WEBVTT NOTE Sentence-level transcript of https://www.youtube.com/watch?v=UHbODjuviAU NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/UHbODjuviAU.html. A cue ends where the next begins, or 2 s after its last word. s1 00:00:01.309 --> 00:00:03.309 [music] s2 00:00:12.720 --> 00:00:13.480 Hi everyone. s3 00:00:13.480 --> 00:00:15.880 Thank you so much for coming to my talk. s4 00:00:15.880 --> 00:00:16.480 I'm Annika. s5 00:00:16.480 --> 00:00:26.440 I'm the founder and CEO of Moshi and I'm going to be talking about a financial investors context that lives in their head and how we can use that when we're building with machines s6 00:00:26.440 --> 00:00:28.280 today. s7 00:00:28.280 --> 00:00:30.080 So I want to do a quick show of hands. s8 00:00:30.080 --> 00:00:36.720 How many people in this room are overwhelmed by the amount of data they have when making a decision? s9 00:00:37.400 --> 00:00:38.680 Okay. s10 00:00:38.680 --> 00:00:47.080 And how many feel like they've got less attention now today with all the amount of data? s11 00:00:47.320 --> 00:00:51.000 And so that's the exact paradox that financial investors have every day. s12 00:00:51.000 --> 00:00:57.720 So in 2025 last year, the world made about 175 zettabytes of data. s13 00:00:57.720 --> 00:01:02.680 That's exponential higher than what we had two decades ago. s14 00:01:02.680 --> 00:01:12.680 So before we were lucky if we had an internet connection, we had newspapers, books, but nowadays we have all of that plus the context that lives in PDFs, subscriptions, s15 00:01:12.680 --> 00:01:21.320 lots of different podcasts and also a stream of never-ending information coming on X and LinkedIn. s16 00:01:21.400 --> 00:01:23.960 So what does that mean for financial investors? s17 00:01:23.960 --> 00:01:27.960 Because more data does not equate to more information. s18 00:01:27.960 --> 00:01:35.120 It actually just gives everyone the same flood of information, but your attention is the flat line across everything. s19 00:01:35.120 --> 00:01:46.920 So for financial professionals who actually are paid on financial synthesis, what happens between the gap between attention and data, it just gets wider and wider and wider. s20 00:01:47.200 --> 00:01:48.080 So who am I? s21 00:01:48.080 --> 00:01:49.840 I've sat on both sides of the table. s22 00:01:49.840 --> 00:01:55.920 So I was uh I spent my career fascinated by data um and I've spent my career working in it. s23 00:01:55.920 --> 00:02:01.732 So I was a derivative trader for nearly a decade and then I moved into data science and machine learning. s24 00:02:01.732 --> 00:02:10.932 Uh so, I got a full bite scholarship to study machine learning, and I got a paper at ICLR uh working in machine learning systems with NASA. s25 00:02:10.932 --> 00:02:12.732 And now I've married those two. s26 00:02:12.732 --> 00:02:14.452 So, those two worlds being Moshi. s27 00:02:14.452 --> 00:02:26.132 Moshi is your 24/7 quant analyst, which works watching the markets 24/7 and understanding what is going on and what matters to a financial investor. s28 00:02:26.132 --> 00:02:30.252 Having sat on both sides, I can understand what the strengths are for both of them. s29 00:02:30.252 --> 00:02:34.012 And that's what I want to go into deep today. s30 00:02:34.372 --> 00:02:37.852 So, I want to talk about the machine that everyone is trying to beat. s31 00:02:37.852 --> 00:02:46.012 So, a human brain, to give you context, runs on about 20 W, which is the power of a dim light bulb. s32 00:02:46.012 --> 00:02:53.892 Uh so, it's 86 billion neurons in the brain, and we're it's making connections that we still can't understand. s33 00:02:53.892 --> 00:02:57.532 And that is on the power of a light bulb. s34 00:02:57.532 --> 00:03:08.132 And the systems that we are trying to imitate, which we're trying to imitate just a sliver of that power, they need a whole power plant to to run. s35 00:03:08.132 --> 00:03:20.732 So, just in terms of efficiency, the human brain is something like multiple millions multiple times more powerful than whatever systems that we're trying to uh whatever we're trying to build. s36 00:03:20.732 --> 00:03:26.812 And so, that actually what that actually means is it it's important for where you're trying to point that technology. s37 00:03:26.812 --> 00:03:35.012 So, at which point in the decision-making uh cycle do you point the