1 00:00:21,333 --> 00:00:24,093 S1: All right. Welcome to unsupervised learning. This is Daniel Miessler. 2 00:00:24,093 --> 00:00:27,633 S1: And today I'm super excited to announce a project I've 3 00:00:27,633 --> 00:00:30,033 S1: been wanting to talk about for a very long time 4 00:00:30,033 --> 00:00:35,703 S1: called substrate. Okay, let's get into the project itself. So 5 00:00:35,703 --> 00:00:39,573 S1: what is it exactly? That is really the question. What 6 00:00:39,573 --> 00:00:43,863 S1: substrate is is an open source framework for human understanding, 7 00:00:43,863 --> 00:00:47,613 S1: meaning and progress. And you might be inclined to say, 8 00:00:47,613 --> 00:00:51,463 S1: what the hell does that mean? And it's a great question. Right? 9 00:00:51,463 --> 00:00:54,223 S1: So the purpose of the project is to make things 10 00:00:54,223 --> 00:01:01,123 S1: that matter to humans more transparent. Discussable. And ultimately, because 11 00:01:01,123 --> 00:01:06,043 S1: they're transparent and discussable, they'll be more fixable. So what 12 00:01:06,043 --> 00:01:08,833 S1: kind of things are we talking about? So we're calling 13 00:01:08,833 --> 00:01:13,243 S1: these substrate components. And these are the components of human meaning. Right. 14 00:01:13,243 --> 00:01:16,363 S1: When we talk about understanding, meaning and progress, these are 15 00:01:16,363 --> 00:01:20,923 S1: the pieces that we're actually talking about. So collections of 16 00:01:20,923 --> 00:01:24,193 S1: things okay. So the first thing is an idea collections 17 00:01:24,193 --> 00:01:28,153 S1: of ideas a list of human novel ideas, problems a 18 00:01:28,153 --> 00:01:32,263 S1: list of our most important human problems, our beliefs, our models, 19 00:01:32,263 --> 00:01:36,493 S1: which are our ways of conceptualizing reality, frames a list 20 00:01:36,493 --> 00:01:40,133 S1: of narratives or lenses for perceiving reality. A list of 21 00:01:40,133 --> 00:01:45,023 S1: solutions that correspond to problems. Information sources. So you'll have, 22 00:01:45,023 --> 00:01:50,333 S1: like New York Times, The Hill, Breitbart, uh, lots of 23 00:01:50,333 --> 00:01:57,683 S1: different sources of information, government sources, individual sources, different media organizations. 24 00:01:57,683 --> 00:02:00,353 S1: It's got to be a comprehensive list, and it's got 25 00:02:00,353 --> 00:02:03,803 S1: to span different political ideologies. That's really important here as 26 00:02:03,803 --> 00:02:08,123 S1: we'll see later people. So this is just individuals, organizations 27 00:02:08,123 --> 00:02:13,913 S1: of different types laws. We're going to collect all different legislation, 28 00:02:14,003 --> 00:02:17,513 S1: starting with the US government, but expanding out to pretty 29 00:02:17,513 --> 00:02:20,393 S1: much the world at some point claims. So this is 30 00:02:20,393 --> 00:02:23,813 S1: a factual claim, a truth claim about the world votes, 31 00:02:23,843 --> 00:02:28,263 S1: a list of votes and results from laws that were 32 00:02:28,263 --> 00:02:33,333 S1: submitted and voted on to basically say, here's what the 33 00:02:33,333 --> 00:02:36,423 S1: votes were from different people. So this is talking about 34 00:02:36,423 --> 00:02:40,863 S1: representatives voting on things, right? Arguments a list of arguments 35 00:02:40,863 --> 00:02:44,943 S1: that have been made in favor or against a particular thing. 36 00:02:44,943 --> 00:02:52,613 S1: Funding sources, lobbyists. So a list of lobbyists in their agendas, missions, donations, 37 00:02:52,613 --> 00:02:57,743 S1: goals and facts. So these are actually claims from up here, 38 00:02:57,743 --> 00:03:01,133 S1: but these are just verified ones. And important to note 39 00:03:01,133 --> 00:03:04,283 S1: that it could become true and then untrue again. So 40 00:03:04,283 --> 00:03:07,133 S1: you have to kind of keep this updated. Now each 41 00:03:07,133 --> 00:03:10,583 S1: of these in this list here, each of these will 42 00:03:10,583 --> 00:03:14,893 S1: be an actual list maintained in a repository within GitHub. 43 00:03:14,923 --> 00:03:17,833 S1: In fact, we already have one of these okay. So 44 00:03:17,833 --> 00:03:19,513 S1: I'm going to click in here. And if we go 45 00:03:19,513 --> 00:03:22,663 S1: into here and look at problems we already have a 46 00:03:22,663 --> 00:03:25,873 S1: list of problems here okay. It's already starting. So we've 47 00:03:25,873 --> 00:03:29,053 S1: got we've got this going and I'm just going to 48 00:03:29,053 --> 00:03:31,513 S1: zoom in a whole bunch for this one. So we've 49 00:03:31,513 --> 00:03:36,013 S1: got like ransomware attacks on US health care systems. Uh, 50 00:03:36,013 --> 00:03:40,643 S1: teen depression in the UK, nuclear weapons development in North Korea. 51 00:03:40,673 --> 00:03:43,733 S1: Each one of them has this PR code so you 52 00:03:43,733 --> 00:03:47,813 S1: can associate directly to it. And that's basically the structure. 53 00:03:47,813 --> 00:03:50,183 S1: And you can see the structure here as well. And 54 00:03:50,183 --> 00:03:52,433 S1: again we have the link here to the GitHub repo 55 00:03:52,433 --> 00:03:54,593 S1: that we just looked at. And they're all part of 56 00:03:54,593 --> 00:03:59,123 S1: this substrate organization which is here. So if you go 57 00:03:59,123 --> 00:04:02,333 S1: into the substrate organization you could see the the main 58 00:04:02,333 --> 00:04:05,933 S1: flow here. And if you go into repositories you see 59 00:04:05,933 --> 00:04:08,963 S1: that we've populated a whole bunch of these, and we've 60 00:04:08,963 --> 00:04:11,213 S1: already got some pretty good activity on some of these 61 00:04:11,213 --> 00:04:14,753 S1: especially problems. And keep in mind, this just came out 62 00:04:14,753 --> 00:04:17,633 S1: like two days ago. So this is brand new okay. 63 00:04:17,633 --> 00:04:21,023 S1: So let's keep going on with the explanation here. So 64 00:04:21,023 --> 00:04:24,593 S1: the structure will allow the entire open source community which 65 00:04:24,593 --> 00:04:31,113 S1: is basically the world to contribute their own problems, claims, sources, frames, goals, etc. 66 00:04:31,113 --> 00:04:35,163 S1: for all those different repositories which are all the different 67 00:04:35,163 --> 00:04:39,033 S1: substrate components. Okay, I think I'm starting to get it, 68 00:04:39,033 --> 00:04:42,363 S1: but I need more. Fair enough. So one way to 69 00:04:42,363 --> 00:04:45,123 S1: think about this is as a way to put handles 70 00:04:45,123 --> 00:04:49,293 S1: onto things that are hard to discuss, right? The whole 71 00:04:49,293 --> 00:04:52,803 S1: reason that I created this project is so that we 72 00:04:52,803 --> 00:04:58,053 S1: can articulate things better to each other, but also to ourselves. Okay, 73 00:04:58,053 --> 00:05:00,693 S1: so let's get into some actual examples here so you 74 00:05:00,693 --> 00:05:02,973 S1: can see what I'm talking about. Let's start with an 75 00:05:02,973 --> 00:05:06,303 S1: argument component okay. This is what this is going to 76 00:05:06,303 --> 00:05:09,393 S1: look like. Think of a common argument we might hear 77 00:05:09,393 --> 00:05:12,303 S1: on any given day about whatever topic. And in this 78 00:05:12,303 --> 00:05:14,673 S1: case it's going to be recycling. So this is just 79 00:05:14,673 --> 00:05:18,073 S1: somebody that you hear say something right. I don't know 80 00:05:18,073 --> 00:05:20,593 S1: why you recycle, man. It's a total waste. It costs 81 00:05:20,593 --> 00:05:23,233 S1: so much to recycle right now. And the programs are 82 00:05:23,233 --> 00:05:26,863 S1: poorly run. So it's not actually benefiting the environment like 83 00:05:26,863 --> 00:05:29,833 S1: I do it if it worked, but it doesn't. And 84 00:05:29,833 --> 00:05:32,773 S1: this is some person watching you put a can in 85 00:05:32,773 --> 00:05:35,683 S1: a recycling bin. So we're confronted by this type of 86 00:05:35,683 --> 00:05:39,803 S1: argument constantly. We're being pitched these different things like oh I, 87 00:05:39,803 --> 00:05:42,983 S1: I would definitely do this because of this boom, that's 88 00:05:42,983 --> 00:05:45,863 S1: an argument. Or I would definitely stop doing that because 89 00:05:45,863 --> 00:05:50,033 S1: of this. That's also an argument. So that's things like recycling, 90 00:05:50,033 --> 00:05:52,613 S1: but it's also things that matter much more about politics 91 00:05:52,613 --> 00:05:55,793 S1: or whatever. So what substrate does this is the most 92 00:05:55,793 --> 00:05:59,093 S1: important thing, right. It takes an argument like this. 93 00:05:59,093 --> 00:05:59,633 S2: And. 94 00:05:59,633 --> 00:06:02,843 S1: Turns it into something like this. And I'm going to 95 00:06:02,843 --> 00:06:05,213 S1: zoom in. It's probably going to be a little bit blurry, 96 00:06:05,213 --> 00:06:10,853 S1: but watch this. This argument consists of these claims, okay, 97 00:06:10,853 --> 00:06:14,633 S1: effort required cost of progress. So it breaks down the 98 00:06:14,633 --> 00:06:19,823 S1: argument into these different components. Limited results okay. Then it 99 00:06:19,823 --> 00:06:25,013 S1: breaks those into further things. Recycling programs require significant effort. 