1 00:00:00,160 --> 00:00:04,320 SPEAKER_02: Once Chat GPT kind of hit the market late 2022, uh 2 00:00:04,320 --> 00:00:07,919 I was curious with a simple question, which is can ChatGPT 3 00:00:07,919 --> 00:00:09,199 beat the SP 500? 4 00:00:09,519 --> 00:00:12,560 SPEAKER_03: You need to have someone in leadership that's AI 5 00:00:12,640 --> 00:00:13,199 pilled. 6 00:00:13,519 --> 00:00:17,519 SPEAKER_01: Only 17% of stocks outperform the SP 500 over a 7 00:00:17,519 --> 00:00:19,199 rolling 10-year period. 8 00:00:19,280 --> 00:00:23,600 So probistically, that stock you pick at random is going to be a 9 00:00:23,600 --> 00:00:26,000 stock that underperforms the SP 500. 10 00:00:26,079 --> 00:00:28,800 SPEAKER_02: How do you look at the entire corpus of every 11 00:00:28,800 --> 00:00:30,719 portfolio you've ever generated? 12 00:00:36,560 --> 00:00:38,960 SPEAKER_00: This conversation is provided for informational and 13 00:00:38,960 --> 00:00:40,320 educational purposes only. 14 00:00:40,479 --> 00:00:43,280 The views expressed are those of the individual speakers as of 15 00:00:43,280 --> 00:00:45,520 the recording date, may differ from the views of their 16 00:00:45,520 --> 00:00:47,600 organizations, and may change without notice. 17 00:00:47,840 --> 00:00:50,479 Nothing discussed constitutes investment, legal, accounting, 18 00:00:50,640 --> 00:00:53,759 or tax advice, an offer or solicitation, or recommendation 19 00:00:53,840 --> 00:00:56,479 or endorsement of any security, investment strategy, or 20 00:00:56,479 --> 00:00:57,280 financial product. 21 00:00:57,520 --> 00:01:00,320 Access to this conversation does not, by itself, create an 22 00:01:00,320 --> 00:01:02,640 advisory, fiduciary, or client relationship. 23 00:01:02,799 --> 00:01:05,120 Any companies, securities, or strategies discussed are 24 00:01:05,120 --> 00:01:06,879 presented solely for illustrative purposes. 25 00:01:07,040 --> 00:01:09,840 The speakers or their affiliates may hold positions in, provide 26 00:01:09,840 --> 00:01:12,159 services to, or otherwise have financial interests in the 27 00:01:12,159 --> 00:01:14,239 companies or securities mentioned, and those interests 28 00:01:14,239 --> 00:01:15,200 may change without notice. 29 00:01:15,359 --> 00:01:17,680 Any forecasts, targets, forward-looking statements, 30 00:01:17,840 --> 00:01:20,640 model outputs, or hypothetical results are based on assumptions 31 00:01:20,640 --> 00:01:22,560 and information available as of the recording date. 32 00:01:22,640 --> 00:01:25,280 They involve inherent risks and uncertainties, may not reflect 33 00:01:25,280 --> 00:01:27,280 actual trading or investment results, and should not be 34 00:01:27,280 --> 00:01:29,280 relied upon as guarantees of future outcomes. 35 00:01:29,519 --> 00:01:31,760 Information obtained from third-party sources is believed 36 00:01:31,760 --> 00:01:34,239 to be reliable, but its accuracy, completeness, and 37 00:01:34,239 --> 00:01:35,840 timeliness are not guaranteed. 38 00:01:36,159 --> 00:01:38,879 All investing involves risk, including volatility and the 39 00:01:38,879 --> 00:01:39,920 possible loss of principle. 40 00:01:40,159 --> 00:01:42,159 Past performance is not indicative of future results, 41 00:01:42,239 --> 00:01:44,640 and no investment strategy can assure a profit or protect 42 00:01:44,640 --> 00:01:45,200 against loss. 43 00:01:45,359 --> 00:01:47,519 Listeners should conduct their own due diligence and consult 44 00:01:47,519 --> 00:01:49,840 their own qualified financial, legal, accounting, and tax 45 00:01:49,840 --> 00:01:51,680 advisors before making any investment decision. 46 00:01:51,920 --> 00:01:54,400 Artificial intelligence systems and other analytical models may 47 00:01:54,400 --> 00:01:56,879 produce incomplete, inaccurate, or inconsistent results. 48 00:01:57,040 --> 00:01:59,200 Any model-generated analysis should be independently 49 00:01:59,200 --> 00:02:01,599 evaluated and should not be used as the sole basis for an 50 00:02:01,599 --> 00:02:02,719 investment decision. 51 00:02:03,760 --> 00:02:07,519 SPEAKER_01: Welcome to the next episode of Invest with AI, where 52 00:02:07,519 --> 00:02:10,400 we explore the intersection of fundamental investing and 53 00:02:10,400 --> 00:02:12,000 artificial intelligence. 54 00:02:12,240 --> 00:02:16,400 We have a great product market fit with the next guest on our 55 00:02:16,400 --> 00:02:20,240 podcast today, Doug Clinton of Intelligent Alpha, who has been 56 00:02:20,240 --> 00:02:22,000 investing with AI. 57 00:02:22,159 --> 00:02:26,240 So excited to have you on the pod, Doug, today and explore 58 00:02:26,240 --> 00:02:29,360 everything you've learned that's worked well, that hasn't worked 59 00:02:29,360 --> 00:02:32,400 well on your journey of investing uh with AI. 60 00:02:32,639 --> 00:02:34,879 Maybe to start, could you tell us a little bit more about 61 00:02:35,039 --> 00:02:37,120 Intelligent Alpha and what you're doing there? 62 00:02:37,680 --> 00:02:38,080 SPEAKER_02: Absolutely. 63 00:02:38,240 --> 00:02:39,199 Brett, good to be with you. 64 00:02:39,280 --> 00:02:40,080 Good to be with UK. 65 00:02:40,560 --> 00:02:44,000 And Intelligent Alpha is a company that I started uh two 66 00:02:44,000 --> 00:02:45,759 years ago now, officially. 67 00:02:46,000 --> 00:02:48,960 And it was born out of an experiment that started three 68 00:02:48,960 --> 00:02:53,120 years ago, where once ChatGPT kind of hit the market late 69 00:02:53,120 --> 00:02:57,360 2022, I was curious with a simple question, which is can 70 00:02:57,360 --> 00:03:01,439 ChatGPT beat the SP 500, which a lot of human managers can't do. 71 00:03:01,599 --> 00:03:04,800 And so we started running some tests to see if that was 72 00:03:04,800 --> 00:03:05,280 possible. 73 00:03:05,439 --> 00:03:08,719 And the short answer was uh things look very promising very 74 00:03:08,719 --> 00:03:09,199 early on. 75 00:03:09,280 --> 00:03:11,120 Maybe it's a little beginner's luck. 76 00:03:11,360 --> 00:03:14,719 But I was excited enough about what I was kind of seeing from 77 00:03:14,719 --> 00:03:18,719 very early LLMs and betting on the trajectory of these LLMs 78 00:03:18,719 --> 00:03:21,840 continuing to get smarter and smarter over time that we wanted 79 00:03:21,840 --> 00:03:24,800 to launch an investment business that was really built around the 80 00:03:24,800 --> 00:03:27,840 idea of harnessing the intelligence from frontier 81 00:03:27,840 --> 00:03:31,120 models, allowing them to do investment analysis and then 82 00:03:31,120 --> 00:03:33,120 ultimately portfolio management. 83 00:03:33,280 --> 00:03:36,719 The one other small piece I'd add about my background and kind 84 00:03:36,719 --> 00:03:39,919 of why we even played in this space in the first place is um 85 00:03:39,919 --> 00:03:44,159 almost a decade ago, I started a venture firm called Deepwater. 86 00:03:44,240 --> 00:03:45,680 I'm still a partner there. 87 00:03:45,759 --> 00:03:50,080 Uh I run kind of uh uh both sides of the coin, being a human 88 00:03:50,080 --> 00:03:53,680 allocator and also relying on AI to allocate at Intelligent 89 00:03:53,680 --> 00:03:54,159 Alpha. 90 00:03:54,319 --> 00:03:57,039 But we invest in early stage technology startups, we invest 91 00:03:57,039 --> 00:04:00,319 now in late stage tech startups and public equity at Deepwater. 92 00:04:00,400 --> 00:04:03,280 And so it was kind of this fun, natural testing ground where 93 00:04:03,280 --> 00:04:06,400 we've been in this world, we've been investing in AI companies, 94 00:04:06,560 --> 00:04:09,759 and Intelligent Alpha was the next logical progression for us 95 00:04:09,759 --> 00:04:11,439 from that uh business at Deepwater. 96 00:04:11,759 --> 00:04:12,800 SPEAKER_01: Super, super interesting. 97 00:04:12,879 --> 00:04:15,520 And I have many, many questions uh about that. 98 00:04:15,919 --> 00:04:21,120 One of the first um is how you thought about just the 99 00:04:21,120 --> 00:04:24,240 fundamentals of large language models, which operate on lexical 100 00:04:24,240 --> 00:04:25,120 intensity, right? 101 00:04:25,279 --> 00:04:27,759 LLMs know that Paris is the capital of France because 102 00:04:27,759 --> 00:04:30,240 they've seen it 14 million times in the training corpus. 103 00:04:30,399 --> 00:04:33,839 And you know, when I've sort of in the past gone and asked 104 00:04:33,920 --> 00:04:38,240 ChatGPT for a portfolio of 10 stocks, it seems to operate on 105 00:04:38,240 --> 00:04:40,160 that same sense of lexical intensity. 106 00:04:40,240 --> 00:04:44,240 The the stocks like NVIDIA that are talked about a lot on the 107 00:04:44,240 --> 00:04:48,079 open web get cited as individual ideas in the portfolio. 108 00:04:48,240 --> 00:04:50,720 That could be a signal or that could not be. 109 00:04:50,879 --> 00:04:54,160 Often sort of over the cycle of markets is probably a negative 110 00:04:54,160 --> 00:04:57,279 signal on retail-driven markets, probably a positive signal. 111 00:04:57,439 --> 00:05:00,800 How have you thought about that concept of lexical intensity, 112 00:05:00,959 --> 00:05:03,519 prevalence in the training corpus, and building a market 113 00:05:03,600 --> 00:05:04,959 beating portfolio? 114 00:05:05,439 --> 00:05:08,639 SPEAKER_02: Yeah, it'd say very early on, uh, one of the things 115 00:05:08,639 --> 00:05:13,600 we sort of realized was in 23, the paradigm for using LLMs 116 00:05:13,600 --> 00:05:16,800 effectively to invest was prompt engineering. 117 00:05:16,959 --> 00:05:19,839 I mean, that was all the rage then, and really understanding 118 00:05:19,839 --> 00:05:23,199 what is the context that you're providing to the models in terms 119 00:05:23,199 --> 00:05:25,120 of the task you want it to output. 120 00:05:25,360 --> 00:05:27,839 And so, like you just said, Brett, if you just say, give me 121 00:05:27,839 --> 00:05:30,560 a stock of 10 portfolios, you're probably gonna get, I would say, 122 00:05:30,720 --> 00:05:34,160 eight of the 10 uh largest companies in the SP 500 would be 123 00:05:34,160 --> 00:05:36,399 my bet, which is fine. 124 00:05:36,560 --> 00:05:38,480 That might be an okay portfolio. 125 00:05:38,639 --> 00:05:42,319 Um, but I think when we think about really using AI to make 126 00:05:42,319 --> 00:05:44,399 intelligent investments, that's not what we want. 127 00:05:44,560 --> 00:05:47,839 And so giving it a different framework where you know, maybe 128 00:05:47,839 --> 00:05:51,360 you narrow the field to small caps and maybe you define that 129 00:05:51,360 --> 00:05:53,600 within certain parameters for market cap. 130 00:05:53,920 --> 00:05:57,839 Maybe you pull in some external data uh from sources where maybe 131 00:05:57,839 --> 00:06:01,600 it's fundamental data, uh, maybe it's your own research about a 132 00:06:01,600 --> 00:06:02,639 handful of companies. 133 00:06:02,800 --> 00:06:05,199 I think that's where it starts to get much more interesting 134 00:06:05,360 --> 00:06:07,839 using these language models and seeing how they sort of 135 00:06:07,839 --> 00:06:11,040 interpret what is all the data that's out there that humans are 136 00:06:11,279 --> 00:06:12,079 looking at, right? 137 00:06:12,240 --> 00:06:16,160 And and language models can process way more qualitative 138 00:06:16,160 --> 00:06:19,439 data and quantitative data than any human analyst can. 139 00:06:19,600 --> 00:06:22,800 So trying to figure out how you can get that right data and the 140 00:06:22,800 --> 00:06:25,920 right selection set into the models is kind of step number 141 00:06:25,920 --> 00:06:26,240 one. 142 00:06:26,480 --> 00:06:29,199 And then I think, as I'm sure you guys have seen too, right, 143 00:06:29,279 --> 00:06:32,560 the frontier has evolved very quickly where it was all about 144 00:06:32,560 --> 00:06:34,639 prompt engineering a couple of years ago. 145 00:06:34,879 --> 00:06:37,439 I think now it's about agentic workflows, you know, 146 00:06:37,600 --> 00:06:38,639 orchestration. 