WEBVTT 1 00:00:02.600 --> 00:00:05.519 Hello, and welcome to This Anthro Life. I'm your host, Adam Gamwell. 2 00:00:06.040 --> 00:00:08.839 You know when you're watching the news or a YouTube video or a documentary 3 00:00:08.839 --> 00:00:12.720 and my talking head says, they're scientific consensus that climate change is man made, 4 00:00:13.279 --> 00:00:16.920 or there is a preponderance of evidence in the literature that meditation is linked 5 00:00:16.960 --> 00:00:20.239 to increased feelings of well being. Have you ever wondered, well? Cool? 6 00:00:20.719 --> 00:00:24.079 But I'd like to see that consensus or maybe be able to follow the 7 00:00:24.079 --> 00:00:28.120 trail of scientific claims made around mindfulness to be better informed. Well, I've 8 00:00:28.120 --> 00:00:32.119 got good news for you. Getting to the consensus behind the claims is what 9 00:00:32.159 --> 00:00:35.880 today's episode is all about. One of the big goals of This Anthro Life 10 00:00:35.960 --> 00:00:40.240 is to amplify the voices, tools and technologies that bring our creative potential to 11 00:00:40.359 --> 00:00:44.719 life. And I'm really excited to be joined on the podcast today by Eric 12 00:00:44.759 --> 00:00:49.520 Olsen. Now. Eric is the CEO of a startup called Consensus. Consensus 13 00:00:49.600 --> 00:00:54.719 is a search engine that uses AI or artificial intelligence to instantly extract aggregate in 14 00:00:54.759 --> 00:01:00.719 distill findings directly from scientific research. Sounds pretty cool, right. It's built 15 00:01:00.719 --> 00:01:06.439 around simple searches like our COVID nineteen vaccinations, safe or the benefits of mindfulness, 16 00:01:06.760 --> 00:01:11.480 or linking together ideas such as poverty reduction and direct cash payments. It 17 00:01:11.519 --> 00:01:15.319 then searches across two hundred million papers, adding more every day, and returns 18 00:01:15.319 --> 00:01:21.200 results that highlight the scientific claims made in each of these different articles. This 19 00:01:21.280 --> 00:01:23.200 is amazing for a few reasons. One, it doesn't tell you what to 20 00:01:23.239 --> 00:01:26.920 think. It simply shows you, at a quick glance, what are the 21 00:01:26.959 --> 00:01:33.000 majority of scientific claims saying about your query. And Two, we're seeing the 22 00:01:33.000 --> 00:01:38.000 emergence of a new wave of natural language processing startups in tech that's aimed at 23 00:01:38.040 --> 00:01:42.400 democratizing scientific knowledge and access. What a time to be alive. Right now, 24 00:01:42.439 --> 00:01:46.280 we're going to get deeper into the why and how of consensus in our 25 00:01:46.319 --> 00:01:49.040 conversation today, So to kick things off, I just want to lay out 26 00:01:49.040 --> 00:01:53.280 two big challenges that tech and services like consensus face for us as listeners to 27 00:01:53.319 --> 00:01:59.359 think about. Whether we're social scientists in academia researchers, or designers and industry, 28 00:01:59.439 --> 00:02:05.000 or good old fashioned public citizen concerned with literacy and public debate. With 29 00:02:05.040 --> 00:02:07.759 the rise of research and information technology, it's surprising when you stop to think 30 00:02:07.800 --> 00:02:13.120 about it, that we haven't also found ways to build consensus around expertise, 31 00:02:14.080 --> 00:02:16.719 and there are a few challenges here at play. One is that Pew reviewed 32 00:02:16.759 --> 00:02:22.680 scientific literature in journals like Nature or Sell if you're in the biological sciences or 33 00:02:22.879 --> 00:02:30.360 American Ethnologist if you're an anthropologist, are behind memberships and paywalls. So if 34 00:02:30.400 --> 00:02:34.000 you're a student at a university, you may never notice this since your department 35 00:02:34.080 --> 00:02:38.120 or university often pays for that subscription. But depending on where you go to 36 00:02:38.199 --> 00:02:40.240 school, you may have noticed that you only have access to some journals but 37 00:02:40.400 --> 00:02:46.960 not others. So this is problem one scientific literature is locked behind a paywall 38 00:02:46.120 --> 00:02:51.800 that most individuals and even some organizations can't afford. Then the second is an 39 00:02:51.800 --> 00:02:55.159 issue around accessibility. If you reviewed literature is often written in heavy jargon, 40 00:02:55.719 --> 00:03:00.639 that is, specialist language that makes it difficult to end time consuming for non 41 00:03:00.759 --> 00:03:05.199 expert audiences or readers to digest. Now, this is more of a stylistic 42 00:03:05.280 --> 00:03:10.360 challenge, whereby academic writers follow along with and reinforce communication styles that work for 43 00:03:10.360 --> 00:03:15.719 the academic system but not for public access. Now sidebar Vita's debated, of 44 00:03:15.719 --> 00:03:20.080 course how well this kind of jargon works for academic systems too, and This 45 00:03:20.120 --> 00:03:22.680 is actually one of the reasons that I started this anthro life back in twenty 46 00:03:22.759 --> 00:03:25.759 nineteen, well nine years ago at the time of recording this episode in late 47 00:03:25.759 --> 00:03:30.039 twenty twenty two. Okay, so these two issues contribute in a big way 48 00:03:30.159 --> 00:03:35.120 to why we haven't seemed to be able to find ways to build consensus around 49 00:03:35.120 --> 00:03:39.919 expertise despite having mountains upon mountains of scientific data and researching our collective human database. 50 00:03:42.120 --> 00:03:46.199 Now consensus is working to change this conversation in really fascinating and impactful ways, 51 00:03:46.800 --> 00:03:51.039 and as well discussed in today's episode, we also need to ask ourselves 52 00:03:51.360 --> 00:03:55.680 and ask of society what we want in need out of access to scientific knowledge. 53 00:03:57.439 --> 00:03:59.960 So we'll be right back after a quick message from this episode sponsor, 54 00:04:00.120 --> 00:04:09.000 and can't wait to dive in with Ericles, you know, to kind of 55 00:04:09.039 --> 00:04:11.479 kick us off. I'd love to get a bit of a sense we kind 56 00:04:11.520 --> 00:04:14.520 of think about your superhero origin story, you know, the bit of your 57 00:04:14.560 --> 00:04:16.199 tale of like how you kind of came into the arena that you are now 58 00:04:16.319 --> 00:04:21.319 around the idea of using natural language processing and aiming to disrupt the search engine 59 00:04:21.319 --> 00:04:24.439 industry, which I love this idea, But tell me, how how did 60 00:04:24.480 --> 00:04:27.160 you get into this space in the first place. Yeah, thanks for having 61 00:04:27.160 --> 00:04:30.240 me, adam a great question. So the real origin of it is, 62 00:04:31.240 --> 00:04:34.519 you know, I come from a family of aconomics and scientists, but I 63 00:04:34.519 --> 00:04:39.680 am not one myself, and it's kind of created this this outsider complex within 64 00:04:39.720 --> 00:04:44.399 myself that made me really really interested in science, but made me an incredible 65 00:04:44.439 --> 00:04:49.959 amateur consuming it. So I found real value in consuming content that had scientists 66 00:04:50.040 --> 00:04:54.560 who are breaking down what the research said about a subject in a way that 67 00:04:54.639 --> 00:04:58.759 a non academic like myself could understand. So it was from getting that value 68 00:04:58.879 --> 00:05:00.600 that, like it was actually a six seven years ago, came up with 69 00:05:00.639 --> 00:05:04.199 the idea of what if there was a way to automate this process and what 70 00:05:04.240 --> 00:05:08.120 if there was a way to get answers to these types of questions that I 71 00:05:08.199 --> 00:05:13.480 have on demand from research from peer viewed sources. And then fast forward five 72 00:05:13.600 --> 00:05:16.839 or six years and I was actually my co founder of Christian Salem who I 73 00:05:16.839 --> 00:05:20.240 pitched the idea to all those years ago, that came back in the middle 74 00:05:20.279 --> 00:05:24.680 of COVID and was like, remember that idea you had, I think the 75 00:05:24.720 --> 00:05:28.079 world really needs this right now, And that's kind of how we got started 76 00:05:28.079 --> 00:05:30.319 on our journey, and then the real light bulb moment for us was after 77 00:05:30.360 --> 00:05:34.120 that we started to dig into what was the state of the technology, could 78 00:05:34.199 --> 00:05:40.399 technology actually solve this problem? And then learning that natural language processing and the 79 00:05:40.439 --> 00:05:44.800 advent of these large language models had really just occurred, and it was really 80 00:05:44.800 --> 00:05:50.519 this kind of perfect why now synergy of societal demand and technological feasibility interesting and 81 00:05:50.600 --> 00:05:54.920 it seemed like on one level two some of the best science and scientific discoveries 82 00:05:54.920 --> 00:05:57.680 are some kind of synchronicity, right where it's like the tech came together with 83 00:05:57.759 --> 00:06:00.240 the moment of the time and then and what people were looking for. It 84 00:06:00.279 --> 00:06:02.959 seems like it's a good triangulate that you know, Yeah, it's lucky in 85 00:06:03.000 --> 00:06:05.319 a lot of ways, right, Like things are more random than than we 86 00:06:05.319 --> 00:06:08.920 wish them to be. And both of those things had me and my co 87 00:06:09.000 --> 00:06:11.319 found or had nothing to do with us, right, there were completely things 88 00:06:11.319 --> 00:06:15.680 out of our control that kind of coalesced to make this possible, you know. 89 00:06:15.720 --> 00:06:18.439 I mean that's it's also like this idea that ethnographers talk about kind of 90 00:06:18.480 --> 00:06:23.480 from human sciences, anthropology, sociology, that some of the best power and 91 00:06:23.519 --> 00:06:28.120 innovation work comes from when we can see those connections of other people might miss, 92 00:06:28.199 --> 00:06:30.160 and so I'm interested in thinking about this idea and in terms of these 93 00:06:30.199 --> 00:06:32.240 pieces that came together, I mean, that's that's actually I think one of 94 00:06:32.360 --> 00:06:36.040 the powerful elements here too is that as we have this kind of perfect storm, 95 00:06:36.040 --> 00:06:39.279 as it were, for this need, I mean, one of the 96 00:06:39.319 --> 00:06:43.160 pieces that stands out you're talking about, you know, six seven years ago 97 00:06:43.639 --> 00:06:46.680 