← Lead With AI
Lead With AI
PhD Student Builds AI That Turns Months of Research Into Minutes and Ranks Top 5 in ChatGPT Store
PhD Student Builds AI That Turns Months of Research Into Minutes and Ranks Top 5 in ChatGPT Store
Loading summary
Transcript60 lines
- [00:00]
Dr. Tamara Nall
What if the best research paper on the planet was one you never found? Damon Burrow, co founder of Scholar AI isn't just fixing search, he's reinventing discovery. Most people are swimming in information but starving for knowledge, he says. So what happens when AI gives you direct access to the minds behind the science, not just the citations? Let's get into it.
- [00:31]
Damon Burrow
Welcome to lead with AI. I'm Dr. Tamara Nall. In each episode, we will take you behind the scenes with visionary leaders shaping the future of AI across public and private sectors. Join us as we explore groundbreaking projects and innovations that are transforming industries and making a real impact on on people's lives. Let's dive in.
- [01:00]
Dr. Tamara Nall
Hi everyone, how are you? It is Dr. T here, your host on Lead with AI and I'm excited to have Damon Burrow, who is the co founder of Scholar AI. Hi, Damon, how are you?
- [01:12]
Damon Burrow
Hi. I'm doing well. Thanks for being here. Appreciate you having me.
- [01:15]
Dr. Tamara Nall
Absolutely. And I am super excited about this conversation. Number one, I have been in academia forever. As a student I taught a little bit and since getting my doctorate, there are some research topics that I'm so interested in. But I feel overwhelmed, drowned in like data and articles and all these scholarly perspectives. So I'm excited for me and for the listeners on talking about Scholar AI. Yeah, yeah, yeah, absolutely. So behind every product has to be a person or a group of people. And we're here talking today. So tell us a little bit about who you are, who is Damon at your core and what led you to build Scholar AI and just talk to us a little bit about that.
- [02:05]
Damon Burrow
Sure, yeah, it's a great question. So I am currently doing a PhD at Duke University in addition to kind of running this business along with my co founders. And so Scholar AI is a product that was for me, so similar to how you outlined this was very much a product born of inspiration, of solving my own problems as I was going through my own academic journey. Very much something that I wish I would have had through undergrad and through some of my other studies. Right beyond that. I was a two sport college athlete. I'm from a very small town in Missouri and I've been building technology for a long time. I got my start in kind of the hardcore research world in undergrad. Originally I was designing algorithms that helped cancer, basically. And so I was ultimately in radiation therapy for cancer patients, trying to make sure that they were being treated better, basically. And that kind of morphed with the emergence of AI and here we are.
- [03:03]
Dr. Tamara Nall
Wow, that's absolutely amazing. And what are you getting your PhD in?
- [03:07]
Damon Burrow
Yeah, so my PhD is in biomedical engineering, so I'm working at the intersection of technology and human health. And so a lot of the work that we've done at Scholar AI helps bolster that. I can cover more area, I can read more papers, I can learn. And as we'll kind of get into later on in the conversation, Scholar AI has helped me perform better experiments, write better grants, write better papers, et cetera.
- [03:32]
Dr. Tamara Nall
Okay, that's amazing. Let's get into it then. So, holy smokes. I love hearing stories about, what do people experience when they are introduced to Scholar AI? What? Where they're like, oh, my God, this really changes everything for me.
- [03:49]
Damon Burrow
Yep. Yeah, it's a great question. So, in truth, whether you find Scholar AI inside of ChatGPT as a GPT, or whether you're using its standalone web app, the experience that you're going to have is very similar to an experience that your listeners may be familiar with inside of the ChatGPT ecosystem. It's a conversational chatbot. You can get into it through other entry points. And some of our more scientifically inclined people, developers, et cetera, may like that entry point better. But for a layperson, for the average user, it's a chatbot, just like chatgpt is. And most of the secret sauce exists under the hood. And what people continue to tell us as they experience it over time is that it understands the scientific problems that they're working on better than ChatGPT. It helps surface better insights, it helps surface faster insights and enables them to go back to the things in which their core competency really resides. It gets. It gets researchers back to the bench, or gets researchers back out into the field, and they get to spend less time reading papers, more time doing the actual science they love.
