --> 00:15:46.840 holy man, those guys' job is hard and our product got better for it 226 00:15:46.120 --> 00:15:50.399 every time because they just saw, how do I provide his ability to the 227 00:15:50.440 --> 00:15:54.879 behavior? How do I show you know this kind of flow? How do 228 00:15:54.000 --> 00:15:58.480 you how do I give them switches to turn features off and on quickly so 229 00:15:58.679 --> 00:16:03.320 they can figure out what's going on with that customer. It's this cross discipline 230 00:16:03.320 --> 00:16:07.399 thing is it's insanely valuable. I think you need a certain amount of depth 231 00:16:07.399 --> 00:16:10.320 to pull it off. But boy, I don't know how you become good 232 00:16:10.320 --> 00:16:12.320 without doing it. Yeah, I mean, I mean there's lots of people 233 00:16:12.440 --> 00:16:15.720 get at their niches. I wouldn't, you know, downplay the person that's 234 00:16:15.759

--> 00:16:18.960 been doing engineering and a very specific thing in a very specific way. There's 235 00:16:18.080 --> 00:16:22.399 very good people doing that. But I would say, if you aspire to 236 00:16:22.440 --> 00:16:25.919 run a company someday, or if you aspire to be a really truly great 237 00:16:25.919 --> 00:16:30.440 engineering leader, having those perspectives will be a lay up on average. As 238 00:16:30.480 --> 00:16:33.320 a way, I would say, yeah, yeah, And every time you 239 00:16:33.320 --> 00:16:37.519 can get a chance to dip into the other spaces, see how decisions are 240 00:16:37.519 --> 00:16:41.440 made, Learn that language, you know how those folks speak and think, 241 00:16:42.360 --> 00:16:47.159 just makes you more capable to helping more people into being better at your own 242 00:16:47.200 --> 00:16:49.639 job. I find it really interesting that you went to Nvidia. I mean, 243 00:16:51.200 --> 00:16:56.159 in Vidia makes the infrastructure that makes a possible. And obviously you can 244 00:16:56.200 --> 00:17:00.799 see the success of Envidia by looking at their stock price over the last few 245 00:17:00.840 --> 00:17:06.079 years, it has gone through the roof the trillion dollar club. Oh my 246 00:17:06.200 --> 00:17:14.000 god. And so what I'm really interested in is you know where Nvidia lies 247 00:17:14.039 --> 00:17:18.440 in the software side of AI, which is what you're doing there. How 248 00:17:18.480 --> 00:17:23.240 does that happen? Well, the real advantage of Nvidia is in fact software, 249 00:17:23.279 --> 00:17:27.200 It's not just hardware, right. Kuda is the moat or one of 250 00:17:27.200 --> 00:17:30.759 the main motes of why in Vidia is successful. It's kind of like how 251 00:17:32.279 --> 00:17:37.440 Windows helped make Microsoft successful because they had backwards compatibility for many years. Well, 252 00:17:37.440 --> 00:17:41.759 it turns out if you wrote your software for Kuda in two thousand and 253 00:17:41.759 --> 00:17:45.000 seven, it still works on every Nvidia device you buy, on every Nvidia 254 00:17:45.119 --> 00:17:49.880 GPU. So explain Kuda for those who don't know. Kuda is a it's 255 00:17:49.920 --> 00:17:55.480 the low level AKA there that allows you to write code and C right or 256 00:17:55.559 --> 00:17:57.359 and there's Python libraries to do it too. But if you need to write 257 00:17:57.400 --> 00:18:03.319 soft it's basically kind of like the assembly layer plus the C layer for every 258 00:18:03.319 --> 00:18:07.640 GPU if you're trying to if you're trying to do multi parallel processing of you 259 00:18:07.640 --> 00:18:11.279 know, primarily graphics. That's why it's called the GPU. But it turned 260 00:18:11.319 --> 00:18:14.319 out, and this is almost kind of they put it out there, and 261 00:18:14.640 --> 00:18:17.680 Jensen Longer CEO, tells the story a lot better than I could possibly tell 262 00:18:17.720 --> 00:18:22.640 it. But you think about how AI researchers found out. It also seems 263 00:18:22.640 --> 00:18:26.799 to be seems to do matrix multiplication very very well well, which is core 264 00:18:26.920 --> 00:18:33.440 to how generative AI works in most AI works in general. So it turns 265 00:18:33.440 --> 00:18:37.839 out that that was in the in the sense of accelerated computing, which is 266 00:18:37.880 --> 00:18:41.359 what we talk about now, which AI is one type of accelerated computing. 267 00:18:41.119 --> 00:18:45.279 We don't have More's law anymore quite the same way we did, right, 268 00:18:45.680 --> 00:18:48.319 you know, I mean it's level by the new More's law. I mean, 269 00:18:48.359 --> 00:18:52.799 it's this kind of massive rise of the ability for multiple parallelism for a 270 00:18:52.920 --> 00:19:00.319 great you know, scale to accelerate computing. So that's kind of the story 271 00:19:00.599 --> 00:19:03.680 of how a thirty year company that kind of you know, most companies don't 272 00:19:03.680 --> 00:19:07.559 become two trillion dollar companies in their thirtieth year. They've become that maybe in 273 00:19:07.559 --> 00:19:11.799 their tenth year, and it wasn't. Yeah. Well, I mean, 274 00:19:12.000 --> 00:19:15.720 and I remember I have a friend who's one of those low level developer types 275 00:19:15.720 --> 00:19:19.519 who's moved from job to job, staying in that And one of the gigs 276 00:19:19.519 --> 00:19:26.759 he did was in the middle Oughts, and he was working for an astronomical 277 00:19:26.799 --> 00:19:33.279 society and he was building software to utilize GPUs to do analytics for identifying asteroids. 278 00:19:33.799 --> 00:19:37.200 So you're literally looking at two pictures of saying what moved kind of thing 279 00:19:37.240 --> 00:19:41.720 rights complicated stuff, and he was hand coding all this stuff because Andrew's that 280 00:19:41.839 --> 00:19:48.079 kind of guy. And Kuda came out and he literally spent a weekend re 281 00:19:48.200 --> 00:19:51.640 implemented with using Kudah. And when I've been wasting my time. This is 282 00:19:51.720 --> 00:19:56.359 just drue and we're talking to v one stuff of kudo. Like, it's 283 00:19:56.359 --> 00:19:59.240 not like they were thinking about Jenner of AI in two thousand and seven. 284 00:20:00.039 --> 00:20:03.559 That's not what they were thinking about, right, It's just astonished. It's 285 00:20:03.599 --> 00:20:08.279 astonishing to think that that library is continued to evolve to become one of the 286 00:20:08.279 --> 00:20:15.759 great underpinnings of this multi parallel processing space we're now running in as hardware people. 287 00:20:15.839 --> 00:20:22.119 Well, I remember talking to Steve Sanderson in at NDC in twenty thirteen. 288 00:20:22.240 --> 00:20:30.680 I think about his project using WebGL to take advantage of GPU processing inside 289 00:20:30.720 --> 00:20:36.240 a browser in twenty thirteen. Yeah, that's the HTML five revolution, right, 290 00:20:36.319 -->