up to today. One of the big topics that comes to mind is things 98 00:06:46.680 --> 00:06:50.319 like the rise of misinformation online as a challenge point in terms of you know, 99 00:06:50.399 --> 00:06:54.680 fake news is the meme or hashtag of the day, right, but 100 00:06:54.759 --> 00:06:58.199 just this idea, I mean, how how did things like this filter into 101 00:06:58.240 --> 00:07:00.040 that that you're kind of thinking process? And I think that this is I 102 00:07:00.079 --> 00:07:03.959 think one of the big challenges that a lot of people face today is I 103 00:07:04.000 --> 00:07:09.439 don't know what information to trust online? And you know, how can technology, 104 00:07:09.480 --> 00:07:13.160 how can tools help us do that? Especially when sometimes those seemingly same 105 00:07:13.199 --> 00:07:15.639 tools are the ones that like make it so we can't trust it in the 106 00:07:15.639 --> 00:07:18.360 first place, right, like algorithms and advertisement. So how do we think 107 00:07:18.360 --> 00:07:21.560 about this kind of model, How can we set this stage of what is 108 00:07:21.560 --> 00:07:26.759 this problem space that we're coming into that we want to solve for. Yeah, 109 00:07:26.879 --> 00:07:29.319 no, no, really, well said across the board. So I 110 00:07:29.360 --> 00:07:31.519 think it might be helpful to like first just like level set on exactly what 111 00:07:31.560 --> 00:07:34.480 it is we're doing, and then I can kind of use that to segue 112 00:07:34.519 --> 00:07:39.000 into answering your question a bit. So, Yeah, we are Consensus, 113 00:07:39.120 --> 00:07:43.279 and we're a new search engine that allows you to type in plain English questions 114 00:07:43.879 --> 00:07:48.680 and be returned relevant findings from pure viewed literature. So basically the ideas instead 115 00:07:48.680 --> 00:07:53.759 of going to Google when you have the question of does magnesium actually help me 116 00:07:53.759 --> 00:07:56.519 sleep? You can type in that exact question. You don't have to do 117 00:07:56.560 --> 00:08:01.120 any special boolean searching or anything. You can actually just type in that question 118 00:08:01.439 --> 00:08:05.680 and our NLP will look through research papers and try to find claims being made 119 00:08:05.759 --> 00:08:09.680 about that question and deliver them to you in this nice aggregated, easy to 120 00:08:09.720 --> 00:08:11.920 consume list. And yeah, to go back to your question. Yeah, 121 00:08:13.120 --> 00:08:15.759 this is another part of our kind of why now story. Obviously, the 122 00:08:16.319 --> 00:08:20.519 societal demand and the clear currents of misinformation and that people were frustrated with their 123 00:08:20.519 --> 00:08:24.000 inability to get good information, but was really you know, a key part 124 00:08:24.000 --> 00:08:28.279 of this is kind of like identifying why that is the case, and you 125 00:08:28.279 --> 00:08:31.160 said it really well yourself that it's like, you know, we want technology 126 00:08:31.160 --> 00:08:33.279 to solve all these problems, but as it turned out that technology was what 127 00:08:33.399 --> 00:08:37.559 was creating a lot of these problems. But there's one real key component of 128 00:08:37.600 --> 00:08:41.600 it too that is driving That is that you know, most of these applications 129 00:08:41.600 --> 00:08:43.120 that we use and websites we visit, and specifically, you know, if 130 00:08:43.120 --> 00:08:48.000 we're talking about Google, the way that we search for consumer information, all 131 00:08:48.039 --> 00:08:52.519 this information has passed through this advertising filter. And that means, you know, 132 00:08:52.000 --> 00:08:56.240 explicitly sometimes just being shown ads. If you typed in downs magnesium helped 133 00:08:56.240 --> 00:09:00.799 me sleep, you'll get a bunch of ads for magnesium supplements. But the 134 00:09:00.919 --> 00:09:05.120 undercurrent of it is that because their incentives are to sell more advertising space, 135 00:09:05.480 --> 00:09:09.840 what they optimize for is continued engagement. So you know, the Google or 136 00:09:09.840 --> 00:09:16.759 Facebook or any of these places misinforming you isn't some giant back channel conspiracy theory 137 00:09:16.799 --> 00:09:20.720 of people in rooms and suits B and mic we want to try to make 138 00:09:20.759 --> 00:09:24.759 Adam believe something that isn't true. It's really just these organizations following their incentives 139 00:09:24.759 --> 00:09:28.879 and optimizing for what's going to making the most money, and that's continued engagement, 140 00:09:28.879 --> 00:09:33.720 because that means continuing selling ads to people's eyeballs. So in turn, 141 00:09:33.879 --> 00:09:37.919 that creates systems that are not designed to give us good information. They're designed 142 00:09:37.960 --> 00:09:41.440 to give us engaging information. And the problem is is that when algorithms try 143 00:09:41.440 --> 00:09:45.120 to learn what's engaging, a lot of times what they find out that what's 144 00:09:45.159 --> 00:09:52.440 engaging is content that is flashy and fiery and controversial. So in turn, 145 00:09:52.519 --> 00:09:56.240 you get shown lots and lots of that that has nothing to do with how 146 00:09:56.320 --> 00:10:00.799 factual the information is. So our real pushback on that is that we think 147 00:10:00.879 --> 00:10:07.639 that the future of search has to be in this premium verticalized subscription model, 148 00:10:07.879 --> 00:10:11.720 because that means that, you know, if we're not selling advertisements, all 149 00:10:11.759 --> 00:10:15.120 of our incentives will just be to deliver you the best product and deliver you 150 00:10:15.159 --> 00:10:18.240 the best information possible. And the hope is is that with NLP taking off 151 00:10:18.240 --> 00:10:22.519 so much, that products can get so good at these services that people will 152 00:10:22.559 --> 00:10:26.799 be willing to pay a small subscription fee for them. And that's when you 153 00:10:26.799 --> 00:10:31.759 can actually see this inflection point of some pushback on these free advertising based tools. 154 00:10:31.080 --> 00:10:33.960 I mean, that's super interesting and it has a bit of that If 155 00:10:35.000 --> 00:10:41.039 I'm a David and Goliath theme to it, right where certainly, how does 156 00:10:41.080 --> 00:10:43.200 that feel? I mean, I think about an exciting bit of challenge point, 157 00:10:43.200 --> 00:10:43.960 you know, how do you kind of a process idea of Like I 158 00:10:45.000 --> 00:10:48.799 think it's a what you're saying is super resonant that there is this ongoing challenge 159 00:10:48.799 --> 00:10:52.200 point that if the way we're presented data is not based on what's most factually 160 00:10:52.240 --> 00:10:56.480 correct or that even necessarily answers my question, But it's based on an advertising 161 00:10:56.519 --> 00:11:01.639 model that has made a multi billion dollar global search industry, how do we 162 00:11:01.679 --> 00:11:05.320 even disrupt them? You know? And so I think it's a really interesting 163 00:11:05.360 --> 00:11:07.320 question. I mean, maybe you have, like to're metaphere, you have 164 00:11:07.360 --> 00:11:09.519 the small stone, maybe have NLP to help us. I mean, that's 165 00:11:09.519 --> 00:11:13.000 exactly what my answer is going to be. Like, I'm not here to 166 00:11:13.039 --> 00:11:16.159 say that we're you know, on the precipice of disrupting Google and making them, 167 00:11:16.639 --> 00:11:20.399 like, you know, really taking away all this all the money that 168 00:11:20.440 --> 00:11:24.559 they're making. But the way to make a giant change is, you know, 169 00:11:24.600 --> 00:11:26.759 to kind of follow that David and Glide example somehow and find your wedge, 170 00:11:26.799 --> 00:11:30.360 find your stone. What is the thing that you can do really well 171 00:11:30.399 --> 00:11:33.799 and start small there. And for us, we think a perfect place to 172 00:11:33.840 --> 00:11:37.919 start is with scientific literature because it's, you know, the information that's the 173 00:11:37.919 --> 00:11:41.440 most valuable source of data in the world, and there's insights to question all 174 00:11:41.480 --> 00:11:43.960 sorts of questions that people have, and they're insights that are really hard to 175 00:11:45.000 --> 00:11:48.159 get by using the currently available tools. So we think it's a great place 176 00:11:48.200 --> 00:11:50.519 to start to maybe you know, disrupt the tiniest bit and then go from 177 00:11:50.519 --> 00:11:54.080 there. Right, you can't start somewhere if you just try to show up 178 00:11:54.080 --> 00:11:56.679 and say, hey, we're going to knock down Google, you have no 179 00:11:56.720 --> 00:11:58.919 shot to do it. You start somewhere, you get really good at that, 180 00:11:58.120 --> 00:12:01.919 you build an awesome product in that space, and then we'll see where 181 00:12:01.919 --> 00:12:03.960 it goes. That makes a good sense, and I think also also wise 182 00:12:05.039 --> 00:12:07.639 right, it's like we do what we can, but like picking that right 183 00:12:07.679 --> 00:12:11.039 movement. So I think it's it's really compelling that scientific literature kind of functions 184 00:12:11.039 --> 00:12:15.519 as that full comment as the piece we're using to then be able to rethink 185 00:12:15.679 --> 00:12:18.840 how search works, you know, and like so it's a very specific example, 186 00:12:18.279 --> 00:12:22.559 and so to kind of set the stage here in terms of the problem 187 00:12:22.600 --> 00:12:24.399 here, I think there's something interesting, right that. So I've gone through 188 00:12:24.399 --> 00:12:26.639 grad school a lot of my colleagues have, and you know, one of 189 00:12:26.639 --> 00:12:30.559 the big pain points, like you don't notice it when you're in school, 190 00:12:31.200 --> 00:12:33.639 is that you know all the journals that you're reading for research are behind paywalls 191 00:12:33.679 --> 00:12:37.600 because usually your school pays for those. But it's interesting, like, as 192 00:12:37.600 --> 00:12:41.799 I spent more time talking with other colleagues across the years in different disciplines or 193 00:12:41.840 --> 00:12:43.440 that went to different schools I had less funding, they wouldn't have access to 194 00:12:43.480 --> 00:12:46.039 some of the same journals I found. And that was interesting to note. 195 00:12:46.039 --> 00:12:50.720 It's this privilege of information and that feels