147 00:06:38,879 --> 00:06:42,000 And I think even beyond where we're at right now, the thing 148 00:06:42,000 --> 00:06:46,160 that we're getting really into is how do you sort of manage 149 00:06:46,800 --> 00:06:48,160 organizational context? 150 00:06:48,319 --> 00:06:51,439 You know, instead of just narrow task context, you know, you're 151 00:06:51,439 --> 00:06:54,560 trying to pick a large cap portfolio, that's one specific 152 00:06:54,560 --> 00:06:56,800 vertical that you might be working on. 153 00:06:56,959 --> 00:07:00,639 How do you look at the entire corpus of every portfolio you've 154 00:07:00,639 --> 00:07:01,360 ever generated? 155 00:07:01,519 --> 00:07:04,079 Or maybe if you have multiple team members generating 156 00:07:04,079 --> 00:07:07,040 different portfolios with their agents, how do you sort of 157 00:07:07,040 --> 00:07:10,480 understand the entire broad knowledge base that's in your 158 00:07:10,480 --> 00:07:13,439 company and make it useful to the models when you're doing 159 00:07:13,439 --> 00:07:14,399 investment work? 160 00:07:15,279 --> 00:07:18,399 SPEAKER_03: Doug, on that point, did um you mention kind of the 161 00:07:18,399 --> 00:07:19,439 evolution from prompt. 162 00:07:19,519 --> 00:07:22,160 And it seems like we went from prompt engineering to content 163 00:07:22,480 --> 00:07:26,399 context engineering to agentic workflows. 164 00:07:26,639 --> 00:07:30,399 And now with kind of the fable-like models, kind of 165 00:07:30,399 --> 00:07:34,079 intent engineering or out, you know, out output pro 166 00:07:34,319 --> 00:07:35,839 output-based prompting. 167 00:07:36,240 --> 00:07:39,759 I'm curious how that last bit, I mean, we're still early in the 168 00:07:39,759 --> 00:07:44,000 fable five days, but um, has that last bit changed your 169 00:07:44,000 --> 00:07:44,879 approach? 170 00:07:46,000 --> 00:07:49,199 SPEAKER_02: Um, I'd say our approach is always changing 171 00:07:49,199 --> 00:07:49,680 slightly. 172 00:07:49,759 --> 00:07:51,279 And I think it has to actually. 173 00:07:51,360 --> 00:07:55,360 And and I think about these models uh in some ways as a 174 00:07:55,360 --> 00:07:58,160 reflection of how alpha changes in markets. 175 00:07:58,319 --> 00:08:01,199 You know, what might generate alpha today probably won't 176 00:08:01,199 --> 00:08:02,560 generate alpha tomorrow. 177 00:08:02,800 --> 00:08:06,480 And whatever you're doing to generate alpha, and that's a 178 00:08:06,480 --> 00:08:09,279 very broad word that probably needs a real definition, but 179 00:08:09,279 --> 00:08:12,399 let's just say outperformance relative to some benchmark right 180 00:08:12,399 --> 00:08:12,720 now. 181 00:08:12,800 --> 00:08:15,600 Um, whatever you're doing to generate that outperformance 182 00:08:15,600 --> 00:08:18,720 today with the models probably won't work tomorrow because 183 00:08:18,720 --> 00:08:21,439 there are other people that are experimenting uh with these 184 00:08:21,439 --> 00:08:24,399 models or finding out creative ways, I think, to uh make 185 00:08:24,399 --> 00:08:25,279 investment decisions. 186 00:08:25,439 --> 00:08:28,160 It might be much more of a human-oriented process than what 187 00:08:28,160 --> 00:08:31,600 we do at Intelligent Alpha, where we rely very much on the 188 00:08:31,600 --> 00:08:33,519 models to kind of work end to end. 189 00:08:33,759 --> 00:08:37,039 Um, but you know, I think we're always trying to change our 190 00:08:37,039 --> 00:08:41,120 approach and see what has sort of maybe stopped working based 191 00:08:41,120 --> 00:08:43,759 on what we were doing in the past and all the data that we 192 00:08:43,759 --> 00:08:45,279 have over the past three years. 193 00:08:45,440 --> 00:08:48,240 And then our intuition, and I think this is where the human 194 00:08:48,240 --> 00:08:51,039 still comes in, sort of our intuition of where the market's 195 00:08:51,039 --> 00:08:52,879 going, what might work in the future. 196 00:08:53,039 --> 00:08:55,840 Um, and in some ways that might be a little bit like how some 197 00:08:55,840 --> 00:08:58,879 quants operate, where a lot of times thinking of the next 198 00:08:58,879 --> 00:09:02,320 factor or thinking of the next sort of data set that you might 199 00:09:02,320 --> 00:09:05,840 believe has alpha is much more of kind of an intuition than 200 00:09:06,240 --> 00:09:07,039 just math. 201 00:09:07,440 --> 00:09:10,080 SPEAKER_01: Yeah, this is this is a huge structural difference 202 00:09:10,080 --> 00:09:12,559 versus building code, for example. 203 00:09:13,120 --> 00:09:16,240 One of the stats I've been I sort of point out to all my 204 00:09:16,240 --> 00:09:21,519 students is only 17% of stocks outperform the SP 500 over a 205 00:09:21,519 --> 00:09:23,200 rolling 10-year period. 206 00:09:23,279 --> 00:09:26,879 So probabilistically that stock you pick at random is going to 207 00:09:26,879 --> 00:09:30,320 be a stock that underperforms the SP 500. 208 00:09:30,639 --> 00:09:33,840 Sort of, you know, people sort of broadly discuss it, 70% of 209 00:09:33,840 --> 00:09:37,519 long-only managers underperform the SP 500. 210 00:09:37,759 --> 00:09:41,919 And so, you know, public markets alpha is truly a power law game. 211 00:09:42,240 --> 00:09:46,080 How do you how do you think about capturing, how do you 212 00:09:46,080 --> 00:09:50,720 think about capturing that power law sort of right tail alpha 213 00:09:50,720 --> 00:09:53,039 essence in the agentic structure? 214 00:09:53,120 --> 00:09:55,440 Because if we're sort of like taking our investment process 215 00:09:55,440 --> 00:09:58,720 and turning that into an agent structure, if I'm a median 216 00:09:58,960 --> 00:10:02,399 investor, I'm actually augmenting a median process, 217 00:10:02,480 --> 00:10:06,320 which structurally does not have the essence of that, of that 218 00:10:06,320 --> 00:10:08,399 alpha capture ability. 219 00:10:09,200 --> 00:10:12,639 SPEAKER_02: I think what's interesting to think in that 220 00:10:12,639 --> 00:10:19,200 context is how much of the intuition coming from the LLM is 221 00:10:19,200 --> 00:10:21,759 valuable versus the intuition coming from the human. 222 00:10:22,080 --> 00:10:25,519 And as I said, with intelligent alpha, the bet we are making is 223 00:10:25,519 --> 00:10:28,480 this long-term bet that over time the models do continue to 224 00:10:28,480 --> 00:10:28,879 get better. 225 00:10:29,039 --> 00:10:30,799 We've seen that in the benchmark data. 226 00:10:30,960 --> 00:10:34,080 And that ultimately probably leaving the models to their own 227 00:10:34,080 --> 00:10:37,759 devices with very little sort of nudges or interpretations or 228 00:10:37,919 --> 00:10:40,799 influences from humans is the best process. 229 00:10:41,039 --> 00:10:44,000 But today, I actually think that's that's a little bit 230 00:10:44,000 --> 00:10:48,320 different, where I still think the creativity of a human to 231 00:10:48,639 --> 00:10:54,399 say, you know, I want to be 30% weighted in some stock because I 232 00:10:54,399 --> 00:10:58,240 have really high conviction that it's going to work for reason X. 233 00:10:58,720 --> 00:11:02,320 Those types of intuitions, and maybe there's some deeper 234 00:11:02,320 --> 00:11:05,360 research that goes on behind the scenes, things that an AI can't 235 00:11:05,360 --> 00:11:08,240 get access to, you know, those things I think are still the 236 00:11:08,240 --> 00:11:11,919 core blocker from just turning the whole thing over to AI. 237 00:11:12,080 --> 00:11:15,200 And so what I would say we've seen work really well right now 238 00:11:15,200 --> 00:11:18,559 is almost this sort of like hybrid model between what would 239 00:11:18,559 --> 00:11:21,279 look like maybe a traditional quant portfolio and a 240 00:11:21,279 --> 00:11:22,960 traditional fundamental portfolio. 241 00:11:23,200 --> 00:11:26,080 And so generally, our portfolios, they don't have a 242 00:11:26,080 --> 00:11:29,519 thousand positions like you might have in a quant portfolio 243 00:11:29,600 --> 00:11:31,440 where it's literally right, you're just looking at numbers, 244 00:11:31,600 --> 00:11:35,039 a lot of small bets, and you're trying to get that 52, 54% 245 00:11:35,440 --> 00:11:36,320 batting average. 246 00:11:36,480 --> 00:11:39,039 You know, we generally have something on the order of a 247 00:11:39,039 --> 00:11:42,240 couple hundred, you know, maybe up to 500, depending on what 248 00:11:42,240 --> 00:11:43,679 strategy we're looking at. 249 00:11:43,840 --> 00:11:47,519 Um, but there are in our portfolios some conviction bets 250 00:11:47,519 --> 00:11:50,480 that we allow the models to make that may look like you know 251 00:11:50,559 --> 00:11:53,519 several percentage points in terms of concentration in the 252 00:11:53,519 --> 00:11:54,399 portfolio. 253 00:11:54,559 --> 00:11:59,360 Um, and AI, I think, has been pretty good in what we've seen 254 00:11:59,360 --> 00:12:03,679 so far in our portfolios at knowing where to be convicted 255 00:12:03,679 --> 00:12:06,159 and and maybe where to dial it back a little bit where it 256 00:12:06,159 --> 00:12:07,279 doesn't have great conviction. 257 00:12:07,360 --> 00:12:10,159 And it's gotten better over time to the point of our thesis. 258 00:12:10,320 --> 00:12:13,759 Um, and so, you know, I think that humans can still play a 259 00:12:13,759 --> 00:12:16,480 really valuable role, which is bringing in that data that the 260 00:12:16,480 --> 00:12:19,919 models just can't get access to, pushing the models to take 261 00:12:19,919 --> 00:12:20,960 really big bets. 262 00:12:21,120 --> 00:12:24,320 But if you're not going to have that human influence, sitting 263 00:12:24,320 --> 00:12:26,960 somewhere between probably a traditional fundamental process 264 00:12:26,960 --> 00:12:29,279 and a traditional quantitative process works really well with 265 00:12:29,279 --> 00:12:30,000 the LLMs. 266 00:12:30,320 --> 00:12:33,360 SPEAKER_03: How important is that data and what role does it 267 00:12:33,360 --> 00:12:33,759 play? 268 00:12:34,000 --> 00:12:38,399 Is it, you know, um, yeah, can you talk a little bit about the 269 00:12:38,639 --> 00:12:42,639 data, the types of data that you bring into it? 270 00:12:43,279 --> 00:12:45,200 SPEAKER_02: Yeah, we kind of think about it in three buckets. 271 00:12:45,279 --> 00:12:47,600 I mean, one is just publicly available data. 272 00:12:47,759 --> 00:12:51,039 So obviously, within the trading data of all these models, and 273 00:12:51,039 --> 00:12:53,840 for all intents and purposes, I kind of think of them as they've 274 00:12:53,840 --> 00:12:56,799 all been trained on essentially the same corpus of internet 275 00:12:56,960 --> 00:12:57,200 data. 276 00:12:57,279 --> 00:12:58,799 So they know everything that's on the internet. 277 00:12:59,120 --> 00:13:01,759 Um, that's kind of piece one in their training data. 278 00:13:01,840 --> 00:13:05,039 And then on top of that, we do bring in a lot of data from 279 00:13:05,039 --> 00:13:08,559 third-party vendors, um, which is just fundamental data, you 280 00:13:08,559 --> 00:13:11,519 know, consensus data, stuff like that, which again is just is 281 00:13:11,679 --> 00:13:12,799 publicly available. 