100 00:06:25,013 --> 00:06:30,783 S1: Recycling programs are expensive. Most rejected materials end up in landfills. Okay. 101 00:06:30,783 --> 00:06:34,683 S1: And then it goes and adds research to all of those. 102 00:06:34,683 --> 00:06:37,083 S1: In this case, it's just pulled it directly out of 103 00:06:37,083 --> 00:06:40,053 S1: the model. But you can actually add AI on top 104 00:06:40,053 --> 00:06:42,033 S1: of this, which we're going to be doing with an 105 00:06:42,033 --> 00:06:45,453 S1: agent framework, either that we build ourselves or that we 106 00:06:45,453 --> 00:06:48,393 S1: use one of the third parties. And these will actually 107 00:06:48,393 --> 00:06:51,723 S1: be tools doing this research down here. Okay. Going and 108 00:06:51,723 --> 00:06:55,413 S1: doing separate lookups to find research. Like you could do 109 00:06:55,413 --> 00:07:00,003 S1: this at really in-depth level for each stage here, especially 110 00:07:00,003 --> 00:07:03,033 S1: this one where you're doing the research and then it 111 00:07:03,033 --> 00:07:06,303 S1: comes back and says supports or supports or weakens. Look 112 00:07:06,303 --> 00:07:12,033 S1: at this. Supports supports weakens weakens partially supports. Then we 113 00:07:12,033 --> 00:07:16,163 S1: have a conclusion ends up being a weak argument. Interesting. 114 00:07:16,163 --> 00:07:18,773 S1: So you could see the reasons this is the most 115 00:07:18,773 --> 00:07:22,493 S1: important thing. You could see how this argument was formed, 116 00:07:22,493 --> 00:07:28,013 S1: how it was researched using what sources. Okay, look at this. Uh, 117 00:07:28,013 --> 00:07:34,463 S1: recycling statistics 2021 National Waste and Recycling Association 2022. So 118 00:07:34,463 --> 00:07:36,773 S1: we actually have source names here, and we can go 119 00:07:36,773 --> 00:07:40,383 S1: into even more depth if we do additional AI work 120 00:07:40,383 --> 00:07:43,953 S1: on this. And that's kind of a foreshadowing for future 121 00:07:43,953 --> 00:07:46,803 S1: section here. So let's just keep going. Again, what this 122 00:07:46,803 --> 00:07:50,523 S1: does is it takes an argument like this recycling example, 123 00:07:50,523 --> 00:07:54,423 S1: and it turns it into a graph like we just saw. Okay, 124 00:07:54,423 --> 00:07:57,183 S1: the most important thing about this graph is that we 125 00:07:57,183 --> 00:07:59,283 S1: could throw it up on a board. We could throw 126 00:07:59,283 --> 00:08:01,083 S1: it on the side of a wall. We could each 127 00:08:01,083 --> 00:08:04,103 S1: view it inside of our AR glasses or whatever in 128 00:08:04,103 --> 00:08:06,893 S1: the in the near future. And now both of us 129 00:08:06,893 --> 00:08:09,683 S1: who are having this discussion can be looking at the 130 00:08:09,683 --> 00:08:14,693 S1: same thing and saying, uh, yeah, I agree with these two, 131 00:08:14,693 --> 00:08:17,453 S1: but I don't agree with this one. And the reason 132 00:08:17,453 --> 00:08:19,673 S1: I don't agree is because I don't trust the source 133 00:08:19,673 --> 00:08:22,373 S1: that it came from. And then you could do things like, okay, 134 00:08:22,373 --> 00:08:25,223 S1: let's grab some content from other sources. Boom. You start 135 00:08:25,223 --> 00:08:28,083 S1: adding other sources to it and maybe it changes based 136 00:08:28,083 --> 00:08:31,353 S1: on those sources. It changes the conclusion of that subpart, 137 00:08:31,353 --> 00:08:34,173 S1: of that argument. And now the whole thing updates. And 138 00:08:34,173 --> 00:08:37,563 S1: the conclusion is maybe, oh, we're not sure. Or maybe 139 00:08:37,563 --> 00:08:40,023 S1: it switches it from yes to no or up to 140 00:08:40,023 --> 00:08:44,103 S1: down or whatever. The point is transparency. Okay. Too many 141 00:08:44,103 --> 00:08:48,753 S1: of our conversations about any topic, it becomes emotional because 142 00:08:48,753 --> 00:08:51,633 S1: it's too difficult to keep all these things in our 143 00:08:51,633 --> 00:08:54,783 S1: minds at once. Okay, a basic argument like it was 144 00:08:54,783 --> 00:08:58,383 S1: just made here about recycling has multiple sub claims in it, 145 00:08:58,383 --> 00:09:01,863 S1: and those individual sub claims need to be backed by data. 146 00:09:01,863 --> 00:09:04,443 S1: And it's very hard to do that just in our 147 00:09:04,443 --> 00:09:07,443 S1: brains when we're in the middle of a conversation, especially 148 00:09:07,443 --> 00:09:09,873 S1: if you're trying to like blast this out. You want 149 00:09:09,873 --> 00:09:12,903 S1: to have this conversation, you want to say, hey, recycling 150 00:09:12,903 --> 00:09:16,463 S1: is good or recycling is bad, or whatever the topic is, 151 00:09:16,463 --> 00:09:19,103 S1: but you try to put that on social media or somewhere. 152 00:09:19,103 --> 00:09:22,823 S1: You're basically writing a text thing, right? Which is trying 153 00:09:22,823 --> 00:09:25,343 S1: to convince someone, which is fine. We've been doing that forever. 154 00:09:25,343 --> 00:09:28,853 S1: But imagine where we could do this graphically, where we 155 00:09:28,853 --> 00:09:32,573 S1: could do this visually and show the connections of how 156 00:09:32,573 --> 00:09:35,933 S1: strong an individual claim is based on how much you 157 00:09:35,933 --> 00:09:39,893 S1: trust an individual source. Right. So that's the power of 158 00:09:39,893 --> 00:09:41,833 S1: this thing. And this is why I'm so excited about it. 159 00:09:41,833 --> 00:09:44,653 S1: And there's many, many more examples we're about to get into. 160 00:09:44,653 --> 00:09:48,583 S1: So each of those objects in that diagram is another 161 00:09:48,583 --> 00:09:52,483 S1: substrate component, right. The claims, the sources, they're all in 162 00:09:52,513 --> 00:09:56,593 S1: there and they're all in there inside of another repository 163 00:09:56,593 --> 00:09:59,623 S1: inside of the substrate project. So here's an example of 164 00:09:59,623 --> 00:10:04,273 S1: a source. New York Times, Associated Press, Breitbart. When people 165 00:10:04,283 --> 00:10:07,013 S1: make truth claims, it's important to be able to fact 166 00:10:07,013 --> 00:10:10,043 S1: check or research those claims to see their support and substrate. 167 00:10:10,043 --> 00:10:13,433 S1: Does this by maintaining a list of those sources that 168 00:10:13,433 --> 00:10:16,373 S1: we may or may not trust. That's the point. Some 169 00:10:16,373 --> 00:10:18,893 S1: people will trust different sources. Some people will not trust 170 00:10:18,893 --> 00:10:22,523 S1: different sources. Right. And that's up to them. Then it 171 00:10:22,523 --> 00:10:25,043 S1: just becomes a question of, okay, well, you should trust 172 00:10:25,043 --> 00:10:28,313 S1: this source and that becomes an argument by itself. Right 173 00:10:28,323 --> 00:10:30,843 S1: Now you can make an argument. Well, you're putting too 174 00:10:30,843 --> 00:10:34,413 S1: much weight onto a particular source or set of sources, 175 00:10:34,413 --> 00:10:37,233 S1: which let's look at a bunch of claims that they've 176 00:10:37,233 --> 00:10:41,193 S1: made recently and let's separately research those. And maybe that 177 00:10:41,193 --> 00:10:44,433 S1: argument is strong enough to make them discount that particular source, 178 00:10:44,433 --> 00:10:47,133 S1: whether that's The New York Times or Breitbart or whatever 179 00:10:47,133 --> 00:10:49,893 S1: it is. Right. So when somebody makes an argument or 180 00:10:49,893 --> 00:10:53,463 S1: a claim within an argument, it can be linked directly 181 00:10:53,463 --> 00:10:56,073 S1: to those sources that you do or do not trust. 