WEBVTT NOTE Sentence-level transcript of https://www.youtube.com/watch?v=vS2IA9pq0Q4 NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/vS2IA9pq0Q4.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 00:00:12.520 --> 00:00:13.120 Welcome everyone. s3 00:00:13.120 --> 00:00:20.320 Looking forward to giving you all a introduction into one of our newest features in Exa, which is version controlled search. s4 00:00:20.320 --> 00:00:23.120 So I'm Ryan, I'm a forward deployed engineer here at Exa. s5 00:00:23.120 --> 00:00:26.000 And today I'm going to give you an overview of two things. s6 00:00:26.000 --> 00:00:27.720 We only have a very short amount of time. s7 00:00:27.720 --> 00:00:32.520 So the first one, I just want to motivate what is Exa, what is agentic web search? s8 00:00:32.520 --> 00:00:42.640 Just because you kind of need to believe in the dream of that to understand why you'd want to use it to do the second thing I'm going to talk about, which is how to backtest financial models using that. s9 00:00:42.640 --> 00:00:48.120 So first, just to understand where I'm at in the room, who here has heard of Exa before or used it? s10 00:00:48.120 --> 00:00:49.680 Okay, I've got a good show of hands. s11 00:00:49.680 --> 00:00:57.360 Who here has ever used like Google back in the day to like do some sort of work or personal use, just like typed into Google to do some sort of search? s12 00:00:57.360 --> 00:01:01.160 Hopefully like everyone in the room, most people had pretty good market share, you might have heard of it. s13 00:01:01.160 --> 00:01:05.160 Last one would be, how many of you now no longer really use Google? s14 00:01:05.160 --> 00:01:11.960 You use ChatGPT or Claude or maybe even Instinct or Muse if you're like at the cutting edge to actually do your searches. s15 00:01:11.960 --> 00:01:13.360 Yeah, most people, right? s16 00:01:13.360 --> 00:01:16.840 So we're going to connect all of these things together. s17 00:01:16.840 --> 00:01:22.600 This last point where a lot of people now are like, "Hey, I don't even really use desktop search anymore. s18 00:01:22.600 --> 00:01:24.400 I kind of ask an agent to do this." s19 00:01:24.400 --> 00:01:26.880 is something that we see more broadly in the market, right? s20 00:01:26.880 --> 00:01:34.600 And the idea is that like agents tend to do a pretty good job at understanding how to search because they're allowed to try multiple, multiple times. s21 00:01:34.600 --> 00:01:35.960 They kind of don't get tired. s22 00:01:35.960 --> 00:01:38.520 They tend to be quite agreeable and work very, very hard. s23 00:01:38.520 --> 00:01:45.680 So when you say like this is the thing that I want, it's going to go out and it's going to do many, many different searches in order to satisfy s24 00:01:45.680 --> 00:01:46.960 like your objective. s25 00:01:46.960 --> 00:01:50.360 And the other thing here is that like they tend to search different to humans. s26 00:01:50.360 --> 00:01:57.120 So as a human, you might put in like a few keywords for the thing and then hope that you get the best results like in the first couple, and otherwise s27 00:01:57.120 --> 00:01:59.880 like maybe you try like a different query and then you'll kind of give up. s28 00:01:59.880 --> 00:02:02.000 Whereas agents will be very verbose. s29 00:02:02.000 --> 00:02:10.000 Um they they tend to describe what they want to search in great detail and will execute many searches in order to understand the space more fully before they respond. s30 00:02:10.000 --> 00:02:14.320 Um and that means that they do way more searches um on on a per task basis. s31 00:02:14.320 --> 00:02:21.400 So, generally the idea is that like, "Hey, nowadays most humans, you're going to be using agents to kind of broker your search tasks." s32 00:02:21.400 --> 00:02:23.120 Um and that's where Exa kind of plugs in. s33 00:02:23.120 --> 00:02:27.720 So, uh when you're working with um like many many agentic systems. s34 00:02:27.720 --> 00:02:41.920 So, whenever Cursor wants to go and search latest documentation to help its coding agent complete its trajectory, whenever you're like talking to Gamma and you're