282 00:13:13,039 --> 00:13:17,279 Um, bucket two for us, um, which is where we we've been trying to 283 00:13:17,279 --> 00:13:21,440 be sort of creative, is how can we use the models to make 284 00:13:21,440 --> 00:13:24,159 estimates about certain things related to companies? 285 00:13:24,320 --> 00:13:27,840 So that may be KPI-related estimates, it may just be strict 286 00:13:27,840 --> 00:13:29,360 earnings-related estimates. 287 00:13:29,600 --> 00:13:33,679 And so they might take some of the publicly available data, um, 288 00:13:33,840 --> 00:13:37,759 and it could include things from Reddit or X, it'll include 289 00:13:37,759 --> 00:13:40,960 things uh in the consensus metrics, et cetera, and then 290 00:13:40,960 --> 00:13:44,320 create an interpretation about that in terms of, you know, we 291 00:13:44,320 --> 00:13:47,519 think Apple will beat this quarter for X reason. 292 00:13:47,679 --> 00:13:51,279 Um that's kind of bucket two for us is think of it as like LLM 293 00:13:51,360 --> 00:13:52,960 sort of augmented data. 294 00:13:53,120 --> 00:13:58,080 And then bucket three is the um, you know, the proprietary sort 295 00:13:58,080 --> 00:14:01,919 of feed on the street research, I would call it, where LLMs, um, 296 00:14:02,080 --> 00:14:05,360 I think there's a path for them to be useful there. 297 00:14:05,519 --> 00:14:08,080 Um, but I think it's really hard to do that today. 298 00:14:08,159 --> 00:14:10,720 And that's, you know, your channel checks, it's calling on 299 00:14:10,720 --> 00:14:14,159 customers, it's going to investment conferences, um, all 300 00:14:14,159 --> 00:14:17,120 the things that are just limitations to a digital system 301 00:14:17,919 --> 00:14:19,679 are kind of that bottleneck. 302 00:14:19,759 --> 00:14:23,360 And to me, that is kind of the longer-term frontier for using 303 00:14:23,360 --> 00:14:26,639 LLMs is can you use them to do channel checks somehow? 304 00:14:26,879 --> 00:14:29,600 Can you imagine a world where actually agents are talking to 305 00:14:29,600 --> 00:14:32,320 other agents doing channel checks, not just agents talking 306 00:14:32,320 --> 00:14:34,720 to humans, which is kind of easy to imagine today? 307 00:14:34,799 --> 00:14:37,200 Um, and just what does that look like in the future? 308 00:14:37,360 --> 00:14:40,320 Because I really think that, I mean, just like it is today, 309 00:14:40,480 --> 00:14:44,480 that probably is the long-term real durable source of alpha 310 00:14:44,639 --> 00:14:47,600 versus kind of just understanding really big context 311 00:14:47,759 --> 00:14:48,879 better than humans right now. 312 00:14:49,279 --> 00:14:51,519 SPEAKER_01: Doug, how have you thought about the sort of the 313 00:14:51,519 --> 00:14:53,200 MCP pathway? 314 00:14:53,360 --> 00:14:57,360 I think the consensus was, you know, four or five months ago, 315 00:14:57,440 --> 00:15:00,879 it was the MCP structure was probably too brittle for 316 00:15:00,879 --> 00:15:02,559 institutional use cases. 317 00:15:02,799 --> 00:15:06,320 I've seen that conversation die down, but there's still a little 318 00:15:06,320 --> 00:15:11,120 bit of debate on what the right data structure looks like to 319 00:15:11,120 --> 00:15:14,399 sort of pipe in the right data to make institutional great 320 00:15:14,399 --> 00:15:14,879 decisions. 321 00:15:15,279 --> 00:15:17,120 SPEAKER_02: Yeah, I think I think it's probably still an 322 00:15:17,120 --> 00:15:18,080 open question. 323 00:15:18,320 --> 00:15:24,720 Um for us, I mean, we use um various MCP uh, you know, 324 00:15:24,960 --> 00:15:28,159 sources that that help us pull in some data. 325 00:15:28,399 --> 00:15:33,360 Um and I think that I think the world broadly is evolving to 326 00:15:33,360 --> 00:15:38,720 this idea that um getting context to the models at the 327 00:15:38,720 --> 00:15:42,320 right time and the right place is super valuable, right? 328 00:15:42,480 --> 00:15:44,399 And you look at like a carbon arc, right? 329 00:15:44,480 --> 00:15:48,639 Paying paying almost per piece uh of data, and you get to 330 00:15:48,639 --> 00:15:50,320 determine what you think is really valuable. 331 00:15:50,399 --> 00:15:53,600 I think we're gonna see a lot more models like that. 332 00:15:53,840 --> 00:15:57,679 And and whether it literally uses, you know, the MCP, that 333 00:15:57,679 --> 00:16:00,720 exact protocol, or if it just looks like an API or whatever. 334 00:16:00,879 --> 00:16:04,639 Um, I don't know how much that matters relative to having just 335 00:16:04,639 --> 00:16:07,840 the right structures, the right payment models, and ultimately 336 00:16:07,840 --> 00:16:11,120 the right data sources and the right ability for the models to 337 00:16:11,120 --> 00:16:13,279 say, I need this piece of data. 338 00:16:13,440 --> 00:16:15,600 I think that's the super valuable piece. 339 00:16:15,679 --> 00:16:20,399 Um and I would say this like models today are, I think 340 00:16:20,399 --> 00:16:21,200 they're pretty good. 341 00:16:21,360 --> 00:16:25,759 If I had to grade them, I'd give them like a B as an analyst sort 342 00:16:25,759 --> 00:16:28,559 of understanding like this is the thing that's probably gonna 343 00:16:28,559 --> 00:16:31,039 move a stock in a given quarter. 344 00:16:31,200 --> 00:16:34,799 Um, their creativity isn't quite there, like a human analyst. 345 00:16:35,039 --> 00:16:37,200 I mean, some of the creative things I've heard for human 346 00:16:37,200 --> 00:16:40,559 analysts doing over the years to try to get an angle on a quarter 347 00:16:40,720 --> 00:16:41,600 are incredible. 348 00:16:41,840 --> 00:16:43,360 But again, think about the future. 349 00:16:43,440 --> 00:16:45,679 Think about a year from now, think about a two years from 350 00:16:45,679 --> 00:16:45,840 now. 351 00:16:45,919 --> 00:16:49,440 I think models will think of those super creative things and 352 00:16:49,440 --> 00:16:51,279 be figuring those things out for us. 353 00:16:51,440 --> 00:16:54,240 Um, and that's the frontier that I'm really interested in trying 354 00:16:54,240 --> 00:16:54,639 to figure out. 355 00:16:54,960 --> 00:16:58,399 SPEAKER_03: Doug, when we uh before we started recording, you 356 00:16:58,399 --> 00:17:02,080 had mentioned the importance of knowledge graphs as it relates 357 00:17:02,080 --> 00:17:02,879 to MCP. 358 00:17:03,039 --> 00:17:05,119 Can you say a little bit more about that? 359 00:17:06,240 --> 00:17:06,799 SPEAKER_02: Yeah. 360 00:17:07,039 --> 00:17:10,400 Um the it's it's funny, okay. 361 00:17:10,640 --> 00:17:13,119 Like the the terms, as you as you mentioned earlier, right, 362 00:17:13,200 --> 00:17:15,759 that have have sort of evolved in like what is the hot thing at 363 00:17:15,759 --> 00:17:16,160 AI? 364 00:17:16,240 --> 00:17:19,920 I mean, it changes every six months-ish or so, it feels like. 365 00:17:20,160 --> 00:17:23,680 Um, and I would say this this concept of sort of the knowledge 366 00:17:23,680 --> 00:17:27,279 graph feels like it very much the frontier right now. 367 00:17:27,440 --> 00:17:32,400 And just trying to figure out, you know, how do you take um all 368 00:17:32,400 --> 00:17:36,000 of the knowledge that you have as an organization or as an 369 00:17:36,000 --> 00:17:40,799 investment team or whatever your your uh your node structure is 370 00:17:41,279 --> 00:17:45,599 and make that entire graph usable by every agent that you 371 00:17:45,599 --> 00:17:47,680 have running on your platform, right? 372 00:17:47,839 --> 00:17:50,480 And that may be hundreds of agents or thousands of agents, 373 00:17:50,559 --> 00:17:52,480 depending on how evolved you are. 374 00:17:52,640 --> 00:17:56,960 Um, but it's a really complex task to let, I mean, imagine 375 00:17:56,960 --> 00:17:59,920 thousands of different agents trying to pull different bits of 376 00:17:59,920 --> 00:18:04,240 context real time, creating new context, right, that other 377 00:18:04,240 --> 00:18:06,319 agents should then have access to. 378 00:18:06,480 --> 00:18:11,200 Um, doing that really well, I think is going to be a source of 379 00:18:11,200 --> 00:18:14,799 alpha for people using language models or just agents in 380 00:18:14,799 --> 00:18:18,319 general, I think, to invest, probably over the next, I'd give 381 00:18:18,319 --> 00:18:19,519 it 12 to 18 months. 382 00:18:19,599 --> 00:18:22,480 I think that'll be a huge advantage before eventually it 383 00:18:22,480 --> 00:18:25,039 probably becomes a little bit more commoditized, whether 384 00:18:25,039 --> 00:18:27,839 that's because some third-party vendor introduces a really 385 00:18:27,839 --> 00:18:30,400 easy-to-use system, which just doesn't exist right now that 386 00:18:30,400 --> 00:18:35,039 I've seen, um, or right, Claude Code and Codex make it so easy 387 00:18:35,039 --> 00:18:38,079 to just say, hey, implement this, and it just works. 388 00:18:38,720 --> 00:18:43,039 SPEAKER_03: It's fascinating too because the sorry, Brett, the 389 00:18:43,359 --> 00:18:46,880 like schemas and ontologies, again, these kind of buzzwords 390 00:18:46,880 --> 00:18:49,359 that no one was talking about six months ago. 391 00:18:49,759 --> 00:18:57,200 But it is this weird idea that you have this data, but you have 392 00:18:57,200 --> 00:19:01,359 this structure that oftentimes is just intuited by your 393 00:19:01,359 --> 00:19:04,559 process, by your analysts, by your PMs, by the industry. 394 00:19:05,200 --> 00:19:08,160 And that might be the gray matter that the LLM hasn't 395 00:19:08,160 --> 00:19:08,880 figured out yet. 396 00:19:09,039 --> 00:19:13,680 And just like watching the race to disentangle that, to 397 00:19:13,680 --> 00:19:16,400 categorize that, to label that, to, you know, to make it 398 00:19:16,400 --> 00:19:20,160 visible, is just uh it's just really fascinating right now. 399 00:19:20,240 --> 00:19:23,519 And and I guess probably early days to know if it's even 400 00:19:23,519 --> 00:19:24,480 effective. 401 00:19:25,200 --> 00:19:25,759 SPEAKER_02: I think so. 402 00:19:25,920 --> 00:19:28,400 And it comes back to this idea of taste. 403 00:19:28,480 --> 00:19:31,759 You know, the you hear the word taste often in AI, which I think 404 00:19:31,759 --> 00:19:35,920 is sort of this nod to longer-term human value, this 405 00:19:35,920 --> 00:19:37,920 sort of instinct that we've talked about. 406 00:19:38,160 --> 00:19:41,519 Um, I do think to some extent, like creating these ontologies 407 00:19:41,519 --> 00:19:44,319 and these knowledge graphs, there is an element of, okay, 408 00:19:44,400 --> 00:19:47,839 fine, like codex point it at something, give it a really 409 00:19:47,839 --> 00:19:51,279 great creative loop, and it'll probably figure something out 410 00:19:51,279 --> 00:19:52,400 that's passable. 411 00:19:52,799 --> 00:19:57,039 Um, but to make it A plus, I think it does need to have that 412 00:19:57,039 --> 00:20:00,480 element of taste from a human who really understands like this 413 00:20:00,480 --> 00:20:03,599 is what it means to generate alpha, not just like what the 414 00:20:03,599 --> 00:20:07,759 internet has taught you as an AI model, but like this is why it's 415 00:20:07,759 --> 00:20:08,400 hard, right? 416 00:20:08,559 --> 00:20:12,319 And this is kind of like maybe even the history of how how 417 00:20:12,319 --> 00:20:16,079 alpha sort of evolves in a sense, and being able to take 418 00:20:16,079 --> 00:20:19,599 that taste and apply that to the ontology, that's where I think 419 00:20:19,599 --> 00:20:22,960 you get this unique value from humans coming in and taking sort 420 00:20:22,960 --> 00:20:26,400 of AI's baseline work, which is like B plus, and then getting it 421 00:20:26,400 --> 00:20:27,519 up to that A level. 