worse when you think about it as 196 00:12:50.720 --> 00:12:54.240 scientific information rights, as peer view literature that we don't have access to. 197 00:12:54.759 --> 00:12:58.039 So I want to think about that with you, like this the impact of 198 00:12:58.039 --> 00:13:03.279 like inaccessible or making accessible scientific information, and I mean, I you know, 199 00:13:03.320 --> 00:13:05.480 we'll just straight up say, I think the current model is broken, 200 00:13:05.559 --> 00:13:07.360 right, like behind a paywall where a lot of research sits. I mean, 201 00:13:07.360 --> 00:13:11.200 this is also interesting in conversation with the fact that at the time of 202 00:13:11.279 --> 00:13:13.600 US recording, the Biden administration just passed this law that at the end of 203 00:13:13.639 --> 00:13:18.879 twenty twenty five into early twenty twenty six, any federally funded scientific research has 204 00:13:18.919 --> 00:13:22.960 to be immediately available upon publication, which is really interesting, and like saying 205 00:13:22.960 --> 00:13:26.759 that feels kind of crazy because it's twenty twenty two and we're still a few 206 00:13:26.799 --> 00:13:30.240 years away from that even being a thing, right, So that's it's crazy 207 00:13:30.279 --> 00:13:33.440 realized. Up until this point, scientific research is still behind a paywall for 208 00:13:33.480 --> 00:13:37.480 the most part, especially the most prestigious things like Nature or sell for you 209 00:13:37.480 --> 00:13:39.440 know, science or biology journals. Anyway, So I'm a rambling now, 210 00:13:39.480 --> 00:13:41.519 But tell me about this process and like how we can think about it, 211 00:13:43.080 --> 00:13:48.000 about this idea in terms of why we need accessible information and like and how 212 00:13:48.000 --> 00:13:50.840 do we think about this the current paradigm today. Yeah, you know, 213 00:13:50.080 --> 00:13:54.679 and when we've talked about like what our problem statement is the broad and way 214 00:13:54.720 --> 00:13:58.120 that we get you know, people's eyeballs to light up and what like the 215 00:13:58.159 --> 00:14:01.440 broad vision of it is this that which is broken. It's really hard to 216 00:14:01.480 --> 00:14:03.440 find good information. Like that's the way to paint it with this this giant 217 00:14:03.679 --> 00:14:09.600 brush. But what we call our sub problem statement has been that the most 218 00:14:09.720 --> 00:14:15.120 valuable and insight filled source of data on the planet is sitting behind paywalls. 219 00:14:15.120 --> 00:14:18.039 But not just sitting down paywalls. Because of that, it's only consumed by 220 00:14:18.039 --> 00:14:22.039 the same people who create it. And there's no way for somebody like myself 221 00:14:22.080 --> 00:14:24.799 to engage with literature unless I'm a part of a university and I'm doing research 222 00:14:24.879 --> 00:14:28.759 myself. And that's a shame on a million different ways, one of them 223 00:14:28.759 --> 00:14:33.440 being like the very obvious, just like economically, the way the system is 224 00:14:33.480 --> 00:14:37.159 just broken because it's many of these things are funded by public dollars, yet 225 00:14:37.399 --> 00:14:39.559 they're not accessible to the public, and that just is incredibly backwards. And 226 00:14:41.159 --> 00:14:45.159 but the good news is the trend is in everyone's favor. Like you said 227 00:14:45.159 --> 00:14:48.840 that, the Biden administration just passed this legislature, and even before that, 228 00:14:48.879 --> 00:14:52.360 the trend has been moving toward making things more open science or making things more 229 00:14:52.360 --> 00:14:58.080 open access. And our data partner they're called Semantic Scholar, and they're run 230 00:14:58.159 --> 00:15:01.720 out of the Allen Institute for Art Official Intelligence in Seattle. It's actually Paul 231 00:15:01.759 --> 00:15:09.000 Allen, Microsoft founders AI Research Arm and they're basically a Google Scholar competitor. 232 00:15:09.039 --> 00:15:13.120 They're a research data aggregator and they try to focus a lot on open access 233 00:15:13.159 --> 00:15:16.679 papers. So a lot of our corpus actually is entirely open access, not 234 00:15:16.720 --> 00:15:18.639 all of it is. We can still do our analysis over the papers, 235 00:15:18.639 --> 00:15:22.120 but if you're to try to go to the full text from some of our 236 00:15:22.120 --> 00:15:26.360 search results, you will run to the paywall unfortunately sometimes, but a good 237 00:15:26.399 --> 00:15:28.480 chunk of our papers are open access, and like you said, that trend 238 00:15:28.559 --> 00:15:33.360 is going to continue to be working in our favor. Very cool thinking about 239 00:15:33.360 --> 00:15:37.000 that. Actually, the idea in terms of I was curious to think about 240 00:15:37.000 --> 00:15:41.759 this and then maybe through Semitic scholar, but how do you approach partnerships in 241 00:15:41.840 --> 00:15:46.600 terms of like working with in terms of how does that consensus plug into partnerships 242 00:15:46.600 --> 00:15:48.519 in this regard in terms of aggregators versus search engine versus kind of do the 243 00:15:48.600 --> 00:15:52.840 NLP the text act that helps put these pieces together? And with that, 244 00:15:52.919 --> 00:15:54.720 I'm curious, you know, so, as an anthropologist, do we get 245 00:15:54.759 --> 00:15:58.440 anthropology journals as part of this as well? Are there? I know they're 246 00:15:58.440 --> 00:16:02.240 not as fancy as oftentimes by your chemistry, journals don't get as much pressed, 247 00:16:02.240 --> 00:16:03.200 but I'm just curious in terms of that, like how do you select 248 00:16:03.279 --> 00:16:07.559 for the kinds of articles journals that are included in the kind of the search 249 00:16:07.639 --> 00:16:11.399 process. Yeah, so to kind of start with, like the partnership's question, 250 00:16:11.720 --> 00:16:18.000 we're incredibly fortunate that we're operating in a space with lots of research organizations 251 00:16:18.039 --> 00:16:22.159 that are nonprofits and that are can be fairly open to partnerships like this. 252 00:16:22.720 --> 00:16:26.080 There is like a big caveat though, that you many times you need to 253 00:16:26.120 --> 00:16:30.360 basically prove that we have mission alignment and stay true to those ideals of promoting 254 00:16:30.399 --> 00:16:33.600 open science and promoting access to scientific literature. That is basically how we got 255 00:16:33.639 --> 00:16:37.360 the partnership with Semantic Scholar was. You know, they're willing to do partnerships 256 00:16:37.399 --> 00:16:41.480 with other research organizations basically without any vetting, but we had to go through 257 00:16:41.480 --> 00:16:45.519 this vetting process as a commercial entity to say, hey, here's how we're 258 00:16:45.559 --> 00:16:48.200 going to be using it, here's how you benefit, and here's how we 259 00:16:48.240 --> 00:16:52.720 push forward your mission. We really tried to work hard to abide by a 260 00:16:52.759 --> 00:16:55.639 lot of those values so we can be attractive to have partner. I mean 261 00:16:55.679 --> 00:16:57.840 one, because we also want to abide by those values, but also it 262 00:16:57.840 --> 00:17:02.759 has the other benefit of being able to partner with a lot of amazing organizations 263 00:17:02.799 --> 00:17:07.119 that have incredible mission statements of promoting access to scientific literature as far as anthropology 264 00:17:07.160 --> 00:17:10.319 journals, I don't know that answer off the top of my head. We 265 00:17:10.400 --> 00:17:12.680 have pretty dawn good coverage. We'vebout two hundred million papers in our corpus, 266 00:17:12.920 --> 00:17:17.240 so I would be shocked if there are not some anthropology journals is a part 267 00:17:17.240 --> 00:17:18.799 of it. Fortunately, I don't have that answer off the top of my 268 00:17:18.839 --> 00:17:22.480 head. And as far as like selecting, it's really whatever we can get 269 00:17:22.480 --> 00:17:26.359 our hands on it. We're not trying to select by domain. The way 270 00:17:26.359 --> 00:17:30.880 that we've trained our algorithms is to try to be domain agnostic of extracting findings 271 00:17:30.880 --> 00:17:34.440 from papers. And I think one of the great surprises of our early product 272 00:17:34.799 --> 00:17:41.319 is how well it does with non medical healthcare related questions I publish. If 273 00:17:41.599 --> 00:17:44.759 you sign up for our product and become a free user, I send out 274 00:17:44.839 --> 00:17:48.319 a newsletter every week that's like, interesting thing I learned this week query of 275 00:17:48.359 --> 00:17:51.079 the week basically, so it's like it's a question that I asked and what 276 00:17:51.160 --> 00:17:53.880 the results were, And this week's was does the death penalty reduce crime? 277 00:17:55.440 --> 00:17:59.519 And our product did an awesome job of servicing research findings about that. So 278 00:18:00.039 --> 00:18:02.960 using that as an example that the domains can be a bit you know less 279 00:18:03.000 --> 00:18:04.559 hard science than you typically think of when you hear a peer of view of 280 00:18:04.559 --> 00:18:08.759 literature, and our product can still do do a pretty decent job of finding 281 00:18:08.799 --> 00:18:14.559 relevant conclusions about those questions. That's super interesting. I think something that stands 282 00:18:14.559 --> 00:18:18.440 out to me here too is a that's really awesome to hear that the algorithm 283 00:18:18.480 --> 00:18:22.759 that you'll put together is able like kind of thinking like a thinking domain agnostically, 284 00:18:22.759 --> 00:18:23.799 I think is really important. It's making me think of, like, 285 00:18:23.920 --> 00:18:27.519 you know, my dreams of being a zocast and journal or that I can 286 00:18:27.519 --> 00:18:33.000 like connect ideas together with double brackets and like be whatever subject agnostic and just 287 00:18:33.000 --> 00:18:34.880 connect ideas as they as a link together and also mean more like how the 288 00:18:34.920 --> 00:18:37.279 human brain works too, So I think that's something else that I think it 289 00:18:37.279 --> 00:18:41.480 gets me. You're really excited about the possibilities that you know, kind of 290 00:18:41.519 --> 00:18:47.279 what the emerging world of natural language processing can do is that we can train 291 00:18:47.359 --> 00:18:51.400 it and train ways that will kind of use the same pathways that our brains 292 00:18:51.480 --> 00:18:52.880 might write, that we don't get stuck in silos of like I only want 293 00:18:52.880 --> 00:18:56.160 to get journals that are from biology or just from anthropology or whatever it is. 294 00:18:56.359 --> 00:19:00.759 But actually the death penalty question is really interesting in terms of probably the 295 00:19:00.759 --> 00:19:03.759 diversity of results that you got back. I'm also thinking of another example. 