182 00:10:56,073 --> 00:10:58,683 S1: And you can see the full argument and all of 183 00:10:58,683 --> 00:11:01,923 S1: its support. In one visual look at this claim. Inflation 184 00:11:01,923 --> 00:11:06,003 S1: fell by 2% under this particular person's term, right? This 185 00:11:06,003 --> 00:11:11,073 S1: argument includes this claim. This claim has this source, the AP, 186 00:11:11,103 --> 00:11:14,943 S1: this claim has this source. New York Times. So this 187 00:11:14,943 --> 00:11:18,133 S1: argument has a particular strength as a result of that. 188 00:11:18,133 --> 00:11:22,783 S1: So argument to claims to sources. And this is why 189 00:11:22,783 --> 00:11:25,633 S1: we're so excited about substrate. It's going to make things 190 00:11:25,633 --> 00:11:29,803 S1: that used to be almost impossible to discuss actually approachable 191 00:11:29,803 --> 00:11:34,333 S1: actually discussable, because we could break it down into its 192 00:11:34,333 --> 00:11:38,353 S1: individual pieces. So before you'd be like, you're just not 193 00:11:38,353 --> 00:11:41,413 S1: able to counter all my arguments and evidence because I'm 194 00:11:41,413 --> 00:11:44,533 S1: too smart. And now you could say, look, here's my argument. 195 00:11:44,533 --> 00:11:46,993 S1: Throw it up on the board. Maybe it's like a 196 00:11:46,993 --> 00:11:49,723 S1: little bit in the future or, you know, Apple Vision 197 00:11:49,723 --> 00:11:52,453 S1: Pro or whatever. And now it's actually a 3D thing. 198 00:11:52,453 --> 00:11:54,343 S1: You could turn it around like a piece of DNA 199 00:11:54,343 --> 00:11:57,733 S1: or something, right? Show me which claim you disagree with 200 00:11:57,733 --> 00:12:01,783 S1: or which source you disagree with that backs up those claims. Right. 201 00:12:01,783 --> 00:12:04,543 S1: So now we can just take this thing apart using 202 00:12:04,543 --> 00:12:07,503 S1: different sources or whatever and change the output. This will 203 00:12:07,503 --> 00:12:11,793 S1: enable far more logical and precise discussions that that is 204 00:12:11,793 --> 00:12:14,433 S1: what we're going for. This is just one example. This 205 00:12:14,433 --> 00:12:17,193 S1: is just the arguments example. So now I want to 206 00:12:17,193 --> 00:12:19,443 S1: take and like back up a little bit and think 207 00:12:19,443 --> 00:12:24,213 S1: about real world use cases for substrate. Overall okay. Sounds 208 00:12:24,213 --> 00:12:27,723 S1: really cool. What do you actually do with it. Exactly. 209 00:12:27,723 --> 00:12:30,343 S1: So this is my favorite. Keep in mind, this is 210 00:12:30,343 --> 00:12:33,163 S1: very early. It literally came out like two days ago. 211 00:12:33,163 --> 00:12:36,733 S1: But we've already got multiple use cases for this thing okay. 212 00:12:36,733 --> 00:12:39,193 S1: So let's get into these okay. The first one here 213 00:12:39,193 --> 00:12:43,723 S1: is describing yourself. This one is massive. Look at this. 214 00:12:43,723 --> 00:12:46,633 S1: This is pretty cool art. This uh, from Midjourney. It's, uh, 215 00:12:46,633 --> 00:12:49,453 S1: part of a design that I did for a previous piece. 216 00:12:49,453 --> 00:12:53,083 S1: But anyway. Yeah. So you can see these are meant 217 00:12:53,083 --> 00:12:56,913 S1: to represent the different connections of like, uh, goals to 218 00:12:56,913 --> 00:13:00,033 S1: mission to problems that she thinks are most important in 219 00:13:00,033 --> 00:13:04,893 S1: the world. Solutions. She thinks, uh, are the answers to 220 00:13:04,893 --> 00:13:09,333 S1: those problems. Different strategies, different projects she's working on. That's 221 00:13:09,333 --> 00:13:12,543 S1: what this, uh, AI art is meant to represent with 222 00:13:12,543 --> 00:13:15,153 S1: this graph. Little function here. But then you can have 223 00:13:15,153 --> 00:13:17,713 S1: labels assigned to that. And keep in mind this is 224 00:13:17,713 --> 00:13:20,353 S1: me looking at her in this coffee shop through my 225 00:13:20,353 --> 00:13:23,803 S1: AR glasses and seeing, okay, first of all, purple is 226 00:13:23,803 --> 00:13:27,583 S1: associated with engineering. Boom. She's an engineer. She's known for 227 00:13:27,583 --> 00:13:29,833 S1: being a friend. She's known for being a writer. So 228 00:13:29,833 --> 00:13:34,003 S1: I'm now seeing this person as she is, which she 229 00:13:34,003 --> 00:13:37,063 S1: has previously defined. So let's get into that. Many people 230 00:13:37,063 --> 00:13:41,413 S1: have trouble describing who they are and what they are about. 231 00:13:41,443 --> 00:13:46,123 S1: This is so critical with substrate, you can basically describe 232 00:13:46,123 --> 00:13:50,713 S1: yourself in any way you want to write text, audio, video, whatever, 233 00:13:50,713 --> 00:13:53,053 S1: even have a conversation with AI, which is not part 234 00:13:53,053 --> 00:13:56,203 S1: of substrate yet. But uh, there's lots of different ways 235 00:13:56,203 --> 00:13:58,783 S1: to do that, and it will be able to both 236 00:13:58,783 --> 00:14:03,823 S1: articulate and visualize you as a person. This is absolutely insane. 237 00:14:03,823 --> 00:14:06,533 S1: People have trouble describing who they are and what they're about. 238 00:14:06,533 --> 00:14:08,603 S1: This is going to help you do that. And if 239 00:14:08,603 --> 00:14:12,593 S1: you share your context or your substrate representations with others, 240 00:14:12,593 --> 00:14:14,483 S1: they'll be able to see what you're about. They'll be 241 00:14:14,483 --> 00:14:17,753 S1: able to see this. Okay. Now, of course, the visuals here, 242 00:14:17,753 --> 00:14:19,493 S1: the AR stuff, I mean that's going to take some 243 00:14:19,493 --> 00:14:23,483 S1: time that that's a separate project, right? That's just visualization 244 00:14:23,483 --> 00:14:26,603 S1: of data that's already there. Substrate is about creating that 245 00:14:26,603 --> 00:14:31,433 S1: data for you for organizations for whatever in these individual components. 246 00:14:31,433 --> 00:14:35,033 S1: So learning a person's values substrate will be a wonderful 247 00:14:35,033 --> 00:14:38,663 S1: way to start learning about someone or something, what they 248 00:14:38,663 --> 00:14:41,423 S1: care about, how they see the world, whatever. So imagine 249 00:14:41,423 --> 00:14:44,663 S1: having something like this available when you're looking at someone 250 00:14:44,663 --> 00:14:47,513 S1: or you're researching them. Okay, check this out. We got 251 00:14:47,513 --> 00:14:50,933 S1: a person here. Okay. Got this guy. It's like the 252 00:14:50,933 --> 00:14:55,323 S1: same coffee shop programmer, gamer, organizer. And you've got this 253 00:14:55,323 --> 00:14:58,053 S1: grid here, right? Or this graph here. So check this out. 254 00:14:58,053 --> 00:15:03,813 S1: Mission merged gaming and programming, continuous learning, community building innovation 255 00:15:03,813 --> 00:15:09,663 S1: and game dev values. Goals, projects, annual Pixel Jam AI, 256 00:15:09,663 --> 00:15:12,843 S1: MPC framework. These are things that he's working on to 257 00:15:12,843 --> 00:15:16,923 S1: further these particular projects. And these are the goals that 258 00:15:16,923 --> 00:15:20,803 S1: he's shooting for, for the mission of merging gaming and programming. 259 00:15:20,803 --> 00:15:24,103 S1: That's insane. So now I know about this person, and 260 00:15:24,103 --> 00:15:26,773 S1: if there's matches, then I could walk up to him 261 00:15:26,773 --> 00:15:29,143 S1: and be like, hey, I hear you're into this, or 262 00:15:29,143 --> 00:15:31,033 S1: I hear here you're into that. And now I have 263 00:15:31,033 --> 00:15:33,943 S1: a conversation to start. Or maybe my eye is having 264 00:15:33,943 --> 00:15:36,733 S1: that conversation with his eye. Whatever. It depends how far 265 00:15:36,733 --> 00:15:38,683 S1: in the future it is. But this will be a 266 00:15:38,683 --> 00:15:42,143 S1: wonderful way to learn about what somebody really cares about 267 00:15:42,143 --> 00:15:44,843 S1: and how they see the world. So check this out. 268 00:15:44,843 --> 00:15:48,713 S1: They believe the most important problems are these three problems. 