422 00:20:27,680 --> 00:20:30,000 And that's where I think you can really get some additional value 423 00:20:30,000 --> 00:20:30,240 from it. 424 00:20:30,640 --> 00:20:33,519 SPEAKER_01: Can we drill into this concept of knowledge graph 425 00:20:33,519 --> 00:20:34,559 a little bit more? 426 00:20:34,720 --> 00:20:36,400 Like, what does it actually look like in practice? 427 00:20:36,480 --> 00:20:39,200 I'd be curious, Kay, your perspective on like how what's 428 00:20:39,200 --> 00:20:41,119 the what's the sort of zero to one? 429 00:20:41,200 --> 00:20:43,920 Like if a client comes in and wants to build a knowledge 430 00:20:43,920 --> 00:20:44,240 graph. 431 00:20:44,720 --> 00:20:45,519 What does that mean? 432 00:20:45,680 --> 00:20:47,599 Is that internal file structure? 433 00:20:47,839 --> 00:20:51,680 How do I get my data out of Excel files and OneNotes and 434 00:20:52,079 --> 00:20:55,279 PDFs and Outlook inbox and Slack? 435 00:20:55,359 --> 00:20:57,440 Like, how do you what's what's an actual process looking? 436 00:20:58,640 --> 00:21:01,839 SPEAKER_03: I mean, I'm going to give you the training wheels 437 00:21:01,839 --> 00:21:06,000 version of this, but um I think that it's a lot of the things 438 00:21:06,000 --> 00:21:10,720 that you just said where you have all this data, but even 439 00:21:10,720 --> 00:21:15,440 something like earnings or EBITDA, that can mean 15 440 00:21:15,440 --> 00:21:18,720 different things to 15 different people, even within a firm, 441 00:21:18,799 --> 00:21:20,480 within a subsector, and so on. 442 00:21:20,960 --> 00:21:26,559 So if you go in and teach the model, like when I say, you 443 00:21:26,559 --> 00:21:30,079 know, for us, the concept of EBITDA means, you know, these 444 00:21:30,079 --> 00:21:33,039 adjustments, you don't never do this, you have to teach the 445 00:21:33,039 --> 00:21:33,440 model that. 446 00:21:33,519 --> 00:21:36,880 And by the way, you have to basically like give it a map 447 00:21:37,039 --> 00:21:39,759 that like these are the ways that we define all of these 448 00:21:39,759 --> 00:21:40,000 things. 449 00:21:40,240 --> 00:21:42,960 So once you give them the map, and so this map can be just like 450 00:21:42,960 --> 00:21:44,720 a markdown file, right? 451 00:21:44,960 --> 00:21:49,279 And once you give them the map, then you have this data, right? 452 00:21:49,359 --> 00:21:53,200 You have let's say 10 portfolio, 10 companies with you know, 10 453 00:21:53,200 --> 00:21:56,559 years of quarterly EBITDA across 10 different sectors. 454 00:21:56,720 --> 00:21:59,759 You have to basically say that, like, well, the way we look at 455 00:22:00,160 --> 00:22:03,200 EBITDA in software is completely different than the way that we 456 00:22:03,200 --> 00:22:05,359 look at it in industrials. 457 00:22:05,759 --> 00:22:08,400 And in doing that, you know, when you want to grab that 458 00:22:08,400 --> 00:22:10,319 number, you have to make this adjustment, you have to look 459 00:22:10,319 --> 00:22:11,839 here, you have to do that. 460 00:22:12,000 --> 00:22:16,160 And so you're basically kind of putting it together so that when 461 00:22:16,160 --> 00:22:19,119 the model goes in and you're like, what's the EBITDA for 462 00:22:19,119 --> 00:22:20,400 Salesforce this year? 463 00:22:20,559 --> 00:22:22,000 It starts to go down this graph. 464 00:22:22,079 --> 00:22:24,799 It's like, well, I know that this, like, the way that Brett 465 00:22:24,880 --> 00:22:26,400 defines EBITDA is this. 466 00:22:26,480 --> 00:22:29,440 And I know that in SaaS, you look at it differently, you go 467 00:22:29,440 --> 00:22:32,079 here, and now I know that there's a Salesforce data here, 468 00:22:32,160 --> 00:22:34,960 but there's this other data I need to add back in or out. 469 00:22:35,200 --> 00:22:39,359 And so it's basically creating that like the term is like 470 00:22:39,359 --> 00:22:43,440 ontology, which is like, again, dumb person here, but you know, 471 00:22:43,519 --> 00:22:47,759 the definition, your definition of what these words mean, so 472 00:22:47,759 --> 00:22:49,519 that you can translate it to the model. 473 00:22:49,839 --> 00:22:52,799 So then you give this, you give it all this data, again, 474 00:22:52,960 --> 00:22:57,039 markdown files, CSV files, um, skills, so on. 475 00:22:57,440 --> 00:23:00,079 Um, and then you want to query it, right? 476 00:23:00,240 --> 00:23:02,079 And so there's all these different ways of querying it, 477 00:23:02,160 --> 00:23:02,240 right? 478 00:23:02,319 --> 00:23:05,519 You could do brute force agentic search, which is like command F 479 00:23:05,680 --> 00:23:08,720 EBITDA, but like still go through the EBITDA knowledge 480 00:23:08,720 --> 00:23:10,480 graph to get to the data point. 481 00:23:10,799 --> 00:23:13,359 Or you can use, like some of our other guests have discussed, 482 00:23:13,519 --> 00:23:16,400 semantic search through RAG and embeddings, where it's like, oh, 483 00:23:16,480 --> 00:23:20,559 okay, you know, you're asking me for uh profitability, which can 484 00:23:20,559 --> 00:23:22,319 mean a lot of different things. 485 00:23:22,559 --> 00:23:25,680 And how does that tie to this definition of EBITDA and 486 00:23:25,680 --> 00:23:27,039 Salesforce and da-da-da? 487 00:23:27,599 --> 00:23:31,119 You can then add like keyword search and you know, different 488 00:23:31,119 --> 00:23:34,559 databases that like have sanitized the data and so on. 489 00:23:34,799 --> 00:23:38,559 So that's conceptually how I'm seeing folks do it, and I work 490 00:23:38,559 --> 00:23:41,440 with much smaller firms, where they're just starting with like 491 00:23:41,440 --> 00:23:44,240 a small group of names, and they're just, you know, we're 492 00:23:44,240 --> 00:23:47,440 gonna do the Salesforce ontology here, and then we might roll 493 00:23:47,440 --> 00:23:51,119 that up to the SaaS ontology over here and just like see if 494 00:23:51,119 --> 00:23:55,200 it works, see if it just like speeds up the process of getting 495 00:23:55,200 --> 00:24:00,480 the right context at the right moment and with the lens that 496 00:24:00,480 --> 00:24:02,400 I'm used to looking at it with. 497 00:24:02,559 --> 00:24:05,119 And then you're getting dev shops that are coming in, and 498 00:24:05,200 --> 00:24:07,920 like the the crazy thing about the dev shops coming in, because 499 00:24:07,920 --> 00:24:12,000 I've like sat in in a few of these conversations, is that a 500 00:24:12,000 --> 00:24:14,880 big chunk of the time is interviewing a PM. 501 00:24:14,960 --> 00:24:17,680 It's like when you say eBITAL, what do you mean? 502 00:24:17,920 --> 00:24:18,240 Right? 503 00:24:18,400 --> 00:24:22,079 Because that's so baked into their heads, but you you need a 504 00:24:22,079 --> 00:24:25,599 translator, and that's been a bit of a challenge with a lot of 505 00:24:25,599 --> 00:24:28,160 firms who are like, like, wait, I thought you were gonna give me 506 00:24:28,160 --> 00:24:28,480 the answer. 507 00:24:28,559 --> 00:24:32,319 It's like, wait, no, you need to tell me how you understand this 508 00:24:32,319 --> 00:24:34,640 concept, and then I'll write code for you. 509 00:24:34,799 --> 00:24:36,559 And then a lot of people are like, well, I don't have time 510 00:24:36,559 --> 00:24:36,880 for that. 511 00:24:36,960 --> 00:24:39,119 Like, I thought you were just gonna do it for me. 512 00:24:39,359 --> 00:24:40,240 I don't know, Doug. 513 00:24:40,640 --> 00:24:41,599 What do you have to add to that? 514 00:24:41,839 --> 00:24:43,599 SPEAKER_02: And it's funny, like, okay, that's what makes it 515 00:24:43,599 --> 00:24:44,240 valuable, right? 516 00:24:44,319 --> 00:24:46,640 Is that you put your stamp on it. 517 00:24:46,799 --> 00:24:49,680 You know, if you just say, well, just give me your general 518 00:24:49,680 --> 00:24:52,559 definition of how you think about EBITDA or whatever the 519 00:24:52,559 --> 00:24:56,880 metric is, um, it sort of defeats the purpose because then 520 00:24:56,960 --> 00:24:59,759 it becomes this sort of generalization instead of 521 00:24:59,759 --> 00:25:02,799 something specific to the knowledge that should be 522 00:25:02,799 --> 00:25:06,000 inherent to your organization that reflects how you think 523 00:25:06,000 --> 00:25:07,920 about investing, how you think about the world. 524 00:25:08,079 --> 00:25:09,839 Um, and it does take a lot of work. 525 00:25:09,920 --> 00:25:12,640 And I think that that's actually, you know, the most 526 00:25:12,640 --> 00:25:15,519 important thing when you think about trying to implement, you 527 00:25:15,519 --> 00:25:17,839 know, whether you're trying to do what we do and really rely on 528 00:25:17,839 --> 00:25:21,200 the LLMs to make investment decisions, or if you're just 529 00:25:21,200 --> 00:25:24,160 trying to build better investment processes that are 530 00:25:24,160 --> 00:25:27,440 still built around humans, it's going to be a big investment in 531 00:25:27,440 --> 00:25:30,000 terms of time, in terms of building the systems. 532 00:25:30,240 --> 00:25:33,200 And uh the thing that I would I would suggest everybody think 533 00:25:33,200 --> 00:25:36,559 about, no matter which end you are on that spectrum, just from 534 00:25:36,880 --> 00:25:39,920 having lived it so far for a couple of years, is that things 535 00:25:39,920 --> 00:25:41,359 are changing so fast. 536 00:25:41,839 --> 00:25:44,720 There's always this tension, I think, in the investment world 537 00:25:45,359 --> 00:25:47,519 for the desire for perfection. 538 00:25:47,599 --> 00:25:53,279 Like we want to nail uh 99% accuracy rate on numbers or you 539 00:25:53,279 --> 00:25:54,640 know, whatever it might be. 540 00:25:54,799 --> 00:25:58,240 And I think we're living in a time right now where you have to 541 00:25:58,240 --> 00:26:02,799 accept a little bit less perfection in the name of speed. 542 00:26:02,960 --> 00:26:06,559 Because if you go and you you go on this like six-month, you 543 00:26:06,559 --> 00:26:09,759 know, really deep, complex process to build some really 544 00:26:09,759 --> 00:26:12,799 elegant knowledge graph, I guarantee you in six months, 545 00:26:12,960 --> 00:26:14,000 something will be different. 546 00:26:14,160 --> 00:26:16,160 The whole paradigm will have changed. 547 00:26:16,240 --> 00:26:19,920 And you'll be like, oh shoot, now we need to build XYZ thing. 548 00:26:20,000 --> 00:26:21,440 That's the new hot thing, right? 549 00:26:21,680 --> 00:26:26,079 And so just don't let yourself get excessively um, you know, 550 00:26:26,240 --> 00:26:27,759 hung up on the details. 551 00:26:28,000 --> 00:26:30,960 Get something that works, that's reliable, and know that it's 552 00:26:30,960 --> 00:26:34,160 going to evolve very quickly over time and evolve with it. 553 00:26:36,079 --> 00:26:36,559 SPEAKER_01: Yeah. 554 00:26:36,880 --> 00:26:39,759 How do you how do you think that sort of obsolescence risk has 555 00:26:39,759 --> 00:26:41,440 been has been massive? 556 00:26:41,599 --> 00:26:44,960 Like, yeah, in in in in three months, is there a vendor where 557 00:26:44,960 --> 00:26:48,079 you just put a listening device in your office and it gathers 558 00:26:48,079 --> 00:26:51,759 context for three weeks and then builds a custom knowledge graph 559 00:26:51,920 --> 00:26:52,720 for you, right? 560 00:26:52,799 --> 00:26:55,279 Like the sort of crazy things are in development. 561 00:26:55,519 --> 00:26:59,440 Now, how do you think about navigating that obsolescence 562 00:26:59,440 --> 00:27:02,240 risk and in building your infrastructure? 