296 00:19:03.799 --> 00:19:07.519 I saw the blog you recently did on the use of psilocybin medication, like 297 00:19:07.559 --> 00:19:11.440 based on Michael Pollan's How to Change Your Mind series, and like that was 298 00:19:11.440 --> 00:19:15.240 a really cool I think example too in terms of a showing the lack of 299 00:19:15.319 --> 00:19:18.119 a better term, the pop culture application of asking of scientific literature. I 300 00:19:18.119 --> 00:19:21.640 mean, that's really what Michael Pollan's work is all about, right and death 301 00:19:21.640 --> 00:19:23.599 thot. He's a great question too in that regard because also the other thing 302 00:19:23.599 --> 00:19:26.559 too is even that space, how do we even know how much is out 303 00:19:26.559 --> 00:19:29.759 there? Right? And so thinking of the power of that, we can 304 00:19:29.880 --> 00:19:32.640 use big data in ways that the human brain doesn't quite do so well. 305 00:19:32.680 --> 00:19:36.720 There are just like literally searching Alexis nexus right or bioarchive, you're only going 306 00:19:36.759 --> 00:19:38.279 to see so much in terms of returned results that you can only read. 307 00:19:38.279 --> 00:19:41.279 But again, this idea in terms of how can we kind of cross those 308 00:19:41.279 --> 00:19:44.920 boundaries, I think it's really exciting. So is that what you've found too? 309 00:19:44.960 --> 00:19:48.799 That is it surprising how good natural language processing has become You've you've kind 310 00:19:48.799 --> 00:19:52.240 have been steeped in it for a while. But you know, how how 311 00:19:52.279 --> 00:19:55.000 are we seeing it evolveal So, you know, even through consensus, right 312 00:19:55.039 --> 00:19:57.000 as you think about partnerships and what you're searching for, how have we seen 313 00:19:57.039 --> 00:20:03.079 that change? Yeah, even we work in the world of NLP, it 314 00:20:03.119 --> 00:20:07.559 continues to pleasantly surprise me every day. I would earnestly say, what you 315 00:20:07.640 --> 00:20:10.759 said is really interesting of like how it kind of operates in some ways the 316 00:20:10.799 --> 00:20:15.000 way that the human brain works. And the real advent of these the new 317 00:20:15.039 --> 00:20:18.880 technologies. They're called large language models, and the unique feature about them is 318 00:20:18.880 --> 00:20:25.240 that they come pre trained and they come with this underlying knowledge of how human 319 00:20:25.319 --> 00:20:29.799 language works. So the big one that you've likely seen the news and listeners 320 00:20:29.839 --> 00:20:33.920 have probably seen on Twitter or something is GPT three open aies model. It 321 00:20:33.079 --> 00:20:37.279 is the part of the underlying technology of DALI, which is that you type 322 00:20:37.279 --> 00:20:41.799 in a prompt and you get back like an art image. People have been 323 00:20:41.880 --> 00:20:47.359 loving that on Twitter, and that is trained on basically like the entire Internet 324 00:20:47.519 --> 00:20:51.559 of articles, of text articles, and the way they teach it to understand 325 00:20:51.599 --> 00:20:56.680 language is they showed examples of articles with certain words blanked out, and they 326 00:20:56.680 --> 00:21:00.480 try to make it predict what word goes in those spaces, and then you 327 00:21:00.519 --> 00:21:03.920 show that literally billions of examples, and over time it can learn how to 328 00:21:04.000 --> 00:21:08.799 accurately predict what word should go here, so it actually understands like the whole 329 00:21:08.839 --> 00:21:12.160 context of what's going on around it. And yeah, it's shown the entire 330 00:21:12.240 --> 00:21:17.480 text before it makes that prediction, so it's trying to understand text holistically in 331 00:21:17.519 --> 00:21:21.519 the way language works. And then what happens is when you try to do 332 00:21:21.559 --> 00:21:23.839 one of these specific tasks, like we're doing that consensus, you do a 333 00:21:23.839 --> 00:21:29.680 process what's called fine tuning, and you give it more custom training data specifically 334 00:21:29.720 --> 00:21:32.599 for the task that you're trying to do, and then you teach it how 335 00:21:32.640 --> 00:21:36.519 to do that task. But it comes with that underlying knowledge first. And 336 00:21:36.559 --> 00:21:37.920 if you think about the way a human would learn how to do a task, 337 00:21:38.039 --> 00:21:41.880 it's very similar to that. Right, if you're being taught something, 338 00:21:41.119 --> 00:21:45.079 you don't come in as this just blank slate. You have all of these 339 00:21:45.079 --> 00:21:48.480 patterns that you've learned about about other tasks you've completed in your life. When 340 00:21:48.480 --> 00:21:49.640 you come to a task, will learn how to do it. So it 341 00:21:51.079 --> 00:21:53.200 really is in some ways I'm mimicking the way a human would learn how to 342 00:21:53.279 --> 00:21:57.200 do something, and in the context of how we do it, Basically, 343 00:21:57.279 --> 00:22:00.000 yeah, there's a pre train model, and then we give it examples of 344 00:22:00.119 --> 00:22:06.799 We hire scientists to go through research papers and mark up our authors making their 345 00:22:06.839 --> 00:22:10.440 claims. So it basically it's just a bunch of ones and zeros next to 346 00:22:10.519 --> 00:22:15.160 sentences where it says zero, this is background information zero, this is methods 347 00:22:15.559 --> 00:22:18.759 zero, this is more background information one when they say these results suggest that, 348 00:22:18.839 --> 00:22:22.720 so on and so forth, and then when you give it enough examples 349 00:22:22.720 --> 00:22:25.599 of that, you can feed that to a model to learn how to pick 350 00:22:25.640 --> 00:22:29.559 out those sentences. Very cool. Even this idea, I think is really 351 00:22:29.559 --> 00:22:33.039 exciting in importance of context, right, and part of it is like, 352 00:22:33.039 --> 00:22:36.400 hey, how do we train models? But then also recognizing that I love 353 00:22:36.400 --> 00:22:37.559 the way you said that too, where it's that we're never just born as 354 00:22:37.559 --> 00:22:41.000 blank slates, right, we always have a set of pattern recognition and context 355 00:22:41.119 --> 00:22:45.319 that like shape who we are. And like the fact that we are able 356 00:22:45.400 --> 00:22:48.119 to build technology that follows that too is exciting. You know. Obviously it 357 00:22:48.119 --> 00:22:52.160 means we got we got to watch it and it makes it super flexible, 358 00:22:52.240 --> 00:22:53.480 right, Like it makes it so you can teach you to do a whole 359 00:22:53.519 --> 00:22:56.559 host of tasks. Yeah, I think that's right on too, And I 360 00:22:56.599 --> 00:22:59.279 think that's that's in some of the most important power. It's like, at 361 00:22:59.319 --> 00:23:03.920 the same time as we're through neuroscience through psychology really coming to understand the plasticity 362 00:23:03.920 --> 00:23:06.440 of the human mind, right and the fact that we can't even reshape neural 363 00:23:06.480 --> 00:23:11.079 pathways for training through different kinds of lifestyle, even that we can see this 364 00:23:11.160 --> 00:23:14.960 kind of happen also on a technological software level is really fascinating that they're kind 365 00:23:14.960 --> 00:23:18.000 of happening simultaneously, which is interesting. And I though that's because we just 366 00:23:18.119 --> 00:23:19.000 learned to see one and that we can see the other one now, or 367 00:23:19.079 --> 00:23:22.559 kind of how that happened. Maybe a chicken and egg kind of problem there. 368 00:23:22.599 --> 00:23:23.519 You know, it's kind of funny to seek some co evolution there, 369 00:23:23.519 --> 00:23:27.000 I suppose. So I'd love to kind of think with this idea. So 370 00:23:27.039 --> 00:23:32.440 as we're training our software and also working with scientists. It's a very cool 371 00:23:32.960 --> 00:23:34.079 point too that we've got humans as part of that. I mean, it's 372 00:23:34.119 --> 00:23:37.640 one of those things that I've conversation I had with Byron Reese on the podcast 373 00:23:37.640 --> 00:23:41.720 a few years ago and more recently he's the CEU of giggo Oman like did 374 00:23:41.720 --> 00:23:45.759 a podcast on Voices and AI and it was really really into the role of 375 00:23:45.799 --> 00:23:49.359 AI and changing how technology works and like obviously going into things like NLP and 376 00:23:49.759 --> 00:23:52.359 you know, is this interesting idea that a lot of times that there's a 377 00:23:52.400 --> 00:23:56.240 common kind of you know, pop cultural notion that AI is something that will 378 00:23:56.279 --> 00:24:00.400 become generalizable, that it will become essentially conscious at some point, right, 379 00:24:00.440 --> 00:24:03.599 And like this you know, filters into our you know, bigger fear narratives 380 00:24:03.599 --> 00:24:07.759 of like the terminator and the matrix and things like that. But realistically, 381 00:24:07.799 --> 00:24:10.000 you know, most models say a I can't do that, and it won't 382 00:24:10.000 --> 00:24:12.440 and even the way that we're talking about this and that AI often at this 383 00:24:12.440 --> 00:24:15.240 point needs to work in partnership with people still, right, Like it's learning 384 00:24:15.279 --> 00:24:18.759 how to interpret context with humans because we're the ones that do without even thinking 385 00:24:18.799 --> 00:24:22.359 about it. Right. I think this is like an important piece for folks 386 00:24:22.359 --> 00:24:26.680 to like pause and think on. Is that AI is super powerful and it 387 00:24:26.880 --> 00:24:30.160 gives our brains a super power to connect ideas. But really it's also to 388 00:24:30.200 --> 00:24:33.759 your point, is based on a lot of how we do interpretation as people 389 00:24:33.079 --> 00:24:37.039 and learns from that. So it is interesting too that it's that as much 390 00:24:37.039 --> 00:24:38.759 as we're students of the way we can connectnowledge AI as a student