269 00:15:48,713 --> 00:15:51,653 S1: This is really cool, and they believe the best solutions 270 00:15:51,653 --> 00:15:55,883 S1: are these solutions to those problems. And they intend to 271 00:15:55,883 --> 00:15:59,933 S1: track progress using the following KPIs. Boom, boom boom. Here 272 00:15:59,933 --> 00:16:02,933 S1: are the KPIs. So imagine if you match up with 273 00:16:02,933 --> 00:16:06,833 S1: somebody across all these different axes right? You match with 274 00:16:06,833 --> 00:16:10,673 S1: them on values, goals, beliefs, preferences. You could find friends, 275 00:16:10,673 --> 00:16:13,493 S1: you can find people to have conversations with. You can 276 00:16:13,493 --> 00:16:16,013 S1: find a mate, you could find a partner. This way 277 00:16:16,013 --> 00:16:18,863 S1: you could find business partners this way. So we're very 278 00:16:18,863 --> 00:16:21,983 S1: excited about the potential to spawn more human connection in 279 00:16:21,983 --> 00:16:26,333 S1: this way. Okay, so the next one here is visualizing arguments. 280 00:16:26,333 --> 00:16:29,273 S1: So when you have a given narrative or rumor or 281 00:16:29,273 --> 00:16:32,723 S1: conspiracy theory going viral, you'll be able to use substrate 282 00:16:32,723 --> 00:16:35,783 S1: to analyze the argument or claim and publish the results. 283 00:16:35,993 --> 00:16:37,973 S1: We already talked about this one. We already showed it. 284 00:16:37,973 --> 00:16:40,523 S1: So I'm just going to show another argument here and 285 00:16:40,523 --> 00:16:42,923 S1: zoom in. Pretty heavy. Look at this. We never went 286 00:16:42,923 --> 00:16:45,353 S1: to the moon. Look at this. Look at this. We 287 00:16:45,353 --> 00:16:48,863 S1: never went to the moon. Contradicted by contradicted by leads 288 00:16:48,863 --> 00:16:52,883 S1: to claim moon landings were faked. Look at this claim. 289 00:16:52,883 --> 00:16:55,353 S1: We never went to the moon. It leads to moon. 290 00:16:55,353 --> 00:17:00,093 S1: Landings were faked by NASA. Okay. And we have actions 291 00:17:00,093 --> 00:17:03,963 S1: which is contradicted by leads to. And then we have 292 00:17:03,963 --> 00:17:06,693 S1: another claim which is the red one. And then look 293 00:17:06,693 --> 00:17:11,133 S1: at this provides attempts to explain demonstrate. So again the 294 00:17:11,133 --> 00:17:14,163 S1: actions or the verbs. And then we have results. So 295 00:17:14,163 --> 00:17:19,093 S1: reflectors on the moon consistently reflect lasers back from the Earth. 296 00:17:19,123 --> 00:17:22,603 S1: I didn't think of this one. I actually dynamically created 297 00:17:22,603 --> 00:17:25,213 S1: this one just just for the launch of the project. 298 00:17:25,213 --> 00:17:28,483 S1: And that was a great one. Okay, what is bouncing 299 00:17:28,483 --> 00:17:31,693 S1: lasers off the moon if there aren't reflectors on the moon, 300 00:17:31,693 --> 00:17:34,483 S1: which means somebody put them there. So I guess maybe 301 00:17:34,483 --> 00:17:37,213 S1: that would take you into aliens, but that's a separate topic. Okay. 302 00:17:37,213 --> 00:17:41,833 S1: Multiple countries have independently verified moon landings. Over £800 of 303 00:17:41,833 --> 00:17:45,163 S1: moon rocks have been studied worldwide, so you've got a 304 00:17:45,163 --> 00:17:48,223 S1: collection of evidence that starts to add up. Conclusion. False. 305 00:17:48,253 --> 00:17:52,753 S1: Overwhelming evidence supports moon landings, and this is just one 306 00:17:52,753 --> 00:17:55,303 S1: level deep. You can go even deeper with sub claims 307 00:17:55,303 --> 00:17:58,153 S1: and sub claims and pointing to the various sources with 308 00:17:58,153 --> 00:18:00,763 S1: actual citations, like you can go as far as you 309 00:18:00,763 --> 00:18:03,523 S1: want with this and have it all stored right there 310 00:18:03,523 --> 00:18:06,473 S1: inside of substrate. So you'll be able to see, for example, 311 00:18:06,473 --> 00:18:10,913 S1: that they're making the following arguments, which include the following claims, 312 00:18:10,913 --> 00:18:14,513 S1: which we fact checked using the following sources which resulted 313 00:18:14,513 --> 00:18:19,253 S1: in the following results which, using the following methodology, leads 314 00:18:19,253 --> 00:18:22,073 S1: us to this conclusion. So I don't know if you 315 00:18:22,073 --> 00:18:25,973 S1: remember Snopes, but Snopes was a basically a rumor denier 316 00:18:25,973 --> 00:18:29,393 S1: or approver. It was like, yes, this sounds like it's 317 00:18:29,393 --> 00:18:31,763 S1: not true, but it's actually true. Or it would be like, 318 00:18:31,763 --> 00:18:33,563 S1: that sounds like it's true, but it's actually not true. 319 00:18:33,563 --> 00:18:36,773 S1: But this is like Snopes, except for in a way 320 00:18:36,773 --> 00:18:41,063 S1: that you can visually explore and you can individually validate 321 00:18:41,063 --> 00:18:46,163 S1: each particular component. So any particular background can now evaluate 322 00:18:46,163 --> 00:18:49,193 S1: this with more transparency than it's ever been possible. And 323 00:18:49,193 --> 00:18:51,713 S1: they could see the pieces. And of course people will 324 00:18:51,713 --> 00:18:54,053 S1: be able to add all their favorite sources. Right. They 325 00:18:54,053 --> 00:18:57,633 S1: could build arguments, they could evaluate the same argument using 326 00:18:57,633 --> 00:18:59,703 S1: a different set of sauces and see if the conclusion 327 00:18:59,703 --> 00:19:02,133 S1: comes up different. So this is why we're excited about 328 00:19:02,133 --> 00:19:04,833 S1: this particular argument piece, which we've spent a lot of 329 00:19:04,833 --> 00:19:07,413 S1: time on, because it's really important. This has the potential 330 00:19:07,413 --> 00:19:12,453 S1: to significantly strengthen our shared understanding of reality, and will 331 00:19:12,453 --> 00:19:14,883 S1: allow us to disagree with each other in a far 332 00:19:14,883 --> 00:19:17,523 S1: healthier way. Here's one for the claim that there's a 333 00:19:17,523 --> 00:19:21,543 S1: tiny teapot orbiting the sun. Okay, so there's a tiny 334 00:19:21,543 --> 00:19:26,163 S1: teapot orbiting the sun investigated by space mission data claimed 335 00:19:26,163 --> 00:19:28,563 S1: the teapot is too small to be detected by our 336 00:19:28,563 --> 00:19:33,423 S1: current instruments. Result no unusual objects detected. No evidence of 337 00:19:33,423 --> 00:19:39,213 S1: teapot found. Support. Support. Conclusion. False. Insufficient evidence to support 338 00:19:39,213 --> 00:19:42,033 S1: the claim, which is exactly what it should say, right? 339 00:19:42,033 --> 00:19:44,423 S1: You can't prove that it's not there, but not being 340 00:19:44,423 --> 00:19:46,553 S1: able to prove that it's not there is not evidence 341 00:19:46,553 --> 00:19:48,863 S1: that it is there. Right. So and that's what we 342 00:19:48,863 --> 00:19:51,833 S1: end up getting to okay. So this is all additive. 343 00:19:51,833 --> 00:19:56,723 S1: It all starts compounding and adding in with itself okay. 344 00:19:56,723 --> 00:20:02,573 S1: So now we're talking about substrate plus I leading to 345 00:20:02,573 --> 00:20:07,433 S1: actual action. Yeah yeah yeah I this I that uh, 346 00:20:07,433 --> 00:20:10,433 S1: totally get it. But this is different. This is not 347 00:20:10,433 --> 00:20:14,933 S1: about I okay. The substrate is not about I. It's 348 00:20:14,933 --> 00:20:18,053 S1: about human meaning and progress. I is just a tool 349 00:20:18,053 --> 00:20:21,233 S1: for helping that along. So think about this with everything 350 00:20:21,233 --> 00:20:24,893 S1: you've heard so far about substrate and what's simultaneously happening 351 00:20:24,893 --> 00:20:27,833 S1: inside the world of I. So context sizes, which is 352 00:20:27,833 --> 00:20:30,143 S1: basically the size of prompts that you can use, are 353 00:20:30,143 --> 00:20:34,283 S1: increasing massively. And inference costs, which is basically a fancy 354 00:20:34,313 --> 00:20:38,363 S1: way of saying the cost to run individual AI pulls 355 00:20:38,363 --> 00:20:42,803 S1: or questions are massively falling. Okay, what this means is 356 00:20:42,803 --> 00:20:46,133 S1: we can basically like chocolate and peanut butter this thing 357 00:20:46,163 --> 00:20:50,483 S1: together with Substrate's ability to have all of these things 358 00:20:50,483 --> 00:20:55,653 S1: stored in very neat, structured ways and being seen together 359 00:20:55,653 --> 00:21:00,153 S1: in graphs combined with AI's ability to hold multiple things 360 00:21:00,153 --> 00:21:03,243 S1: in its mind at once, and then perform really cool 361 00:21:03,243 --> 00:21:08,793 S1: actions and answer questions and make recommendations. So this combination 362 00:21:08,793 --> 00:21:13,473 S1: is absolutely insane. So we could feed AI with our goals. Okay, 363 00:21:13,473 --> 00:21:16,263 S1: us as an individual, us as a county, us as 364 00:21:16,263 --> 00:21:20,953 S1: a city, us as a company. Us as a country. Okay, goals, KPIs, 365 00:21:20,953 --> 00:21:25,873 S1: risks and have it help us untangle these, discuss them, 366 00:21:25,873 --> 00:21:30,403 S1: debate about them, vote on them, whatever, and then take action. Okay, 367 00:21:30,403 --> 00:21:34,153 S1: so here's some of the examples that we're most excited 368 00:21:34,153 --> 00:21:38,173 S1: about for for this combination of substrate plus I. So 369 00:21:38,173 --> 00:21:43,783 S1: first one science automated hypothesis to results workflows. So one 370 00:21:43,783 --> 00:21:46,243 S1: big problem with science is that it takes so long 371 00:21:46,273 --> 00:21:49,033 S1: to do science. And this is why it takes so long. 372 00:21:49,033 --> 00:21:51,373 S1: Because look at look at all the different things you 373 00:21:51,373 --> 00:21:53,953 S1: have to go through. It's hard to come up with ideas. 