563 00:27:02,640 --> 00:27:06,720 SPEAKER_02: Yeah, we've we've built so much stuff that we no 564 00:27:06,720 --> 00:27:07,519 longer use. 565 00:27:07,759 --> 00:27:11,599 Um, and literally some of it has lasted a month, two months. 566 00:27:11,839 --> 00:27:15,920 Um and I think like with the emergence of the coding tools, 567 00:27:16,000 --> 00:27:18,640 that has accelerated it, you know, because it is so easy 568 00:27:18,640 --> 00:27:22,000 right now to write, not even just like basic code. 569 00:27:22,079 --> 00:27:25,200 I mean, to write pretty decent code, like very functional 570 00:27:25,200 --> 00:27:28,480 programs, even for people who aren't like high-level 571 00:27:28,480 --> 00:27:32,319 engineers, um, is is pretty easy if you know how to use the 572 00:27:32,559 --> 00:27:33,440 systems well. 573 00:27:33,680 --> 00:27:36,720 Um now, writing something that scales to an organization of 574 00:27:36,720 --> 00:27:39,519 maybe 100 people, a thousand people, that is a different sort 575 00:27:39,519 --> 00:27:40,480 of challenge. 576 00:27:40,720 --> 00:27:44,240 But um, I I think that we should assume that that reality 577 00:27:44,240 --> 00:27:47,599 continues to be persistent, which is technology will evolve 578 00:27:47,599 --> 00:27:48,240 really fast. 579 00:27:48,400 --> 00:27:52,640 So build your systems as kind of quick as you can to work really 580 00:27:52,640 --> 00:27:54,319 well for the paradigm today. 581 00:27:54,559 --> 00:27:58,000 Take advantage of everything you can today, but be ready to be 582 00:27:58,000 --> 00:28:01,440 nimble and don't have a lot of you know sunk cost in something 583 00:28:01,440 --> 00:28:03,359 where you say, well, we can't change that because we just 584 00:28:03,359 --> 00:28:06,640 spent$2 million building the system, but there is this better 585 00:28:06,640 --> 00:28:06,799 thing. 586 00:28:07,119 --> 00:28:08,960 I mean, it's almost like NVIDIA chips, right? 587 00:28:09,119 --> 00:28:14,079 Like you buy your H100s and then the B100 or B200, right, is way 588 00:28:14,079 --> 00:28:14,720 more efficient. 589 00:28:14,799 --> 00:28:18,160 You save more money, you know, per token that you create by 590 00:28:18,160 --> 00:28:19,839 just investing in the better chips. 591 00:28:20,000 --> 00:28:22,400 So you just have to be prepared to do that, just like the data 592 00:28:22,400 --> 00:28:25,039 centers are doing with their with their infrastructure. 593 00:28:25,440 --> 00:28:27,039 SPEAKER_03: Yeah, a funny story on this. 594 00:28:27,119 --> 00:28:30,960 Uh, there was an AI vendor that was very hot on it's kind of an 595 00:28:30,960 --> 00:28:37,119 enterprise search uh in 2025 before the MCP takeoff moment, 596 00:28:37,279 --> 00:28:40,000 which I would say is like December of last year. 597 00:28:40,240 --> 00:28:43,839 Um, and everyone was into this vendor, I don't want to name it, 598 00:28:43,920 --> 00:28:45,279 uh, but I was into this vendor. 599 00:28:45,440 --> 00:28:47,359 I did tons of calls with them last year. 600 00:28:47,519 --> 00:28:50,240 And then when the connectors hit, it went dark. 601 00:28:50,400 --> 00:28:53,119 No one wanted to talk to this vendor at all. 602 00:28:53,680 --> 00:28:55,359 Last week, a friend texted me. 603 00:28:55,440 --> 00:28:59,119 He's like, hey, I actually stayed on that vendor, and they 604 00:28:59,119 --> 00:29:02,559 have incorporated some of these elements of knowledge graph, 605 00:29:02,799 --> 00:29:05,759 different semantic search, permissioning is a big deal in 606 00:29:05,759 --> 00:29:07,039 the knowledge graph. 607 00:29:07,279 --> 00:29:10,079 Uh and it's really good. 608 00:29:10,480 --> 00:29:12,480 That's what the front, what the text told me. 609 00:29:12,559 --> 00:29:15,359 So even in the vendor cycle, you see it. 610 00:29:15,440 --> 00:29:16,079 It's wild. 611 00:29:16,319 --> 00:29:19,680 Like I would not want to be building and selling on that 612 00:29:19,680 --> 00:29:21,039 side of the fence. 613 00:29:21,440 --> 00:29:22,640 SPEAKER_02: Yeah, I agree. 614 00:29:22,880 --> 00:29:25,440 And it I think that's actually a great point, too, Kay, because 615 00:29:25,440 --> 00:29:28,240 if you think about um in the investment space, the ability to 616 00:29:28,240 --> 00:29:29,920 control your own destiny too. 617 00:29:30,160 --> 00:29:32,880 Um, this is just a philosophy that we have in Intelligent 618 00:29:32,880 --> 00:29:36,240 Alpha, but to the extent that we can sort of build something 619 00:29:36,240 --> 00:29:40,559 ourselves or use open source and modify ourselves very quickly, 620 00:29:40,720 --> 00:29:44,240 um, we've tried to very much do that and not rely too much on 621 00:29:44,240 --> 00:29:46,720 third-party vendors because of what you just described. 622 00:29:46,799 --> 00:29:50,640 Like, you know, the vendor could be awesome when you install them 623 00:29:50,640 --> 00:29:52,720 day one and something new could come out. 624 00:29:52,799 --> 00:29:55,759 And by no fault of their own, just the fact that they bet on a 625 00:29:55,759 --> 00:29:58,400 certain idea or a certain technology, maybe they fall 626 00:29:58,400 --> 00:29:59,279 behind a little bit. 627 00:29:59,359 --> 00:30:01,440 And six months later, maybe they're awesome again. 628 00:30:01,599 --> 00:30:04,559 Like it's just really hard to not be able to sort of control 629 00:30:04,559 --> 00:30:07,039 your own destiny given how fast things are moving. 630 00:30:07,200 --> 00:30:10,319 Um, so that's the other piece for us is just philosophically, 631 00:30:10,400 --> 00:30:13,359 kind of how do you think about how quickly and and what kinds 632 00:30:13,359 --> 00:30:15,759 of investments you want to make in building your technology 633 00:30:15,759 --> 00:30:16,480 infrastructure? 634 00:30:16,640 --> 00:30:18,880 And is it something that you want to have a lot of control 635 00:30:18,880 --> 00:30:22,319 over or are you okay with relying on good third-party 636 00:30:22,319 --> 00:30:22,880 vendors for that? 637 00:30:23,200 --> 00:30:26,160 SPEAKER_01: Doug, how have you thought about model routing? 638 00:30:26,319 --> 00:30:28,400 You know, it sort of sort of feels like every few weeks you 639 00:30:28,400 --> 00:30:31,759 get a new model and you know, open the open AI models were 640 00:30:31,759 --> 00:30:32,240 ascended. 641 00:30:32,400 --> 00:30:35,680 You know, this sort of H126 has been all about Claude. 642 00:30:35,839 --> 00:30:39,119 Now, this sort of conversation shifting back to Kimmy and Deep 643 00:30:39,119 --> 00:30:42,960 Seek and the open source models and token efficiency argument, 644 00:30:43,039 --> 00:30:43,680 et cetera. 645 00:30:43,920 --> 00:30:46,960 How have you thought about evaluating that systematically? 646 00:30:47,119 --> 00:30:49,519 And when do you make the decision to plug in a new model 647 00:30:49,519 --> 00:30:52,319 and sort of pivot pivot your routing structure? 648 00:30:52,799 --> 00:30:56,000 SPEAKER_02: Yeah, we do a ton of benchmarking internally. 649 00:30:56,079 --> 00:30:58,880 We've actually started to publish some of this uh work as 650 00:30:58,880 --> 00:31:00,079 well externally. 651 00:31:00,160 --> 00:31:01,519 Um, literally today. 652 00:31:01,599 --> 00:31:05,839 Um, so it'll be a few days, uh, I guess maybe prior to when this 653 00:31:05,839 --> 00:31:07,440 gets uh launched. 654 00:31:07,680 --> 00:31:12,160 But we launched something called the IA 500, which is a nod to 655 00:31:12,160 --> 00:31:14,000 the SP 500, obviously. 656 00:31:14,160 --> 00:31:17,279 But it's a it's a stock picking benchmark where we have a 657 00:31:17,279 --> 00:31:21,440 quarterly rebalance structure, just like most uh indexes, and 658 00:31:21,440 --> 00:31:26,480 we test 10 of the top uh models, top meaning you know, most 659 00:31:26,480 --> 00:31:27,200 influential. 660 00:31:27,359 --> 00:31:29,920 It would be all the names that you would expect to kind of be 661 00:31:29,920 --> 00:31:30,720 in that group. 662 00:31:30,880 --> 00:31:35,200 And we look at their ability to pick a portfolio of about 200 663 00:31:35,200 --> 00:31:38,960 stocks, and then we actually ensemble all of the picks from 664 00:31:38,960 --> 00:31:42,960 the 10 models into a single portfolio of about 500 stocks. 665 00:31:43,200 --> 00:31:47,279 And uh it's performed quite well versus the SP 500 over the past 666 00:31:47,279 --> 00:31:50,319 year since we've been kind of putting this data together. 667 00:31:50,400 --> 00:31:53,359 Um so you can check it out at IA500.com. 668 00:31:53,440 --> 00:31:56,400 But the the kind of answer to your question though, Brett, is 669 00:31:56,400 --> 00:32:00,400 we're always doing these tests on an ongoing basis to see what 670 00:32:00,400 --> 00:32:02,000 models sort of stand out where. 671 00:32:02,240 --> 00:32:06,640 And what I would tell you is this there um in our view, there 672 00:32:06,640 --> 00:32:11,599 is a separation that is notable between the closed source models 673 00:32:11,599 --> 00:32:14,319 and the open weight models that we see on most of our 674 00:32:14,319 --> 00:32:15,200 benchmarks. 675 00:32:15,440 --> 00:32:18,160 In general, I would say when we're talking about investment 676 00:32:18,160 --> 00:32:22,240 tasks, not just stock picking, uh, the the closed source models 677 00:32:22,400 --> 00:32:26,240 and GPT has consistently been the best one that we test in 678 00:32:26,240 --> 00:32:27,359 this uh section. 679 00:32:27,599 --> 00:32:31,279 Um, they always have a little bit of outperformance as a group 680 00:32:31,279 --> 00:32:33,920 relative to uh the open weight models. 681 00:32:34,000 --> 00:32:36,079 Um we do see some superiority there. 682 00:32:36,160 --> 00:32:39,279 Uh, and I know, like you said, Claude has been uh probably the 683 00:32:39,279 --> 00:32:40,640 hottest one this year. 684 00:32:40,880 --> 00:32:43,920 And uh this is a contrarian take, but but I do think that 685 00:32:43,920 --> 00:32:48,960 GPT is probably a better model overall for the stock-related 686 00:32:48,960 --> 00:32:49,680 work that we do. 687 00:32:49,839 --> 00:32:52,160 It generally is our top-performing model in most of 688 00:32:52,160 --> 00:32:53,279 the things we look at. 689 00:32:53,519 --> 00:32:55,359 Um, so that's piece number one. 690 00:32:55,440 --> 00:32:57,599 On the open source stuff, I actually think that to your 691 00:32:57,599 --> 00:33:00,960 point is is really interesting because Kimmy's had a moment. 692 00:33:01,039 --> 00:33:04,960 Uh Nematron is is actually, I think, quite good as well on the 693 00:33:04,960 --> 00:33:05,839 American side. 694 00:33:06,000 --> 00:33:09,839 Um, GLM, I know, was was hot before Kimi came out with K3. 695 00:33:10,400 --> 00:33:14,480 Um I think we're gonna see kind of the same thing there in this 696 00:33:14,480 --> 00:33:19,039 open source space, which is uh just like the closed source or 697 00:33:19,039 --> 00:33:23,440 the closed source models, very rapidly sort of pushing the 698 00:33:23,440 --> 00:33:28,000 boundary up, um, but maybe not quite hitting the closed source. 699 00:33:28,079 --> 00:33:32,240 That's sort of my bet, is that we'll see a persistent gap 700 00:33:32,240 --> 00:33:36,000 between closed and open, uh, mainly because I think a lot of 701 00:33:36,000 --> 00:33:38,960 it is, you know, based on distillation, how fast some of 702 00:33:38,960 --> 00:33:42,400 the the uh the open source stuff is moving, which is totally 703 00:33:42,400 --> 00:33:42,640 fine. 704 00:33:42,799 --> 00:33:45,440 But if you really want to be on the frontier, and I think you do 705 00:33:45,440 --> 00:33:48,880 want that in the investment space where 1% edge is 706 00:33:48,880 --> 00:33:51,680 important, I think you really do have to spend more time looking 707 00:33:51,680 --> 00:33:53,119 at the closed source models. 