technology, which is AI, which we're what we're building? s38 00:03:35.012 --> 00:03:45.812 Because if the most efficient reasoning engine sits between your two ears, then and it's running on a 20 W light bulb, where does machine learning actually sit in that? s39 00:03:45.812 --> 00:03:56.012 And so, do you try and replace it, which is what AI is trying to what I feel that AI is trying to do every day, or do you actually feed it? s40 00:03:56.532 --> 00:04:01.572 So, let's take a few steps back to understand how a trader's brain or a financial investor actually thinks. s41 00:04:01.572 --> 00:04:03.774 So, what do people actually want? s42 00:04:03.774 --> 00:04:07.294 Any investor you speak to is looking for the same thing, edge. s43 00:04:07.294 --> 00:04:09.094 What is edge? s44 00:04:09.094 --> 00:04:12.374 So, edge is in practice just three things. s45 00:04:12.374 --> 00:04:15.494 I want to know what's going on before anyone else is. s46 00:04:15.494 --> 00:04:17.094 I want to know my blind spots. s47 00:04:17.094 --> 00:04:20.814 So, tell me the second and third order derivatives of what's happening. s48 00:04:20.814 --> 00:04:26.974 And also, I want to have the conviction on what I'm doing so that I can size up on a trade that I believe in. s49 00:04:26.974 --> 00:04:30.574 All those three together give you edge. s50 00:04:30.574 --> 00:04:32.094 So, now here's the problem. s51 00:04:32.094 --> 00:04:41.214 If you're building in this space and everyone has access to the same models, and they have access to the same systems, then what is edge, right? s52 00:04:41.214 --> 00:04:43.254 Is it a skills file? s53 00:04:43.254 --> 00:04:46.334 Is it a whole bunch of MCP connectors? s54 00:04:46.334 --> 00:04:48.854 Is it just a better prompt? s55 00:04:48.854 --> 00:04:51.254 Is it a better model? s56 00:04:51.254 --> 00:04:53.814 Every Everyone has access to all of that. s57 00:04:53.814 --> 00:04:59.214 So, if everyone has access to all of these things, then what is edge? s58 00:04:59.214 --> 00:05:01.814 So, the edge is actually what you can't download. s59 00:05:01.814 --> 00:05:04.254 It what It's what sits in an investor's head. s60 00:05:04.254 --> 00:05:05.854 And how do they get that? s61 00:05:05.854 --> 00:05:12.534 It's decades of working and learning from some of the best traders in the industry and sitting next to them. s62 00:05:12.534 --> 00:05:14.214 And here's the concerning part. s63 00:05:14.214 --> 00:05:20.374 What happens when you have an investor and everything lives in their head when they move firms or they retire? s64 00:05:20.374 --> 00:05:23.654 That IP leaves with them. s65 00:05:24.254 --> 00:05:34.574 So, I think before we even start to think about how AI fits in into all of this, we have to actually understand what a financial investor is doing. s66 00:05:34.574 --> 00:05:38.374 They're keeping up with the markets, which are now moving 24/7. s67 00:05:38.374 --> 00:05:42.054 They're trying to understand what is going on and how they relate to each other. s68 00:05:42.054 --> 00:05:45.214 So, those billions of different data connections. s69 00:05:45.214 --> 00:05:50.294 They're trying to compare and understand the historical significance of the event happening right now. s70 00:05:50.294 --> 00:05:55.214 What happened in the past, why it happened, and what the market did when it happened. s71 00:05:55.214 --> 00:06:00.094 And trying to piece all of that together, and then run it on the models that they have. s72 00:06:00.094 --> 00:06:03.014 And then the last part is actually make a decision. s73 00:06:03.014 --> 00:06:06.855 So, there's a lot that takes place before actually making a financial decision. s74 00:06:06.855 --> 00:06:11.695 So, I just want to keep the last bit with you, which is the decision-making. s75 00:06:12.215 --> 00:06:15.655 So, let's take think about what AI actually changed. s76 00:06:15.655 --> 00:06:28.575 So, before, if you worked in finance, you'd have a team of quants, data scientists, data engineers, traders, who are all working together to build this decision-making engine. s77 00:06:28.575 --> 00:06:33.695 So, it's a whole stack of very expensive and very qualified people. s78 00:06:33.695 --> 00:06:39.055 However, with AI, these tools made you believe that you actually don't need any of these people. s79 00:06:39.055 --> 00:06:46.015 What you need is one person, a