- [04:52]
Dr. Tamara Nall
That's amazing. So we're talking like days, weeks and months of research literally in minutes. Is that what we're talking about?
- [04:58]
Damon Burrow
That's right. And kind of the things that are powering that under the hood is very early, early, early days. ChatGPT didn't give you sources. Now, ChatGPT does give you sources, but what our technology does is it connects your query, the thing that you're putting into that chatbot, with an entire knowledge graph that's been built over years. Lots and lots of training data goes into this knowledge graph that basically says, this user is asking for X and we're going to give them X in a way that is enriched over what Google Scholar or over what ChatGPT just running in the Standard version can do. So it's not just about surfacing insights faster, it's about servicing higher quality insights in less time.
- [05:38]
Dr. Tamara Nall
Wow. Wow. So it's so like when we're, when we're peeking in the brain of it all or under the hood, you're saying it's more than sources.
- [05:45]
Damon Burrow
Yeah, exactly. And so these large language models operate through, typically through what is called vector search. That's very different than your Google Scholar keyword search. Right. Where you're just highlighting times, you know, very kind of superficial words in a paper, that kind of thing. What our system does is it actually grabs insights out of various parts of the paper that don't just match via keywords, but also match semantically. Meaning what did the user actually intend with their question, not just what did they write? And our insights, the extraction of the insights that we're able to get, actually go deeper than just keywords and text as well. And we wrote some scripts again early on that enabled us not just to query the written part of all the papers, but the graphs, the tables, the figures, etc.
- [06:32]
Dr. Tamara Nall
Wow. Okay. Yeah, this would have been helpful back in 2022 for me.
- [06:37]
Damon Burrow
That's right.
- [06:37]
Dr. Tamara Nall
Oh my gosh, that's amazing. And kind of you talked about the holy smokes moment, talk to us about. And everybody kind of, I feel like as we're developing products, as we're advancing technology, we as founders, as creators even, we have a moment where it's like, oh my God, like, I can't believe that what I am working on delivers that. So talk to us about that time where you had that moment where you got chills. If you have a victory dance that you want to like get up and do that and to take us back to that moment, tell us what is it like, quantify it for us, visualize it for us.
- [07:13]
Damon Burrow
Yeah, yeah, it's a great question. I don't dance. Not well, at least I'll skip that part. But no, no, I have, I have a story, actually. It perfectly encapsulates kind of what you're, what you're alluding to. And so the devices that we build in the lab here are similar to some of the diagnostic tests that some of your viewers and listeners may have been exposed to in the past. You go to the clinic, you go to the hospital, etc, you might get a blood test for one reason or the other. We build those technologies so that people can take them into their houses very much like a blood glucose sensor for people that are living with maybe chronic illness or for people that Maybe can't arrive at a healthcare facility, maybe they can't drive, maybe they're not ambulatory, etc. One of the challenges that I was trying to overcome with the chip that we were developing was I needed a surface that was, or I need a fluid rather, that would make the fluid that we were working with, specifically blood, in this case, sufficiently wettable that we could control the motion of that blood drop. Because what we do is we get the blood drop on a chip, we drive it to completion. That way we can actually get the results out very much like a blood glucose sensor does. Slightly different mechanism, but, but much the same kind of core principle. And the fluids that we had in the lab, none of which were working, the combination of the blood and the surface we were trying to get the blood to transport through, they were incompatible, basically. And so I needed to find something that made the blood that we were trying to work with compatible with the kind of current architecture of the device that we had. Didn't know about it. Normally that process would have looked through several hours of reading different material, that kind of thing. Instead, I was able to query directly with scalar A and said, hey, this is the problem that I'm having. Here's the goal, here is what I know I need. Help me find a short list of possible, you know, surfactants, basically different detergents that I could use possibly in this kind of blood cocktail that would allow me to actually get a functional chip. And lo and behold, the first hit on Scholar AI. I went to it, I bought it immediately from Sigma, went into the lab, and now we have a functional chip that we wouldn't have had otherwise.