like, "Hey, it'd be great if I had just this like SEC data from like Apple in like 2018 s35 00:02:41.920 --> 00:02:44.240 to complete my slide, like that would make it brilliant." s36 00:02:44.240 --> 00:02:52.640 It's going to use Exa to go out, search the full web, so we have our own crawling infrastructure, our own database, our own embedding model, our own retrieval architecture s37 00:02:52.640 --> 00:02:55.440 to get exactly what it needs to answer its query. s38 00:02:55.440 --> 00:03:04.720 Um this is like a couple of the logos of like the bigger companies, but because this is AI engineer, like here are a few like kind of archetypes that we work across the financial space. s39 00:03:04.720 --> 00:03:13.680 I work with a lot of hedge funds, asset managers, sovereign wealth funds, VCs across like a bunch of different tasks because they also need like the latest information. s40 00:03:13.680 --> 00:03:16.440 So, if you're a hedge funder, you're going to be trading on information. s41 00:03:16.440 --> 00:03:18.080 You're looking in the market for alpha, right? s42 00:03:18.080 --> 00:03:24.680 And for information that allows you to understand how you could position your your organization better. s43 00:03:24.680 --> 00:03:34.640 If you're a um uh an insurer, you're probably understanding like what counterparties you're working with or who you're insuring to make sure that ultimately you're making the determination. s44 00:03:34.640 --> 00:03:38.959 So, you want to make sure you're making it on the best information and that your risk models are the most accurate. s45 00:03:38.959 --> 00:03:41.840 Or even if you're in VC, you're like, "I need the hottest startup. s46 00:03:41.840 --> 00:03:44.200 I need to know the in next instant before they blow up." s47 00:03:44.200 --> 00:03:47.280 And they're not necessarily going to be in like some structured data set. s48 00:03:47.280 --> 00:03:50.239 Like I need to go and find them wherever they exist in the web. s49 00:03:50.239 --> 00:03:53.800 Um that's what they come to Exa to kind of support them for. s50 00:03:53.800 --> 00:04:04.920 Um and then the the kind of to wrap up the last part on like why Exa and the reason why I think now you'd be like, "Okay, I'm I'm building agents or I'm interested in agents and I definitely need a generative web search. s51 00:04:04.920 --> 00:04:06.200 Why Exa?" s52 00:04:06.200 --> 00:04:15.400 Um and it's because like we're a lab um in the sense that we're very very interested in the search market um and in the ways in which agents search and we're hyper focused on that problem. s53 00:04:15.400 --> 00:04:20.880 There's three like core tenants and then we'll talk about how that goes into versioning to segue into the next part of the talk. s54 00:04:20.880 --> 00:04:23.040 But first and quite obviously is like quality, right? s55 00:04:23.040 --> 00:04:31.640 So, you want to make sure that when the agent does that um the tool call and the retrieval, it gets what it needs cuz the fewer tool calls that it needs to make, s56 00:04:31.640 --> 00:04:37.120 the more cost-effective on your inference and also just like the more likely it is to succeed on its overall trajectory. s57 00:04:37.120 --> 00:04:39.640 You're not going to bring in something that like misaligns it. s58 00:04:39.640 --> 00:04:43.080 Um uh we're really really interested in the quality space. s59 00:04:43.080 --> 00:04:56.560 We uh recently released like a new benchmark um which we think more accurately assesses the um the the performance of like the search tools separately to the inference because these two things can be confounded, right? s60 00:04:56.560 --> 00:05:00.560 Like what agent or what model is calling the search and what search is connected to what model. s61 00:05:00.560 --> 00:05:03.600 So, that you can kind of have a look at that to see our latest thinking there. s62 00:05:03.600 --> 00:05:04.840 Secondly, on cost. s63 00:05:04.840 --> 00:05:07.040 So, it's not just the line item of the tool call, right? s64 00:05:07.040 --> 00:05:11.600 Like how much does one search cost because agents can broker many different searches. s65 00:05:11.600 --> 