few API calls, and you've replaced that whole stack of people. s80 00:06:46.015 --> 00:06:48.495 But, here's the gap that no one actually talks about. s81 00:06:48.495 --> 00:06:57.735 Once you spent your weekend building a dashboard, the moment you actually want to productionize that and share it with one more person, you now have to think about security, s82 00:06:57.735 --> 00:07:03.975 a database, you have to think about storage, reliability, and the biggest thing in all is credits. s83 00:07:03.975 --> 00:07:06.415 How do you actually run the systems? s84 00:07:06.415 --> 00:07:09.175 And so, what we've seen is pricing has changed. s85 00:07:09.175 --> 00:07:14.575 People are no longer pricing by tools, but they're pricing for every API call. s86 00:07:14.575 --> 00:07:19.855 So, now if everyone is building the same tools, what actually makes it different? s87 00:07:19.855 --> 00:07:22.055 And so, there's two types of tools in finance. s88 00:07:22.055 --> 00:07:26.735 One is the tools that you pull data from, and one the tools you push data to. s89 00:07:26.735 --> 00:07:29.135 And so, if you're prompting a model, you are the context. s90 00:07:29.135 --> 00:07:36.455 You're explaining to it, you're arguing with a model that's frozen in time, which is telling you, "Hey, go to bed at 2:00 a.m. It's late now." s91 00:07:36.455 --> 00:07:38.815 when actually it's 3:00 p.m. on a Wednesday. s92 00:07:38.815 --> 00:07:45.415 And all of that, if you think about comparing it to reasoning with a human, it's absolutely no comparison. s93 00:07:45.415 --> 00:07:48.495 And then you've got the type of tool which pushes data to you. s94 00:07:48.495 --> 00:07:53.935 And that's the tool where it's flooding you with information, and that's a data engineering problem. s95 00:07:53.935 --> 00:08:03.535 So, when you have these two tools, and you've got different problems that you're solving for, one is context, and then the other one is the amount of data. s96 00:08:03.535 --> 00:08:11.217 You've got to think about what the traders actually using, and what what they think about in each of these different tools. s97 00:08:11.297 --> 00:08:20.737 And the one thing that it we're seeing is quietly breaking is that in finance, when you join a bank or you join a fund, you sit next to a trader or a team s98 00:08:20.737 --> 00:08:23.417 and you absorb everything that's going on. s99 00:08:23.417 --> 00:08:26.217 And you learn from them and you just listen to everything. s100 00:08:26.217 --> 00:08:30.217 You have one ear on the desk and one ear around everything else that's going on. s101 00:08:30.217 --> 00:08:32.897 And it's consistency that pays off. s102 00:08:32.897 --> 00:08:42.217 And if everyone is going to be running their own teams as one-person pods, then what happens to that future generation of context that's coming in? s103 00:08:42.217 --> 00:08:46.617 And that's the real cost that right now people are not pricing in. s104 00:08:46.617 --> 00:08:51.177 And so in finance, you can't just hand the decision-making loop to a team of agents. s105 00:08:51.177 --> 00:08:51.577 Why? s106 00:08:51.577 --> 00:08:53.657 Because the other side of the trade is a client. s107 00:08:53.657 --> 00:08:57.217 If it's in a commodity trade, it's a cargo, a physical cargo. s108 00:08:57.217 --> 00:09:00.657 If it's a company, it's a team of people. s109 00:09:00.657 --> 00:09:08.537 And so what happens in finance when something goes wrong is you need a human to go and stand up and say, "I made the decision for this reason." s110 00:09:08.537 --> 00:09:12.457 If an agent leaks your API key, all it says is, "Oops, I'm sorry. s111 00:09:12.457 --> 00:09:13.377 I shouldn't have done that. s112 00:09:13.377 --> 00:09:14.617 Please rotate it." s113 00:09:14.617 --> 00:09:25.377 And you can't have that in finance because um like an agent is just self-correcting and the one small error is just exponentially getting bigger. s114 00:09:25.897 --> 00:09:35.617 So, this is the question that I've been needing to, which is if the brain is the most powerful tool we have, why are we spending everything trying to push it out of the loop? s115 00:09:35.617 --> 00:09:41.217 Why don't we flip that and instead of replacing it, why don't we just feed it? s116 00:09:41.257 --> 00:09:49.657 And so if you think about what's in a trader's head, they've got everything leading up to