WEBVTT

NOTE Sentence-level transcript of https://www.youtube.com/watch?v=59AA5kIoqjA

NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/59AA5kIoqjA.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
00:00:12.559 --> 00:00:14.240
How's everybody doing?

s3
00:00:14.240 --> 00:00:15.599
Good.

s4
00:00:15.599 --> 00:00:16.320
All right.

s5
00:00:16.320 --> 00:00:17.680
Pretty crazy times we live in.

s6
00:00:17.680 --> 00:00:18.880
Oh, my microphone.

s7
00:00:18.880 --> 00:00:20.320
Um, okay.

s8
00:00:20.320 --> 00:00:25.199
I am very excited to tell you all about perfect search built for AI agents.

s9
00:00:25.199 --> 00:00:28.400
Uh, and I'm going to say a lot of crazy things in this talk, so bear with me.

s10
00:00:28.400 --> 00:00:38.399
uh they are all true and I will tell you what is exa uh the story of exa so how we got here and then where we're going okay cool

s11
00:00:38.399 --> 00:00:50.079
if you take away anything from this talk it is this uh slide uh this is showing web searches per day over the past 30 years uh from humans and AIS obviously it was all humans until

s12
00:00:50.079 --> 00:00:59.600
around uh 2020s and now we're in 2026 and actually this year we expect the number of searches from AI systems to exceed that of humans.

s13
00:00:59.600 --> 00:01:00.399
Pretty crazy.

s14
00:01:00.399 --> 00:01:03.600
And then in the next few years, it should be a thousand times more.

s15
00:01:03.600 --> 00:01:10.799
So AI, AI systems, AI products, whatever AI system you use, sum together will will search a thousand times more than humans.

s16
00:01:10.799 --> 00:01:13.680
That's pretty uh pretty crazy world that that we're getting into.

s17
00:01:13.680 --> 00:01:18.240
And the entire ecosystem of search is changing because of it.

s18
00:01:18.240 --> 00:01:21.200
So Exa is the search engine for AI agents.

s19
00:01:21.200 --> 00:01:27.119
You know, we were the first ones to be like we were the first search engine for AI and now things are getting uh kind of wild.

s20
00:01:27.119 --> 00:01:31.520
uh we now serve over 5,000 companies over 400,000 developers.

s21
00:01:31.520 --> 00:01:33.280
I see some customers in the audience.

s22
00:01:33.280 --> 00:01:37.439
Uh we serve you know very diverse set of agents from coding agents like cursor.

s23
00:01:37.439 --> 00:01:44.000
So if you use cursor at some point the cursor agent decides to search for the latest technical documentation or the news it'll be using ex under the hood.

s24
00:01:44.000 --> 00:01:45.840
We serve go to market agents like HubSpot.

s25
00:01:45.840 --> 00:01:51.040
So uh we help uh their users get you know really high quality lists of companies to sell to.

s26
00:01:51.040 --> 00:01:52.799
We serve all sorts of financial agents.

s27
00:01:52.799 --> 00:01:57.920
Uh I was in New York a few weeks ago and pretty much everyone there is now building financial agents and they need the best financial data.

s28
00:01:57.920 --> 00:02:05.280
Um yeah and really just a huge diversity of agents from uh from labs to to to YC startups.

s29
00:02:05.280 --> 00:02:07.600
Okay, but how do we get here?

s30
00:02:07.600 --> 00:02:15.599
So what's the why of EXO that to me that's always been the most important and there are a lot of ways to say the problem but the short way is just misinformed anarchy.

s31
00:02:15.599 --> 00:02:16.800
This is what we're trying to avoid.

s32
00:02:16.800 --> 00:02:18.480
We want to create the opposite of this.

s33
00:02:18.480 --> 00:02:19.920
Uh so what what does this mean?

s34
00:02:19.920 --> 00:02:23.440
Well, this is the internet or it's a visual depiction of the internet.