of us. 391 00:24:38.799 --> 00:24:45.480 Also, we're going to take a quick break. Just wanted to let 392 00:24:45.519 --> 00:24:48.039 you know that we're running ads to support the show down. We'll be right 393 00:24:48.079 --> 00:24:56.599 back. Yeah. I was on a podcast the other weeks. We were 394 00:24:56.599 --> 00:25:00.559 talking about NLP and he was talking about it as like he was reading an 395 00:25:00.640 --> 00:25:03.200 article that there are gonna have to work to like fight plagiarism in schools because 396 00:25:03.200 --> 00:25:07.240 students could use AI generators to write papers, which totally can happen, right, 397 00:25:07.279 --> 00:25:11.640 and I can do that now, But I was saying my response to 398 00:25:11.640 --> 00:25:14.559 them was, if you're a student that's trying to do that, you know 399 00:25:14.599 --> 00:25:18.640 there is like a prerequisite step in some ways that you have to get a 400 00:25:18.680 --> 00:25:22.799 bunch of training data of a papers show up model to then learn how to 401 00:25:22.799 --> 00:25:25.680 do an a paper. So the work that it would be required to you 402 00:25:25.720 --> 00:25:27.559 know, find tune a model to do exactly what you needed it to do 403 00:25:27.640 --> 00:25:30.880 for your class, you know, is way more of the way more work 404 00:25:30.920 --> 00:25:34.039 than just writing the paper yourself. Now, with that said, like a 405 00:25:34.119 --> 00:25:40.680 generalizable generation model could write you like a paper that could pass as a human 406 00:25:40.680 --> 00:25:42.920 paper, but it probably isn't very good for your class. And in order 407 00:25:42.960 --> 00:25:45.759 to really get it specifically at the task in the way that you would need 408 00:25:45.759 --> 00:25:48.640 it written for your class, you would probably need examples of that. So 409 00:25:48.880 --> 00:25:52.599 a prerequisite for having a model spit you out a papers is to write the 410 00:25:52.599 --> 00:25:57.079 a papers yourself. That's a good point, right, It's like you can 411 00:25:57.200 --> 00:26:03.119 plagiarize, but right, yeah, you can hire someone else to do exactly 412 00:26:03.200 --> 00:26:04.279 right. But that's I think it's really interesting, and I think that's right 413 00:26:04.319 --> 00:26:07.599 on where there's can be a fear of adopting technology, especially one like that 414 00:26:07.680 --> 00:26:11.160 feels like it gets close to home, right, like how am I writing? 415 00:26:11.200 --> 00:26:14.720 You know? But it's it's based kind of in the similar concern that 416 00:26:14.759 --> 00:26:17.359 we kind of talked about up top, where there's there's a lot of mistrust 417 00:26:17.440 --> 00:26:18.720 online on one level, like I don't know if I can trust the information. 418 00:26:19.200 --> 00:26:22.319 And so we've also seen, you know, in the past ten years, 419 00:26:22.640 --> 00:26:26.039 the rise of plagiarism checkers, like in word, Microsoft Word has one 420 00:26:26.079 --> 00:26:30.440 you know, Google will look forward to. And so I've even seen AI 421 00:26:30.559 --> 00:26:34.240 writers that then also plagiarized check itself, which is interesting because it doesn't know 422 00:26:34.279 --> 00:26:37.519 if it's plagiarizing until it looks and says, okay, I wrote the thing. 423 00:26:37.559 --> 00:26:41.079 Oops, yep, I borrowed that from Wikipedia, which I think is 424 00:26:41.119 --> 00:26:44.720 interesting as well. So like it doesn't know it's plagiarizing until then it writes, 425 00:26:44.799 --> 00:26:47.400 then it checks, you know oftentimes or the ones that I've seen in 426 00:26:47.400 --> 00:26:49.480 anyway, But I think that's that's a really interesting question too in terms of 427 00:26:49.759 --> 00:26:55.240 there's the the work that we have to do in terms of providing more access 428 00:26:55.279 --> 00:26:57.519 to scientific literature and making it more accessible and easier easier for folks to get, 429 00:26:57.599 --> 00:27:02.119 especially if for lay people. Right when we already looked into this question 430 00:27:02.160 --> 00:27:06.480 of the paywall problem, But then how do we make this usable for your 431 00:27:06.519 --> 00:27:08.480 average lay your lay person too who's never going to get a master's at PC, 432 00:27:08.559 --> 00:27:11.480 he doesn't need one, doesn't moment, but they're interested in like yet 433 00:27:11.680 --> 00:27:15.720 is our COVID nineteen vaccines effective? Right? What are the benefits of mindfulness? 434 00:27:15.720 --> 00:27:18.480 There's some examples on your website. I thought they were the good thought 435 00:27:18.559 --> 00:27:22.519 starters. So how do we think about that, Like, how do we 436 00:27:22.559 --> 00:27:26.000 help build this in a way that's useful for everybody. Yeah, it's a 437 00:27:26.039 --> 00:27:33.039 great question, and I think that it's it's finding the balance of not oversimplifying 438 00:27:33.119 --> 00:27:38.519 and perverting meaning while trying to use these tools to make things consumable and digestible 439 00:27:38.519 --> 00:27:41.759 and easy to use. I think a real big emphasis on that last one 440 00:27:41.920 --> 00:27:47.279 of easy to use, because that says nothing of the digestibility some too, 441 00:27:47.319 --> 00:27:49.319 always of the text that you're showing. So I'll kind of tie this all 442 00:27:49.319 --> 00:27:53.640 together that what we're doing is we're not generating any text, because we think 443 00:27:53.680 --> 00:27:59.640 that is a fairly dangerous road to go down of trying to summarize what a 444 00:27:59.759 --> 00:28:03.440 sign this is saying without any checks on what does the source text actually say, 445 00:28:03.480 --> 00:28:06.880 without being able to show that to the user. So we want to 446 00:28:06.920 --> 00:28:10.680 do our best to always just extract the information and show them a real quote 447 00:28:10.680 --> 00:28:14.599 from the literature. So I'm going with all this is that that is how 448 00:28:14.640 --> 00:28:17.680 you use the tool to make it easy. You try to find the sentence 449 00:28:18.039 --> 00:28:21.240 where they're answering the question and then just extract it. And you might be 450 00:28:21.279 --> 00:28:25.319 sacrificing a little bit of digestibility because you know, scientists still love to write 451 00:28:25.359 --> 00:28:27.640 in jargon, and that sentence maybe jargon filled but we kind of think that, 452 00:28:27.759 --> 00:28:30.599 you know, this is by no means perfect, and we're going to 453 00:28:30.640 --> 00:28:33.359 continue to iterate on it, but we found that to be this kind of 454 00:28:33.440 --> 00:28:37.480 nice balance of using these tools to make things easier by finding the answers, 455 00:28:37.799 --> 00:28:42.839 but not going overboard and not oversaturating and oversimplifying it by doing all this like 456 00:28:42.880 --> 00:28:48.000 super advanced generation, where we could be totally perverting the underlying meaning of what 457 00:28:48.039 --> 00:28:52.519 the scientist is trying to say. So I think in total, my answer 458 00:28:52.640 --> 00:28:55.839 is like trying to find that balance of where can you use these tools to 459 00:28:55.880 --> 00:29:02.680 circumnavigate things and make things easier while not go overboard and oversaturated and oversimplifying. 460 00:29:03.240 --> 00:29:07.119 Yeah, that's fundamentally important too. And even the software itself, I actually 461 00:29:07.119 --> 00:29:11.119 really appreciated this. It has a beta tag on the website and it says, 462 00:29:11.759 --> 00:29:12.640 do your due diligence from this work. You know, it's like, 463 00:29:12.640 --> 00:29:15.240 don't just take this as gospel, but but you know, still be a 464 00:29:15.279 --> 00:29:18.759 scientist. Still be scientific, and you're thinking and say, look at this, 465 00:29:18.799 --> 00:29:21.559 and then continue to look around as part of your process. I think 466 00:29:21.559 --> 00:29:23.680 that's actually really fascinating too, because I think you're one hundred percent right where 467 00:29:23.680 --> 00:29:26.519 it's like in the in the world of view X and usability, it's like, 468 00:29:26.519 --> 00:29:27.519 we want to make things easy to use for our users, want them 469 00:29:27.519 --> 00:29:30.960 to be frictionlessen to be a good experience. But especially in this case, 470 00:29:32.000 --> 00:29:34.839 like when we're looking at also scientific data, I think it's really really fascinating 471 00:29:34.839 --> 00:29:37.279 and important that the point that you've noted that it's we have to make sure 472 00:29:37.279 --> 00:29:41.960 that we're not either oversimplifying the complexity of an answer, but then also that 473 00:29:42.000 --> 00:29:45.039 means we're doing interpretation of it right, and that then may change what we 474 00:29:45.079 --> 00:29:48.079 are then changing the meaning of if we change the wording to make it sound 475 00:29:48.119 --> 00:29:52.319 more or less jargoning, we realize we're adding a second layer there in terms 476 00:29:52.319 --> 00:29:56.160 of interpreting what we're seeing. And that's that's an interesting and like big conundrum 477 00:29:56.200 --> 00:30:00.079 that you know, whole different like can of worms as it were, but 478 00:30:00.160 --> 00:30:03.839 interesting to think about. Yeah, we're actually training some de jargonizing models. 479 00:30:03.920 --> 00:30:07.519 But we were just having this conversation earlier today. By how we want to 480 00:30:07.559 --> 00:30:10.799 you know, we're still a ways away from having that in the product, 481 00:30:11.160 --> 00:30:14.279 but our idea for how we'd want that in the product is a little button 482 00:30:14.319 --> 00:30:17.880 that lets you toggle that on. So you're you're basically saying, I am 483 00:30:17.920 --> 00:30:22.680 acknowledging that now I want to see the slightly more simplified version of these claims. 