00:05:15.880 Um so, the idea is both like, "Hey, what's the value to this to complete your trajectory?" s66 00:05:15.880 --> 00:05:23.480 But also like what's the impact of this on the total cost of that agent completing its task cuz that's actually the thing that you care about, right? s67 00:05:23.480 --> 00:05:25.800 You're generally less opinionated about one tool call. s68 00:05:25.800 --> 00:05:28.560 It's like, "How much did this overall agent take?" s69 00:05:28.560 --> 00:05:36.960 Uh and that's why like we're we we worked on things like say dynamic highlights, which means that for every single result, we're not we're not just returning full pages. s70 00:05:36.960 --> 00:05:46.720 Uh we're not even just taking one page, which is our previous highlight features, and figuring out what are the exact right snippets on that page that we need to send back to the context window to complete the trajectory. s71 00:05:46.720 --> 00:05:56.920 We now do that across all of the search results, which means that you can actually get higher quality results at 95% fewer tokens than you would at full page text. s72 00:05:56.920 --> 00:06:06.440 We have a blog about this, it's very interesting, but basically it means you'll save a lot of money downstream on all of the input tokens you'll be spending on inference for the agent to complete its trajectory. s73 00:06:06.440 --> 00:06:12.000 Last one on latency, so because we own our own infrastructure, we can do very very low latency search. s74 00:06:12.000 --> 00:06:17.200 Um I'm just going to touch on the other side of the spectrum, which is like very long-running deep research. s75 00:06:17.200 --> 00:06:28.200 So we have ExaAgent Ultra, which we released um like not too long ago, maybe a few weeks ago, which is like state-of-the-art on um a lot of benchmarks around like very very long-running s76 00:06:28.200 --> 00:06:28.640 um search. s77 00:06:28.640 --> 00:06:31.040 And that's because we can do a lot of things under the hood. s78 00:06:31.040 --> 00:06:39.680 We have obviously we own the search harness, but we can also be opinionated about that how that interacts with the agent to harness do these like very very long-running research task. s79 00:06:39.680 --> 00:06:41.919 So highly recommend you check those things out. s80 00:06:41.919 --> 00:06:45.760 Um this is just a segway to be like, "Okay, Exa really cares about this space. s81 00:06:45.760 --> 00:06:47.600 I think that their search is going to be good." s82 00:06:47.600 --> 00:06:51.760 To motivate the next part, which is really interesting is the topic of this talk, which is versioning. s83 00:06:51.760 --> 00:06:54.480 So just recently we released ExaSnapshot. s84 00:06:54.480 --> 00:06:59.880 Uh it's 400 billion plus snapshots of the web, um and you can search them by date. s85 00:06:59.880 --> 00:07:05.040 Um and that's really really interesting in the um in the financial services industry. s86 00:07:05.040 --> 00:07:09.360 We'll get into that, but first I'm just going to talk a little bit about like what this basically means. s87 00:07:09.360 --> 00:07:17.240 So uh we're we're a search company, so we're constantly uh searching across the web for the latest information that we can add to our index. s88 00:07:17.240 --> 00:07:20.760 Um and every single time we do it, we check it for we're checking for freshness. s89 00:07:20.760 --> 00:07:26.320 Because ultimately you want to have the latest live up-to-date information at the point at which you're searching. s90 00:07:26.320 --> 00:07:34.320 But because we're doing all of those checks, it means that we can capture a version of those URLs at the point where we checked it and have like a version history. s91 00:07:34.320 --> 00:07:35.800 Um and you can access that in two ways. s92 00:07:35.800 --> 00:07:40.720 One is like, "I know the URL I want to find as it was at or before a given date." s93 00:07:40.720 --> 00:07:43.560 And so you can give us a URL and you say, "This is the date that I want it. s94 00:07:43.560 --> 00:07:49.320 I want it at that date or before and serve it to me as you would have if I had requested it at that point." s95 00:07:49.320 --> 00:07:52.920 The second one, which is even more exciting, is this concept of version search. s96 00:07:52.920 --> 00:08:02.120 So, where a search API, and the idea is like, "Hey, what would happen if I had executed that search but at a prior point, right?" s97 00:08:02.120 --> 00:08:06.600 So, I only get results that would have been served at that point or before. s98 00:08:06.600 --> 00:08:09.640 Um that has very like interesting implications. s99 00:08:09.640 --> 00:08:12.680 I'm just going to do one like very small example here. s100 00:08:12.680 --> 00:08:18.800 So, OpenAI recently I say as of like October 1st. s101 00:08:18.800 --> 00:08:24.080 So, on the left-hand side here, you'll see a snapshot search as it was in October 1st. s102 00:08:24.080 --> 00:08:28.840 And on the right-hand side was a live search as of like a couple of days ago when I made this slide. s103 00:08:28.840 --> 00:08:41.280 Um but uh at that time in early October, like there was some discussions that like OpenAI had a like annualized like run rate around like the 70 billion, right? s104 00:08:41.280 --> 00:08:46.080 Um and that was leading to certain ways in which the market was projecting how and where they would be valued. s105 00:08:46.080 --> 00:08:49.720 And then more recently, there was some and that reporting was originally by Axios. s106 00:08:49.720 --> 00:08:56.280 And then more recently, there were some publications that were reporting like, "Actually, it's closer to 50 billion for fun accounting reasons, right?" s107 00:08:56.280 --> 00:09:03.280 Like it turns out that they were counting gross revenue when actually like some of their revenue was coming from cloud providers. s108 00:09:03.280 --> 00:09:07.120 And technically, the way that you should account for that would be that there would be a vendor pass-through. s109 00:09:07.120 --> 00:09:09.000 So, net annualized revenue is better. s110 00:09:09.000 --> 00:09:12.600 That's how what some what some other players were reporting, etc., etc., right? s111 00:09:12.600 --> 00:09:21.839 But basically, it meant that the the the net annualized run rate was like kind of down forecast from 70 billion down to 50 billion, right? s112 00:09:21.839 --> 00:09:25.960 Now, this is just like a financial thing that some of you may or may not find interesting. s113 00:09:25.960 --> 00:09:37.320 But the so what for version search is like, "Let's say you're in the interest of creating a product or a solution or an agent that is made to that that will that its job is to to find these things, right?" s114 00:09:37.320 --> 00:09:41.680 You can now go back and be like, "Okay, before everyone figured it out, right? s115 00:09:41.680 --> 00:09:51.040 Before the market and before some financial analyst published the think piece that was like slamming down on this particular thing and everyone knew about it, what information existed on the internet, right? s116 00:09:51.040 --> 00:09:55.440 Like was there an SEC filing that would have revealed this thing in an accounting footnote? s117 00:09:55.440 --> 00:10:05.480 Like was there some other reporting from some other vendor that maybe is like you would consider is like less reputable, but actually was able to find that kind of thing and and maybe you should consider them differently." s118 00:10:05.480 --> 00:10:11.640 Um you can you can potentially like trace these things before like the rest of the market found it. s119 00:10:11.640 --> 00:10:15.960 But most importantly, like not only can you test out that you can backtest that, right? s120 00:10:15.960 --> 00:10:25.440 You could then adjust the ways in which you're setting up your model, you can adjust the ways that you're setting up your agent to be able to be like, "Could it have found this discrepancy?" s121 00:10:25.440 --> 00:10:30.960 Uh and more importantly, could it have found that discrepancy in a way that is generalizable, right? s122 00:10:30.960 --> 00:10:42.960 Let's like create a set of like like 50 sample sets, see if it can perform against that, check it against the holdout and be like, "Okay, I think that my model is now going to be pretty good at finding