374 00:21:53,953 --> 00:21:56,683 S1: It's hard to design experiments. It's hard to find funding 375 00:21:56,713 --> 00:21:59,863 S1: to do the experiments. It's hard to interpret the results. 376 00:21:59,863 --> 00:22:02,203 S1: It's hard to publish the results, and it's hard to 377 00:22:02,203 --> 00:22:04,963 S1: get those results in front of the right people who 378 00:22:04,963 --> 00:22:07,673 S1: actually have influence and might want to do something. So 379 00:22:07,673 --> 00:22:10,313 S1: now imagine that we have our list of problems, our 380 00:22:10,313 --> 00:22:13,733 S1: list of proposed experiments, all within substrate, a list of 381 00:22:13,733 --> 00:22:17,933 S1: funding sources, also within substrate. They're all there. Now I 382 00:22:17,963 --> 00:22:21,383 S1: can help us to do almost every step in that 383 00:22:21,383 --> 00:22:25,013 S1: difficulty chain we just talked about above. Okay. So I 384 00:22:25,043 --> 00:22:29,543 S1: can help us come up with or collect ideas and hypotheses. 385 00:22:29,543 --> 00:22:32,093 S1: It can help us design experiments. It can help us 386 00:22:32,093 --> 00:22:34,943 S1: collect and evaluate the best funding sources, because it's got 387 00:22:34,943 --> 00:22:37,193 S1: the full list right there. So it could be like, okay, 388 00:22:37,193 --> 00:22:39,923 S1: based on this, based on what your goals and your 389 00:22:39,923 --> 00:22:42,383 S1: mission and what you're trying to do here, these are 390 00:22:42,383 --> 00:22:45,413 S1: the types of groups that are most likely to want 391 00:22:45,413 --> 00:22:48,503 S1: to fund you. Okay. Requesting funding now that it knows 392 00:22:48,503 --> 00:22:50,903 S1: those groups, maybe it knows how to contact them because 393 00:22:50,903 --> 00:22:53,483 S1: that information is inside a substrate. Okay. So we can 394 00:22:53,483 --> 00:22:56,433 S1: write the perfect pitch for you to get funding, help. 395 00:22:56,433 --> 00:22:59,193 S1: You set up the experiments, and we're going to need 396 00:22:59,193 --> 00:23:01,353 S1: some humans and or robots to help with that part. 397 00:23:01,353 --> 00:23:05,943 S1: But whatever. Running and monitoring the experiments, interpreting the results, 398 00:23:05,943 --> 00:23:09,723 S1: all of this really, really right in the wheelhouse of AI, 399 00:23:09,753 --> 00:23:13,323 S1: writing the paper and sharing the paper also possible with 400 00:23:13,323 --> 00:23:16,113 S1: AI and getting better all the time. So in other words, 401 00:23:16,113 --> 00:23:19,893 S1: we're talking about hypothesis to propose experiment to looking at 402 00:23:19,893 --> 00:23:24,333 S1: funding sources, to acquiring funding, to running experiments, to publishing 403 00:23:24,333 --> 00:23:28,773 S1: the results, to actually making progress. So in the beginning, 404 00:23:28,773 --> 00:23:31,083 S1: this is still going to require a lot of human help, right? 405 00:23:31,083 --> 00:23:35,193 S1: Especially at the idea and the running of the experiment phases. 406 00:23:35,193 --> 00:23:37,683 S1: But over time, I will get better at that as well. 407 00:23:37,683 --> 00:23:41,553 S1: But what we're talking about ultimately with this right here 408 00:23:41,553 --> 00:23:46,073 S1: is the acceleration of science. Right. Because this cycle right 409 00:23:46,073 --> 00:23:48,173 S1: here is leading to a whole bunch of failures and 410 00:23:48,173 --> 00:23:50,693 S1: making us move on to do something else instead. But 411 00:23:50,693 --> 00:23:52,913 S1: this whole pipeline, and I want to give credit to 412 00:23:52,913 --> 00:23:56,573 S1: Joseph Thacker for thinking about a very similar thing at 413 00:23:56,573 --> 00:24:00,353 S1: a similar time, like a year ago, talking about experiments 414 00:24:00,353 --> 00:24:04,853 S1: and hypotheses and testing and stuff like that. So a 415 00:24:04,853 --> 00:24:09,313 S1: lot of people probably thinking very similar things along these lines. 416 00:24:09,313 --> 00:24:11,983 S1: But we are so excited about this. The idea of 417 00:24:11,983 --> 00:24:16,303 S1: just being able to experiment and advance our knowledge forward 418 00:24:16,303 --> 00:24:18,823 S1: through the most powerful mechanism that we're aware of, which 419 00:24:18,823 --> 00:24:24,103 S1: is science accountability. Okay, this one's insane. Monitoring crime and 420 00:24:24,103 --> 00:24:27,943 S1: corruption okay. So the reason it's so easy to get 421 00:24:27,943 --> 00:24:30,853 S1: away with corruption and crime right now is because there 422 00:24:30,853 --> 00:24:36,113 S1: aren't enough people watching gangs, cartels, embezzlers, dirty politicians. They're 423 00:24:36,113 --> 00:24:40,223 S1: actually dropping evidence all the time. There's receipts, there's like travel, 424 00:24:40,223 --> 00:24:44,363 S1: there's tickets, there's cameras. There's lots of different ways to 425 00:24:44,363 --> 00:24:47,093 S1: know that a particular person was in a particular place, 426 00:24:47,093 --> 00:24:50,603 S1: and they're not even being that careful, because it's actually 427 00:24:50,603 --> 00:24:53,243 S1: so difficult to go and collect that stuff and bring 428 00:24:53,243 --> 00:24:56,133 S1: it together into a narrative. So it usually takes a 429 00:24:56,133 --> 00:25:00,063 S1: major journalist team or a massive law enforcement operation to 430 00:25:00,063 --> 00:25:04,383 S1: dump thousands of hours of highly skilled work to collect 431 00:25:04,383 --> 00:25:06,873 S1: all this different evidence. Okay, then you have to do 432 00:25:06,873 --> 00:25:10,203 S1: the analysis. Then you have to formulate the conclusions. Then 433 00:25:10,203 --> 00:25:12,033 S1: you have to document all of this. Then you have 434 00:25:12,033 --> 00:25:13,563 S1: to like take it to the media. You have to 435 00:25:13,563 --> 00:25:16,833 S1: get in front of people. And most crime and corruption 436 00:25:16,833 --> 00:25:20,613 S1: slips by because nobody is simply watching. There aren't enough journalists. 437 00:25:20,613 --> 00:25:24,093 S1: There aren't enough law enforcement teams who have the skills 438 00:25:24,093 --> 00:25:26,043 S1: to do this stuff. And even if they had the skills, 439 00:25:26,043 --> 00:25:28,773 S1: they don't have the resources and the time. So substrate 440 00:25:28,773 --> 00:25:32,253 S1: plus AI versus dirty politician, this is a use case. 441 00:25:32,253 --> 00:25:35,313 S1: So let's take substrate with some AI added on. And 442 00:25:35,313 --> 00:25:38,043 S1: let's think about a dirty politician who is taking massive 443 00:25:38,043 --> 00:25:41,283 S1: gifts from a particular lobbyist. Let's say this is some 444 00:25:41,283 --> 00:25:46,193 S1: dirty Democrat, some dirty Republican. Independent doesn't matter. This has 445 00:25:46,193 --> 00:25:48,683 S1: nothing to do with that. Problem is, there are so 446 00:25:48,683 --> 00:25:52,853 S1: many donations. There are so many lobbyists, so many representatives, 447 00:25:52,853 --> 00:25:56,813 S1: so many actual laws and bills and so many votes. 