708 00:33:53,440 --> 00:33:53,759 unknown: Yeah. 709 00:33:54,000 --> 00:33:55,440 SPEAKER_01: Okay, how have you thought about this? 710 00:33:55,599 --> 00:33:59,759 I mean, you you've spent a lot of time with cloud specific 711 00:33:59,759 --> 00:34:03,759 training, cloud co-working in particular, which has been sort 712 00:34:03,759 --> 00:34:08,719 of the default UI for you know mid to small size funds, super 713 00:34:08,880 --> 00:34:11,760 powerful, particularly when you sort of set it up with the right 714 00:34:11,760 --> 00:34:13,920 connectors and the right skills. 715 00:34:14,079 --> 00:34:18,239 If funds have adopted Cloud Desktop, Cloud Cowork, Code, et 716 00:34:18,239 --> 00:34:21,840 cetera, and you see this model evolution, what what what's 717 00:34:21,840 --> 00:34:22,079 next? 718 00:34:22,239 --> 00:34:25,280 What's the counterpunch to that, to that evolution? 719 00:34:25,760 --> 00:34:29,280 SPEAKER_03: It's I folks are scratching their heads right 720 00:34:29,280 --> 00:34:29,440 now. 721 00:34:29,679 --> 00:34:33,840 Um, I think that the the smaller funds that I work with are 722 00:34:34,239 --> 00:34:37,199 they're not ready to go to the open source route. 723 00:34:37,360 --> 00:34:40,400 I mean, a lot of these funds use managed, you know, managed IT 724 00:34:40,559 --> 00:34:40,960 services. 725 00:34:41,119 --> 00:34:44,800 Like they don't even have a dedicated IT person in-house, 726 00:34:45,039 --> 00:34:47,920 let alone a software engineer in-house. 727 00:34:48,239 --> 00:34:51,840 So I think the open source question is is not for the kind 728 00:34:51,840 --> 00:34:55,599 of the long tail of of smaller funds without the technical 729 00:34:55,599 --> 00:34:56,400 expertise. 730 00:34:56,880 --> 00:35:03,119 Um, I think that uh there was a huge pull into Claude. 731 00:35:03,199 --> 00:35:09,360 I think some folks are still using Chat GPT uh web because 732 00:35:09,360 --> 00:35:13,119 they haven't like Codex was very scary, you know, just the name 733 00:35:13,119 --> 00:35:14,400 intimidated people. 734 00:35:14,639 --> 00:35:17,199 But Codex is having its co-work moment. 735 00:35:17,360 --> 00:35:20,880 I mean, it had it, and it's very, very good. 736 00:35:21,119 --> 00:35:23,440 The new codex or chat GPT work, whatever. 737 00:35:23,840 --> 00:35:25,599 I don't know, I don't even know what to call it anymore. 738 00:35:26,079 --> 00:35:26,880 They need to fix that. 739 00:35:27,039 --> 00:35:28,480 They've got a branding problem, you're right. 740 00:35:28,719 --> 00:35:30,000 They've got a big branding problem. 741 00:35:30,079 --> 00:35:32,400 I still call it codex just so people know what I'm talking 742 00:35:32,400 --> 00:35:32,639 about. 743 00:35:33,199 --> 00:35:36,400 Um, and and I think some folks are starting, the folks that 744 00:35:36,400 --> 00:35:42,079 kept their Chat GPT subscription open are starting to dabble. 745 00:35:42,320 --> 00:35:47,039 I've found that in in the folks I talk to, the in the 746 00:35:47,039 --> 00:35:52,480 intelligence so much of the way LMs are being used is is like 747 00:35:52,480 --> 00:35:55,920 the end-to-end workflow or as much of the workflow that you 748 00:35:55,920 --> 00:35:56,639 can capture. 749 00:35:56,880 --> 00:36:00,800 So, yes, obviously you want the model that is 2%, 3% better. 750 00:36:00,960 --> 00:36:04,159 But if the harness isn't there, then that raw intelligence 751 00:36:04,159 --> 00:36:07,280 doesn't really matter to my clients. 752 00:36:07,679 --> 00:36:11,599 And so they are kind of poking around on codecs. 753 00:36:11,840 --> 00:36:15,039 I think that until they fix the branding problem though, no 754 00:36:15,039 --> 00:36:18,320 one's really pulled the trigger because they no one understands 755 00:36:18,320 --> 00:36:22,480 what codex is, but it's really cowork. 756 00:36:22,800 --> 00:36:25,920 Uh, and I think that's that's the problem that people haven't 757 00:36:25,920 --> 00:36:26,559 uh had. 758 00:36:26,639 --> 00:36:30,000 And I think that just the I the issue of cost, I think one of 759 00:36:30,000 --> 00:36:33,519 the things that's really interesting about the ChatGB, 760 00:36:33,679 --> 00:36:38,079 the codex versus co-ork conversation is that as Doug 761 00:36:38,239 --> 00:36:41,920 said, the just the 5.6 model series is very good. 762 00:36:42,000 --> 00:36:43,679 It's very good at agentic work. 763 00:36:43,760 --> 00:36:48,239 I can't speak at the the nuances of financial analysis, the edges 764 00:36:48,239 --> 00:36:52,000 of financial analysis, but it's also significantly cheaper. 765 00:36:52,719 --> 00:36:56,880 And uh, I just as a as a tangent, I have got a$200 Claude 766 00:36:57,039 --> 00:37:02,320 Max plan that I I I personally that I get close to using up um 767 00:37:02,880 --> 00:37:03,760 regularly. 768 00:37:04,000 --> 00:37:09,440 And I've got a$20 GPT plan and they're generous on their 769 00:37:09,440 --> 00:37:14,559 resets, but it I it takes a long time for me to hit the$20 limit 770 00:37:14,559 --> 00:37:16,800 on the GPT uh plan. 771 00:37:16,960 --> 00:37:19,280 So I think that folks are trying to figure it out. 772 00:37:19,440 --> 00:37:23,039 I think eventually they'll get to a dual place, but that's 773 00:37:23,039 --> 00:37:26,880 really like an IT and a mind share question until there's 774 00:37:26,880 --> 00:37:31,039 like a 50%, 20% delta in performance that's not there at 775 00:37:31,039 --> 00:37:31,679 the moment. 776 00:37:32,639 --> 00:37:33,280 SPEAKER_01: Yeah. 777 00:37:33,920 --> 00:37:36,800 Question for you, Doug, and curious your perspective on this 778 00:37:36,880 --> 00:37:37,280 too, K. 779 00:37:37,440 --> 00:37:42,559 Like in in other areas, we've seen the sort of uh wrapper 780 00:37:42,559 --> 00:37:45,440 businesses become ascendant again, like Harvey and Lagora. 781 00:37:45,679 --> 00:37:48,159 It sort of solves a lot of these, like you know, model 782 00:37:48,159 --> 00:37:51,920 router and data wholesale aggregation, you know, 783 00:37:52,079 --> 00:37:53,440 commercial questions. 784 00:37:53,679 --> 00:37:56,960 Um, we've seen that in certain areas of finance, like 785 00:37:56,960 --> 00:38:00,000 investment banking with Rogo really sort of ascendant. 786 00:38:00,320 --> 00:38:03,760 We haven't seen that yet in investing yet. 787 00:38:04,000 --> 00:38:07,599 Is that due to just the heterogeneity of our craft? 788 00:38:07,760 --> 00:38:10,320 Or do you think that's just a timing issue that we'll start to 789 00:38:10,320 --> 00:38:13,119 see that moment where you see some leaders? 790 00:38:13,280 --> 00:38:16,239 Uh, you know, Alpha Sense is sort of most adjacent to that. 791 00:38:16,400 --> 00:38:19,679 Facts building out an AI workspace, et cetera. 792 00:38:19,840 --> 00:38:24,239 Um Bloomberg's sort of perpetually 18 months behind the 793 00:38:24,239 --> 00:38:24,880 frontier. 794 00:38:24,960 --> 00:38:27,440 But I'm curious if you've you know heard any interesting 795 00:38:27,440 --> 00:38:30,400 vendor stories you think we'll start to see, you know, sort of 796 00:38:30,400 --> 00:38:33,679 the right you you workspace being this finance-specific 797 00:38:33,679 --> 00:38:34,320 workspace. 798 00:38:36,000 --> 00:38:40,320 SPEAKER_02: I haven't heard any um sort of under-the-radar 799 00:38:40,320 --> 00:38:42,159 companies or workspaces yet. 800 00:38:42,239 --> 00:38:45,519 Um, but I wouldn't be surprised if somebody does try something 801 00:38:45,519 --> 00:38:48,559 here, and maybe it is an AlphaSense or somebody like 802 00:38:48,559 --> 00:38:48,719 that. 803 00:38:48,800 --> 00:38:50,960 It would sort of make sense they might try. 804 00:38:51,199 --> 00:38:54,400 Um, we've talked to, and and I know Kay, you've talked to a ton 805 00:38:54,400 --> 00:38:55,760 of people in this space too. 806 00:38:55,920 --> 00:38:59,599 Um, we've talked to a handful of pretty big asset managers uh 807 00:38:59,599 --> 00:39:03,119 over the past year just about, you know, other ways we could 808 00:39:03,119 --> 00:39:04,239 help them solve problems. 809 00:39:04,320 --> 00:39:07,679 And and one of the things that always is an issue or a sticking 810 00:39:07,679 --> 00:39:11,199 point with us working with them in any capacity is how do you 811 00:39:11,199 --> 00:39:13,840 get through uh compliance-related issues? 812 00:39:14,000 --> 00:39:17,360 How do you get through um sort of data sovereignty-related 813 00:39:17,519 --> 00:39:20,559 issues and making sure that, you know, if they're sharing data 814 00:39:20,559 --> 00:39:23,599 with us in some way, our systems aren't using that data. 815 00:39:23,760 --> 00:39:26,480 That data is not available to any of their competitors, maybe 816 00:39:26,559 --> 00:39:27,679 that are using the same system. 817 00:39:27,760 --> 00:39:32,000 And so um, these are all, I think, solvable problems, but I 818 00:39:32,000 --> 00:39:36,719 think they're um they're easier to solve technically than they 819 00:39:36,719 --> 00:39:39,519 are to solve organizationally, if that makes sense. 820 00:39:39,760 --> 00:39:43,039 And I wonder if that level of headache is the thing that's 821 00:39:43,039 --> 00:39:46,880 maybe holding us back more than you know, just the reality that 822 00:39:46,880 --> 00:39:48,320 the technology is probably ready. 823 00:39:48,400 --> 00:39:51,280 It's you know, are the institutions ready and and are 824 00:39:51,840 --> 00:39:55,119 um very large financial institutions, which which I 825 00:39:55,119 --> 00:39:58,159 would argue and I hope you guys would agree are are maybe not 826 00:39:58,159 --> 00:40:01,920 the most aggressive as it pertains to uh being innovative, 827 00:40:02,159 --> 00:40:05,360 are just really slow to try to figure it out, too. 828 00:40:05,679 --> 00:40:06,320 unknown: Yeah. 829 00:40:07,599 --> 00:40:11,840 SPEAKER_03: I I would say that I have seen a small resurgence of 830 00:40:11,840 --> 00:40:15,360 kind of like the wrapper vertical tools, like this first 831 00:40:15,679 --> 00:40:18,559 pass tool or something that just gives me all the investor 832 00:40:18,559 --> 00:40:20,480 transcripts and so on, like cleans it up. 833 00:40:20,559 --> 00:40:23,360 You know, the three steps that you had to do in Claude to get 834 00:40:23,360 --> 00:40:26,320 to sanitize and get that data into the working format. 835 00:40:26,559 --> 00:40:29,519 And people are more open to those, but it's a little bit 836 00:40:29,519 --> 00:40:32,960 aside of your uh outside of your question, Brett. 837 00:40:33,199 --> 00:40:39,840 In terms of the platform, um I think that people wanna people 838 00:40:39,840 --> 00:40:43,679 kind of want the iPhone of all this, just like you push a 839 00:40:43,679 --> 00:40:44,159 button. 840 00:40:44,320 --> 00:40:44,639 Yeah. 841 00:40:45,440 --> 00:40:48,559 And there's a recognition that the minute it becomes the 842 00:40:48,559 --> 00:40:54,000 iPhone, then you've flattened out the edge into the product. 843 00:40:54,800 --> 00:40:55,119 Right. 844 00:40:55,199 --> 00:40:57,760 And so you need to have an iPhone that's more like an 845 00:40:57,760 --> 00:41:01,679 Android where you can customize this widget and this widget and 846 00:41:01,679 --> 00:41:03,920 this widget and this data source. 847 00:41:04,239 --> 00:41:07,199 But then you're kind of back at the same starting problem. 848 00:41:07,599 --> 00:41:11,599 And so I think that people are um it's a big change management 849 00:41:11,599 --> 00:41:11,920 question. 850 00:41:12,000 --> 00:41:15,679 And I think one of them, one of the questions that I think 851 00:41:15,679 --> 00:41:19,440 larger firms will ask is like, what will get people to use the 852 00:41:19,440 --> 00:41:19,920 thing? 