00:20:41.240 the CAMAS compute and all of that sort of thing. I've certainly been 291 00:20:41.279 --> 00:20:48.680 banging on the drum about Moore's law ending for some time now. Was TMSC 292 00:20:48.839 --> 00:20:52.359 talking about the three nanometer process like, folks, you're running out of atoms 293 00:20:52.640 --> 00:20:57.880 just cannot get much smaller than that, right center. Well, and also 294 00:20:59.480 --> 00:21:03.440 that it's not just that we went parallel, it's that we're also doing specialized 295 00:21:03.519 --> 00:21:07.720 compute. So I mean for the long we you know, go back a 296 00:21:07.759 --> 00:21:14.559 few years and we had math coprocessors for our CPUs, and then but when 297 00:21:14.599 --> 00:21:18.240 you move up the stack. Obviously, the GPU was the first real alternative 298 00:21:18.240 --> 00:21:23.640 to the CPU for major workloads. Now they're talking about MPUs, right, 299 00:21:23.680 --> 00:21:29.839 they start a neural processing unit, although I personally don't I mean, I 300 00:21:29.920 --> 00:21:32.759 haven't dug far enough it is to say how different is an MPU from a 301 00:21:32.839 --> 00:21:37.079 GPU. Really, I think if you really kind of break it down, 302 00:21:37.000 --> 00:21:41.799 there's going to be a lot of different kinds of we'll just call it asterix 303 00:21:41.079 --> 00:21:45.759 PU, right, you know. And there's as we've gotten more advanceds of 304 00:21:45.799 --> 00:21:47.920 technology. I mean, I don't know if it's widely known, but we 305 00:21:48.000 --> 00:21:52.160 use a lot of AI to develop chips. Sure it's SOLF, right, 306 00:21:52.440 --> 00:21:55.960 and a lot of that's going to come up with lots of different designs. 307 00:21:56.400 --> 00:21:59.400 Like if you can imagine if you can compile your program down to something that 308 00:21:59.440 --> 00:22:03.200 actually runs on hardware where you know your normal instruction set is xor and yeah, 309 00:22:03.279 --> 00:22:06.839 you know your normal kind of assembly language instructions you might have learned early 310 00:22:06.839 --> 00:22:10.000 in your career. What if it's the software, right, What if you 311 00:22:10.039 --> 00:22:11.839 had a special chip for every kind of little bit of software that you were 312 00:22:11.839 --> 00:22:15.000 going to run be it generated bay, I be it whatever kind of future 313 00:22:15.079 --> 00:22:22.039 architecture around neural processing or other things, because the cost to build specialized chips 314 00:22:22.079 --> 00:22:23.279 is going down as well, right, so so much so, there's like, 315 00:22:23.440 --> 00:22:26.319 I think in New York there's a there's a you know, somebody has 316 00:22:26.319 --> 00:22:29.680 like a food cart, but instead of a food cart, it will make 317 00:22:29.720 --> 00:22:33.519 a die. You can actually literally change the software on the die and make 318 00:22:33.599 --> 00:22:37.519 it build like a string copy function that runs purely in hardware or something. 319 00:22:37.599 --> 00:22:41.119 Right, you know, I like going out for chips, but this is 320 00:22:41.160 --> 00:22:45.599 a different kind of chip. Do not get this one in mayonnaise? That 321 00:22:45.720 --> 00:22:52.039 is wrong. My first album with my brother Franklin Brothers nineteen ninety nine, 322 00:22:52.480 --> 00:22:56.319 we program I mostly me programmed drums, you know, drum samples and stuff 323 00:22:56.799 --> 00:23:02.079 for all the drums, and we gave him album credits. Chip Franklin, 324 00:23:03.759 --> 00:23:07.680 I didn't know you had a third brother. I don't. I don't. 325 00:23:07.720 --> 00:23:11.359 Well, depends on your definition of brother. I guess. You know the 326 00:23:11.440 --> 00:23:15.960 count Ai. You know it's maybe a family member, I guess. But 327 00:23:15.279 --> 00:23:19.119 it's a little wild, certainly. I mean, I think Gethub nailed it 328 00:23:19.160 --> 00:23:22.680 with the term co pilot, just like it's an assistant. Right, it's 329 00:23:22.720 --> 00:23:27.839 an automation tool. You're still the pilot. It's your fault when things go 330 00:23:27.920 --> 00:23:33.000 wrong, And that co pilot has become a buzzword for any kind of a 331 00:23:33.200 --> 00:23:36.200 I've stopped trying to drink every time they say it because I'm out of the 332 00:23:36.240 --> 00:23:38.160 keynote in five minutes, right, like you've done. I have a provocative 333 00:23:38.240 --> 00:23:44.160 question, what if you become the co pilot? Yeah? I don't think 334 00:23:44.200 --> 00:23:48.319 it's all that prodocative because I haven't haven't seen a piece of software that can 335 00:23:48.400 --> 00:23:52.039 that has direction yet. Yeah, but you know, so like a lot 336 00:23:52.039 --> 00:23:55.960 of the things that we're doing now these days are so I have a project. 337 00:23:55.960 --> 00:23:57.640 So we talked about autonomous observability agents. We want to go there? 338 00:23:59.160 --> 00:24:03.839 Yeah there, All right, what does that do? Because a lot of 339 00:24:03.880 --> 00:24:07.160 people are like that sounds like a cool title. What does that actually mean? 340 00:24:07.720 --> 00:24:11.000 Well, one of the things we discovered and so in my day job, 341 00:24:11.039 --> 00:24:14.759 which is we do GPU allocations. So we got to figure out all 342 00:24:14.799 --> 00:24:18.960 the internal external groups that are going to use GPUs and take the scarce resource 343 00:24:18.000 --> 00:24:21.559 and allocate them in a manner that is not just good for a video but 344 00:24:21.559 --> 00:24:22.920 good for a customer base. All that stuff right, all the good things. 345 00:24:25.039 --> 00:24:26.839 It turns out we have more data sources than we can possibly just kind 346 00:24:26.839 --> 00:24:32.880 of automatically do engineer or like human do data engineering to understand as quickly as 347 00:24:32.920 --> 00:24:36.039 we need to. And so it came to us, what if we

built 348 00:24:36.039 --> 00:24:41.319 a system that could do texts where you can say words in human in English 349 00:24:41.359 --> 00:24:45.720 or whatever human language you want, convert that into either you know, druid 350 00:24:45.799 --> 00:24:51.480 squol or elastic search queries or any other thing. Right, So we're not 351 00:24:51.480 --> 00:24:53.200 trying to be dev in the software developer per se. We're not trying to 352 00:24:53.440 --> 00:24:57.839 write and try our programs. It turns out AI is extraordinarily good at writing 353 00:24:57.960 --> 00:25:02.880 short programs that all the well known patterns. And there's a lot of training 354 00:25:02.960 --> 00:25:07.920 data on how to write SQ well in the corpus of models today, and 355 00:25:07.960 --> 00:25:11.039 you can do a little bit of tuning to make it right. You know, 356 00:25:11.079 --> 00:25:14.599 code that runs against rest endpoints, code that runs against well known things 357 00:25:14.640 --> 00:25:18.279 like elastic search. You know we have in fact, you know, Jensen 358 00:25:18.319 --> 00:25:22.319 talked about it in his latest keynote, doing text to a back and SAP. 359 00:25:22.559 --> 00:25:26.839 So if you want to just ask questions of your database, ask questions 360 00:25:26.880 --> 00:25:33.440 of your large ERP system right. Being able to do that. What if 361 00:25:33.440 --> 00:25:37.759 I can ask questions about my GPU telemetry? What if I can get answers 362 00:25:37.839 --> 00:25:41.400 back about my GPU telemetry? Wouldn't it be great? And I had this? 363 00:25:41.799 --> 00:25:44.680 The story I can tell is, you know, one of my frustrations 364 00:25:44.680 --> 00:25:47.640 when I was, you know, VP of engineering at New Relic was I'd 365 00:25:47.680 --> 00:25:51.039 be asked to do things like, hey, what happened with that incident last 366 00:25:51.119 --> 00:25:53.759 night? A lot of times it wouldn't be as accurate as I would like 367 00:25:53.880 --> 00:25:56.960 because I had to go at eight am, I had to call up a 368 00:25:57.000 --> 00:26:00.640 bunch of directors and ask, hey, what have last night? And they're 369 00:26:00.680 --> 00:26:06.480 tired, right, they're they're they're exhausted. They they spent the night fighting 370 00:26:06.799 --> 00:26:10.559 the fire, right, And so you know, you think about AI's a 371 00:26:10.599 --> 00:26:14.039 hallucinating Well, nobody hallucinates like somebody that's not gotten very much sleep, right, 372 00:26:14.160 --> 00:26:19.000 So so you've hit you have this dynamic. And so I walk into 373 00:26:19.039 --> 00:26:23.119 these meetings half the time right, half the time completely off base. What 374 00:26:23.160 --> 00:26:29.400 I wanted was a system where I can ask questions, what the F happened 375 00:26:29.480 --> 00:26:32.559 last night? Right, and add out the F if you want, no, 376 00:26:33.440 --> 00:26:36.799 No, I don't think and then I get an answer back, and 377 00:26:36.920 --> 00:26:40.440 I can then say, oh great, that took three seconds or even took 378 00:26:40.440 --> 00:26:42.880 thirty seconds because it thought a lot about the question. We allowed it to 379 00:26:42.920 --> 00:26:47.799 ruminate a little bit and confer with other experts in the system about the answer, 380 00:26:48.960 --> 00:26:52.160 and then I don't give me a considered answer and then be able to 381 00:26:52.200 --> 00:26:55.400 help me ask the next question. You see this in perplexity today. By 382 00:26:55.440 --> 00:26:59.240 the way, you see in perplexity you have this capability where you ask a 383 00:26:59.319 --> 00:27:02.039 question and gives you the next five good questions as a consequence of that, 384 00:27:02.640 --> 00:27:06.039 I want that against observability back. Then if I could go back in time, 385 00:27:06.440 --> 00:27:10.119 right, so I could spend that two hours interrogating the system rather than 386 00:27:10.160 --> 00:27:14.319 waiting for everybody to call with bad information. Right, I mean, I 387 00:27:14.359 --> 00:27:17.680 mean part of me wonders, you know, is in my my time in 388 00:27:17.680 --> 00:27:21.759 that role is like you had to root cause analysis problem. These folks were 389 00:27:21.799 --> 00:27:25.839 fighting the fire and they weren't getting to the summary phase before you needed to 390 00:27:25.880 --> 00:27:30.119 talk to leadership about it, like you need you needed to finish root cause 391 00:27:30.119 --> 00:27:33.160 analysis and you didn't have time or or you know, there needed to be 392 00:27:33.160 --> 00:27:41.519 more automation around it. The number times that the IRC log for strange loops 393 00:27:41.519 --> 00:27:45.559 save my bacon. You know that the you know, the firefight actually happened 394 00:27:45.599 --> 00:27:48.279 over I R C because we were all remote, and that you could you 395 00:27:48.319 --> 00:27:52.240 want to know what happened? Read it. Read how we spent four hours 396 00:27:52.359 --> 00:27:57.440 getting to knowing what went wrong and fifteen minutes fixing it in back the day 397 00:27:57.480 --> 00:28:00.240 that I would the first thing I would do when I wake up every morning 398 00:28:02.160 --> 00:28:06.519 was look at the backscroll in slack one of any incident mansion rooms, and 399 00:28:06.759 --> 00:28:08.839 you would get some you'd get some ideas about it, but it still wasn't 400 00:28:08.880 --> 00:28:12.160 like great. Sometimes you didn't resolve to root cause you just resolved to how 401 00:28:12.200 --> 00:28:15.359 do we fix a thing? You've got to be up again? And then 402 00:28:15.359 --> 00:28:17.680 you went to bed. Yeah, he didn't actually get your root cause. 403 00:28:17.960 --> 00:28:22.359 But and I think these gen AI tools are really good at summarizing those long 404 00:28:22.440 --> 00:28:25.960 logs like that seems to be one of the strengths are at although it would 405 00:28:25.960 --> 00:28:30.960 also argue it's very hard

to test if they're bad at it. Yeah, 406 00:28:30.119 --> 00:28:34.799 Like you ask it to summarize forty pages of notes and it gives you a 407 00:28:34.839 --> 00:28:37.400 paragraph back. How do you know if it's a good paragraph or a bad 408 00:28:37.440 --> 00:28:41.640 paragraph. Hm, good point, right, because you didn't read the forty 409 00:28:41.680 --> 00:28:45.599 pages that was the point. So how would you know if it was bad? 410 00:28:45.640 --> 00:28:48.359 And to me, the biggest battle I have with all of the neuralnet 411 00:28:48.400 --> 00:28:52.799 stuff right now is testability. How do you know it's correct or what to 412 00:28:52.880 --> 00:28:57.000 what degree it is correct? Well, this is called you build a robust 413 00:28:57.039 --> 00:29:03.440 evil set in any of these models that you're doing. Be It and all 414 00:29:03.440 --> 00:29:06.880 the big generative AI companies. I mean there's leader boards like how much they 415 00:29:06.880 --> 00:29:11.599 test work in the email. This gets really tricky and this is a very 416 00:29:11.920 --> 00:29:17.640 new area, right because I think there's maybe a small number of people in 417 00:29:17.640 --> 00:29:21.079 the world right now that are even attempting to do text to SQL or text 418 00:29:21.119 --> 00:29:26.440 to other computer language where you're having to do the same thing summarization, but 419 00:29:26.119 --> 00:29:30.680 instead of doing kind of traditional RAG or having it in the model, you're 420 00:29:30.680 --> 00:29:33.559 pulling new data in every time, right, You're asking what happened last night 421 00:29:34.079 --> 00:29:37.680 from systems of record that would be able to tell you from the telemetry what's 422 00:29:37.680 --> 00:29:42.240 actually happening. And so it's it's a new area. We're we're developing it 423 00:29:42.240 --> 00:29:44.839 as we go, but a lot of what we do, a lot of 424 00:29:44.880 --> 00:29:48.480 our work, and our team is less programming more finding a good evail set 425 00:29:48.480 --> 00:29:52.160 and validating, so we have confidence in the systems that are telling us what 426 00:29:52.200 --> 00:29:55.559 we think is happening. And I think that's one of the things that folks 427 00:29:55.559 --> 00:29:59.720 who really struggling with as from the development perspective, is recognizing how data driven 428 00:29:59.759 --> 00:30:03.119 all this actually is that a good evail set is a set of data you 429 00:30:03.119 --> 00:30:07.559 can test effectively against that without Without that, no mon of code is going 430 00:30:07.559 --> 00:30:10.960 to save you. And gentleman, I need to interrupt for one moment for 431 00:30:11.079 --> 00:30:18.880 this very important message and we're back. It's Dot a Rock So I'm Richard 432 00:30:18.920 --> 00:30:21.519 Campbell. That's Carl Franklin. Hey, hey, hey, talking to our 433 00:30:21.519 --> 00:30:26.359 friend Aaron Erickson, who has been no madding and has found himself landed at 434

00:30:26.640 --> 00:30:30.039 Nvidia, which is very cool, dude, Like, congratulations, what an 435 00:30:30.079 --> 00:30:33.279 amazing company. And it was it was cool before jen a I took off 436 00:30:33.319 --> 00:30:37.240 and you had to run the whole place. But why are you folks building 437 00:30:37.400 --> 00:30:45.079 autonomous agents? Like aren't you happy selling shovels during the gold Rush? Because 438 00:30:45.039 --> 00:30:52.200 because we can no we so we have a we have a need to increase 439 00:30:52.960 --> 00:30:56.359 utilization of our GPU fleet. We have a need to be able to monitor 440 00:30:56.400 --> 00:31:00.920 our own GPU environment. The reason I got so are doing this is because 441 00:31:00.920 --> 00:31:03.799 we had a need in that category and it trued out. You know, 442 00:31:03.799 --> 00:31:07.880 if you work for an AI company and you find AI to be useful in 443 00:31:07.920 --> 00:31:11.400 solving a problem, you should probably do that. And so we are doing 444 00:31:11.440 --> 00:31:15.160 that well. And I mean you had before the break that little lead in. 445 00:31:15.200 --> 00:31:18.200 It's like what if you were what if you were the copilot? Like 446 00:31:18.599 --> 00:31:22.039 to me, the autonomous agent is still the copilot. It's just got a 447 00:31:22.079 --> 00:31:26.960 better instruction set. It's not it's not a gopher, you know. It's 448 00:31:26.000 --> 00:31:29.680 not ask a question, get a response. Ask a question, get a 449 00:31:29.680 --> 00:31:33.319 response. It's more of a here's a problem space that needs to work on, 450 00:31:33.440 --> 00:31:37.599 report back routinely. Like it's almost a if you were equating it to 451 00:31:37.759 --> 00:31:41.759 a junior employee. And I loathe to do that because I do everything I 452 00:31:41.759 --> 00:31:47.680 can to avoid anthro memorphizing software. Now that it's become a real problem, 453 00:31:48.079 --> 00:31:49.960 that's the only way we can think of it, though, Yeah, except 454 00:31:51.000 --> 00:31:53.960 that it makes human wrong right like, it makes us think. It makes 455 00:31:55.039 --> 00:31:59.079 us think that this thing has intent. It gives we as soon as you 456 00:31:59.119 --> 00:32:05.920 give it an agent, then it's then its response has credibility more than it 457 00:32:06.039 --> 00:32:09.240 deserves. Like you. But do you think that us being the copilot means 458 00:32:09.240 --> 00:32:14.640 the AI is driving the train? Well? Is that the implication? That's 459 00:32:14.920 --> 00:32:21.279 the idea? So and you and you assist the AI when it needs help 460 00:32:21.519 --> 00:32:23.880 from somebody when it doesn't have the answer. Is that what you're talking about? 