448 00:25:56,813 --> 00:25:59,693 S1: But guess what? It's all public. We're talking about the 449 00:25:59,693 --> 00:26:03,593 S1: US here. This is all public information. It is required 450 00:26:03,593 --> 00:26:06,533 S1: by law that all the stuff is posted. The lobbyist 451 00:26:06,533 --> 00:26:10,463 S1: groups must actually register themselves. The donations that they make 452 00:26:10,463 --> 00:26:14,093 S1: to any particular representative, they have to be public records 453 00:26:14,093 --> 00:26:17,663 S1: of meetings. I believe those are also public, and so 454 00:26:17,663 --> 00:26:22,703 S1: are all the votes that representatives make on bills where 455 00:26:22,703 --> 00:26:26,453 S1: lobbyists have been donating and trying to influence. So a 456 00:26:26,453 --> 00:26:29,543 S1: nonprofit or even just a project, a small project that 457 00:26:29,543 --> 00:26:31,853 S1: comes out of substrate or whatever, just a bunch of 458 00:26:31,853 --> 00:26:34,523 S1: open source people could use AI to collect all of 459 00:26:34,523 --> 00:26:37,823 S1: these different things, continuously put them in substrate, and they're 460 00:26:37,823 --> 00:26:40,313 S1: already going in substrate because we're about to start dumping 461 00:26:40,313 --> 00:26:43,283 S1: all the laws, all the different voting records, all the 462 00:26:43,283 --> 00:26:48,053 S1: different histories for each lobbyist and also for each representative, 463 00:26:48,053 --> 00:26:51,323 S1: all that goes right into substrate to inspect. And then 464 00:26:51,323 --> 00:26:54,593 S1: I can ingest all of that at any given moment 465 00:26:54,593 --> 00:26:57,903 S1: and basically tell us this. Here are all the bills 466 00:26:57,903 --> 00:27:00,423 S1: written by that person. Here are all the summaries of 467 00:27:00,423 --> 00:27:03,393 S1: those bills. Basically, what are they trying to do? Here's 468 00:27:03,393 --> 00:27:08,733 S1: who those bills helped and who they hurt. Right? Here 469 00:27:08,733 --> 00:27:12,003 S1: are all the lobbyists that care about those particular issues. 470 00:27:12,003 --> 00:27:16,563 S1: Here are all the donations that those groups made to 471 00:27:16,563 --> 00:27:20,673 S1: those representatives or that particular representative. And here's how the 472 00:27:20,673 --> 00:27:24,213 S1: representative voted on every single bill. Then guess what? The 473 00:27:24,213 --> 00:27:27,453 S1: AI is really good at, which we already know. Okay. 474 00:27:27,453 --> 00:27:31,443 S1: Perform a comprehensive analysis of all legislation created and voted 475 00:27:31,443 --> 00:27:34,503 S1: on from Bill Myers, Senator from Arkansas. I made that up. 476 00:27:34,503 --> 00:27:37,593 S1: Hopefully there is no one like that cross-referenced with every 477 00:27:37,593 --> 00:27:41,133 S1: single donation ever made to him, every dinner he's ever 478 00:27:41,133 --> 00:27:45,163 S1: attended with them, every gift he's ever received, etc. finally, 479 00:27:45,163 --> 00:27:47,263 S1: give me your assessment of whether or not he is 480 00:27:47,263 --> 00:27:51,463 S1: being unduly influenced by this lobbyist and give me your 481 00:27:51,463 --> 00:27:54,583 S1: reasons for this conclusion. Then it comes back with something 482 00:27:54,583 --> 00:27:59,293 S1: like this assessment. This is a compromised politician reasoning. Osint 483 00:27:59,293 --> 00:28:03,913 S1: reveals you can have Osint going and researching various things 484 00:28:03,913 --> 00:28:07,843 S1: that this person is doing. Again, all public, all legal. Right? 485 00:28:07,843 --> 00:28:11,243 S1: We're not talking about nasty stuff here. We're talking about legal. 486 00:28:11,243 --> 00:28:16,913 S1: Public documents reveals that he was illegally gifted a small 487 00:28:16,913 --> 00:28:20,063 S1: yacht last year, which he tweeted about and later deleted. 488 00:28:20,063 --> 00:28:22,853 S1: He's had 31 dinners in the last 18 months with 489 00:28:22,853 --> 00:28:27,203 S1: them totaling over like almost $15,000. Osint reveals that the 490 00:28:27,203 --> 00:28:32,483 S1: Lobbyist's president used considerable influence to get Bill Meyer's daughter 491 00:28:32,483 --> 00:28:35,813 S1: admitted to an exclusive private school that she wasn't actually 492 00:28:35,813 --> 00:28:39,263 S1: qualified for. Every vote he's made about this particular issue 493 00:28:39,293 --> 00:28:42,443 S1: has been in the direction that the lobbyist actually wants 494 00:28:42,443 --> 00:28:46,613 S1: to happen, and previous votes before they started the relationship, 495 00:28:46,613 --> 00:28:50,543 S1: the politician used to vote in the opposite direction. Therefore, 496 00:28:50,573 --> 00:28:53,903 S1: the conclusion from this I. And by the way, you 497 00:28:53,903 --> 00:28:56,663 S1: published the algorithm for the AI as well. We've got 498 00:28:56,663 --> 00:29:00,353 S1: another project called fabric, where you publish the actual thoughts 499 00:29:00,353 --> 00:29:04,433 S1: and directions given to the AIS so that they're inspectable. Right? 500 00:29:04,433 --> 00:29:06,833 S1: You you don't want people to think, oh, they came 501 00:29:06,833 --> 00:29:09,383 S1: to that conclusion because it's biased. No. You can look 502 00:29:09,383 --> 00:29:12,353 S1: and see the actual instructions given to the AI to 503 00:29:12,353 --> 00:29:15,503 S1: evaluate this thing impartially. So now you could trust this 504 00:29:15,503 --> 00:29:18,833 S1: thing with a very high rate of confidence, because you 505 00:29:18,833 --> 00:29:21,213 S1: could see how it's thinking and you could see all 506 00:29:21,213 --> 00:29:24,753 S1: the data sources, just like before with arguments. So basically, 507 00:29:24,753 --> 00:29:30,303 S1: the incredibly important objects of legislation, votes, etc. are all 508 00:29:30,303 --> 00:29:34,623 S1: things that can be monitored and collected and stored within substrate. 509 00:29:34,623 --> 00:29:39,303 S1: So that's that one next one here. Leadership. This is absolutely, 510 00:29:39,303 --> 00:29:43,953 S1: absolutely powerful okay. I've been a consultant for a couple 511 00:29:43,953 --> 00:29:46,933 S1: of decades now and I've worked with hundreds of startups, 512 00:29:46,933 --> 00:29:51,823 S1: so many large corporations, and a big, big problem for 513 00:29:51,823 --> 00:30:00,043 S1: most organizations. And that includes governments, families, individuals, startups, corporations. 514 00:30:00,043 --> 00:30:03,883 S1: Everything is that they don't have clarity. Just like a person. 515 00:30:03,883 --> 00:30:06,163 S1: It's hard to know what they think the issues are, 516 00:30:06,163 --> 00:30:09,463 S1: what they specifically plan on doing, and how they plan 517 00:30:09,463 --> 00:30:13,003 S1: to measure progress. So we see this with business leaders 518 00:30:13,003 --> 00:30:16,093 S1: and we see this with politicians. So with substrate, we 519 00:30:16,093 --> 00:30:19,543 S1: intend to make it so that check this out. Every 520 00:30:19,543 --> 00:30:22,783 S1: leader will need to have a full detailed plan that 521 00:30:22,783 --> 00:30:27,043 S1: has the following components. Imagine if everyone had this when 522 00:30:27,043 --> 00:30:30,433 S1: they were pitching some plan for the high school, or 523 00:30:30,433 --> 00:30:33,073 S1: some plan to be a principal, or some plan to 524 00:30:33,083 --> 00:30:35,993 S1: be a community leader, or some plan to be a 525 00:30:35,993 --> 00:30:38,993 S1: politician or whatever it is. Here's what I think the 526 00:30:38,993 --> 00:30:42,293 S1: problems are. Remember, this is a problem object inside of 527 00:30:42,293 --> 00:30:44,483 S1: substrate so you can actually look at them. Here's what 528 00:30:44,483 --> 00:30:47,363 S1: I think the problems are. And coming off of the 529 00:30:47,363 --> 00:30:50,363 S1: problem here's what I think the solutions are. Here are 530 00:30:50,363 --> 00:30:54,683 S1: my proposed strategies for accomplishing that. Here are the KPIs. 531 00:30:54,683 --> 00:30:57,923 S1: These are the metrics. Here's how actually going to measure myself. 