this kind of feature." s123 00:10:42.960 --> 00:10:48.640 And that's something that's kind of new in this in the in the web search space and that you can backtest that. s124 00:10:48.640 --> 00:11:00.160 Um this is like a like a little diagram just to kind of motivate the idea, but at any like live search that you do with like any existing search product, the idea is it's live, so it has all of the information. s125 00:11:00.160 --> 00:11:06.280 So, in this particular little example, um we're we're trying to predict whether company got acquired or not. s126 00:11:06.280 --> 00:11:10.400 Um in this case, if you do a live search, you're going to find like that the deal was announced, right? s127 00:11:10.400 --> 00:11:13.040 Because that happened, it got released everywhere. s128 00:11:13.040 --> 00:11:18.880 You could um be like, "Hey, I'm trying to train a model that's able to predict whether this company got acquired or not." s129 00:11:18.880 --> 00:11:26.839 So, I'm going to go like 1 day before the announcement, 7 days before the announcement, a month before the announcement and see whether I could or could not predict it, right? s130 00:11:26.839 --> 00:11:36.920 Uh and you can build a set of those companies um and basically train your model um or whatever system it is that it's trying to do this prediction, and see like was it able to reliably s131 00:11:36.920 --> 00:11:39.600 like predict this particular feature. s132 00:11:39.600 --> 00:11:49.160 Um So, I'm going to spend a little bit of time here just talking about how different systems are currently being back-tested with Snapshot. s133 00:11:49.160 --> 00:11:51.520 So, there's like three kind of archetypes here. s134 00:11:51.520 --> 00:11:56.120 The first one is I'm operating like a very traditional model. s135 00:11:56.120 --> 00:12:00.320 Maybe it's ML, maybe it's even more simple than that, like a decision tree or something. s136 00:12:00.320 --> 00:12:07.840 Uh and all I want to do is just gather unstructured data from the version web into some sort of structured set of information. s137 00:12:07.840 --> 00:12:15.800 So, it could be something like software prices, it could be some sentiment analysis, it could be like the terms and conditions of a particular site. s138 00:12:15.800 --> 00:12:20.840 Whatever it is that like powers your model, not all information is continually available. s139 00:12:20.840 --> 00:12:31.880 Like if you're trying to understand software prices, like, you know, the the page of that software price is not necessarily stored in a versioned um a way for that s140 00:12:31.880 --> 00:12:35.480 website in a way that you could reliably access. s141 00:12:35.480 --> 00:12:45.680 Or potentially like you have some like, sentiment analysis that you want to do, but like that's going to be mired in like what point was that sentiment released in time, right? s142 00:12:45.680 --> 00:12:48.240 And you want to be able to get like a reliable set of those signals. s143 00:12:48.240 --> 00:12:52.960 And then you can just transform it into some array of signals that goes into your model and you can predict it. s144 00:12:52.960 --> 00:12:55.280 That's like the sort of basic space. s145 00:12:55.280 --> 00:13:01.240 And some people are just like enriching data that they wouldn't have had access to for their existing models. s146 00:13:01.240 --> 00:13:04.880 So, that's just like sort of like the status quo plus one. s147 00:13:04.880 --> 00:13:08.000 The interesting space is in like the next two areas. s148 00:13:08.000 --> 00:13:11.720 So, the first one is this idea of like adding an agentic feature. s149 00:13:11.720 --> 00:13:15.520 And the And And here is like, okay, well, I have this existing model. s150 00:13:15.520 --> 00:13:23.680 It performs pretty well, but I have an idea that like if I add one or two like agentic features to this model, it would be able to perform better. s151 00:13:23.680 --> 00:13:26.280 So, let's say something like an adverse media scan. s152 00:13:26.280 --> 00:13:35.360 Necessarily like adverse media is going out into the web, searching across it for all of the information that exists, potentially