- [09:17]
Dr. Tamara Nall
Wow. Okay. That gave me chills. That is amazing. And while you don't dance, I'll do one for you. Oh, my God. I mean, that's actually. Oh, my God, that is amazing. And so I'm curious, like, what happens next? I mean, like, this is breakthrough. Is this now going to be fed to like some large healthcare company or diagnostic company, or is it still like just what you're working on in the lab for your research or.
- [09:43]
Damon Burrow
Yeah, it's a great question. And we do have some commercial aspirations for this test. The very specific test that I'm building is built to help Oregon transplant recipients. So if you get in your kidney, the very short story is you go on a lifetime regimen of medication that helps you not reject that kidney and that helps you still be able to fight off other infections. And so the blood dosage that you need versus the blood dosage that I need might be very different because of our diet, where we live, our body style, et cetera. And those doses are very tightly controlled. Those people need a companion diagnostic device like the one that I'm building in order to avoid symptoms and get good outcomes. And so we have commercial ambitions for the test that's being built. That said, it's a research prototype. At this point, we've received some funding based on the progress that we've made. There are several checkpoints to get until we receive FDA approval or otherwise. And yes, to answer your question directly, could be new company or could be licensed through an existing company, such that you kind of gain distribution and you get to market faster and those kind of things, all of which tbd. But we are, we are optimistic, I would say.
- [10:52]
Dr. Tamara Nall
Well, that's amazing. And you know, I have VC companies that listen to the podcast, so I.
- [10:57]
Damon Burrow
Know who to talk if you want to. Yeah, that's right.
- [11:01]
Dr. Tamara Nall
So let's talk about the ethics. So, yeah, you know, stellar AI can take weeks and months worth of research and literally get it to you in. In minutes, if not seconds. What are like, the ethical boundaries that you and your co founders have to keep in mind? I mean, some people, you know, might say, I don't like the word devil's advocate, but okay, some type of advocate. Some people might say, well, how so? We have people who are working on their masters, their PhD, their doctorates, are doing any type of research, trying to be an authority in something, and then scholar AI just like, gets it to them. How then are they going to be the authority when they didn't really, like, work for it? How do you respond to that? How do you stay on top of those. That ethical boundaries?
- [11:45]
Damon Burrow
Yeah, I mean, number one, I think it's an excellent question, and I think it is something that the field, not just us, is grappling with. Right. There are studies that are coming out that basically say kind of your brain on ChatGPT is not as good. Personally, I think it's probably not as good down some vectors and probably better in others. Right. Like, if it hadn't been for scholar AI in the example that I just outlined, it's possible that I'm still looking for a fluid that's sufficiently wedding. You know, I'm kind of of the opinion that AI is more similar to a calculator and that people are going to use this as a tool to increase leverage. That said, it absolutely has to be kind of implemented and rolled out with precision and with some thoughtfulness. The way that we've approached that internally that I could speak to directly at Scholar AI is we've kind of taken two approaches. Number one, from the first earliest days of Scholar AI, one of the critical problems we were solving for. It's mostly a solved problem now, but back then it was not is providing sour with all the material. So if some of your viewers yourself remember in the earliest days of ChatGPT, it would just make things up. Right? It would hallucinate rampantly. Hallucinations are not a completely solved problem, but largely are. And so the very first thing we did was here's this information that you asked for and here's the exact paper or abstract conference presentation, et cetera, that that information came from so that it could be human verified. So the humans are still responsible for going ultimately to the source material, verifying that it is real and then verifying that it does work. The good thing is in science we're never taught to believe it just on first pass. And so all these scientists are still having to go to the lab, still having to validate their own experiments, et cetera. And so we do think about that quite a lot. The other thing, and the second point that I would like to make here with Scholar AI is that we also, rather than approaching this in a purely question and answer style, what we do is we try our best. It doesn't always work because these large language models are a little bit difficult to wrangle sometimes what we try to do is present countering narratives in the output from scholarlyi. Right. So if you ask an open ended question with any amount of subjectivity, if you ask the question what is two plus two? It's going to give you the answer four. There's not really any ambiguity. But if you said what is the best research path to try to solve this difficult challenge? It might outline two different paths to you make sufficiently strong arguments for path A and then path B and leave it up to the user as a thoughtful person to say I'm more aligned with path B because I think that argument is stronger, or I'm more aligned with path A because that argument is stronger. So kind of this thesis antithesis and in summary approach, whereas instead of just giving you the answer, we are giving you strong arguments for both sides and then allowing again, thoughtful, smart people to make the determination as to which of those sides they more align with.