s35
00:02:23.440 --> 00:02:29.120
You know, you got a bunch of pages, you got blog posts, you got company websites, you got images, you got tweets, you got all sorts of things.

s36
00:02:29.120 --> 00:02:33.440
Uh, and the web is really really big, right?

s37
00:02:33.440 --> 00:02:35.120
This is showing a few thousand pages.

s38
00:02:35.120 --> 00:02:37.360
The web is, you know, on the order of a trillion pages.

s39
00:02:37.360 --> 00:02:40.319
So, uh, you know, a million times bigger than this.

s40
00:02:40.319 --> 00:02:43.360
And it contains a huge amount of the world's information.

s41
00:02:43.360 --> 00:02:52.720
Uh and that means that you know if this is just readily available you could find if you could go to any link and just get all the world's information then I'm sure we all walk around with like deep understanding of everything right

s42
00:02:52.720 --> 00:02:57.760
obviously not uh it's messy uh and so uh it's crazy and it can't fit in our heads.

s43
00:02:57.760 --> 00:03:03.360
So we need information tools that could help synthesize this or filter it into the the things you need to know.

s44
00:03:03.360 --> 00:03:07.360
And you know we have a tool called Google and it's a pretty solid tool.

s45
00:03:07.360 --> 00:03:13.280
It could get you things like the Costco homepage or information about Taylor Swift or whatever you want to search, but it's not perfect, right?

s46
00:03:13.280 --> 00:03:16.239
And I love this example where you type shirts without stripes.

s47
00:03:16.239 --> 00:03:18.400
Uh, and if you notice, you get shirts with stripes.

s48
00:03:18.400 --> 00:03:19.200
Why is it doing that?

s49
00:03:19.200 --> 00:03:22.959
Well, it's not it's not trying to be a database of the world's information that gives you exactly what you want.

s50
00:03:22.959 --> 00:03:24.640
It's kind of like a recommendation engine.

s51
00:03:24.640 --> 00:03:27.040
Um, and you can see this with more examples.

s52
00:03:27.040 --> 00:03:32.640
Um, so find me everyone in Singapore who works on AI search and any blog post or research paper they've written.

s53
00:03:32.640 --> 00:03:36.319
I bet you've never typed in anything like this to Google because you know it's not going to work, right?

s54
00:03:36.319 --> 00:03:46.000
you're going to get like uh links and documents that contain some of those words, but not like actually a database result of all, you know, 412 people who match

s55
00:03:46.000 --> 00:03:47.680
and and it gets pretty serious, right?

s56
00:03:47.680 --> 00:03:49.360
Like I'm a citizen.

s57
00:03:49.360 --> 00:03:51.040
I'm trying to be informed about what's going on in the world.

s58
00:03:51.040 --> 00:03:56.159
I want to find the most important US news across all media or, you know, US news articles, whatever it is.

s59
00:03:56.159 --> 00:03:57.760
You just don't trust Google to give you that, right?

s60
00:03:57.760 --> 00:03:59.680
It's just going to get it's just going to recommend some things.

s61
00:03:59.680 --> 00:04:01.920
It's it's almost akin to like social media in the sense.

s62
00:04:01.920 --> 00:04:04.480
It's kind of like a recommendation engine.

s63
00:04:04.480 --> 00:04:06.000
Okay.

s64
00:04:06.000 --> 00:04:11.200
So that means that no one really has a complete understanding of anything.

s65
00:04:11.200 --> 00:04:16.320
I actually when I walk around like SF or wherever I'm walking around, I see people I often think like no one knows what's going on in the world.

s66
00:04:16.320 --> 00:04:17.199
Everyone's like this.

s67
00:04:17.199 --> 00:04:18.720
Uh including myself, you know.

s68
00:04:18.720 --> 00:04:23.040
Uh you could imagine uh this person on the right on their phone like trying to find a new job.