doing it in an agentic way, right? s153 00:13:35.360 --> 00:13:44.120 Where you're doing multiple searches, you're exhausting multiple different trajectories, and then you're going to take all of that and then condense it into some sort of structure. s154 00:13:44.120 --> 00:13:48.280 Um Uh another example here would be to say like equity research in much the same way. s155 00:13:48.280 --> 00:13:50.040 Like, "Hey, I want to go out. s156 00:13:50.040 --> 00:14:00.600 I want to look at the full universe of companies and I want to find like potentially all of their like like estimated uh like rack revenue or head count or all of the features that matter." s157 00:14:00.600 --> 00:14:08.480 Um and the idea is that you'd be using something like adverse search at the point at which you're doing a prediction life to go out and gather all of that information s158 00:14:08.480 --> 00:14:11.120 to then send into your um traditional model. s159 00:14:11.120 --> 00:14:16.000 And you can now use like adverse snapshot to back test those particular features. s160 00:14:16.000 --> 00:14:21.320 So, you can go back and be like, "Hey, if I was to do this like adverse media scan like 1 month before." s161 00:14:21.320 --> 00:14:30.600 Or maybe I've already like onboarded a bunch of people into my portfolio like 4 months ago, what would I have said at the point where they were going to enter my portfolio? s162 00:14:30.600 --> 00:14:33.080 And would how is that tracked against today? s163 00:14:33.080 --> 00:14:39.480 Um and that kind of motivates like especially with a lot of financial services customers like you don't now just have to believe that like search will be better. s164 00:14:39.480 --> 00:14:44.280 You can actually go back and back test that against your existing processes and say, "Hey, it's better." s165 00:14:44.280 --> 00:14:48.560 Um so, that's like the maybe the next evolution like one step plus plus. s166 00:14:48.560 --> 00:14:53.880 Um on the far right-hand side here um is kind of like maybe the future. s167 00:14:53.880 --> 00:14:57.440 So, there's some like startups that are working in this prediction space. s168 00:14:57.440 --> 00:15:04.040 And the idea here is that the full sort of like simulation or prediction is being managed by an agent. s169 00:15:04.040 --> 00:15:08.760 So, in the middle is like an ML model that's maybe doing a prediction or some algorithm or decision tree. s170 00:15:08.760 --> 00:15:13.480 On the right-hand side like it is like an agent that's just going off and doing a bunch of things. s171 00:15:13.480 --> 00:15:18.600 It could be searching internal knowledge, it could be accessing data feeds, it could be doing whatever it is. s172 00:15:18.600 --> 00:15:24.000 But one of the things that it may do that that is to tool call uh and access the web. s173 00:15:24.000 --> 00:15:33.160 And the idea is like now you don't have to try and create some internal corpus that tries to represent what it would be like to access the full web. s174 00:15:33.160 --> 00:15:41.160 You can just be like, "Hey, I want to make sure that I can control that feature so I know that future information is not just going to leak directly into my model." s175 00:15:41.160 --> 00:15:47.160 Like I'm trying to predict company acquisitions through some very very insane like agentic process that's accessing everything. s176 00:15:47.160 --> 00:15:54.840 You can now be like, "All right, let's just make sure that's not just going to have one web search that finds that it was acquired, comes back and predicts it with like 100% s177 00:15:54.840 --> 00:15:55.960 accuracy. s178 00:15:55.960 --> 00:15:59.560 Um so that's kind of the the sort of right-hand side there. s179 00:15:59.560 --> 00:16:06.640 Um I'm just going to give like a little example of like the middle um just to round off um this this this kind of discussion. s180 00:16:06.640 --> 00:16:20.240 Um so uh yesterday we ran like um a kind of extended workshop on snapshot uh and we had a bunch of around about 100 people in the room that were kind of uh running through this the same example we had before about trying to predict company acquisitions. s181 