853 00:41:20,159 --> 00:41:20,480 Right. 854 00:41:20,639 --> 00:41:23,119 And if it's giving them an iPhone, then give them a damn 855 00:41:23,119 --> 00:41:23,679 iPhone. 856 00:41:24,159 --> 00:41:26,400 But I think that there are always going to be firms that 857 00:41:26,400 --> 00:41:29,199 are gonna say, like, I want you to have the iPhone and I want 858 00:41:29,199 --> 00:41:32,639 you to have Chrome because I want you to take that widget 859 00:41:32,960 --> 00:41:34,960 because we never do the widget that way. 860 00:41:36,000 --> 00:41:36,719 Yeah, yeah. 861 00:41:36,880 --> 00:41:39,199 SPEAKER_01: Yeah, and then the sort of that that intersects the 862 00:41:39,199 --> 00:41:42,960 sort of uh obsolescence or the bitter lesson dynamic. 863 00:41:43,039 --> 00:41:47,519 Like I look back to our fall 25 curriculum where we were 864 00:41:47,519 --> 00:41:51,119 teaching people about chat GPT projects and downloading 865 00:41:51,119 --> 00:41:54,079 documents, uploading them, creating system prompts, all of 866 00:41:54,079 --> 00:41:57,039 this, which is like completely obsolete now. 867 00:41:57,119 --> 00:41:57,199 Yeah. 868 00:41:57,360 --> 00:42:00,960 So how do you deal with the change management problem while 869 00:42:01,039 --> 00:42:04,079 the sort of like the baseline of what needs to be changed is 870 00:42:04,079 --> 00:42:08,480 constantly evolving and continually up for debate? 871 00:42:08,559 --> 00:42:09,440 It's uh it's a hairy. 872 00:42:09,840 --> 00:42:11,119 Lots of work for people like us. 873 00:42:11,280 --> 00:42:13,119 It's a hairy, it's a hairy problem. 874 00:42:14,000 --> 00:42:17,599 SPEAKER_02: I I think, you know, and I think a lot of um a lot of 875 00:42:17,599 --> 00:42:20,480 finance organizations, larger ones, even mid-sized ones, I 876 00:42:20,480 --> 00:42:24,719 think they want to try to build this themselves in certain ways 877 00:42:24,800 --> 00:42:27,599 and for certain reasons, compliance related, right, 878 00:42:27,760 --> 00:42:29,280 control related, all these things. 879 00:42:29,440 --> 00:42:32,559 But working with somebody who knows and has a vision for it, I 880 00:42:32,559 --> 00:42:33,199 think is super important. 881 00:42:33,360 --> 00:42:35,840 Kay, I know you work with a lot of companies that that are 882 00:42:35,840 --> 00:42:38,079 trying to figure this problem out. 883 00:42:38,400 --> 00:42:41,920 Um, but the but the thing I think every firm needs to think 884 00:42:41,920 --> 00:42:44,079 about is they all need to operate a little bit like a 885 00:42:44,079 --> 00:42:47,280 startup, you know, and and to the extent that you're not 886 00:42:47,280 --> 00:42:50,960 willing to maybe break a few rules and and really try to 887 00:42:50,960 --> 00:42:54,960 figure things out um in a unique way because things are changing 888 00:42:54,960 --> 00:42:55,360 so fast. 889 00:42:55,440 --> 00:42:58,559 I think it will put you behind the firms that are willing to 890 00:42:58,559 --> 00:43:02,079 take the risk, invest in the technology, invest in, you know, 891 00:43:02,159 --> 00:43:04,800 maybe even some of the governance risk and and whatever 892 00:43:04,800 --> 00:43:08,880 else comes with that and understands what this new world 893 00:43:08,880 --> 00:43:11,519 looks like with LLMs being a much bigger part of all 894 00:43:11,519 --> 00:43:12,880 investment-related processes. 895 00:43:13,119 --> 00:43:14,880 SPEAKER_01: Yeah, the tricky part is if you look at like the 896 00:43:14,880 --> 00:43:18,880 top 100 hedge funds, you know, certainly near the top of that 897 00:43:18,880 --> 00:43:19,519 stack. 898 00:43:19,679 --> 00:43:23,119 I think most funds will build that in-house, but even quite 899 00:43:23,119 --> 00:43:26,400 large funds, you know, multi-eight-figure, you know, 900 00:43:26,559 --> 00:43:30,079 AUM, billion AUM funds, like they're many of these teams, the 901 00:43:30,079 --> 00:43:33,440 investment teams are still quite small, you know, 10 to 15 people 902 00:43:33,599 --> 00:43:37,039 and you know, relatively small internal IT teams. 903 00:43:37,280 --> 00:43:41,119 This is a big lift to build these, not just to build the 904 00:43:41,119 --> 00:43:44,400 first version, but to sort of navigate this obsolescence risk 905 00:43:44,400 --> 00:43:45,760 and eval structures. 906 00:43:45,840 --> 00:43:50,079 And um it's a it's not a it's not an immaterial expense, it's 907 00:43:50,079 --> 00:43:54,800 not an immaterial um you know, corporate strategy shift. 908 00:43:55,760 --> 00:43:58,880 SPEAKER_03: And and to to just piggyback off what Doug said, 909 00:43:59,039 --> 00:44:00,719 the need for a Vision, right? 910 00:44:00,800 --> 00:44:04,400 You you you cut it outside of the citadels where, like, you 911 00:44:04,400 --> 00:44:08,159 know, there's hundreds of people that are paid a lot of money to 912 00:44:08,159 --> 00:44:09,119 solve this problem. 913 00:44:09,280 --> 00:44:13,519 But once you get to that tier below, you need to have someone 914 00:44:13,519 --> 00:44:15,599 in leadership that's AI pilled. 915 00:44:16,079 --> 00:44:19,920 Because they will not break the thing that needs to be broken. 916 00:44:20,079 --> 00:44:21,679 They will not take the risk that needs. 917 00:44:22,000 --> 00:44:25,360 I mean, there are, I still talk to very successful firms like we 918 00:44:25,360 --> 00:44:28,079 have web search turned off on cloud. 919 00:44:28,320 --> 00:44:29,199 Like, what? 920 00:44:29,440 --> 00:44:32,719 Like, I didn't even know that that was a feature, uh, let 921 00:44:32,719 --> 00:44:34,320 alone in 2026. 922 00:44:34,800 --> 00:44:35,119 Right? 923 00:44:35,280 --> 00:44:39,440 Like, you need someone with that vision because from that vision, 924 00:44:39,519 --> 00:44:42,960 you're gonna nudge compliance, you're gonna take a governance 925 00:44:42,960 --> 00:44:46,880 risk, you're gonna pay some money to give some analyst who's 926 00:44:46,960 --> 00:44:51,280 you know also AI pilled a bunch of fable tokens to go go figure 927 00:44:51,280 --> 00:44:52,079 some stuff out. 928 00:44:52,239 --> 00:44:56,719 And that that is a hard, those are a lot of hard pieces. 929 00:44:56,880 --> 00:44:59,280 You know, it's like playing chess, like with all these 930 00:44:59,280 --> 00:45:02,480 pieces to move around, and someone added, and then the 931 00:45:02,480 --> 00:45:06,000 chess table board is vibrating at the same time. 932 00:45:06,079 --> 00:45:08,239 So, like the piece that you thought was in the corners 933 00:45:08,480 --> 00:45:13,360 actually have moved over one cell by you know, within you 934 00:45:13,360 --> 00:45:14,480 know, a quarter. 935 00:45:16,320 --> 00:45:17,920 SPEAKER_01: Yeah, yeah. 936 00:45:18,639 --> 00:45:21,920 Yeah, all right, Doug, I didn't want to be asked, I asked Fable. 937 00:45:22,480 --> 00:45:25,440 Build me a stock portfolio of 10 stocks. 938 00:45:25,760 --> 00:45:26,159 All right. 939 00:45:27,360 --> 00:45:30,480 Uh compute substrate is 35% of the portfolio. 940 00:45:30,639 --> 00:45:36,239 Nvidia is a 12% position, TSM 10, AVGO 8%, VRT 5%. 941 00:45:36,639 --> 00:45:39,679 I didn't I didn't tell them make an AI portfolio, I just said 942 00:45:39,920 --> 00:45:42,000 build me a portfolio of 10 stocks. 943 00:45:42,239 --> 00:45:45,760 The distribution abstraction layers, 35%, Microsoft, Meta, 944 00:45:45,920 --> 00:45:47,840 Amazon, and now are another 35%. 945 00:45:48,639 --> 00:45:52,559 The token beneficiaries, Palantir, uh, Spotify, another 946 00:45:52,559 --> 00:45:52,880 20%. 947 00:45:53,199 --> 00:45:56,239 And then it's giving me a 10% cash or hedge. 948 00:45:56,480 --> 00:45:59,920 So without any prompting, it built like a super concentrated, 949 00:46:00,000 --> 00:46:02,639 high octane AI portfolio. 950 00:46:02,800 --> 00:46:06,480 No biotech, no consumer staples, no REITs, no banks. 951 00:46:06,800 --> 00:46:11,039 Um so yeah, over the last 36 months, that probably shows like 952 00:46:11,039 --> 00:46:14,800 massive alpha, but how do you like, you know, sort of when the 953 00:46:14,800 --> 00:46:17,679 tide goes out, this portfolio is getting absolutely smacked, 954 00:46:17,840 --> 00:46:18,159 right? 955 00:46:18,559 --> 00:46:19,119 SPEAKER_02: Yep. 956 00:46:19,440 --> 00:46:22,719 Yeah, and I think that goes right back to thinking about 957 00:46:22,719 --> 00:46:26,480 context and and what can you give the system to help it 958 00:46:26,480 --> 00:46:27,679 understand the current regime. 959 00:46:27,840 --> 00:46:30,159 And if you think about how Fable, right? 960 00:46:30,239 --> 00:46:33,199 If you just use Fable for a general task today, the 961 00:46:33,199 --> 00:46:38,559 knowledge cutoff, I believe, on Fable is about April of 26. 962 00:46:39,119 --> 00:46:41,119 Um, so think about the window, right? 963 00:46:41,280 --> 00:46:46,719 The last kind of information the model had was Q1, ostensibly, of 964 00:46:46,719 --> 00:46:47,280 this year. 965 00:46:47,519 --> 00:46:51,440 AI trade was ripping then before we had the pullback kind of uh 966 00:46:51,440 --> 00:46:53,760 March and then ripped right back after that. 967 00:46:54,000 --> 00:46:57,199 Um, so the model is probably thinking, okay, well, that's 968 00:46:57,199 --> 00:46:59,760 what the world looks like, and so I want to have this really 969 00:46:59,760 --> 00:47:01,920 super AI aggressive portfolio. 970 00:47:02,159 --> 00:47:07,519 Uming that if you put anything into a model, finance related or 971 00:47:07,519 --> 00:47:10,880 otherwise, that the model is going to have some sort of bias 972 00:47:10,880 --> 00:47:15,440 related to its most up-to-date context, I think is piece number 973 00:47:15,440 --> 00:47:18,239 one is understanding that you're probably gonna get something 974 00:47:18,239 --> 00:47:21,280 that looked really smart based on something that happened in 975 00:47:21,280 --> 00:47:21,760 the past. 976 00:47:21,840 --> 00:47:24,800 And your job is to figure out how you can get it current 977 00:47:24,800 --> 00:47:28,159 information so that it understands the real world and 978 00:47:28,159 --> 00:47:30,079 hopefully can update its priors. 979 00:47:30,239 --> 00:47:33,119 Um, and that's what we try to do at Intelligent Alpha. 980 00:47:33,199 --> 00:47:36,000 You know, we would uh, in that example, right, if we were just 981 00:47:36,000 --> 00:47:38,880 gonna do a really simple thing with Fable, uh, we'd probably 982 00:47:38,880 --> 00:47:43,280 give it some current information about recent financial reports, 983 00:47:43,440 --> 00:47:47,440 earnings reports, uh, what consensus expectations look 984 00:47:47,440 --> 00:47:49,519 like, um, recent transcripts. 985 00:47:49,599 --> 00:47:52,639 And so then hopefully the model would say, okay, look, the AI 986 00:47:52,639 --> 00:47:54,000 trade has been playing out. 987 00:47:54,159 --> 00:47:57,840 Maybe it would even know we had this July pullback, and uh 988 00:47:57,840 --> 00:48:00,559 hopefully it would adapt in the right way to go forward. 989 00:48:00,719 --> 00:48:03,519 I mean, me as a human, I'm still pretty bullish on the AI trade. 990 00:48:03,599 --> 00:48:07,039 I think after the July washout, um, if you look back at the 991 00:48:07,039 --> 00:48:10,480 dot-com era, it looks a lot like 1998, kind of LTCM. 992 00:48:10,559 --> 00:48:12,480 There's there's kind of some vibes there. 993 00:48:12,719 --> 00:48:16,480 And uh the world got pretty crazy after that in terms of the 994 00:48:16,480 --> 00:48:17,280 dot-com trade. 