461 00:32:24.480 --> 00:32:30.359 Correct? Correct? So what you do, and I stated this publicly, 462 00:32:30.559 --> 00:32:36.440 you create an org structure of agents. So there is a director agent. 463 00:32:36.559 --> 00:32:39.359 The director agent has a goal. It has a goal of a metric, 464 00:32:39.440 --> 00:32:45.759 almost like giving it an OKR in an organization structure, and its job 465 00:32:45.839 --> 00:32:52.000 is to analyze data. Work with analyst agents that might understand a domain might 466 00:32:52.039 --> 00:32:55.920 be I understand GPS need to run in this temperature range. I understand financial 467 00:32:55.960 --> 00:32:59.880 transactions need to operate in a certain manner. You know, you can apply 468 00:32:59.880 --> 00:33:01.480 to a lot of different domains. And this is a pattern. I'm not 469 00:33:01.480 --> 00:33:07.559 really talking about anything secret here. And then you have worker agents. Their 470 00:33:07.640 --> 00:33:12.079 job is to take human language and convert it into database queries or rest calls 471 00:33:12.200 --> 00:33:16.319 or whatever else, so that confer with themselves. Right, maybe human assistant 472 00:33:16.359 --> 00:33:20.839 it might call you back to say can you clarify this yes or no? 473 00:33:21.039 --> 00:33:23.920 Is this true? Right? It might actually ask you questions and then figure 474 00:33:23.920 --> 00:33:28.480 out an action plan as a result of the analysis. And so the action 475 00:33:28.559 --> 00:33:32.880 plan might be realoicate these GPUs from here to here. It might be hey, 476 00:33:34.119 --> 00:33:37.240 human go take a look at this data. Tell me what you think 477 00:33:37.279 --> 00:33:39.240 of it. Right, there's all sorts of ways you can think about this, 478 00:33:39.400 --> 00:33:44.119 But what it starts to do is once you reach a certain threshold of 479 00:33:44.200 --> 00:33:47.839 capability, it becomes more of its driving things. It's driving the details and 480 00:33:47.880 --> 00:33:52.240 you're kind of monitoring it right with the real data about how is this actually 481 00:33:52.279 --> 00:33:58.759 improving the system? Now you've said like you've given an emission and it's it's 482 00:33:59.079 --> 00:34:01.559 essentially going down the checklist of the right way to execute on that mission and 483 00:34:01.640 --> 00:34:05.960 responding to the impadences that he gets back to it from it. And I 484 00:34:06.000 --> 00:34:07.760 would hope you said monitoring is the key, right. I mean, there's 485 00:34:07.800 --> 00:34:13.039 got to be a CEO that's watching it and not allowing it to make decisions 486 00:34:13.119 --> 00:34:19.800 until they're understood and passed off by human io. So have you been to 487 00:34:19.840 --> 00:34:24.639 San Francisco lately? No, but I suspect you have. So I have 488 00:34:24.719 --> 00:34:29.239 an app on my phone, I can call a way Mo self driving car 489 00:34:29.400 --> 00:34:32.639 with no driver in it. Yeah. Yeah, And occasionally, if it 490 00:34:32.679 --> 00:34:38.159 gets into a jam, it will notify somebody at a real, real human 491 00:34:38.239 --> 00:34:42.800 Hey, I'm in a traffic jam or I'm stuck in some unusual condition and 492 00:34:42.840 -->

00:34:45.840 they take over. Yes, somebody put a traffic cone on my hood. 493 00:34:46.280 --> 00:34:50.760 Yes. Yes. And so if you've seen these videos, I'm sure you 494 00:34:50.800 --> 00:34:52.320 probably have. Not everybody has. I mean, there's people that have never 495 00:34:52.320 --> 00:34:55.320 seen this and haven't heard this is actually happening. But it's this kind of 496 00:34:55.320 --> 00:34:59.239 scaling of self driving cars. Like we started out with cruise control, we 497 00:34:59.400 --> 00:35:04.760 moved to smarter cruise control, we move to kind of full self driving. 498 00:35:04.800 --> 00:35:07.880 Is like the biggest manifestation of this, But there's this kind of face shift 499 00:35:07.880 --> 00:35:13.760 that happens between hey, it helps me till now we're in San Francisco, 500 00:35:13.880 --> 00:35:16.239 it's I help it when it calls me. And it's just kind of an 501 00:35:16.280 --> 00:35:20.639 evolution of how these things happen. And so it's a lot of investment, 502 00:35:20.920 --> 00:35:22.719 a lot of evals, a lot of monitoring on the way there. But 503 00:35:22.760 --> 00:35:28.719 once you get to that threshold where you're monitoring the thing and it's right ninety 504 00:35:28.800 --> 00:35:32.679 nine percent of the time, especially in domains where it being wrong isn't catastrophic. 505 00:35:32.840 --> 00:35:36.679 Like you may not do this with financial transactions tomorrow, but you might 506 00:35:36.760 --> 00:35:42.559 do this with certain other maybe kind of lower risk things, right, And 507 00:35:42.599 --> 00:35:44.920 this is where it starts, and then you start to get more and more 508 00:35:45.039 --> 00:35:50.840 of this stuff. Yeah, I don't know if we're building on a house 509 00:35:50.880 --> 00:35:57.960 of cards here, but I certainly think that there are routine tasks in throughout 510 00:35:58.119 --> 00:36:01.360 business that can be off greated from a set of rules. You know, 511 00:36:01.480 --> 00:36:07.280 we've made fact box long ago, uh Tier one tech support. More and 512 00:36:07.360 --> 00:36:10.519 more automation happening on that because there is this set of did you turn it 513 00:36:10.559 --> 00:36:13.599 off and on again? Did you check this? That? You know, 514 00:36:13.800 --> 00:36:17.119 like it's a checklist of things as soon as you've got a person flipping through 515 00:36:17.159 --> 00:36:22.519 a book, you know, going back and forth on the standard set of 516 00:36:22.599 --> 00:36:25.079 questions like that could be software. It just seems like layers of automation. 517 00:36:25.559 --> 00:36:31.840 I've just always concerns me when when we get a sense of consdering more intent 518 00:36:32.039 --> 00:36:37.639 than that that it is still a set of instructions, although often those sets 519 00:36:37.639 --> 00:36:44.960 of instructions are based on a parametric response to data coming in. You know, 520 00:36:45.000 --> 00:36:46.960 there's a reason why we turn the we turn the lights red when it's 521 00:36:47.000 --> 00:36:52.440 past a certain threshold, right, right, But there's so there's a good 522 00:36:52.519 --> 00:36:55.079 job or a book called forgive Me. I'm going to quote a book name. 523 00:36:55.639 --> 00:37:00.519 It's not a curse word, but it's called bulkt jobs of the book. 524 00:37:01.199 --> 00:37:07.360 And it turns out there's a lot of roles that I think maybe aren't 525 00:37:07.360 --> 00:37:13.559 that intellectually challenging, but for whatever reason, we still keep around where you 526 00:37:13.559 --> 00:37:17.960 know, I call these kind of like low level intellectual robots. And if 527 00:37:19.000 --> 00:37:22.199 you can kind of automate a lot of the low level intellectual roboticness out of 528 00:37:22.280 --> 00:37:24.679 jobs, and you think of like what happens at the DMV, I don't 529 00:37:24.719 --> 00:37:28.119 know that there's a lot of creativity going on at the DMV. And I 530 00:37:28.119 --> 00:37:30.440 don't know that you want a lot of creativity going out on the DMV. 531 00:37:30.920 --> 00:37:32.599 I don't know that you want a lot of creativity in your finance department. 