s69
00:04:23.040 --> 00:04:25.120
They're looking for biotech companies to work for.

s70
00:04:25.120 --> 00:04:28.240
Uh are they going to get like all the possible biotech companies that match?

s71
00:04:28.240 --> 00:04:32.080
No. Uh so there there's always going to be like this unknown of what's out there.

s72
00:04:32.080 --> 00:04:44.479
or uh you know maybe this other person with the headphones uh maybe they're trying to yeah try they want to understand what's going on uh in in in in in some region of the world they're just not going to have a deep understanding they can't not only can they not find the information they might not be able

s73
00:04:44.479 --> 00:04:55.120
to trust the information so we basically are in a world where we live without this like key information infrastructure that is so critical and I think this is extremely important so important that if we don't fix this problem

s74
00:04:55.120 --> 00:05:03.199
I believe we get a world that looks like this a dystopia I'm not kidding uh this is AI generated uh version of San San Francisco in 2035.

s75
00:05:03.199 --> 00:05:07.680
Um, and it's basically a world where no one no person really understands what's going on.

s76
00:05:07.680 --> 00:05:14.720
If people don't understand what's going on, then as we have this crazy AI technology that we all are talking about today that's coming and the world is getting way more powerful.

s77
00:05:14.720 --> 00:05:16.000
There's going to be conflict and all these things.

s78
00:05:16.000 --> 00:05:18.160
If if people don't know what's going on in the world, this is very bad.

s79
00:05:18.160 --> 00:05:19.440
Like we will be manipulated.

s80
00:05:19.440 --> 00:05:22.479
We will make really bad decisions as individuals and as a society.

s81
00:05:22.479 --> 00:05:26.160
And I think this if we, you know, if we could fix this problem, it'll be way better.

s82
00:05:26.160 --> 00:05:34.479
Um, and when I think about all the possible problems that there are that are really important and neglected, to me, like solving information is the most important neglected problem.

s83
00:05:34.479 --> 00:05:36.320
Okay, so that's the why of XA.

s84
00:05:36.320 --> 00:05:38.560
Uh, quick story of how we got here.

s85
00:05:38.560 --> 00:05:44.639
So, I I I've been thinking about this problem for a very long time, way before even 2021 when we started, uh, even since high school.

s86
00:05:44.639 --> 00:05:53.440
Uh, but I think what was really cool is that in 2021, it suddenly became possible in our eyes to build a new type of search engine because transformers had gotten really good.

s87
00:05:53.440 --> 00:05:55.759
So this is a time when like GB3 had recently come out.

s88
00:05:55.759 --> 00:05:56.800
GB3 was like magical.

s89
00:05:56.800 --> 00:05:59.520
You know, you you type in a paragraph of text and it fully understands you.

s90
00:05:59.520 --> 00:06:03.360
At the same time, you know, as we saw with Google, it's like it doesn't fully understand you.

s91
00:06:03.360 --> 00:06:10.400
And so what if you could combine uh the power of GB3 with a search engine and maybe you could have perfect search over the world's information.

s92
00:06:10.400 --> 00:06:17.520
And actually the thought experiment that always drove me was like, you know, if we take a query, a complex query, and a document, and we run GB3 over it, and we say, does this match?

s93
00:06:17.520 --> 00:06:19.600
It'll be it'll do a really good job of saying, does it match?

s94
00:06:19.600 --> 00:06:24.240
Now do that over a trillion documents for every search and you get a perfect search engine or near perfect.

s95
00:06:24.240 --> 00:06:26.720
Uh the problem is that would cost like $10 million per search.

s96
00:06:26.720 --> 00:06:28.720
So it really becomes an interesting optimization problem.

s97
00:06:28.720 --> 00:06:31.759
How do you like billionx or trillionx reduce the cost of that?

s98
00:06:31.759 --> 00:06:36.319
So that's kind of like the ideas that started Exa is actually the first day of Exa 2021.