995 00:48:17,360 --> 00:48:19,760 And so I'm not saying that's exactly what's going to happen, 996 00:48:19,920 --> 00:48:22,800 but I think the AI trade is is kind of far from over. 997 00:48:22,880 --> 00:48:26,000 And and uh I'd be curious to see if the machine would agree with 998 00:48:26,000 --> 00:48:26,320 me. 999 00:48:26,559 --> 00:48:27,199 unknown: Yeah. 1000 00:48:27,679 --> 00:48:29,840 SPEAKER_01: Yeah, I'd say at like a high level, like I've 1001 00:48:30,079 --> 00:48:34,480 been pretty skeptical of this sort of new emerging AI hedge 1002 00:48:34,480 --> 00:48:38,239 fund, AI asset management business where you let the 1003 00:48:38,239 --> 00:48:39,360 machines pick the stock. 1004 00:48:39,519 --> 00:48:42,639 I mean, the last few weeks have been a little bit more 1005 00:48:42,639 --> 00:48:43,840 open-minded to it. 1006 00:48:43,920 --> 00:48:47,199 I sort of agree with this sort of context being a critical 1007 00:48:47,199 --> 00:48:50,159 piece and this decomposition of the investment process into 1008 00:48:50,159 --> 00:48:51,360 actual signal. 1009 00:48:51,519 --> 00:48:54,639 Um part of it's just like the recent model, like you know, 1010 00:48:54,800 --> 00:48:57,039 Fable's not perfect, but the Fables whole series, you know, 1011 00:48:57,199 --> 00:49:01,280 Kimmy K3 series of models like are pretty surprising me to the 1012 00:49:01,280 --> 00:49:04,880 upside on a few of these pretty complicated tasks in terms of 1013 00:49:04,880 --> 00:49:07,920 adherence to a to a complicated skills architecture. 1014 00:49:08,159 --> 00:49:11,599 Um so it's gonna be an interesting nine to 18 months to 1015 00:49:11,599 --> 00:49:14,480 see what uh see see see what emerges. 1016 00:49:14,639 --> 00:49:18,719 Uh what advice would you give for for those you know watching 1017 00:49:18,719 --> 00:49:22,079 this pod who are thinking about maybe uh you know building an 1018 00:49:22,079 --> 00:49:24,559 asset management business with with AI? 1019 00:49:25,440 --> 00:49:28,400 SPEAKER_02: Yeah, I think just like in investing, right? 1020 00:49:28,559 --> 00:49:31,760 You want to look six, 12, 18 months out what you think the 1021 00:49:31,760 --> 00:49:33,440 world is gonna look like then. 1022 00:49:33,760 --> 00:49:36,079 Um, because ultimately, if you're right about your 1023 00:49:36,079 --> 00:49:38,159 prediction about the world, hopefully you're you're right 1024 00:49:38,159 --> 00:49:40,239 about what stocks you buy ahead of that. 1025 00:49:40,480 --> 00:49:44,719 And um thinking about building with these models, um, we're 1026 00:49:44,719 --> 00:49:47,599 always trying to figure out, okay, we know they're going to 1027 00:49:47,599 --> 00:49:48,079 get better. 1028 00:49:48,239 --> 00:49:49,280 I think that's a given. 1029 00:49:49,360 --> 00:49:53,039 If anybody is following the AI space in 12 months, these models 1030 00:49:53,039 --> 00:49:54,239 would be more capable. 1031 00:49:54,400 --> 00:49:57,519 Um, but like what are the bottlenecks today that we think 1032 00:49:57,519 --> 00:50:01,119 will be uh unleashed with additional capabilities? 1033 00:50:01,280 --> 00:50:04,239 How can we prepare ourselves to be ready to sort of take 1034 00:50:04,239 --> 00:50:08,159 advantage of those uh things that get unlocked as soon as 1035 00:50:08,159 --> 00:50:10,400 possible and stay at the forefront? 1036 00:50:10,480 --> 00:50:13,599 Because I think uh to the extent you can stay at the forefront. 1037 00:50:13,679 --> 00:50:15,840 I mean, there's this great quote that I say all the time 1038 00:50:15,920 --> 00:50:19,119 internally from Paul Bookheit, um, who was at Google and then Y 1039 00:50:19,280 --> 00:50:21,519 Combinator, he said, if you're in the lead, if you just keep 1040 00:50:21,519 --> 00:50:24,239 running faster than everybody else, no one could ever catch 1041 00:50:24,239 --> 00:50:24,480 you. 1042 00:50:24,639 --> 00:50:27,679 I think that's true no matter what you're trying to do with 1043 00:50:27,679 --> 00:50:28,239 these models. 1044 00:50:28,400 --> 00:50:31,760 If you can really stay on the edge and keep experimenting, 1045 00:50:31,920 --> 00:50:34,400 don't worry about perfection, even though that's really hard 1046 00:50:34,400 --> 00:50:37,039 to say in the investment world, but just keep iterating. 1047 00:50:37,199 --> 00:50:39,840 I think that that will put you in a really good place to find 1048 00:50:39,840 --> 00:50:42,639 ways to generate alpha with these models over time. 1049 00:50:43,039 --> 00:50:43,519 Yeah. 1050 00:50:43,760 --> 00:50:45,599 SPEAKER_01: Yeah, I've been a hater on a lot of these things, 1051 00:50:45,679 --> 00:50:48,320 or at least trying to provide some skeptical, reasoned, 1052 00:50:48,400 --> 00:50:49,760 empirical pushback. 1053 00:50:49,840 --> 00:50:53,679 And one of my friends says, Brett, evidence is a lagging 1054 00:50:53,679 --> 00:50:55,199 indicator in AI, right? 1055 00:50:55,280 --> 00:50:58,719 You sort of have to operate and build the system on faith that 1056 00:50:58,719 --> 00:51:01,440 the thing you can't do today will be possible in three months 1057 00:51:01,519 --> 00:51:02,559 or six months. 1058 00:51:02,719 --> 00:51:07,519 Um I'm trying to take that to heart as I think about you where 1059 00:51:07,519 --> 00:51:09,679 we could be in the spring of 27. 1060 00:51:09,760 --> 00:51:12,000 It's also becoming easier because the evidence of where 1061 00:51:12,000 --> 00:51:14,559 we're at today versus six or nine months ago, if I sort of 1062 00:51:14,559 --> 00:51:18,960 pull up my outputs of what I could do in summer of 25 versus 1063 00:51:19,280 --> 00:51:22,400 summer of 26, it's a fundamentally different, 1064 00:51:22,559 --> 00:51:26,400 different human, uh fundamentally different sort of 1065 00:51:26,400 --> 00:51:27,840 intelligence source today. 1066 00:51:28,000 --> 00:51:32,000 Um so where we're at next summer kind of hurts my brain a little 1067 00:51:32,000 --> 00:51:33,519 bit to to think about. 1068 00:51:33,679 --> 00:51:36,800 Any thoughts, uh any sort of closing thoughts or comments, uh 1069 00:51:37,039 --> 00:51:37,840 uh Kay? 1070 00:51:39,280 --> 00:51:43,679 SPEAKER_03: Yeah, I think um that that the ability to be 1071 00:51:43,679 --> 00:51:45,920 uncomfortable with uncertainty, right? 1072 00:51:46,000 --> 00:51:51,199 And kind of like leaning into it versus kind of uh holding on to 1073 00:51:51,199 --> 00:51:52,320 the determinism. 1074 00:51:52,480 --> 00:51:56,320 But you know what I got from from our talk with Doug today is 1075 00:51:56,639 --> 00:51:59,119 just it's it's messy, right? 1076 00:51:59,280 --> 00:52:02,480 And I think a lot of the conversation, it's an industry 1077 00:52:02,480 --> 00:52:05,199 that doesn't like to be messy for the right reasons. 1078 00:52:05,360 --> 00:52:08,159 Like there's a there's significant penalty penalties 1079 00:52:08,159 --> 00:52:10,559 for being messy in this industry. 1080 00:52:11,039 --> 00:52:14,320 But to the extent that you can kind of bring in that that 1081 00:52:14,320 --> 00:52:16,800 experimentation, which is like what the three of us on this 1082 00:52:16,800 --> 00:52:20,800 episode and many of our clients are doing, you know, again, 1083 00:52:21,360 --> 00:52:25,440 those pockets will lead you to kind of stay, you know, skate to 1084 00:52:25,440 --> 00:52:26,719 where the puck is going. 1085 00:52:27,280 --> 00:52:27,679 SPEAKER_01: Yeah. 1086 00:52:27,920 --> 00:52:30,880 Yeah, one of the hard hardest questions I get is almost a 1087 00:52:30,880 --> 00:52:31,280 simple one. 1088 00:52:31,360 --> 00:52:32,960 It's like, hey, Brett, I want to set this up. 1089 00:52:33,039 --> 00:52:33,760 What do I do? 1090 00:52:33,920 --> 00:52:35,199 You know, what vendor should I use? 1091 00:52:35,360 --> 00:52:36,239 I'm like, I don't know. 1092 00:52:36,320 --> 00:52:39,360 It's kind of like still three, three years in or four years, 1093 00:52:39,519 --> 00:52:43,599 almost four years into the Chat GPT moment, it's still kind of a 1094 00:52:43,599 --> 00:52:44,800 hard question. 1095 00:52:44,880 --> 00:52:46,960 There's not yet an obvious answer. 1096 00:52:47,039 --> 00:52:50,559 It's a little bit more of a process, a journey of 1097 00:52:50,559 --> 00:52:54,079 experimentation and learning, which is a little bit 1098 00:52:54,079 --> 00:52:57,280 unsatisfying, but the comfort with uncertainty is probably the 1099 00:52:57,280 --> 00:52:57,840 exact way. 1100 00:52:57,920 --> 00:52:58,960 I think you nailed it. 1101 00:52:59,119 --> 00:53:00,719 Nailed it, nailed it, nailed it, Kay. 1102 00:53:00,800 --> 00:53:02,480 What what what say you, Doug? 1103 00:53:03,119 --> 00:53:05,599 SPEAKER_02: Yeah, I think on that, on that point, Brett, the 1104 00:53:05,760 --> 00:53:08,800 the best way I always try to tell people to get involved with 1105 00:53:08,800 --> 00:53:11,280 these models for any purpose, whether it's finance or 1106 00:53:11,280 --> 00:53:14,239 otherwise, is you have to find something that you're really 1107 00:53:14,239 --> 00:53:17,280 curious about, and you just get lost in the models doing. 1108 00:53:17,440 --> 00:53:22,159 And that could be designing uh a new clothing line, it could be 1109 00:53:22,159 --> 00:53:25,440 writing some piece of content, it could be trying to pick, you 1110 00:53:25,440 --> 00:53:28,400 know, some super aggressive AI-related stock portfolio, 1111 00:53:28,559 --> 00:53:29,440 whatever it is. 1112 00:53:29,519 --> 00:53:33,519 Um, the more that you can just get lost using the models and 1113 00:53:33,519 --> 00:53:36,719 then understanding like the intricacies of how they work, 1114 00:53:36,960 --> 00:53:40,239 understanding where you can push the limits and the edges and 1115 00:53:40,239 --> 00:53:43,280 what you can have it actually build for you, that's the best 1116 00:53:43,280 --> 00:53:46,079 thing you can do, is just kind of learn it by feel by doing 1117 00:53:46,079 --> 00:53:47,760 something that's really fun for you. 1118 00:53:47,920 --> 00:53:50,800 And then you can take that curiosity, all those learnings 1119 00:53:50,800 --> 00:53:53,519 that you have, and apply them to something that might be a little 1120 00:53:53,519 --> 00:53:55,360 bit more structured and intentional. 1121 00:53:55,440 --> 00:53:57,840 Um, but until you kind of get through that and you just 1122 00:53:57,840 --> 00:54:00,480 experience the models, you know, unfettered, right? 1123 00:54:00,559 --> 00:54:01,840 And just let them go wild. 1124 00:54:02,000 --> 00:54:04,480 I think it's really hard to kind of think about the structured 1125 00:54:04,480 --> 00:54:06,000 thing because you just don't even really know what they're 1126 00:54:06,000 --> 00:54:07,039 capable of yet. 1127 00:54:07,280 --> 00:54:07,920 unknown: Yeah. 1128 00:54:08,719 --> 00:54:11,599 SPEAKER_01: Well, thank you so much for uh for for joining us 1129 00:54:11,599 --> 00:54:11,840 today. 1130 00:54:11,920 --> 00:54:14,159 This has been really fascinating, and uh I'm super 1131 00:54:14,159 --> 00:54:17,119 excited to see, you know, where where you're at on this in six 1132 00:54:17,119 --> 00:54:18,480 months or nine months or 12 months. 1133 00:54:18,559 --> 00:54:19,599 Maybe we'll come back in. 1134 00:54:19,679 --> 00:54:22,800 It's like, you know, we got the keys, we figured it out now. 1135 00:54:22,960 --> 00:54:25,840 So uh thank you so much for for being with us, Doug. 1136 00:54:25,920 --> 00:54:30,079 This was really fun, uh, fun and uh in an interesting time and an 1137 00:54:30,079 --> 00:54:30,880 interesting space. 1138 00:54:31,360 --> 00:54:31,760 SPEAKER_02: Absolutely. 1139 00:54:31,840 --> 00:54:32,639 Thank you guys too. 1140 00:54:33,119 --> 00:54:34,000 SPEAKER_03: Thanks, Doug.