532 00:37:32.639 --> 00:37:37.519 In some domains, there's some kinds of software development that aren't terribly creative. 533 00:37:37.559 --> 00:37:42.400 I don't ever want to write another get rest point, get you know, 534 00:37:42.480 --> 00:37:45.880 rest endpoint again that gets data from a simple thing. I know that can 535 00:37:45.920 --> 00:37:49.679 be automated for the most part, So we use copilot for that today. 536 00:37:49.760 --> 00:37:52.679 Right, We're going to find more and more things that we just use copilot 537 00:37:52.679 --> 00:37:55.400 for a lot. That's that's low level of intellectual labor, right, and 538 00:37:55.440 --> 00:38:00.320 we will start t all to make that more well and and well. Every 539 00:38:00.320 --> 00:38:05.000 time we've ever done this, it frees up resources for doing more creative work. 540 00:38:05.599 --> 00:38:09.920 Right. The reality, a lot of travel agents lost their jobs as 541 00:38:09.960 --> 00:38:17.559 the airline industry and the hospitality industry moved online, but travel exploded far more 542 00:38:17.599 --> 00:38:22.679 people traveled, all those industries got bigger, and a net more jobs were 543 00:38:22.719 --> 00:38:27.719 made than were lost from travel agents. You know that that pattern consists. 544 00:38:27.920 --> 00:38:31.599 It happens over and over again. So the theory being these you know, 545 00:38:31.760 --> 00:38:37.199 menial jobs, these menial tasks that can be automated, should be automated because 546 00:38:37.239 --> 00:38:39.519 that person is capable of more. Yeah, it's the only way you can 547 00:38:39.559 --> 00:38:43.400 look at it, because you can be mad that this is happening, and 548 00:38:43.920 --> 00:38:47.880 lots of people are and I have a lot of you know, but reality, 549

00:38:47.960 --> 00:38:52.719 it's just how we manage disruption is the important of this, right, 550 00:38:52.840 --> 00:38:57.480 Like, yeah, we should be able to manage this disruption compassionately. But 551 00:38:58.159 --> 00:39:00.800 it doesn't mean, you know, the alter of his not having disruption, 552 00:39:00.920 --> 00:39:06.400 isn't there. I referenced the the Luddite movement. Yeah, you know, 553 00:39:06.679 --> 00:39:09.039 and for those who don't mean we've heard the term. But originally, when 554 00:39:09.079 --> 00:39:17.039 the first automated the first steam powered looms were deployed in Scotland, the folks 555 00:39:17.039 --> 00:39:22.880 that were used to weaving by hand freaked out and smashed them. And here's 556 00:39:22.920 --> 00:39:27.440 something that people don't know about that movement. But the Lunites weren't anti technology. 557 00:39:27.960 --> 00:39:30.320 In fact, they were the technologists of the day, right. They 558 00:39:30.320 --> 00:39:34.719 were the high tech. They were using high tech machinery. They understood it, 559 00:39:35.440 --> 00:39:39.239 but it was it was really about you know, jobs, right, 560 00:39:39.360 --> 00:39:45.559 yeah, And the reality, of course was that was a communication between the 561 00:39:45.559 --> 00:39:50.840 employer and employees. The result was that ultimately rebuilt the machines and had all 562 00:39:50.880 --> 00:39:54.039 those people running them. Right. The bigger thing that happened as they automated 563 00:39:54.079 --> 00:40:00.239 the weaving of cloth is that people bought more clothes because the price of them 564 00:40:00.280 --> 00:40:04.480 fell dramatically. And not only that all of those people keep their jobs, 565 00:40:04.519 --> 00:40:07.119 but a whole lot of more people got imployed because they made a lot more 566 00:40:07.199 --> 00:40:13.800 cloth. And you know, the industry exploded. The idea of owning different 567 00:40:13.800 --> 00:40:15.800 garments for every day, not just a good set of clothes and a working 568 00:40:15.800 --> 00:40:20.519 set of clothes, came out of being able to automate cloth weaving. So 569 00:40:20.639 --> 00:40:22.519 you want to know one of the most ironic things about what's happening right now, 570 00:40:22.599 --> 00:40:32.559 I'm sure you do, and that is so now engineers are automating engineering. 571 00:40:32.599 --> 00:40:37.079 And you see this with some of the responses that Devin the software developer. 572 00:40:37.079 --> 00:40:44.440 I found that fascinating because who here in software development hasn't automated the living 573 00:40:44.480 --> 00:40:46.639 crap out of somebody else's jobs, a lot of other people's jobs. We've 574 00:40:46.679 --> 00:40:52.679 been doing that for the entire what do you think CICD pipelines are, right, 575 00:40:52.400 --> 00:40:55.920 We've been doing this for the entire history of computing. Okay, normally 576 00:40:55.960 --> 00:41:00.320 there is nobody. I'm sorry this might make some people mad. I'm getting 577 00:41:00.320 --> 00:41:04.360 older, so I kind of don't care, as we all are, right, 578 00:41:04.719 --> 00:41:07.880 we all age one. But the reality is if you're mad that, 579 00:41:08.000 --> 00:41:13.280 like programmers are automating programmers jobs, well, sorry, I guess that's just 580 00:41:13.400 --> 00:41:15.079 kind of like what we've been doing, and I guess it's coming back and 581 00:41:15.079 --> 00:41:17.719 it's kind of funny. But like when people get mad at this, like 582 00:41:17.719 --> 00:41:21.400 they're like, oh my god, Devin, it's like, yeah, sorry, 583 00:41:21.840 --> 00:41:24.320 we did it to ourselves. By the way. By the way, 584 00:41:24.679 --> 00:41:30.679 you have so much opportunity now if you can learn how to construct these systems 585 00:41:30.719 --> 00:41:34.239 together, use this. So you know, we built this thing that in 586 00:41:34.320 --> 00:41:37.880 video inference microservice just a pattern just being video, it could be an inference 587 00:41:37.960 --> 00:41:40.840 microservice. So now instead of building a micro service that knows one domain well 588 00:41:40.880 --> 00:41:45.800 and you build it in code, you build ALM that knows one domain well, 589 00:41:45.119 --> 00:41:52.000 that takes in requests in human and generates data for you in whatever form 590 00:41:52.039 --> 00:41:54.199 that you need. But it's the same idea, right, it's taking some 591 00:41:54.239 --> 00:41:59.440 of these same archeittral patterns that we've known as engineers and using it to build 592 00:41:59.480 --> 00:42:04.039 bigger, smarter, more interesting systems that then let us be more creative. 593 00:42:04.239 --> 00:42:07.639 Yeah, why did you think that the methods for using machine learning tools weren't 594 00:42:07.679 --> 00:42:12.440 going to improve? Right? Really, come on, like, that's of 595 00:42:12.480 --> 00:42:16.440 course they are the Especially in a young in technology like this, it evolves 596 00:42:16.800 --> 00:42:21.880 even faster than the more mature technologies, where a lot of the stuff's one 597 00:42:21.920 --> 00:42:24.679 point zero and just becoming to two is a big jump. Maybe three is 598 00:42:24.840 --> 00:42:29.000 kind of where we're talking about these first few versions of how we build this 599 00:42:29.039 --> 00:42:31.960 stuff. Of course is going to be massively transforming. Well, but it's 600 00:42:32.280 --> 00:42:35.920 funny the way it's happened in AI, where it's kind of like it's very 601 00:42:35.960 --> 00:42:37.840 quiet for periods of time, right, you know, you have researchers doing 602 00:42:37.880 --> 00:42:40.440 things. It kind of feels like it's it's quiet, it's quiet, and 603 00:42:40.480 --> 00:42:46.039 then boom, something like tchike happens and then massive, massive change. I 604 00:42:46.039 --> 00:42:49.159 think ken Beck had a quote saying, I don't know how this is going 605 00:42:49.239 -->