s99
00:06:36.319 --> 00:06:40.080
It's actually by the way our 5y year anniversary as of uh as of as of yesterday.

s100
00:06:40.080 --> 00:06:41.919
So um yeah been a great time.

s101
00:06:41.919 --> 00:06:46.160
Um yeah I wish I took a better selfie here but this was the first first day.

s102
00:06:46.160 --> 00:06:53.840
Um and uh and uh it exo is basically built on the idea that look like uh traditional search engines they use keywords.

s103
00:06:53.840 --> 00:06:58.560
Keywords are are very efficient and and they get you they can handle simple queries.

s104
00:06:58.560 --> 00:07:05.680
But if you want to handle more complex queries, you just need to to to use neural networks and particularly embeddings are a way of like encoding.

s105
00:07:05.680 --> 00:07:09.680
You can't run like I said like a neural network over every document uh for every query.

s106
00:07:09.680 --> 00:07:13.840
But you can pre-process every document into some sort of structure like an embedding.

s107
00:07:13.840 --> 00:07:18.240
Uh and then you could use and then that that captures a lot of the intelligence of neural network and then you could use those embeddings.

s108
00:07:18.240 --> 00:07:26.960
Of course embeddings have al their own problems and often you want to combine embeddings and keywords but certainly embeddings are a big part of of the picture and that's how you can handle shirts without stripes.

s109
00:07:26.960 --> 00:07:28.160
We can handle those kind of clear.

s110
00:07:28.160 --> 00:07:30.479
Uh another way of saying it is just stack more layers.

s111
00:07:30.479 --> 00:07:31.840
Um very bitter lesson pill.

s112
00:07:31.840 --> 00:07:34.639
This is we are very early on to being uh bitter lesson.

s113
00:07:34.639 --> 00:07:35.680
I don't know if you know this meme.

s114
00:07:35.680 --> 00:07:37.195
If you don't it probably looks really weird.

s115
00:07:37.195 --> 00:07:37.840
[laughter]

s116
00:07:37.840 --> 00:07:38.400
Okay.

s117
00:07:38.400 --> 00:07:40.800
Um but anyway, so we were very brutal left field.

s118
00:07:40.800 --> 00:07:41.840
So we did some crazy things, right?

s119
00:07:41.840 --> 00:07:43.199
We we raised a couple million dollars.

s120
00:07:43.199 --> 00:07:44.560
We spent half of it on a GPU cluster.

s121
00:07:44.560 --> 00:07:45.919
That was crazy at the time.

s122
00:07:45.919 --> 00:07:57.520
Um we did a huge amount of research for for for really years just heads down and and we did a lot of we were very new to search to be honest like we're just really obsessed with the problem and so we invented a lot of new stuff that I still haven't seen

s123
00:07:57.520 --> 00:07:59.360
uh uh even today.

s124
00:07:59.360 --> 00:08:01.680
Um and so just yeah history of Exo.

s125
00:08:01.680 --> 00:08:06.319
So 2022 so this is like a year and a half after starting uh we we were called Metaphor at the time.

s126
00:08:06.319 --> 00:08:07.919
Um some of you might know it.

s127
00:08:07.919 --> 00:08:10.879
uh we launched our first uh search engine to to the world.

s128
00:08:10.879 --> 00:08:13.840
Um and it was pretty exciting like it was a new way of doing search.

s129
00:08:13.840 --> 00:08:15.919
A lot of people were really excited about it.

s130
00:08:15.919 --> 00:08:18.960
The next big thing that happened two weeks later was CHPT came out.

s131
00:08:18.960 --> 00:08:20.879
Uh and that really changed the world.

s132
00:08:20.879 --> 00:08:24.000
Uh and like this is what San Francisco looked like at the time if you remember.

s133
00:08:24.000 --> 00:08:26.319
Um and this is the exit team at the time.

s134
00:08:26.319 --> 00:08:29.520
Uh all right.