- [14:30]
Dr. Tamara Nall
Yeah, you know, and I love that. And then as you were talking, I was also kind of thinking, you know, Scholar AI removes all of the grunt work, meaning the months and hours, hundreds and Thousands of hours it would take me to do the research. But even if I had to take all that time, I still must become an expert on what I found, excuse me, and do critical thinking and kind of think through, through it. And so basically it just shortens the time of research. But like you said in the very beginning, now I have more time to be out in the field and you know, do surveys or do evaluations or be in the lab, et cetera. Like you said, like with your issue or your problem, it could, you could still be looking for it, but now you can, and now you can move on to more value added discovery, if you will.
- [15:20]
Damon Burrow
That's right, that's right. No.
- [15:23]
Dr. Tamara Nall
Yep. Awesome. So let's talk about the big future. So fast forward, how do you see Scholar AI really contributing to the everyday person?
- [15:34]
Damon Burrow
Yeah, it's a good question. So right now we help tens of or hundreds of thousands, depending on the month. More, more so in the school year, less so in the summer months, just because of the cyclical nature of students, et cetera, you know, do research, right? Look up clinical trials, look up papers that they may reference to plan their next experiment or even help perform their protocols, that kind of thing. The other thing that we've worked really, really hard on at Scholar AI is fundamentally these large language models are kind of regression machines. They give you the average of kind of the best consensus answer, if you will. That doesn't necessarily lead itself to scientific discovery, right? The leading edge of knowledge is not consensus. Sometimes it may be where the consensus is driving towards, but it's not always kind of average or the median. And so we've kind of taken a zoomed out approach, kind of looked across a metaphorical field and said kind of where are the green spaces, right? Where, where are the unknown? And how do we help tailor the tool that is Scholar AI? Help put it in the hands of the most creative and thoughtful people and help them discover new insights, help them push the boundaries of kind of what is known. And some of that is just a trial and error approach. And as I kind of alluded to earlier with the description of the product, in addition to just the standard chat interface, we also make the Scholar AI research agent, as we call it, available as an API. We give that out to the scientific community and we let them build tools because they are creative people, right? Scientists are experimenters kind of in by design and they want to build things and so we kind of allow them to choose their own adventure with the things that we've already built for them. Being supplied by the API. They can plug into the API, they can build their own tools. So we've had people build new types of drug discovery tools. I don't know that those have actually led to any the discovery of new drugs. But again, that's an avenue where people are exploring and they're able to take our work, they're able to push it further with their own ingenuity, and we want to continue to be enabling for that. And as kind of a final thought here, we think it's probably going to be partnerships because we see that Scholar AI has expertise in the domain that is sciences. But the reality is the AI tools are going across productivity. And so we want to see how can we plug into the clinical trial environment. Right. How can we be apprised of new drugs as they're going through development? How can we help those researchers ask questions that they themselves don't know how to answer yet? And so it's kind of moving into that more proactive space and moving more into the cutting edge space versus just being able to answer kind of a question that while it may have taken you time, you certainly could have done that through enough investigation of reading through the scientific material yourself, going through channels like Google Scholar or those kind of things.