00:42:51.639 to change, but I know it's going to change a lot about Yeah, 606 00:42:51.679 --> 00:42:55.320 a lot of things how I write code, and I think the best response 607 00:42:55.360 --> 00:42:59.599 to a lot of this is wonder yeah, you know, and and just 608 00:42:59.679 --> 00:43:02.320 kind of being even as you get into your later years in your career, 609 00:43:02.400 --> 00:43:05.400 you know, I'm kind of rounding third base in my career, I think 610 00:43:05.400 --> 00:43:07.760 at this point, and I still every day I wake up, I'm like, 611 00:43:07.800 --> 00:43:09.880 what can we do with this stuff? How can we build on these 612 00:43:09.920 --> 00:43:14.960 with these tools that we have to do things that are incredible for our customers, 613 00:43:14.960 --> 00:43:19.280 that are incredible for new Like I look at what Nvidiot does. I'm 614 00:43:19.280 --> 00:43:22.000 sorry if I'm if I sound like a fanboy, I am, you know, 615 00:43:22.320 --> 00:43:24.760 guilty. So it's a lot. But things we're doing around protein discovery, 616 00:43:25.079 --> 00:43:29.199 things that our customers are doing around protein discovery, that we're never even 617 00:43:29.280 --> 00:43:34.599 possible pre transformers, pre the transformer model. There's going to be diseases, 618 00:43:34.639 --> 00:43:39.719 care there's going to be technologies built that make humanity better in every way possible. 619 00:43:39.760 --> 00:43:45.480 There's also going to be, let's be real technologies that make things terrible 620 00:43:45.599 --> 00:43:47.679 in a lot of ways. And you know, I don't. The thing 621 00:43:47.719 --> 00:43:51.400 is, you can't stop it though. You know, there's no world where 622 00:43:51.519 --> 00:43:53.440 you can suddenly, oh, GPU's got to throw them away like they're here. 623 00:43:53.559 --> 00:43:57.440 And even even if technology doesn't move at all from what we have now, 624 00:43:57.719 --> 00:44:00.679 so only started to use these tools in various ways, and you open 625 00:44:00.719 --> 00:44:04.239 the door to philosophy. So let's go here. Yeah. One of the 626 00:44:04.360 --> 00:44:07.039 questions that is always on my mind is, Okay, there are a lot 627 00:44:07.079 --> 00:44:13.800 of skilled labors in the workforce who can make the adjustment. There are way 628 00:44:13.840 --> 00:44:19.000 more unskilled laborers that just need a job, and so how do you see 629 00:44:19.000 --> 00:44:22.719 the future playing out for them? I wish I knew the answer that question, 630 00:44:22.800 --> 00:44:25.719 but I don't know the answer that question. If I'm honest, I 631 00:44:25.719 --> 00:44:30.880 can speculate. I can speculate that I think you know. The answer is 632 00:44:30.920 --> 00:44:35.239 sometimes everybody's going to talk about UBI. That sounds kind of terrible. Universal 633 00:44:35.440 --> 00:44:37.440 even if you had a good UBI. Yes, universal basic can come like 634 00:44:37.440 --> 00:44:40.599 Okay, maybe they don't do anything. I think we have to make the 635 00:44:40.599 --> 00:44:45.280 world more prosperous so that we can give people the opportunities to be creative. 636 00:44:45.599 --> 00:44:47.719 Yeah, would be the way. If I could be the dictator of the 637 00:44:47.760 --> 00:44:52.119 world, right, how I would arrange that. Yeah, that's the goal. 638 00:44:52.760 --> 00:44:54.239 Yeah, for Aaron to be the dictator of the world. Sure, 639 00:44:54.360 --> 00:44:58.000 yes, that's exactly what I meant. Richard, Thank you. I have 640 00:44:58.039 --> 00:45:00.599 a better idea. Why don't we just have a super smart AI be the 641 00:45:00.679 --> 00:45:04.480 president. I know it sounds a little ridiculous, but you know, so 642 00:45:04.559 --> 00:45:13.400 are the presidents regardless of you're more ridiculous. Yeah, ridiculous. As a 643 00:45:13.440 --> 00:45:19.440 service, we admittedly are at a dark phase in some of this stuff, 644 00:45:19.480 --> 00:45:22.719 but I'm pretty sure I wish to will want people to be the compassionate ones 645 00:45:23.239 --> 00:45:29.039 that are helping set our values and are setting the rules for the software to 646 00:45:29.480 --> 00:45:37.639 do their things. Yeah, I know I'm such a party pooper. Fine, 647 00:45:38.960 --> 00:45:44.639 I can't have president GPT. I mean, you know, I would 648 00:45:44.679 --> 00:45:49.480 love I would love somebody to do an experiment at least elect g In fact, 649 00:45:49.480 --> 00:45:52.239 that was one of the first things I asked when auto GPT came out, 650 00:45:52.239 --> 00:45:54.360 which is like one of the first agent systems that somebody built. Almost 651 00:45:54.400 --> 00:45:58.920 immediately when CP four was live, They're like, what if we put this 652 00:45:58.920 --> 00:46:00.800 in a loop and what if we So I asked it, how do you 653 00:46:00.800 --> 00:46:06.119 elect yourself mayor of San Francisco? Because you know, it just kind of 654 00:46:06.159 --> 00:46:09.039 sucks here in that manner, and it immediately knew, oh, I'm going 655 00:46:09.119 --> 00:46:13.840 to have to get a straw candidate because AIS can't get elected. But if 656 00:46:13.880 --> 00:46:17.199 I can control a candidate and they just trust me with the decisions, I 657 00:46:17.199 --> 00:46:21.760 don't know, it still sounds ystopian. I mean maybe we should stop watching 658 00:46:22.519 --> 00:46:25.199 and being like, I wonder this would be great products. Yeah, here's 659 00:46:25.239 --> 00:46:29.599 the real issue, Eron, is that we decided to attach the term artificial 660 00:46:29.639 --> 00:46:35.960 intelligence to this technology. Yeah, because that has had seventy years of science 661 00:46:36.000 --> 00:46:38.760 fiction, you know, perverting working against it. You know, the first 662 00:46:38.760 --> 00:46:43.280 time the public hears the phrase artificial intelligence is in two thousand and one a 663 00:46:43.360 --> 00:46:47.159 Space Odyssey and Hal tries to kill

everybody, like we have set a path, 664 00:46:47.199 --> 00:46:52.239 and now you want to hang that moniker on this software. Yeah, 665 00:46:52.360 --> 00:46:57.679 like it was. It's a mistake because it's hard enough to introduce new automation 666 00:46:57.840 --> 00:47:04.719 into society, but to introduce new automation to society with decades of telling me 667 00:47:04.800 --> 00:47:07.039 it's going to kill everybody, like, it's kind of a dumb thing to 668 00:47:07.119 --> 00:47:10.000 have done. I'm going to take it up with the branding department of AI 669 00:47:10.159 --> 00:47:15.239 and see if we can come over with a better moniker. Good luck. 670 00:47:15.320 --> 00:47:17.519 Unfortunately, it's probably stuck. I think. I think once it's like out 671 00:47:17.559 --> 00:47:21.079 there, it's kind of just gonna be a thing. But Genie's out of 672 00:47:21.119 --> 00:47:24.000 the bottom. It's it's a little late now, right. And the joke, 673 00:47:24.079 --> 00:47:29.199 of course, is that Minski came up with the term to convince the 674 00:47:29.320 --> 00:47:34.679 US Army to give him money. It has always been a marketing term to 675 00:47:34.920 --> 00:47:37.760 raise money. That's the sort of reality of it. Now, every so 676 00:47:37.840 --> 00:47:42.719 often you build some software and it's happened several times. You can call a 677 00:47:42.840 --> 00:47:46.159 Mai Winters if you want, but the iteration is the same. Science works 678 00:47:46.199 --> 00:47:49.880 on the problem for a while until it gets to a certain level where the 679 00:47:49.960 --> 00:47:53.639 engineers take over and apply it, and once the applications are done, it's 680 00:47:53.679 --> 00:47:58.119 just software, right, right, are they? The joke term I've used 681 00:47:58.119 --> 00:48:00.800 for this