532 00:30:57,923 --> 00:31:01,043 S1: And how about this fire me if I don't get 533 00:31:01,043 --> 00:31:05,093 S1: the KPIs to this number by this date. So at 534 00:31:05,093 --> 00:31:07,343 S1: the end of the three and a half years when 535 00:31:07,343 --> 00:31:10,883 S1: I'm going back up for reelection, I expect whatever the 536 00:31:10,883 --> 00:31:13,403 S1: thing is, uh, the number of kids who don't go 537 00:31:13,403 --> 00:31:16,733 S1: hungry in this high school, the literacy level in this 538 00:31:16,733 --> 00:31:20,213 S1: high school, the unemployment rate, whatever it is, doesn't matter 539 00:31:20,213 --> 00:31:23,803 S1: if I don't improve that number by X amount. By 540 00:31:23,803 --> 00:31:26,533 S1: this date, you have my permission to vote me out. 541 00:31:26,563 --> 00:31:30,943 S1: Imagine having that level of clarity and accountability for any 542 00:31:30,943 --> 00:31:34,903 S1: leader trying to get any job doing anything. So really 543 00:31:34,903 --> 00:31:38,413 S1: excited about that one. Okay, next one here. This is 544 00:31:38,413 --> 00:31:40,603 S1: the best one okay. I saved the best for last. 545 00:31:40,603 --> 00:31:43,183 S1: These are all adding on each other. Watch this. Look 546 00:31:43,183 --> 00:31:46,753 S1: at this. I did a post a while back about 547 00:31:46,753 --> 00:31:51,073 S1: how companies are essentially graphs of algorithms, and I encourage 548 00:31:51,073 --> 00:31:52,933 S1: you to go check that out. I've got the link 549 00:31:52,933 --> 00:31:55,783 S1: down here below, but it's like, okay, you have a 550 00:31:55,783 --> 00:31:59,203 S1: company that processes bad images come in, they send them 551 00:31:59,203 --> 00:32:02,263 S1: to you on the website, and you then do a 552 00:32:02,263 --> 00:32:04,873 S1: number of things to that photo, right? You do a 553 00:32:04,873 --> 00:32:07,873 S1: high quality scan, you repair it, you stylize it, you 554 00:32:07,873 --> 00:32:10,183 S1: caption it, you send it to the user. Then there's 555 00:32:10,183 --> 00:32:13,603 S1: a marketing group. A marketing group has an idea. It 556 00:32:13,603 --> 00:32:15,943 S1: shares it with a team. They decide on the best version. 557 00:32:15,943 --> 00:32:18,433 S1: They do this, they do a final vote. They do 558 00:32:18,433 --> 00:32:22,543 S1: a launch campaign. Okay, or the uploading process. Visit the 559 00:32:22,543 --> 00:32:25,813 S1: website create account, click the upload link. These are all 560 00:32:25,813 --> 00:32:29,503 S1: individual steps okay, these are just algorithms. And this is 561 00:32:29,503 --> 00:32:32,863 S1: a graph of algorithms where they're all connected. And you 562 00:32:32,863 --> 00:32:35,813 S1: can break these into smaller and smaller pieces until you 563 00:32:35,813 --> 00:32:39,473 S1: eventually see the world in this way. Okay. I'm just 564 00:32:39,473 --> 00:32:42,263 S1: going to click into this thing just to highlight this 565 00:32:42,263 --> 00:32:45,353 S1: a little bit. This is what it ends up looking like. Okay. 566 00:32:45,353 --> 00:32:48,833 S1: It looks like this is what your companies are going 567 00:32:48,833 --> 00:32:52,673 S1: to look like before too long okay. Not your companies. 568 00:32:52,673 --> 00:32:56,213 S1: Everyone's company okay. AI consultancies are going to come in. 569 00:32:56,213 --> 00:32:58,683 S1: They're going to see the world in this way. Everything 570 00:32:58,683 --> 00:33:02,163 S1: is a process. Everything is a set of components. Okay? 571 00:33:02,163 --> 00:33:05,043 S1: You got human decisions here. You got human teams here. 572 00:33:05,043 --> 00:33:07,383 S1: Maybe the red are all the places that it's human 573 00:33:07,383 --> 00:33:09,963 S1: and AI wants to replace that. Maybe the red is 574 00:33:09,963 --> 00:33:12,723 S1: it's a little bit too manual. Maybe the red is 575 00:33:12,723 --> 00:33:16,593 S1: the efficiency numbers are too low or it takes too long. 576 00:33:16,593 --> 00:33:18,693 S1: Doesn't matter. You could drill in. You could see this 577 00:33:18,693 --> 00:33:21,033 S1: is actually a human team. They actually need more people. 578 00:33:21,033 --> 00:33:24,093 S1: We need to hire more people, etc. the point is, 579 00:33:24,093 --> 00:33:27,663 S1: you can look at any company or any process as 580 00:33:27,663 --> 00:33:29,973 S1: a series of these components, which is part of a 581 00:33:29,973 --> 00:33:33,243 S1: larger process which has a flow. Okay. That's the important 582 00:33:33,243 --> 00:33:36,273 S1: part here. Okay. Yeah. This is the piece here. Companies 583 00:33:36,273 --> 00:33:38,853 S1: are just graphs of algorithms. But think of it this 584 00:33:38,853 --> 00:33:41,163 S1: way I don't think I went far enough with that. 585 00:33:41,163 --> 00:33:45,333 S1: Everything can be conceptualized in this way. Everything can be 586 00:33:45,333 --> 00:33:49,173 S1: conceptualized in this way as a process. So essentially what 587 00:33:49,173 --> 00:33:51,963 S1: you have is you have a current state of things, right? 588 00:33:51,963 --> 00:33:54,963 S1: State of the universe, but smaller down to a scale 589 00:33:54,963 --> 00:33:57,423 S1: that we're dealing with, like a company or a process 590 00:33:57,423 --> 00:33:59,823 S1: or a team or a department. Okay. Then we have 591 00:33:59,823 --> 00:34:03,693 S1: an action or event, like a decision is made. Um, 592 00:34:03,693 --> 00:34:07,413 S1: you know, customer does a particular thing. They buy a thing. 593 00:34:07,413 --> 00:34:10,143 S1: We need to pay out an insurance policy, whatever the 594 00:34:10,163 --> 00:34:12,263 S1: thing is. Right? And then that results in a new 595 00:34:12,263 --> 00:34:17,843 S1: state of things. So previous state action thing happens a cause. 596 00:34:17,843 --> 00:34:19,943 S1: And then you have like an effect which is the outcome, 597 00:34:19,943 --> 00:34:22,463 S1: which is the result. And if we add human components 598 00:34:22,463 --> 00:34:26,393 S1: into this like people's jobs or making decisions, and we 599 00:34:26,393 --> 00:34:28,313 S1: do this for like running a business or a country 600 00:34:28,313 --> 00:34:30,923 S1: or family, we have additional pieces, right? We have people, 601 00:34:30,923 --> 00:34:33,983 S1: we have decisions, we have strategies. We have lessons learned. 602 00:34:33,983 --> 00:34:37,733 S1: We have conclusions. We have reasons. All of these are 603 00:34:37,733 --> 00:34:42,023 S1: substrate components. There aren't infinite numbers of these. These are 604 00:34:42,023 --> 00:34:45,143 S1: all look, what was the decision made? There aren't that 605 00:34:45,143 --> 00:34:49,163 S1: many decisions okay. Buy more stock. Um, hire more people. 606 00:34:49,163 --> 00:34:52,103 S1: This is the type of decision in there, right? Again, 607 00:34:52,103 --> 00:34:54,653 S1: which we can connect and see visually. And what that 608 00:34:54,653 --> 00:34:57,443 S1: results in is a way to tie together all of 609 00:34:57,443 --> 00:35:00,743 S1: this into much larger graphs, graphs that we could use 610 00:35:00,743 --> 00:35:05,093 S1: to describe the operations of anything a family, company, or 611 00:35:05,093 --> 00:35:08,813 S1: even a country. So here's one for a small company. 612 00:35:08,813 --> 00:35:11,993 S1: And I made this live. This is actual like I 613 00:35:11,993 --> 00:35:14,603 S1: put in some fake company stuff. Look at this. No 614 00:35:14,603 --> 00:35:18,773 S1: customers website visitor quality lead. Yes. No. Okay. Goes down. 615 00:35:18,773 --> 00:35:21,383 S1: Look at this. Breaks down the thing. This is how 616 00:35:21,383 --> 00:35:24,243 S1: we're about to see our companies. We're about to see 617 00:35:24,243 --> 00:35:27,123 S1: all of anything that humans do as a set of 618 00:35:27,123 --> 00:35:30,303 S1: processes like this. And it's fine if you want to 619 00:35:30,303 --> 00:35:34,533 S1: resist this, but trust me, this is coming because the 620 00:35:34,533 --> 00:35:38,553 S1: consultancies are coming. I got to stop for you. 40% 621 00:35:38,553 --> 00:35:43,113 S1: of McKinsey's business right now is already in AI consulting. 40%. 622 00:35:43,113 --> 00:35:44,973 S1: This thing happened like a year and a half ago. 623 00:35:44,973 --> 00:35:48,603 S1: They are already doing almost half their business doing this. 624 00:35:48,603 --> 00:35:51,333 S1: They're doing this. They're coming in. They're looking at companies 625 00:35:51,333 --> 00:35:54,033 S1: and saying, what can I optimize? What can I improve? 