s135
00:08:29.520 --> 00:08:39.839
Um, but uh everything was saved when we got this uh this this message message on on Twitter from this person whatever uh who wanted an API access to our search engine.

s136
00:08:39.839 --> 00:08:42.640
And that was really weird because we were never thinking that oh this is going to be an API.

s137
00:08:42.640 --> 00:08:44.320
We were just trying to build a better search engine in Google.

s138
00:08:44.320 --> 00:08:45.440
We'll figure out how to make money later.

s139
00:08:45.440 --> 00:08:46.800
Uh then someone asked us for an API.

s140
00:08:46.800 --> 00:08:48.080
We were like no we don't have an API.

s141
00:08:48.080 --> 00:08:48.560
Sorry.

s142
00:08:48.560 --> 00:08:55.279
Uh but then like we started getting uh more requests for API access including uh you know my roommate uh who lived downstairs.

s143
00:08:55.279 --> 00:09:05.920
And then we very quickly realized wait a sec like there's a business model like okay well what we realized was AIS need search because like the the argument is basically look even GB5

s144
00:09:05.920 --> 00:09:16.720
gigantic model it's tiny in comparison to the internet right so like these systems always need to search you're never going to have GB 6 GB 7 it's not going to be able to like just know everything about the world it needs to be connected to a retrieval engine

s145
00:09:16.720 --> 00:09:27.040
uh and and that was a really interesting insight because the these things now need a search API uh right and so we pretty quickly realized okay wait AI is going to search the web in In fact, they're going to search the web way more than humans.

s146
00:09:27.040 --> 00:09:31.279
Um, and they're going to search in very different ways.

s147
00:09:31.279 --> 00:09:36.080
Um, so, uh, this is example of what humans, uh, this looks like when humans search, right?

s148
00:09:36.080 --> 00:09:37.040
They search simple queries.

s149
00:09:37.040 --> 00:09:39.120
This is what Google was made for, like SpaceX news.

s150
00:09:39.120 --> 00:09:40.000
It's good at that.

s151
00:09:40.000 --> 00:09:42.240
Um, but an AI system is very different, right?

s152
00:09:42.240 --> 00:09:53.120
It kind of looks like this information guzzler creature that's like insane and and like it would be crazy if the same search engine that was optimal for humans was also optimal for these AI systems.

s153
00:09:53.120 --> 00:10:02.399
So anyway, uh, you know, we realized, okay, uh, if we build a search, uh, API for these AI agents, uh, or it wasn't called AI agents at the time, it was just AIS for LM.

s154
00:10:02.399 --> 00:10:04.880
Uh, then, uh, that we could make money from that.

s155
00:10:04.880 --> 00:10:07.760
Uh, that's a nice business model, and we think it's going to grow really fast.

s156
00:10:07.760 --> 00:10:12.560
Uh, and also the beautiful thing is it matches our initial our our mission, which is perfect search, right?

s157
00:10:12.560 --> 00:10:14.640
Like AI systems really want perfect search.

s158
00:10:14.640 --> 00:10:16.079
They don't want SEO.

s159
00:10:16.079 --> 00:10:17.680
Uh, they don't want ads.

s160
00:10:17.680 --> 00:10:22.240
They just want almost like a database of the world's information, which is what we were always trying to build.

s161
00:10:22.240 --> 00:10:27.600
So we built the first search for LMS and yeah actually like in 2023 we said soon AIs will search more than humans.

s162
00:10:27.600 --> 00:10:30.880
Uh you know uh three years later it's now happening.

s163
00:10:30.880 --> 00:10:32.560
Um okay.

s164
00:10:32.560 --> 00:10:32.720
Yeah.

s165
00:10:32.720 --> 00:10:36.079
And so then next couple years we we built a lot of really crazy stuff.

s166
00:10:36.079 --> 00:10:38.480
Uh it's way more complex than just embedding search.