484 00:30:22.000 --> 00:30:26.480 And you know, you're very acutely aware that you should be doing more 485 00:30:26.799 --> 00:30:30.079 due diligence when you're interpreting one of those results with a little like toggle bar 486 00:30:30.119 --> 00:30:33.480 at the top right of the screen. And yeah, I appreciate you calling 487 00:30:33.480 --> 00:30:36.319 out the beta dag because yeah, we think that's that's important. We're still 488 00:30:36.319 --> 00:30:38.240 early, We by no means do everything perfectly, and even if we did 489 00:30:38.319 --> 00:30:41.759 extract all these answers perfectly, like, it's still is important. If you're 490 00:30:41.759 --> 00:30:45.920 going to be, you know, making an action item in your life because 491 00:30:45.920 --> 00:30:48.039 of one of these answers, you should probably pop up in the paper and 492 00:30:48.119 --> 00:30:51.799 read all about it. Like it's it's great to be able to get a 493 00:30:51.920 --> 00:30:55.559 landscape of evidence really quickly. That's why we built the product. But context 494 00:30:55.599 --> 00:30:59.000 is always incredibly important. You should do your due diligence when you're when you're 495 00:30:59.000 --> 00:31:02.599 going through these find And then yeah, to your point about friction less, 496 00:31:02.640 --> 00:31:06.079 Yeah, that is we've tried to make the product really easy to use, 497 00:31:06.119 --> 00:31:11.480 because that's a way to make it approachable without having any risk of anything to 498 00:31:11.519 --> 00:31:15.119 do with perverting the signs, right, like justin the log inflow and the 499 00:31:15.119 --> 00:31:18.880 way you can search and the try searching buttons and all of those things, 500 00:31:18.880 --> 00:31:22.039 like, right, those are things that are completely outside of the information. 501 00:31:22.079 --> 00:31:26.079 We're showing their usability in a much more like aesthetic way. So we're trying 502 00:31:26.119 --> 00:31:27.759 to do everything we can to be really good at those things, because those 503 00:31:27.759 --> 00:31:32.799 are ways we can make this more approachable for anybody without having to worry if 504 00:31:32.799 --> 00:31:36.359 we're dumbing down the signs too much. That's a great point and I think 505 00:31:36.400 --> 00:31:38.759 super valuable too, because you know, one thing I have to hear too 506 00:31:38.839 --> 00:31:42.240 is that you know, working in consumer research, whether it's working and doing 507 00:31:42.279 --> 00:31:47.480 kind of design research with different clients, that there can be this I don't 508 00:31:47.559 --> 00:31:48.440 fear is too strong of a word, but just you know, if folks 509 00:31:48.440 --> 00:31:52.759 are not used to scientific literature research, like it's a scary space, right 510 00:31:52.759 --> 00:31:55.559 because it is. It's jargony, it's behind this weird pay well, it's 511 00:31:55.640 --> 00:31:57.240 complex, you know, how do I understand it? And so I think 512 00:31:57.240 --> 00:32:00.400 there's this, you know, to your point, like a real value in 513 00:32:00.440 --> 00:32:04.839 making the experience, the digital experience of coming even to the search itself welcoming 514 00:32:04.920 --> 00:32:07.400 Pogle scholar hasn't changed in twenty years. It's the same intervas for the best 515 00:32:07.440 --> 00:32:10.400 decade. That's so true. That's so true. And it's hell to try 516 00:32:10.400 --> 00:32:14.319 to start something on there. Right, you can't get like internal parts of 517 00:32:14.359 --> 00:32:15.559 papers. You can get titles, but you know, there's nothing in terms 518 00:32:15.599 --> 00:32:19.880 of really finding a way in so to me, also this really fills a 519 00:32:19.880 --> 00:32:22.200 big gap like that, right, in terms of running that question of access. 520 00:32:22.559 --> 00:32:28.079 But then also the rise also in you know, natural language itself. 521 00:32:28.119 --> 00:32:30.920 I mean, this is something that I as a consumer side, came into 522 00:32:30.079 --> 00:32:32.799 being interested in this, Like with calendars, right, meet with Eric on 523 00:32:32.839 --> 00:32:35.680 Tuesday at two pm, you know, like I could type that in then 524 00:32:35.680 --> 00:32:37.119 it would at an event to my calendar. I thought was super cool, 525 00:32:37.160 --> 00:32:42.519 you know, I think then being able to do this and asking scientific questions 526 00:32:42.680 --> 00:32:45.920 is incredible, right, what a really interesting way of having to then think, 527 00:32:45.960 --> 00:32:49.960 okay, you know, how do I formulate my hypothesis in a three 528 00:32:50.000 --> 00:32:52.359 part question you know for a paper, way harder than just saying, yeah, 529 00:32:52.400 --> 00:32:55.319 what are the benefits of mindfulness? And then kind of returning results in 530 00:32:55.559 --> 00:33:00.319 that space I think is really interesting and important because that also plays it, 531 00:33:00.359 --> 00:33:01.920 I think, an important role in accessibility. Now we're folks like they have 532 00:33:01.920 --> 00:33:05.440 a point of access, they may have not hit otherwise. Yeah, we 533 00:33:05.519 --> 00:33:08.599 think there's three main differentiators between us and a search engine like Google scholar, 534 00:33:08.640 --> 00:33:12.759 and we think it's like the three things that are the worst part of that 535 00:33:12.839 --> 00:33:15.119 experience of using Google scholars. So one is exactly like you said, of 536 00:33:15.119 --> 00:33:20.319 the actual trying to get information and related to your question, Google Schullar does 537 00:33:20.319 --> 00:33:22.039 not allow you to type in a plain English question. If you type in 538 00:33:22.079 --> 00:33:25.359 your research question, you're not going to likely get back very relevant results. 539 00:33:25.759 --> 00:33:31.079 You have to do this like you know, kind of insider exclusive boolean searching 540 00:33:31.160 --> 00:33:34.640 to get it to come up with what you want. And then there's that 541 00:33:34.720 --> 00:33:37.240 next part of all they do is just give you links, which you know, 542 00:33:37.279 --> 00:33:42.400 if you're doing a giant lit review and you're a PhD and you know 543 00:33:42.480 --> 00:33:45.119 you have to go dedicate you know, forty eight hours of or fifty hours 544 00:33:45.119 --> 00:33:49.759 of your next week to do that, fine, like whatever, But you 545 00:33:49.799 --> 00:33:52.720 know, your everyday consumer who just wants to ask a question and get it 546 00:33:52.759 --> 00:33:55.839 from good sources. You're not about to go through all those papers and spend 547 00:33:55.839 --> 00:33:59.599 considerable time. So we wanted to make a way that makes it easier to 548 00:33:59.599 --> 00:34:02.480 get information from those papers where we're not just delivering new links, we're actually 549 00:34:02.559 --> 00:34:07.319 surfacing the insights to your question. Now, if you're then objective is to 550 00:34:07.359 --> 00:34:09.559 go deeper and this is a better way to surface papers to look into. 551 00:34:09.840 --> 00:34:13.960 Great. But if it's also just a way to quickly get an understanding of 552 00:34:14.000 --> 00:34:16.800 what the research says about a subject or a question, also great, Right, 553 00:34:17.159 --> 00:34:20.960 And then the last one is a little more subjective. But just their 554 00:34:20.960 --> 00:34:23.000 interface sucks, and they're going to always continue to try to build it to 555 00:34:23.039 --> 00:34:29.320 be a real product experience and not just some clunky academic search engine. It 556 00:34:29.400 --> 00:34:31.920 matters though, I mean, like Apple really kicked up a consumer design language 557 00:34:31.920 --> 00:34:35.280 as needing to be important, you know, with iPhone, and so I 558 00:34:35.360 --> 00:34:37.960 mean through today, it's it's like, it's interesting to see that it's been 559 00:34:37.000 --> 00:34:42.239 such a broad respect now for things like color palette and typography and as matter, 560 00:34:42.360 --> 00:34:44.960 right, and like, where is the affordances that I can easily see 561 00:34:44.960 --> 00:34:46.280 I can click on on a website make a difference, you know, And 562 00:34:46.360 --> 00:34:50.880 like again, having these come together around scientific research, I think is like 563 00:34:50.920 --> 00:34:53.440 that itself is also somewhat revolutionary. I guess I'm trying to get at like 564 00:34:53.639 --> 00:34:59.920 that we don't often have accessibility front and center as a point of value as 565 00:35:00.000 --> 00:35:02.920 a mission orientation for scientific research, which I think is super important in that 566 00:35:02.960 --> 00:35:06.840 regard. I think that thing stands out that I'd love to hear about is 567 00:35:07.400 --> 00:35:09.719 you know, we talked a bit about semantic scholar, but then you've also 568 00:35:09.719 --> 00:35:13.800 talked about I've seen your website like other partnerships like with sie Score, and 569 00:35:13.960 --> 00:35:16.159 this is important because it's how do we build trust in what we're finding? 570 00:35:16.239 --> 00:35:20.199 And so this partnership tell me a bit about this too, or if there's 571 00:35:20.199 --> 00:35:22.599 other ones that like this that are kind of helped put together the bigger web 572 00:35:22.639 --> 00:35:25.480 of how do we build the tools and the trust. No, it's a 573 00:35:25.480 --> 00:35:30.599 great question. Yeah, we're extremely honored and privileged to be able to partner 574 00:35:30.639 --> 00:35:34.880 with size Score. So they are they're actually another private company, but similar 575 00:35:34.920 --> 00:35:37.599 We have a similar mission statement and are after the same things. And what 576 00:35:37.679 --> 00:35:45.719 they've actually done is it's the largest analysis of scientific methods ever conducted. And 577 00:35:45.760 --> 00:35:49.800 they did an analysis I believe it was one point five million papers and looked 578 00:35:49.840 --> 00:35:53.320 at the actual methods that these studies undertook, what percentage of them were randomized, 579 00:35:53.480 --> 00:35:59.679 blinded, had ample statistical power, had diverse population sets, and they 580 00:36:00.159 --> 00:36:05.280 graded the papers and then use that to grade the journals that they came from. 581 00:36:05.599 --> 00:36:09.760 So the whole thing is just a giant pushback on traditional bibliographic ranking systems. 