626 00:35:54,033 --> 00:35:57,123 S1: So that's pretty cool. We can create that. But that's 627 00:35:57,123 --> 00:35:59,553 S1: not the full power of this yet. Check this out. 628 00:35:59,553 --> 00:36:02,643 S1: The smarter the AI gets, it's not only going to 629 00:36:02,643 --> 00:36:06,603 S1: see this and describe it and make some recommendations, but no, 630 00:36:06,603 --> 00:36:09,753 S1: it can optimize the stuff. In other words, this is 631 00:36:09,753 --> 00:36:12,133 S1: just the current state. What about future state? What about 632 00:36:12,133 --> 00:36:15,643 S1: recommended state? Should this company merge departments? Where can we 633 00:36:15,643 --> 00:36:18,073 S1: add more people? What kind of people should we hire? 634 00:36:18,073 --> 00:36:21,673 S1: Which processes here are inefficient which can be replaced by AI? 635 00:36:21,673 --> 00:36:24,133 S1: That's going to be the number one question somebody is asking. 636 00:36:24,133 --> 00:36:26,593 S1: Where can we use more human decision making? If we 637 00:36:26,593 --> 00:36:30,103 S1: wanted to grow, where would that happen? Now imagine this 638 00:36:30,103 --> 00:36:35,063 S1: for a family, a corporation, a church, a city, a county, etc. 639 00:36:35,063 --> 00:36:37,553 S1: and keep in mind, the more data that you have here, 640 00:36:37,553 --> 00:36:40,463 S1: the better, right? You can. You can pull in all 641 00:36:40,463 --> 00:36:44,543 S1: the data about conversion rates, churn rates, like whatever data 642 00:36:44,543 --> 00:36:47,543 S1: you have that comes in. That's part of this whole equation. 643 00:36:47,543 --> 00:36:50,903 S1: So here's an example for a security team because my 644 00:36:50,903 --> 00:36:54,803 S1: background is in security. So it currently takes 3.5 business 645 00:36:54,803 --> 00:36:59,003 S1: days to complete a security assessment. Delays in security assessment 646 00:36:59,003 --> 00:37:02,873 S1: turnaround are the number one complaint in the engineering survey. 647 00:37:02,873 --> 00:37:06,863 S1: So the security group sends a survey to engineering and says, hey, 648 00:37:06,863 --> 00:37:09,263 S1: what do you love and what do you hate about security? 649 00:37:09,263 --> 00:37:11,603 S1: This is their number one complaint. It takes too long 650 00:37:11,633 --> 00:37:14,423 S1: to get stuff back. So you're slowing us down. So 651 00:37:14,423 --> 00:37:17,393 S1: if we switch to the new flex scan model, which 652 00:37:17,393 --> 00:37:19,943 S1: is a new model for doing security assessments in a 653 00:37:19,943 --> 00:37:23,923 S1: more flexible way using fewer generalist security testers were able 654 00:37:23,923 --> 00:37:27,673 S1: to complete type B and type C assessments 94% faster. 655 00:37:27,703 --> 00:37:30,673 S1: This will give our senior testers two extra days to 656 00:37:30,673 --> 00:37:33,823 S1: do a high impact security assessments, and this will also 657 00:37:33,823 --> 00:37:37,393 S1: likely make engineering much happier with security and make them 658 00:37:37,393 --> 00:37:40,903 S1: more likely to cooperate on our goals. So there's multiple 659 00:37:40,903 --> 00:37:44,623 S1: steps the full articulation and breakdown of how a process 660 00:37:44,623 --> 00:37:47,853 S1: is currently running. Visualization of that process to help with 661 00:37:47,853 --> 00:37:51,723 S1: human understanding, and then analysis by AI of how to 662 00:37:51,723 --> 00:37:55,863 S1: optimize the process to optimize the stated goals of the entity. 663 00:37:55,863 --> 00:37:59,313 S1: That's the three pieces. And remember, the AI will also 664 00:37:59,313 --> 00:38:02,583 S1: have access to the mission of the organization as well, 665 00:38:02,583 --> 00:38:05,193 S1: and its goals and its strategies and its team members 666 00:38:05,193 --> 00:38:08,283 S1: and its projects and its budget. Everything. So we'll have 667 00:38:08,283 --> 00:38:11,863 S1: the full context on how resources are being spent relative 668 00:38:11,863 --> 00:38:14,473 S1: to the desired outcomes, and we'll be able to see 669 00:38:14,473 --> 00:38:17,833 S1: how the actual KPIs that we care about, the actual numbers, 670 00:38:17,833 --> 00:38:21,853 S1: the actual metrics, are moving as we adjust these. So 671 00:38:21,853 --> 00:38:24,253 S1: it'll be able to do all sorts of recommendations hiring 672 00:38:24,253 --> 00:38:28,453 S1: new people, hiring people with specific skills, using more AI 673 00:38:28,453 --> 00:38:32,653 S1: and high volume and low creativity areas, adjusting strategies based 674 00:38:32,653 --> 00:38:36,223 S1: on goals and market conditions. Canceling projects like this one 675 00:38:36,223 --> 00:38:38,803 S1: and that one. Like they're not even related to the goals. 676 00:38:38,803 --> 00:38:41,173 S1: They're costing way too much. Way too much of the 677 00:38:41,173 --> 00:38:43,333 S1: team is working on them. Let's get rid of those. 678 00:38:43,333 --> 00:38:47,293 S1: This project over here, this human powered project, it's way better. 679 00:38:47,293 --> 00:38:50,713 S1: It's producing way more revenue. It's way more tied to 680 00:38:50,713 --> 00:38:55,453 S1: our actual goals and KPIs. So we're taking those people 681 00:38:55,453 --> 00:38:58,633 S1: and putting them over here instead. So ultimately we're talking 682 00:38:58,633 --> 00:39:03,203 S1: about the ability to continuously analyze and optimize any system 683 00:39:03,203 --> 00:39:06,653 S1: using full knowledge of its goals and progress. And the 684 00:39:06,653 --> 00:39:09,683 S1: more data about the system it has, the better it 685 00:39:09,683 --> 00:39:12,143 S1: will actually work. And the smarter the AI gets, the 686 00:39:12,143 --> 00:39:15,683 S1: better it actually works. Completely insane. All right, to summarize, 687 00:39:15,683 --> 00:39:18,803 S1: the world is hard to understand, and things that are 688 00:39:18,803 --> 00:39:21,863 S1: hard to understand are hard to discuss and improve. The 689 00:39:21,863 --> 00:39:24,733 S1: goal of Substrate is to address this problem by making 690 00:39:24,733 --> 00:39:29,173 S1: the things humans care about more visible. Discussable and Improvable. 691 00:39:29,173 --> 00:39:32,713 S1: The framework is open source, lives on GitHub. At its core, 692 00:39:32,713 --> 00:39:36,583 S1: it's a collection of crowdsourced lists of the things humans 693 00:39:36,583 --> 00:39:39,283 S1: care about and that make up our discourse and society. 694 00:39:39,283 --> 00:39:42,403 S1: One major problem that people and organizations have is not 695 00:39:42,403 --> 00:39:45,493 S1: knowing and or being able to communicate what they are about. 696 00:39:45,493 --> 00:39:48,613 S1: Using this framework, people and organizations will be able to 697 00:39:48,613 --> 00:39:51,763 S1: articulate their values and purpose more clearly, which will not 698 00:39:51,763 --> 00:39:54,793 S1: only help them, but everyone that they interact with. Substrate 699 00:39:54,793 --> 00:39:58,213 S1: is magnified by I because I can or will soon 700 00:39:58,243 --> 00:40:01,213 S1: be able to hold all of substrate in its mind 701 00:40:01,213 --> 00:40:03,253 S1: at once. And from there, we'll be able to ask 702 00:40:03,253 --> 00:40:06,463 S1: all sorts of meaningful questions such as what does that 703 00:40:06,463 --> 00:40:10,063 S1: person or organization about? Are we pursuing the best path 704 00:40:10,063 --> 00:40:12,793 S1: towards our goals, or what are the most critical mistakes 705 00:40:12,793 --> 00:40:15,823 S1: I'm currently making? Ultimately, this will allow us to take 706 00:40:15,823 --> 00:40:19,123 S1: action on these things. What action should I do right 707 00:40:19,123 --> 00:40:21,643 S1: now to optimize this workflow? What should I do right 708 00:40:21,643 --> 00:40:24,493 S1: now to achieve the best possible outcome that's aligned with 709 00:40:24,493 --> 00:40:28,183 S1: my goals? In short, substrate is a way to better 710 00:40:28,183 --> 00:40:32,503 S1: understand and optimize the things that we care about as humans. 711 00:40:32,503 --> 00:40:35,233 S1: So we would love for you to get involved. I 712 00:40:35,233 --> 00:40:38,353 S1: want to give thanks to people who are already involved 713 00:40:38,353 --> 00:40:41,563 S1: and are like, really, really into this. Along with myself, 714 00:40:41,563 --> 00:40:46,393 S1: Jonathan Dunn, Jason Haddox, Clint Gibler, Joseph Thacker, Joel Parrish 715 00:40:46,393 --> 00:40:49,993 S1: and Robert Hansen, some of the few that are already 716 00:40:49,993 --> 00:40:53,563 S1: involved and excited about this. And if you are interested, 717 00:40:53,563 --> 00:40:57,253 S1: please go to the Substrate Project and, uh, let us 718 00:40:57,253 --> 00:41:00,523 S1: know that you'd like to contribute. Thanks for your time.