s167
00:10:38.480 --> 00:10:41.920
It combines all sorts of systems some of which are included here.

s168
00:10:41.920 --> 00:10:43.920
And now you know we're a much bigger team.

s169
00:10:43.920 --> 00:10:45.279
Um and that's how we got here.

s170
00:10:45.279 --> 00:10:46.160
Okay cool.

s171
00:10:46.160 --> 00:10:52.640
So just a quick like what can you do with EXA and then I'll talk about where where we're going how we're going to get the perfect search.

s172
00:10:52.640 --> 00:10:57.120
So the present so right now yeah we're the highest quality information for AI agents.

s173
00:10:57.120 --> 00:10:59.600
Um and you could do all sorts of things.

s174
00:10:59.600 --> 00:11:03.040
So for example uh a lot of people like really complex queries.

s175
00:11:03.040 --> 00:11:07.279
Uh so you want to find every startup funded by YC work and AI give me their batch and status.

s176
00:11:07.279 --> 00:11:08.480
You could do that with Exa.

s177
00:11:08.480 --> 00:11:17.920
Uh, and you could make this like arbitrarily complex like a lot of people don't aren't aware of this, but you could just use Exa and just find really like any uh list of companies, people,

s178
00:11:17.920 --> 00:11:21.839
uh, like like blog posts, news articles that you want.

s179
00:11:21.839 --> 00:11:22.959
Uh, it will take some time.

s180
00:11:22.959 --> 00:11:27.839
So, it'll take, you know, maybe not seconds, might take minutes, but you'll get the information you want.

s181
00:11:27.839 --> 00:11:30.720
At the same time, we also have the fastest search API in the world.

s182
00:11:30.720 --> 00:11:34.399
So, we have a 200 millisecond search endpoint, and that's what it feels like.

s183
00:11:34.399 --> 00:11:35.600
So, it's super fast.

s184
00:11:35.600 --> 00:11:36.800
It's way too fast for humans, right?

s185
00:11:36.800 --> 00:11:37.760
But we're not serving humans.

s186
00:11:37.760 --> 00:11:50.079
uh we're serving AI systems and so like for example we serve some voice agents and uh if you're a voice agent and you talk to the voice agent and it wants to do a search underneath the hood uh you know every millisecond counts you wanted to do a search really fast so that it could go you know

s187
00:11:50.079 --> 00:11:55.680
process it with an LLM and then output the best uh audio back to the customer.

s188
00:11:55.839 --> 00:11:59.760
We also have cool things like uh super efficient token extraction.

s189
00:11:59.760 --> 00:12:04.399
So everyone's talking about uh the compute crunch and how everyone's spending way too much uh on LM.

s190
00:12:04.399 --> 00:12:05.920
Well, actually could help there.

s191
00:12:05.920 --> 00:12:12.160
Uh because you know when the LM makes a query, it wants to get the just the information it needs, just the tokens it needs.

s192
00:12:12.160 --> 00:12:18.480
And so we take uh the documents, we give you 10 documents and then we'll give you only the most important like 100 tokens from those documents.

s193
00:12:18.480 --> 00:12:22.399
And that will save you a lot of downstream LLM costs.

s194
00:12:22.399 --> 00:12:26.720
We also, you know, some people they don't necessarily want uh, you know, snippets from each document.

s195
00:12:26.720 --> 00:12:27.920
They actually want structured output.

s196
00:12:27.920 --> 00:12:35.279
So hey uh you like let's say you're building a recruiting AI agent and you want to find you know all the engineers who recently left their uh big lab job.

s197
00:12:35.279 --> 00:12:41.040
Give me like uh you know uh the the most cited uh uh paper that they've written.

s198
00:12:41.040 --> 00:12:43.519
Give me the the college they went to and and the year they graduated.

s199
00:12:43.519 --> 00:12:44.639
We'll just give you that as structured output.

s200
00:12:44.639 --> 00:12:46.959
It makes it really easy.