582 00:36:09.960 --> 00:36:13.559 The one everyone knows about its impact factor, right, which just is 583 00:36:13.760 --> 00:36:16.880 a measure of influence and incitation counts of journals, And this is basically a 584 00:36:16.960 --> 00:36:21.559 pushback on that. And they actually published their own paper that showed that it 585 00:36:21.639 --> 00:36:27.599 was completely uncorrelated with impact factor, which is a pretty big red flag for 586 00:36:27.639 --> 00:36:30.639 those who are using an impact factor. And it makes sense, right because 587 00:36:30.639 --> 00:36:34.280 it's actually looking at what the researchers within these journals we're doing. It still 588 00:36:34.400 --> 00:36:37.719 is a proxymetric, like just because something comes from a journal that typically has 589 00:36:37.760 --> 00:36:42.400 really high standards doesn't mean that a given paper doesn't. And that is a 590 00:36:42.440 --> 00:36:45.079 slight limitation of those tags that we have in the product that it getting a 591 00:36:45.079 --> 00:36:49.920 good score just means it's from a good journal. It doesn't inherently mean that 592 00:36:49.920 --> 00:36:52.400 that paper is better. It makes it much more likely than it is, 593 00:36:52.480 --> 00:36:55.039 but doesn't inherently mean it, but it is a real step forward in terms 594 00:36:55.079 --> 00:37:00.760 of these proxy metrics, and then the future is being able to take some 595 00:37:00.840 --> 00:37:05.639 proxy metrics also looking at the study itself and looking like, what were the 596 00:37:05.679 --> 00:37:08.199 actual methods of this paper, kind of rolling that up into one and saying, 597 00:37:08.239 --> 00:37:10.920 you know, what is the rigor criteria of this score? What is 598 00:37:10.960 --> 00:37:15.159 the reproducibility of those findings? As something we are actively working on and really 599 00:37:15.159 --> 00:37:20.320 hope we can get into the product. That sounds amazing and super interesting and 600 00:37:20.400 --> 00:37:23.840 important, especially because I'm trying to think that my partner is a microbiologist and 601 00:37:23.880 --> 00:37:28.079 so that's how I learned about bioarchive, you know. But this is an 602 00:37:28.079 --> 00:37:31.280 important point in terms of like the reproduceability and that as like a product roadmap 603 00:37:31.320 --> 00:37:34.920 feature. I think it's super interesting too in terms of how could we like 604 00:37:35.440 --> 00:37:37.920 ask these broader questions of methodology, which is something else again not really done. 605 00:37:37.920 --> 00:37:43.760 So I appreciate how up top we're talking about this idea that we're working 606 00:37:43.800 --> 00:37:45.920 with scientists with some zeros and some ones to kind of mark where we're talking 607 00:37:45.960 --> 00:37:50.800 about claims. And then this is like kind of another out of the second 608 00:37:50.840 --> 00:37:53.079 But there's another step in terms of bringing in the questions of methodology, how 609 00:37:53.119 --> 00:37:54.880 those are discussed, what do they look like, and how do we rank 610 00:37:54.920 --> 00:37:59.039 that, and then also what might that mean for things like reproduceability of an 611 00:37:59.039 --> 00:38:01.880 experiment for interesting and can we automate it? Yeah, can we automate that, 612 00:38:02.000 --> 00:38:05.599 like sign me up? That sounds that sounds really cool, and maybe 613 00:38:05.599 --> 00:38:07.920 we can use that to then do a little debunking of some of the behavioral 614 00:38:07.920 --> 00:38:12.760 science that like exactly. We've actually met with the group, the Center for 615 00:38:12.800 --> 00:38:16.079 Open Science, and they've done some awesome work in this world. And the 616 00:38:16.079 --> 00:38:20.199 head of the Center for Open Science is Brian Nosek, and he is actually 617 00:38:20.239 --> 00:38:23.880 most famous for one of the things. He's an amazing, amazing figure in 618 00:38:23.880 --> 00:38:28.320 this space, but one of the things he's most famous for is debunking the 619 00:38:28.519 --> 00:38:32.840 psychology study about power poses and how if you stand with your arms and legs 620 00:38:32.880 --> 00:38:37.719 really wide, you're more likely to succeed in an interview. If you stand 621 00:38:37.760 --> 00:38:39.800 like that beforehand and then go in, you're more likely to speak well or 622 00:38:39.800 --> 00:38:44.880 speak confidently, do well in interview with public speaking. And he showed that 623 00:38:44.880 --> 00:38:49.639 that was not reproducible, and then that was a kind of bogus finding he's 624 00:38:49.679 --> 00:38:54.039 been a huge pioneer and in debunking some of these behavioral psychology studies, and 625 00:38:54.079 --> 00:38:59.679 we've talked with him about these reproducibility metrics and how we could actually incorporate it. 626 00:38:59.719 --> 00:39:02.599 And another example. We're not as formal of a partnership where we're using 627 00:39:02.599 --> 00:39:06.760 the data like we are with side score in the product, but it's a 628 00:39:06.800 --> 00:39:10.000 team that we meet with nearly monthly and continue to talk about these things and 629 00:39:10.239 --> 00:39:14.960 our privilege to be associated with them. Very cool. This is just like 630 00:39:15.000 --> 00:39:17.320 a giant nerd fest of awesome that I know. All the listeners right now 631 00:39:17.320 --> 00:39:20.199 we're being like, man, how do I don't get into this? This 632 00:39:20.239 --> 00:39:23.119 is great to be able to plug into these questions. We are totally free 633 00:39:23.159 --> 00:39:27.719 for now, and it is totally free to sign up and create an account. 634 00:39:27.719 --> 00:39:30.280 So hide over a Consensus Dot app and create an account and ask some 635 00:39:30.320 --> 00:39:34.679 research questions and yeah, giant nerd fest and awesome. We might have to 636 00:39:34.800 --> 00:39:37.440 put that as like a tagline in the products some one. You're there here, 637 00:39:37.440 --> 00:39:40.719 folks, Giant nerd fest of awesome. Same, No, totally, 638 00:39:40.760 --> 00:39:43.719 I think, And that's great. Actually I just signed up myself too, 639 00:39:43.760 --> 00:39:45.920 so I'm excited to be digging around and see what we can we can find 640 00:39:45.960 --> 00:39:50.400 here. So I think kind of as a what's next question, I mean, 641 00:39:50.440 --> 00:39:52.639 you're kind of pointing a little bit or hinting a bit about what might 642 00:39:52.639 --> 00:39:55.039 be down the product roadmap, but you know this bigger question too of like 643 00:39:55.159 --> 00:40:01.280 we're around the precipice of this wave of NLP based technologies and startups, and 644 00:40:01.320 --> 00:40:05.000 so where are we going? Like this is a really interesting space, right, 645 00:40:05.159 --> 00:40:07.199 I think we're in the next frontier in terms of what tech can help 646 00:40:07.280 --> 00:40:09.400 us do around NLP. And so what are you most excited for in this 647 00:40:09.440 --> 00:40:13.519 space, Like whether with consensus or just what else you're seeing that's kind of 648 00:40:13.519 --> 00:40:15.400 shaping the industry or what we could do in the next five ten years. 649 00:40:16.199 --> 00:40:21.559 You know, I'll give you an answer that applies a bit to consensus as 650 00:40:21.559 --> 00:40:23.920 well as like as a more broad answer of like what is to common apologies 651 00:40:23.920 --> 00:40:28.599 that this isn't totally revolutionary and is actually what we've somewhat talked about before. 652 00:40:28.679 --> 00:40:34.039 But what I'm probably most excited for is how these will be used to build 653 00:40:34.559 --> 00:40:38.880 really really amazing specialized search tools because of the ability that you know, I 654 00:40:38.920 --> 00:40:42.280 think they're going to start to pop up in certain verticals. You know, 655 00:40:42.800 --> 00:40:45.639 I'm biased because we're popping up in the science vertical, but there'll be other 656 00:40:45.639 --> 00:40:50.159 ones and other verticals that will build use these tools to build amazing information retrieval 657 00:40:50.400 --> 00:40:54.239 information synthesis engines where you can type in questions or do all these different types 658 00:40:54.280 --> 00:40:59.559 of searches and instantly be returned interesting results. And I am just so excited 659 00:40:59.559 --> 00:41:01.840 to see the innovation with products like that, because I think that is our 660 00:41:01.840 --> 00:41:07.239 opportunity to really change the landscape of how we take in information. If these 661 00:41:07.239 --> 00:41:10.599 products can actually become good enough that they can operate on subscription models, they 662 00:41:10.599 --> 00:41:14.800 have a chance to change the way that the world consumes information. And as 663 00:41:14.840 --> 00:41:16.760 far as yeah, like what's to come with consensus, You know, I've 664 00:41:16.760 --> 00:41:21.800 talked about some of this that reproduce a vability score is something we're actively working 665 00:41:21.840 --> 00:41:24.960 on. And another thing we're really excited about is trying to do like I 666 00:41:25.079 --> 00:41:30.760 kind of mentioned that like toggle button of potentially de jargonizing, is trying to 667 00:41:30.800 --> 00:41:35.239 figure out the right ways to introduce some synthesis into art results. So right 668 00:41:35.280 --> 00:41:38.159 now we're pulling out the quotations, and we want that feature to always be 669 00:41:38.199 --> 00:41:42.360 available to people because it's really valuable, it's nice and raw. We're not 670 00:41:42.400 --> 00:41:45.400 perverting any meanings, but we do want to experiment with ways to synthesize some 671 00:41:45.440 --> 00:41:50.559 of these results and give people a need maybe even quicker snapshot of the landscape 672 00:41:50.559 --> 00:41:53.559 of evidence just on one summary screen. A lot of questions to answer there, 673 00:41:53.599 --> 00:41:57.159 and there's gonna be a whole other things to consider when we do that, 674 00:41:57.199 --> 00:42:00.400 of making sure we're not perverting meanings and representing things that shouldn't be represented. 