the uh to the trade itself. s117 00:09:49.657 --> 00:09:53.337 But before that, there's a whole bunch of other work that takes place. s118 00:09:53.337 --> 00:09:58.657 They're reading, they're gathering information, they're cross-checking, uh they're watching the market. s119 00:09:58.657 --> 00:10:03.897 Because we have so much information now, the biggest question people have is what is the truth? s120 00:10:03.897 --> 00:10:06.697 What is what is actually the different sources saying? s121 00:10:06.697 --> 00:10:09.017 What is the divergence of information? s122 00:10:09.017 --> 00:10:11.258 How can I trust what is out there? s123 00:10:11.258 --> 00:10:15.458 And once I've done all of that, then I've actually got the decision. s124 00:10:15.458 --> 00:10:28.538 So, if we split it into two different parts, we understand that some of it, well, is all the grind, the reading, the gathering, the data cleaning, and some of it is judgment, which is the decision. s125 00:10:28.538 --> 00:10:30.978 And machines are incredible at the grind. s126 00:10:30.978 --> 00:10:32.738 They're incredible at cleaning data. s127 00:10:32.738 --> 00:10:35.138 They're incredible at gathering data. s128 00:10:35.138 --> 00:10:38.578 And then the human is incredible at the decision making. s129 00:10:38.578 --> 00:10:43.258 So, why not just split those into two separate parts? s130 00:10:43.378 --> 00:10:46.338 And this is what brings us to Moshi, which is what we've built. s131 00:10:46.338 --> 00:10:50.378 It's a 24/7 quant analyst for institutional traders. s132 00:10:50.378 --> 00:11:02.298 And what that means is we are running a team of six different agents, which are built in a deterministic manner, which are watching the market 24/7, more data than is humanly possible, s133 00:11:02.298 --> 00:11:14.618 plus the investors' data, which lives in PDFs, emails, different subscriptions, understanding all of that in a way that is making the the trader a better decision maker. s134 00:11:14.618 --> 00:11:18.698 So, Moshi has read 2.4 billion data points. s135 00:11:18.698 --> 00:11:24.818 And in the 4 seconds it took you to read that sentence, Moshi has learned 300 new things. s136 00:11:25.298 --> 00:11:30.818 And in those 300 new things that it's learned, that's new context for the model. s137 00:11:30.818 --> 00:11:36.578 The issue with that is new context is worthless if you don't actually know, is it trustworthy? s138 00:11:36.578 --> 00:11:38.618 Where does it come from? s139 00:11:38.618 --> 00:11:42.378 Do I even value this analyst's opinion? s140 00:11:42.378 --> 00:11:48.498 Because in finance, even if it's a really confident answer, if it's wrong, then it's worth nothing. s141 00:11:48.498 --> 00:11:49.458 And so, what do we do? s142 00:11:49.458 --> 00:11:50.138 We measure it. s143 00:11:50.138 --> 00:12:01.898 We measure the hallucination, we measure every claim, we see how it's traceable, and we also give the user the assumptions that we've had and how we've led to that decision. s144 00:12:01.898 --> 00:12:10.818 Because if an investor asks you about your opinion, they don't just want to know what you think, they want to know how you came to that reasoning. s145 00:12:11.458 --> 00:12:18.019 And so, what we do is we leave the decision-making with the human because we believe that that is what humans do best. s146 00:12:18.019 --> 00:12:20.499 There's no machine more powerful than the brain. s147 00:12:20.499 --> 00:12:24.539 But everything else that you need to run is what we give you. s148 00:12:24.539 --> 00:12:33.139 So, we believe that every single firm is running these desks of quants, of data engineers, of data scientists, and that's what we give you. s149 00:12:33.139 --> 00:12:36.379 Because it's impossible to run that per human. s150 00:12:36.379 --> 00:12:40.379 It's impossible to run your own quant and data science team, and it also doesn't make sense. s151 00:12:40.379 --> 00:12:44.219 Economics-wise, it just is too much to run per person. s152 00:12:44.219 --> 00:12:45.139 And so, that's what we do. s153 00:12:45.139 --> 00:12:54.459 We run an institutional-grade analyst how it works in bigger institutions to actually give you that conviction to make better trades. s154 00:12:54.459 --> 00:12:56.459 So, what does that actually buy you? s155 00:12:56.459 --> 00:13:05.699 It buys you more efficiency, up to 35% more, so that you