s201
00:12:46.959 --> 00:12:51.360
And yeah, I think uh one takeaway here is we're not building one search engine.

s202
00:12:51.360 --> 00:12:52.800
Like perfect search is not one thing.

s203
00:12:52.800 --> 00:12:57.200
It's actually we have 5,000 search engines for each of our 5,000 customers, right?

s204
00:12:57.200 --> 00:13:09.279
uh we want to build our system so it's super flexible because we want every business to be super optimized and that's I think a beautiful thing because like we don't want to declare what is the perfect search we want you to almost you know tell us what exactly do you want do you want super fast do you

s205
00:13:09.279 --> 00:13:22.079
want uh you know the highest possible quality even though it takes minutes do you want to search only over these thousand domains do you want to never search over those thousand domains do you want to search over within this time window do you want to never get product pages some people ask us for that like so there's

s206
00:13:22.079 --> 00:13:32.800
all sorts of things that you could do with X it's very flexible customizable And yeah, I mean our our search quality is really good for these AI agents because we've spent years, you know, doing research

s207
00:13:32.800 --> 00:13:35.200
uh into how do you build a new type of search engine for agents.

s208
00:13:35.200 --> 00:13:40.320
Uh it's even better than Google which was built for humans, which which makes sense.

s209
00:13:40.800 --> 00:13:44.160
Okay, a cool thing that we released recently uh was exon.

s210
00:13:44.160 --> 00:13:50.079
Uh so, you know, agents, they don't really care whether the information is from the public web or from private data sources.

s211
00:13:50.079 --> 00:13:51.360
They just want the truth, right?

s212
00:13:51.360 --> 00:13:55.519
And so it's always been obvious to us that you know we want to assemble all the world's information.

s213
00:13:55.519 --> 00:13:57.920
It's perfect search over all the world's information not just the web.

s214
00:13:57.920 --> 00:14:05.760
And so now we have a system where uh data providers can actually uh partner with Exa so that developers can then get the data from those providers.

s215
00:14:05.760 --> 00:14:17.600
So we're basically creating like a new market a new economy for the for agents uh where you know if you have high valuable data you could get paid uh from all the developers who want their agents to access that data.

s216
00:14:17.600 --> 00:14:19.120
So we're creating this beautiful marketplace.

s217
00:14:19.120 --> 00:14:19.839
I think it's really cool.

s218
00:14:19.839 --> 00:14:29.120
It's like a it's like a free market like the uh the data providers can decide how much they think their content is worth and then developers can decide what data they want

s219
00:14:30.160 --> 00:14:35.040
and then you could do cool things like this uh research AI info companies monthly website visitors similar web.

s220
00:14:35.040 --> 00:14:44.000
Oh went too fast but you could get like basically this is combining uh information from the public web and also information from similar web uh which is not publicly available.

s221
00:14:44.000 --> 00:14:46.320
So you can do queries like that right now.

s222
00:14:46.320 --> 00:14:47.040
Okay.

s223
00:14:47.040 --> 00:14:49.199
So that's where XA is right now.

s224
00:14:49.199 --> 00:14:57.360
But the future has always been super exciting to me and and the goal has always been perfect information and now we know it's for AI agents.

s225
00:14:57.360 --> 00:15:00.320
So instead of a world like this, no bad.

s226
00:15:00.320 --> 00:15:03.760
Uh we want the we want search to kind of feel like this.

s227
00:15:03.760 --> 00:15:07.040
It's actually really hard to to describe what perfect information feels like.

s228
00:15:07.040 --> 00:15:13.680
Uh best way you could say is like literally any information query you have it just works no matter how complex that is.

s229
00:15:13.680 --> 00:15:19.360
Um, another way you could think about it is like it's as if you did a year of research in a second.

s230
00:15:19.360 --> 00:15:26.160
So imagine no matter what you're looking for, whether it's people or or companies or news, imagine you you spent a whole year, you spent all of 2026 just doing research for it.