has been artificial intelligence is what you call a piece of technology when 682 00:48:00.840 --> 00:48:05.760 it doesn't work, right, Because as soon as it does work, it 683 00:48:05.920 --> 00:48:12.519 gets a new name. It becomes planning and optimization software, it becomes vision 684 00:48:12.559 --> 00:48:16.000 image recognition, it becomes large language model. Right as long as there you 685 00:48:16.000 --> 00:48:20.159 know, I do a lot of investing. If you call it AI technology, 686 00:48:20.199 --> 00:48:22.599 you just told me you have stuff that doesn't work. When you call 687 00:48:22.639 --> 00:48:28.280 it something else, then okay, let's take a look at it. That's 688 00:48:28.320 --> 00:48:30.639 a great point because I think a lot of people think they just got to 689 00:48:30.679 --> 00:48:32.400 slap an AI, you know, peanut butter in my product and it gets 690 00:48:32.440 --> 00:48:36.119 better. Well, I have a failed startup to say that does not, 691 00:48:36.639 --> 00:48:39.480 yeah, work that way, And I think the best products they use AI 692 00:48:39.599 --> 00:48:43.559 don't mention AI at all. They just solve a problem. No, because 693 00:48:44.239 --> 00:48:45.760 I think it's plumbing. I think it's just going to be part of the 694 00:48:45.800 --> 00:48:52.199 software. The ones that do It's questionable whether what we think of as software 695 00:48:52.199 --> 00:48:57.239 developers is AI, where's involved at all? It's just algorithmic programming, right. 696 00:48:57.320 --> 00:49:00.079 I mean there's some they all want to slap the AI monicor on it 697 00:49:00.119 --> 00:49:06.480 when there may may not be any you know, any machine learning, any 698 00:49:07.559 --> 00:49:08.960 that kind of thing. It could be a series of statements. But listen, 699 00:49:09.320 --> 00:49:15.840 without a doubt, there is a software technology using probabilistic language analytics that 700 00:49:15.039 --> 00:49:22.320 is getting effective results. And it's and that's it right Like now, there's 701 00:49:22.480 --> 00:49:27.840 an interesting area to think about about how we revise UX's and how we can 702 00:49:27.920 --> 00:49:32.960 put software in more places like this feeds into the other stories of ubiquitous computing, 703 00:49:34.599 --> 00:49:37.159 and you know the fact that we're making them in a small form factic 704 00:49:37.199 --> 00:49:38.119 we want to put on our shirt and so forth. It speaks to you, 705 00:49:38.440 --> 00:49:42.880 I want an even more personal compute device than the slab of black glass 706 00:49:42.880 --> 00:49:45.920 that Apple gave me. You know, we are ripe for change in this 707 00:49:45.960 --> 00:49:52.800 particular space. But I think part of the problem we're having right now is 708 00:49:53.000 --> 00:50:01.880 this Ultron versus Jarvis kind of science fiction crap that's infected our thinking around an 709 00:50:01.960 --> 00:50:07.519 interesting piece of software that we can do some good with well, and the 710 00:50:07.559 --> 00:50:09.400 thing that cheching be tenailed wasn't the fact that it's AI. Yeah, it's 711 00:50:09.400 --> 00:50:14.960 great. What chech tenailed was what people want is I want to ask questions 712 00:50:14.960 --> 00:50:20.159 to a box and get moderately or ideally perfectly accurate answers to my questions, 713 00:50:20.679 --> 00:50:23.119 regardless of what they are. I debate the irebade the accurate part. They 714 00:50:23.119 --> 00:50:30.599 wanted answers. They wanted responses that we're spoken confidently right, like we're and 715 00:50:30.639 --> 00:50:34.480 apparently many people are happy with the answers even if they are not right. 716 00:50:34.679 --> 00:50:38.920 That was sufficient. Now I when I'm doing these AI talks, I bring 717 00:50:39.000 --> 00:50:44.320 up Alpha fold two because I think it's one of the most important pieces of 718 00:50:44.519 --> 00:50:47.000 machine learning software that almost nobody knows about. Is this the Wolfram thing? 719 00:50:47.400 --> 00:50:51.360 No, this is al fold exactly comes from deep mind. These are the 720 00:50:51.360 --> 00:50:55.480 guys

who solved go okay, but then they reapplied the technology to protein folding. 721 00:50:57.000 --> 00:50:59.679 And again it's like, this is pretty esoteric stuff. But there's a 722 00:50:59.679 --> 00:51:04.039 few things said. I find it important and only a it's doing protein. 723 00:51:04.159 --> 00:51:08.599 It's computing protein fold results that are taking years to even validate, Like that's 724 00:51:08.800 --> 00:51:12.400 so we're not even sure if it's all right. But everything that we've been 725 00:51:12.440 --> 00:51:15.599 able to prove so far, it's getting far better results than anything else we've 726 00:51:15.639 --> 00:51:20.760 ever done. But also like the competition, the CASP competition, which is 727 00:51:20.880 --> 00:51:24.480 very much like the ImageNet competition that kicked off vision recognition in the twenty tens, 728 00:51:25.239 --> 00:51:30.519 The CASP competitions were going on for a few years to what four years 729 00:51:30.559 --> 00:51:32.960 ago was the first time alpha fole jumped in on it, and they open 730 00:51:34.039 --> 00:51:37.360 source the whole frickin' thing. The next two year one it was like half 731 00:51:37.440 --> 00:51:42.480 alpha fol, but this year's a CASP event. It's all alpha fol. 732 00:51:43.239 --> 00:51:46.880 Everybody's using it, Like every scientist looked at this and said this is a 733 00:51:46.880 --> 00:51:51.280 better way to work on protein problems, and they've all adopted it. Like 734 00:51:52.440 --> 00:51:55.920 this is where we're seeing automation and new technology at its best. There's nobody 735 00:51:57.000 --> 00:52:00.719 trying to make up billion dollars off of this yet. Right when products start 736 00:52:00.760 --> 00:52:04.000 to appear from it, then things might get ugly. But right now, 737 00:52:04.079 --> 00:52:10.960 in the research phase, everything has been shared that this non deterministic algorithm for 738 00:52:12.119 --> 00:52:19.880 computing behavior at the very low level of biology is open to every scientist who's 739 00:52:19.920 --> 00:52:22.840 interested in this space, and they're all testing each other, they're all competing 740 00:52:22.880 --> 00:52:28.440 with each other to make it better. We went from media remember what you 741 00:52:28.440 --> 00:52:30.440 know it was the XKCD where it's like identify a burden this picture. It's 742 00:52:30.440 --> 00:52:34.000 like, I need a huge research budget too. Now it's a trivial thing 743 00:52:34.119 --> 00:52:37.840 you can do on your phone. We may be up against that for understanding 744 00:52:37.880 --> 00:52:43.679 the fundamentals of biology and complex biological organisms with alpha fold Like, it's so 745 00:52:43.840 --> 00:52:46.760 profound and it's not being I love that it's not being celebrated in some respects 746 00:52:47.519 --> 00:52:52.880 because it's people too busy working. You know, this current detonation around LMS 747 00:52:53.119 --> 00:52:58.360 is about the scientists moving on right. The Hinton's Suskevards of the world are 748 00:52:58.440 --> 00:53:01.119 largely saying, hey, jen Ai has gone as far as it go and 749 00:53:01.159 --> 00:53:05.760 that's like, Okay, we figured out a concrete Now you engineers can play 750 00:53:05.920 --> 00:53:09.159 with it, and we as engineers now get to change the world with it. 751 00:53:09.280 --> 00:53:13.159 Sam Malton in a recent talk I literally just listened to this morning, 752 00:53:13.239 --> 00:53:15.119 but it was a really good talk. I think he was an MIT or 753 00:53:15.119 --> 00:53:19.760 something. He talks about how great we're going to get GP five. But 754 00:53:19.800 --> 00:53:22.840 what we want to be able to distill is the ability to figure out the 755 00:53:22.840 --> 