can actually spend time to have the headspace to have more conviction on the trades that you want to do. s156 00:13:05.699 --> 00:13:11.259 So, if you speak to a lot of traders, they'll say to you, "Hey, I don't want to speak to anyone all day. s157 00:13:11.259 --> 00:13:12.939 I just want to trade." s158 00:13:12.939 --> 00:13:20.819 So, why are you giving these people just LLMs to talk to and to debate, "Hey, is FOMC today or is it tomorrow?" s159 00:13:20.819 --> 00:13:23.219 And then the LLM is saying it's tomorrow when it's today. s160 00:13:23.219 --> 00:13:25.979 That's just a complete waste of time. s161 00:13:25.979 --> 00:13:29.979 And so, that's the opposite of what investors want. s162 00:13:29.979 --> 00:13:34.459 They want something that can give them the power to reason more and use their brain. s163 00:13:34.459 --> 00:13:35.579 So, Mosaic isn't you. s164 00:13:35.579 --> 00:13:37.099 Mosaic isn't a model. s165 00:13:37.099 --> 00:13:44.979 It's It is you on steroids, which is the same instinct but fed by all the data that lives in different parts. s166 00:13:44.979 --> 00:13:51.659 So, that you can actually spend your time decision-making instead of debating when FOMC is. s167 00:13:52.019 --> 00:13:59.219 In In terms of the scale, we watch over 2.4 billion data points, and it's growing every millisecond. s168 00:13:59.219 --> 00:14:07.539 And so, that's decades of derivatives trading knowledge around the clock, which is what it No No one person can build. s169 00:14:07.539 --> 00:14:09.299 And then it's the pushing of the data. s170 00:14:09.299 --> 00:14:14.539 So, when you're pulling data from an LLM, you are giving it the context on an old model. s171 00:14:14.539 --> 00:14:23.381 When you are getting pushed new data, it's forcing your brain to think about new things and make connections in ways that you didn't initially think about. s172 00:14:23.381 --> 00:14:34.701 So, that means that you're then thinking about these blind spots which you've been told about, which you are then trying to understand, okay, well, I wouldn't have caught this, so what am I going to do with this new information? s173 00:14:34.701 --> 00:14:43.781 And then it's the deep dive, the ability to deep dive into one thing with a 360° view of information that you wouldn't have had before. s174 00:14:43.781 --> 00:14:49.181 And so we do that with the big data and all of the specific data. s175 00:14:50.341 --> 00:15:01.981 And so, how does it differ from a tool that you're building on the weekend when a tool that you need to build 24/7 on every market on big data, which is running s176 00:15:01.981 --> 00:15:07.261 in a way that you can understand the context, which is reliable and trustworthy, that becomes infrastructure. s177 00:15:07.261 --> 00:15:20.021 It's no longer a weekend tool because you need it to be something that other people can trust and you can trace back because that's what you have with a human, which is someone being able to stand up and say, "Hey, I made this decision because of XYZ." s178 00:15:20.021 --> 00:15:25.381 Not calling an agent who's never who's just going to apologize to you. s179 00:15:25.621 --> 00:15:36.861 So, the one question that I think we haven't answered yet, which is if real edge lives in a human brain, how much of a machine can it actually how much can a machine actually capture? s180 00:15:36.861 --> 00:15:45.661 And I don't think we have an answer to that yet, um and I think whoever answers it is not building a product, they're building a new way of thinking in finance, s181 00:15:45.661 --> 00:15:48.501 and that's what we're trying to explore. s182 00:15:49.301 --> 00:15:58.621 So, we've been building very quietly with uh a few of the biggest players in the trading industry in finance, and so this is the first time we're opening up to talk to the public s183 00:15:58.621 --> 00:16:08.341 uh because we believe that every single person should have access to institutional grade analysis without having to run their own data science or quant team. s184 00:16:08.341 --> 00:16:10.981 Um so, if that's interesting to you, come talk to me. s185 00:16:10.981 --> 00:16:15.981 I love talking about everything to do with data and finance and financial engineering. s186 00:16:15.981 --> 00:16:20.582 Um and yeah, before you trade, we want people to talk to Moshi. s187 00:16:20.582 --> 00:16:22.782 Thank you.