00:16:20.240 --> 00:16:27.720 Um one thing that was kind of interesting is that um the this was like some like one of the example features from the workshop from one of the people s182 00:16:27.720 --> 00:16:30.720 um that uh that that got a relatively high score. s183 00:16:30.720 --> 00:16:39.680 But a thing that was interesting here is like um the just to just to motivate so the first one was this idea that like, "Hey, within the past year did an investor s184 00:16:39.680 --> 00:16:43.680 or a significant shareholder publicly call for a supported sale of the company as a whole?" s185 00:16:43.680 --> 00:16:51.480 And the second one, which is the interesting one, is like, "Hey, did this company's share price um was at least 60% lower than high share price during the past s186 00:16:51.480 --> 00:16:51.839 5 years?" s187 00:16:51.839 --> 00:16:53.880 Which the idea is like it's kind of dipped down. s188 00:16:53.880 --> 00:16:58.960 Um the last bit that's interesting is like, "Hey, is the company named Dominion?" s189 00:16:58.960 --> 00:17:02.480 Now this is a little bit of overfitting for people that know like backtesting. s190 00:17:02.480 --> 00:17:06.800 Um like and the idea here is like this could be something that like an agent could do, right? s191 00:17:06.800 --> 00:17:11.920 If it's trained on like a live set, um it can potentially just overfit to the training data that you have. s192 00:17:11.920 --> 00:17:24.640 But the idea with snapshot is like not only can you go back and test for these things, um you can go back and like set up your your your e-vals and your identic harness in such a way that you can detect these things, but you can also try and check that it generalizes. s193 00:17:24.640 --> 00:17:30.840 And this is something that would be very, very hard to do if you are gated by like testing live, like in production. s194 00:17:30.840 --> 00:17:36.920 And you can only change the ways in which these features are described over time and check against real time, right? s195 00:17:36.920 --> 00:17:45.040 you might be waiting for months to try and see how and why your identic setup um, actually is predictive or is overfitting in different ways. s196 00:17:45.040 --> 00:18:02.160 Um, that is like you can describe these like quite complex features like um, these ones are like maybe you could imagine trying to recreate in um, uh, in like some sort of like uh, like non-semantic search, but you can go as crazy as you want within the within existing like s197 00:18:02.160 --> 00:18:03.640 uh, like extra semantic search world. s198 00:18:03.640 --> 00:18:16.520 So, you can say things like, "Hey, I want to see if there was any adverse media against like reputable VPs that are operating like um, uh, that are operating within VC funds that have like been on the AI Engineer podcast s199 00:18:16.520 --> 00:18:17.840 and they've said something bad about it, right? s200 00:18:17.840 --> 00:18:25.120 Like you can go as as hard as you want on these features that you think are going to be predictive cuz that's what core extra search allows, but then you can back test it, right? s201 00:18:25.120 --> 00:18:27.400 and see if it's performative or not. s202 00:18:27.400 --> 00:18:30.240 Um, then I'm just going to kind of like recap. s203 00:18:30.240 --> 00:18:38.840 So, the first thing that we talked about here is like um, using Exa because it's quality cost-effective identic web search across every latency. s204 00:18:38.840 --> 00:18:49.000 Um, we have lots of blogs around this, so encourage you if you've not tried Exa um, to have a go at this um, because the first thing you want to know is like, "Hey, do I actually think that this is going to add some benefit to my model?" s205 00:18:49.000 --> 00:18:59.160 The second thing as we talked about is like, "Hey, using Exa snapshot to be able to like back test these financial models or whichever model of risk risk model prediction model with these web search features s206 00:18:59.160 --> 00:19:04.760 in order to see like, hey, does adding it uh, add value to to you and your identic processes?" s207 00:19:04.760 --> 00:19:07.040 Thank you so much.