00:53:29.039 reasoning part separate from the architecture that does the kind of like large language more 756 00:53:29.159 --> 00:53:31.440 data. Somehow, somehow we knew we added more data and we got something 757 00:53:31.440 --> 00:53:37.239 that looks like reasoning, and we're not entirely sure it's real reasoning or what 758 00:53:37.400 --> 00:53:40.599 it really is. But if we start to understand the architecture of reasoning in 759 00:53:40.639 --> 00:53:45.840 a much more nuanced way, I think that's where the next big breakthrough is 760 00:53:45.840 --> 00:53:47.519 going to come from. At least that's based on IPI today. I mean, 761 00:53:47.519 --> 00:53:52.039 it can change tomorrow with whatever papers. I mean, but once we 762 00:53:52.079 --> 00:53:55.440 get better kind of synthetic reasoning and we understand that architecture, that's going to 763 00:53:55.440 --> 00:54:00.599 be probably the next I certainly look at the smell perspective and say, are 764 00:54:00.639 --> 00:54:02.760 we still just making it bigger so because we don't actually understand it, or 765 00:54:02.840 --> 00:54:07.960 I'm hoping that jat that the GPT five is much more reasoned, that it's 766 00:54:07.280 --> 00:54:10.599 we're not automatically bigger, that we're more focused in areas that we're starting to 767 00:54:10.639 --> 00:54:17.119 do more discrimination, you know, effectively understanding testing your language in deeper layers 768 00:54:17.119 --> 00:54:22.159 before you present it. I don't know if we're there yet. There is 769 00:54:22.199 --> 00:54:24.840 a set of code smells that you see in every rapidly evolving piece of technology, 770 00:54:24.880 --> 00:54:29.079 and one of them is right sizing, and you are starting to see 771 00:54:29.480 --> 00:54:32.679 movement towards right sizing. But what you're describing then is, rather than wait 772 00:54:32.719 --> 00:54:39.320 for the next research breakthrough from the scientists, that engineers iteratee and find incremental 773 00:54:39.360 --> 00:54:45.360 improvements in iteration that occasionally can lead to another breakthrough. Well, and a 774 00:54:45.360 --> 00:54:49.400 lot of this. You talked about

anthromorphizing earlier in the episode, and I 775 00:54:49.480 --> 00:54:54.280 want to call out there's a great researcher Ethan Mallock, I think professor Wertn 776 00:54:54.360 --> 00:54:58.280 or something like that, and he wrote a great book called co Intelligence, 777 00:54:58.320 --> 00:55:02.199 and he talks about actually the virtuousness of using a limited form of anthromopore that 778 00:55:02.440 --> 00:55:08.119 I can't even say the word right, but talking in human and optimizing your 779 00:55:08.119 --> 00:55:12.639 prompts that way. So we saw that even just asking it to think step 780 00:55:12.639 --> 00:55:15.440 by step, if you do a chech fee question and say and think step 781 00:55:15.440 --> 00:55:19.559 by step, people have demonstrated to get better results. What I really look 782 00:55:19.599 --> 00:55:24.800 forward to is the application of industrial organizational psychology to using LMS and seeing a 783 00:55:24.920 --> 00:55:30.679 different techniques such as how people talk to each other, how organizations work, 784 00:55:30.719 --> 00:55:35.880 and do organizations and lms, when you have multiple agents working with each other, 785 00:55:36.440 --> 00:55:39.280 share some of those same properties, which on a conjecture basis, we 786 00:55:39.280 --> 00:55:45.039 haven't proven this, but hypothetically they should because lms are trained on people, 787 00:55:45.079 --> 00:55:49.400 and they're also trained on people's interactions, and that's the important part. It's 788 00:55:49.400 --> 00:55:53.559 not that's speaking to it more like a human makes it more human. It 789 00:55:53.599 --> 00:55:58.800 was trained on human data. Yeah, so obviously it's going to work more 790 00:55:58.800 --> 00:56:02.119 effectively if you work with the data set as it is. And again I'm 791 00:56:02.159 --> 00:56:08.599 always balancing this what came from us, and it is the reason that things 792 00:56:08.599 --> 00:56:14.320 work the way they do, versus imbibing any of that capability on the software 793 00:56:14.360 --> 00:56:17.000 itself. It's software. It's a reflection, which is why you always think. 794 00:56:17.320 --> 00:56:21.880 This is why you always think chet GPT and ask talk to it nicely, 795 00:56:22.039 --> 00:56:24.960 because studies are shown when you ask somebody nicely for something, you get 796 00:56:25.000 --> 00:56:30.000 a better result than being a jerk to it. And there's the most anthropomorphy. 797 00:56:30.119 --> 00:56:36.039 Does this mean my wife should stop telling sch Alexa to shut up instead 798 00:56:36.039 --> 00:56:39.239 of saying stop she does that? And I don't think there's an LM and 799 00:56:39.320 --> 00:56:44.719 Alexa yet, So you don't have nothing to worry about I forgot what Alexa 800 00:56:44.800 --> 00:56:46.360 was. I thought you were actually talking about a child. Okay, now 801 00:56:46.440 --> 00:56:50.599 no Alexa, Yeah, no, no, no, I have I'm starting 802 00:56:50.639 --> 00:56:52.960 a new meme where if you put in front of anything that you don't like, 803 00:56:53.239 --> 00:56:58.760 that's a diminutive. So you know this coffee, Oh my god. 804 00:56:58.840 --> 00:57:01.440 Well, plus insaying the app word because folks are sometimes playing this in open 805 00:57:01.480 --> 00:57:06.159 air shire's their devices, which makes them sad. And that's been a running 806 00:57:06.199 --> 00:57:08.559 joke in dotam Rocks too. So you got a lot of mileoget of that. 807 00:57:08.760 --> 00:57:12.599 Sounds like you're working on some amazing stuff, Aeron. I can't wait 808 00:57:12.639 --> 00:57:15.440 to see some of it in the field. Yeah, like the emergence of 809 00:57:15.519 --> 00:57:20.679 autonomous agents. I hear the noise in a lot of different places. Not 810 00:57:20.760 --> 00:57:23.239 the product yet, but obviously it's a goal. And I would categorize the 811 00:57:23.320 --> 00:57:25.719 show as a geek out. I mean, we really went a lot of 812 00:57:25.719 --> 00:57:30.480 places that wasn't real, you know, practical day to day programming, and 813 00:57:31.000 --> 00:57:35.199 these are great shows to do from time to time. So thanks Erin. 814 00:57:35.519 --> 00:57:38.159 I'm so happy to be here. It's been fifteen years. I feel like 815 00:57:38.199 --> 00:57:42.480 maybe I've got a few more wrinkles. Yeah, maybe, y'all do too, 816 00:57:42.639 --> 00:57:45.400 but I feel like maybe a little bit more Risdom as well, or 817 00:57:45.440 --> 00:57:49.719 maybe not, but you know, I'm excited to see y'all's It's great to 818 00:57:49.719 --> 00:57:52.639 be on the show. And good luck in the future and that means everybody, 819 00:57:53.000 --> 00:57:57.400 and thank you for listening to dot Rocks and we'll see it next time. 820 00:57:58.039 --> 00:58:22.440 But dot net Rocks is brought to you by Franklin's Net and produced by 821 00:58:22.519 --> 00:58:28.480 Pop Studios, a full service audio, video and post production facility located physically 822 00:58:28.519 --> 00:58:32.440 in New London, Connecticut, and of course in the cloud online at pwop 823 00:58:32.719 --> 00:58:37.440 dot com. Visit our website at d O T N E, t R 824 00:58:37.480 --> 00:58:42.280 O c k S dot com for RSS feeds, downloads, mobile apps, 825 00:58:42.440 --> 00:58:45.719 comments, and access to the full archives going back to show number one, 826 00:58:45.920 --> 00:58:50.760 recorded in September two thousand and two. And make sure you check out our 827 00:58:50.760 --> 00:58:54.039 sponsors. They keep us in business. Now go write some code. See 828 00:58:54.079 --> 00:59:05.280 you next time. You got Jvan in summer time that means home. Then 829 00:59:05.360 --> 00:59:07.840 my Texas a lie present Hall