s231
00:15:26.160 --> 00:15:27.199
You get that in a second.

s232
00:15:27.199 --> 00:15:30.240
That that should give you a sense of what perfect uh information feels like.

s233
00:15:30.240 --> 00:15:36.160
And you do that for every search, all the, you know, crazy number of searches that AI agents are going to make.

s234
00:15:36.160 --> 00:15:38.560
And so yeah, we want to move really fast at Exo.

s235
00:15:38.560 --> 00:15:50.079
like we're we're moving extremely like basically every basically a quarter or two quarters now we have as much progress as we did the past 5 years and it keeps being like that it's exponential growth and so yeah our ambitions for 2027 are pretty crazy

s236
00:15:50.079 --> 00:16:00.800
uh we want uh the world to be like this where everyone walks around with like deep understanding of what's going on I think it's particularly important because you know things like the 2028 presidential election are coming out are coming soon and I would love for

s237
00:16:00.800 --> 00:16:08.756
you know the entire world to have access to nearperfect information so that everyone is very informed going into that election you kind of can feel the gravity

s238
00:16:08.756 --> 00:16:09.199
[laughter]

s239
00:16:09.199 --> 00:16:11.839
of what we're doing here of of perfect information.

s240
00:16:11.839 --> 00:16:13.440
I encourage others to try to do it too.

s241
00:16:13.440 --> 00:16:14.639
Um it's very important for the world.

s242
00:16:14.639 --> 00:16:15.600
It's like key information.

s243
00:16:15.600 --> 00:16:17.600
It's key infrastructure.

s244
00:16:17.600 --> 00:16:21.199
And yeah, you have to like that that slide I showed at the beginning, it's not the whole picture, right?

s245
00:16:21.199 --> 00:16:25.360
If you play it out, we're talking a thousand times more searches from AI systems than humans.

s246
00:16:25.360 --> 00:16:26.800
This you can't even capture that on a graph.

s247
00:16:26.800 --> 00:16:32.000
Like this is not that's like 20 times a thousand times be all the way up there like top of a building or something, right?

s248
00:16:32.000 --> 00:16:35.199
So it's crazy what's coming and it's really happening.

s249
00:16:35.199 --> 00:16:45.839
Um basically like you know humans on average search a few times a day on Google but when everyone has you know AI assistants and every software product you use has AIs in it every interaction you're doing it's going to be grounding

s250
00:16:45.839 --> 00:16:52.560
itself in in search so it's going to be a huge number of searches and if each of those searches are you know as as true as possible as near perfect

s251
00:16:52.560 --> 00:16:57.040
uh then then then the world looks like this in 2035.

s252
00:16:57.120 --> 00:17:03.360
Um yeah I I do think that if I do think we're basically our our future is limited by ourselves.

s253
00:17:03.360 --> 00:17:05.919
Like we're basically getting into a world where our technology is so good.

s254
00:17:05.919 --> 00:17:11.120
It's really just a matter of like can we coordinate and just decide together that like on sensible things, right?

s255
00:17:11.120 --> 00:17:16.000
Like if you look at San Francisco, there's amazing things happening here and there's really stupid things happening here at the same time.

s256
00:17:16.000 --> 00:17:19.760
Like that's that's just coordination problems and coordination comes from the information we consume.

s257
00:17:19.760 --> 00:17:21.839
Uh so that's what we're working on.

s258
00:17:21.839 --> 00:17:24.480
You all have a role to play in also getting to this world.

s259
00:17:24.480 --> 00:17:29.360
Uh so thank you all for for for working on whatever passion you're working on and thanks for listening to me.

s260
00:17:29.360 --> 00:17:29.760
All right.

s261
00:17:29.760 --> 00:17:30.632
Thank you.

s262
00:17:30.632 --> 00:17:32.632
[applause]

s263
00:17:45.042 --> 00:17:47.042
[music]