675 00:42:00.599 --> 00:42:04.880 But there's definitely going to be a place for that in these NLP products 676 00:42:04.880 --> 00:42:09.440 are only getting better at large scale synthesis across a bunch of different sets of 677 00:42:09.559 --> 00:42:13.880 results. Yeah, that makes me super excited too. It makes me think 678 00:42:13.880 --> 00:42:16.119 too, because there's been a you know, like explosion of read this book 679 00:42:16.119 --> 00:42:20.920 in five minutes summary book services, you know, and oftentimes those are then 680 00:42:20.920 --> 00:42:23.800 written by people right there, they're like human made synthesis synthesizes, And that's 681 00:42:23.800 --> 00:42:27.800 where it all started, right is like we saw, like whatever products you've 682 00:42:27.840 --> 00:42:30.719 seen that are like manually curated products around like tech summarization, those are going 683 00:42:30.760 --> 00:42:35.840 to become NLP companies in the future. So like Blankest is a perfect example, 684 00:42:35.920 --> 00:42:37.880 right, it's the book summaries. Those are manually curated right now, 685 00:42:38.280 --> 00:42:43.039 no world where in some order of time that isn't going to be just done 686 00:42:43.079 --> 00:42:45.199 by machines, And like, those are the types of products that we're going 687 00:42:45.239 --> 00:42:47.800 to see pop up. A really interesting thing that I've thought it all before 688 00:42:47.920 --> 00:42:52.760 is like the place that we're seeing a lot of these models take off early 689 00:42:52.920 --> 00:42:58.440 right now, and these applications are in generating like subjective content in some ways. 690 00:42:58.800 --> 00:43:02.559 So like one of the most prominent applications we're seeing is using these these 691 00:43:02.639 --> 00:43:07.159 language models to generate marketing copy. And then the other really famous one, 692 00:43:07.159 --> 00:43:12.159 obviously is this art generation. And I was saying, it's kind of funny 693 00:43:12.199 --> 00:43:15.760 that, like, if you think back to a decade ago when we were 694 00:43:15.800 --> 00:43:20.000 all hypothesizing about what AI was going to do, we said, everyone always 695 00:43:20.000 --> 00:43:23.719 thought the place that they'll come last are for roles that are truly subjective and 696 00:43:23.840 --> 00:43:27.400 like art based, right, Like everyone's like, oh, you know, 697 00:43:27.599 --> 00:43:29.920 if you're a painter, like there's no AI is going to come for your 698 00:43:30.000 --> 00:43:34.119 job. And it turned out that that head prediction is like completely wrong. 699 00:43:34.440 --> 00:43:38.199 And I think it's wrong for two reasons. One because when you're generating something 700 00:43:38.280 --> 00:43:43.599 like an art image or even like a marketing slogan, you have a lot 701 00:43:43.599 --> 00:43:46.320 of leeway to kind of like do it in this like party trick way, 702 00:43:46.480 --> 00:43:51.719 and there aren't real consequences for what you're delivering back, right, Like when 703 00:43:51.800 --> 00:43:55.079 you type of prompt into Dala and you say, what generate me this random 704 00:43:55.119 --> 00:43:59.280 image, It doesn't really matter what it spits back to you. The fact 705 00:43:59.320 --> 00:44:01.960 that it's spitting back as something relevant to your prompt, like that creates the 706 00:44:02.000 --> 00:44:06.199 moment that they're seeking to creative. Oh very cool, right, But it 707 00:44:06.280 --> 00:44:09.599 isn't like there aren't these stakes of did they answer this question perfectly and find 708 00:44:09.599 --> 00:44:14.119 the relevant information inside a scientific document? And the same is kind of true 709 00:44:14.159 --> 00:44:16.000 for marketing copy, right, Like, there isn't like a perfect way to 710 00:44:16.079 --> 00:44:20.440 generate a marketing slogan. You just kind of throw things out the wall and 711 00:44:20.519 --> 00:44:22.639 see what sticks. So guess what a machine is really good at doing throwing 712 00:44:22.679 --> 00:44:27.079 things in a wall and seeing what sticks. Right. And then I was 713 00:44:27.079 --> 00:44:30.800 talking about this in my co founder and he brought up another really interesting reason 714 00:44:30.880 --> 00:44:36.719 why this is happening is because the Internet is filled with just subjective content and 715 00:44:36.760 --> 00:44:40.000 it creates these giant training corpuses to build these models. So like the Internet 716 00:44:40.159 --> 00:44:45.719 is literally just filled with marketing copy right, like where you have titles to 717 00:44:45.880 --> 00:44:47.800 articles, so you could say, all right, here you have all this 718 00:44:47.920 --> 00:44:51.960 training data already built for you, where it's here's the text of the article, 719 00:44:52.239 --> 00:44:54.199 here's the header. Okay, build me a model that generates headers for 720 00:44:54.320 --> 00:44:58.679 articles, right, Like, anybody with NLP skills could build you that model 721 00:44:58.719 --> 00:45:00.519 and not that long of a time with the giant trainings that data to do 722 00:45:00.559 --> 00:45:05.000 it on. And the same goes for these images, right Google image. 723 00:45:05.000 --> 00:45:08.599 All the images across the Internet have these captions and titles of what they are, 724 00:45:08.840 --> 00:45:13.199 Like if you'd Google search something it has whatever, your images will have 725 00:45:13.239 --> 00:45:15.280 a little like description of it. Well, there's your training data. You 726 00:45:15.320 --> 00:45:19.400 have all these pictures and all of the prompts. Now you can train a 727 00:45:19.400 --> 00:45:22.639 model that can take in a prompt and spit out an image. So anyway, 728 00:45:22.880 --> 00:45:27.519 I think it's really funny that everyone thought that the more subjective thing, 729 00:45:27.559 --> 00:45:31.000 like it's going to be the hardened like one or zero tasks that AI does 730 00:45:31.079 --> 00:45:35.840 first, because like those all things were automated and all the subjective things that 731 00:45:35.920 --> 00:45:37.920 make us so uniquely human they'll never come for that, and then that was 732 00:45:37.960 --> 00:45:43.280 completely wrong. Watch them, watch them come, you know, it was 733 00:45:43.440 --> 00:45:45.880 funny. It was actually only recently I feel I feel like I got duped 734 00:45:45.880 --> 00:45:49.119 that I realized or found out that, you know, when you're doing those 735 00:45:49.159 --> 00:45:52.360 like captift phrases, like you know, either like the pictures of a boat, 736 00:45:52.400 --> 00:45:55.840 we're just training an AI that's always are all street pictures because it's training 737 00:45:55.840 --> 00:45:59.119 in self driving cars. That's why they try to make you pick out what 738 00:45:59.159 --> 00:46:01.960 the stop sign is, or what the motorcycles are, or what the stoplight 739 00:46:02.119 --> 00:46:07.199 is. It's training data for self driving cars. Does that count as publicly 740 00:46:07.280 --> 00:46:12.519 funded research? You know, it's probably an argument for that. How do 741 00:46:12.559 --> 00:46:14.679 I gotta cut of that? You know? Eric, thanks so much for 742 00:46:14.760 --> 00:46:15.800 joining me on the POT today. That's been It's been really great. I'm 743 00:46:15.800 --> 00:46:22.039 super excited to have listeners check out consensusal but link and guessing folks to kind 744 00:46:22.039 --> 00:46:23.480 of test it out and see and get their thoughts of what it's like, 745 00:46:23.480 --> 00:46:28.320 because I'm really curious to see the human science community dive in and see what 746 00:46:28.320 --> 00:46:30.760 we can find in there. Again, taking some thoughts with the benefits of 747 00:46:30.760 --> 00:46:34.639 mindfulness that the question around death penalty thing is really interesting as well, so 748 00:46:35.079 --> 00:46:37.199 an inspiration for for folks to kind of dive on in. But thanks much 749 00:46:37.239 --> 00:46:40.719 taking the time of talking with us today, and yeah, we wish you 750 00:46:40.760 --> 00:46:44.159 the best luck and hope we can keep in contact and you know, dive 751 00:46:44.239 --> 00:46:45.719 on in. His consensus keeps on building. Absolutely, this has been a 752 00:46:45.719 --> 00:46:52.119 blast. Thank you so much, Adam awesome. Thanks thanks again to Eric 753 00:46:52.119 --> 00:46:54.679 Olsen from Consensus for joining me on the podcast today. I am really excited 754 00:46:54.679 --> 00:46:59.159 at the possibilities of making it easier, faster, and more digestible to see 755 00:46:59.320 --> 00:47:02.000 what the latest sientific consensus is around all sorts of ideas. And in case 756 00:47:02.000 --> 00:47:06.920 you forgot, Consensus is currently freewill in beta, so definitely check it out, 757 00:47:07.199 --> 00:47:08.440 give it a spin and let me know what you fine. Really curious 758 00:47:08.440 --> 00:47:12.360 that the new kinds of connections and possibilities that we could find, so I'd 759 00:47:12.360 --> 00:47:15.400 love to see what you're looking up and the different ideas that you're bringing together. 760 00:47:15.679 --> 00:47:17.519 Now. In fact that this Anthro Life and Consensus did a little collaboration 761 00:47:17.519 --> 00:47:20.320 that I'd love to share with you. You can check it out on the 762 00:47:20.360 --> 00:47:23.000 latest post from the TL newsletter on sub stack. The link to Subscribe is 763 00:47:23.039 --> 00:47:25.679 in the show notes below, and as I've been mentioning for a little while 764 00:47:25.760 --> 00:47:29.079 now, it's been a goal of mine to do more writing and expand the 765 00:47:29.079 --> 00:47:31.719 TL mind share across mediums. So I hope you'll check it out and subscribe, 766 00:47:31.920 --> 00:47:35.039 and if you get something out of this Anthro Life, please take a 767 00:47:35.079 --> 00:47:37.800 moment to give the show a review on your podcast listening of choices, but 768 00:47:37.880 --> 00:47:39.599 that's an option, and share it with a friend who you know we'll love 769 00:47:39.639 --> 00:47:43.760 it too. It's funny, despite podcasting being a thing for years now, 770 00:47:43.800 --> 00:47:46.119 if the culmus around twenty years, the best way to discover new shows is 771 00:47:46.159 --> 00:47:50.760 still through human connections and sharing with your friends. So I'd be honored if 772 00:47:50.760 --> 00:47:52.519 you find the show worthwhile to have you share it with a friend, and 773 00:47:52.599 --> 00:47:55.239 hey, let me know about it. Should me a message at this anthor 774 00:47:55.280 --> 00:47:59.800 life at Gmail or getting contact through the TL website, and maybe we'll have 775 00:47:59.800 --> 00:48:01.840 some new podcast love stories to share. An air never a bad thing, 776 00:48:01.920 --> 00:48:06.039 right and is always to thank you so much for joining and being a part 777 00:48:06.039 --> 00:48:09.519 of this conversation. The invitation remains open to pitch and submit ideas for episodes 778 00:48:09.599 --> 00:48:13.719 or blog posts, and that includes things I can make that you can make 779 00:48:13.920 --> 00:48:15.280 or that we can make together. So if you have an idea, let's 780 00:48:15.280 --> 00:48:19.000 get in touch. Much love my friends. Check out consensus dot app, 781 00:48:19.039 --> 00:48:20.880 give it a spin, and I can't wait to see you next time. 782 00:48:20.920 --> 00:48:23.280 On the next episode of This Anthrow Life, I'm Adam Gamwell, we'll see 783 00:48:23.320 --> 00:48:23.599 you soon.