- [18:21]
Dr. Tamara Nall
Got it. Wow, that's amazing. So basically, if I want to use Scholar AI for myself and just use it as it is, as fic, or if I wanted to build something on it and make it a little bit more specific to my needs or in a certain industry or vertical, I can do that as well.
- [18:37]
Damon Burrow
That's right.
- [18:39]
Dr. Tamara Nall
Amazing. So to that point, if somebody wanted to try Experience build, how do they do that? How do we get to Scholar AI?
- [18:48]
Damon Burrow
Yeah, perfect question. The easiest entry point is directly in ChatGPT. So anyone that is familiar with the ChatGPT interface on the left sidebar, there's an option for GPTs. If you scroll down to the research and analysis tab, we're number three on the GPT store. We kind of fluctuate between two and two and five. Call it depending on. Again, thank you. Thank you. Yes, on seasonal, kind of the seasonal seasonality of students being again kind of in semester or out of it. And then also via that. We also have our own standalone web app which is very much the same experience. They can find that at app scholarai IO if anybody wants to follow me directly, I'm happy to point them in that direction. And then for those experimenters, for those enthusiasts that want access to the API, we have a self sign up. Also on ScholarAI IO you can, you can find all the documentation for the API there, or depending on the access that they want, if they really want to see under the hood, if they really want to kind of make it their own, we work with those people one on one. We have a team that helps do that. And so if they want to reach out to me, we can absolutely get them in contact with the right people. And then depending on the exact tool they kind of want to build, we can kind of help them hit the ground running.
- [19:59]
Dr. Tamara Nall
Awesome. And what's the best way to get in contact with you? Is there email or LinkedIn?
- [20:02]
Damon Burrow
Absolutely. So yeah, yeah, so LinkedIn is great. I'm just Damon Burrow on LinkedIn. That's a great way to find me. The other way is email and I'm just Damon D a M o n@scholarai IO.
- [20:13]
Dr. Tamara Nall
Wow, amazing. Absolutely amazing. So we have, that's a lot of ways to get in contact with Scholar AI. No excuses. Build your own. It's convenient. I mean, pretty much everybody uses Chat GPT at least as one of maybe a number of LLMs. But so to go there and to know that y' all are, you know, 1 to 5, depending on the season, is a huge accomplishment. Like, kudos to you, really.
- [20:37]
Damon Burrow
Yeah, thank you.
- [20:38]
Dr. Tamara Nall
Yeah, absolutely. So my last guest has a question for you, and that question is, having built a business through the emergence and rapid growth of AI, what are some things you know now that you wish you knew when you started?
- [20:54]
Damon Burrow
Yeah, that's a great, that's a great question. I think, having been a part of the earliest days. So we, we were one of the first plugins on ChatGPT. We were, we were there early days, we were solving problems for the users of ChatGPT. Right then it became very difficult to predict what the foundation model providers, you know, the OpenAI's of the world, the anthropics, Google's, even meta in the open source world, those kind of things were going to do and so we've tried to remain ahead of them. I think that knowing what we know now, we probably would have gone to scientists themselves earlier than we did and said, how can we build specialized tools for you specifically that maybe weren't going to kind of overlap so much with what ChatGPT was, was doing? How do we help amplify ChatGPT? How do we help build in collaboration with the progress of that financial model, such that every time you were using ChatGPT, you were also using Scholar AI, and it wasn't so much of an either or Choice of that way you got all the good out of ChatGPT without most of the bad. Right. And you got all the good out of scholar AI. And so I think I would have found ways that while the entry point of the GPT is very easy, I think that running a scholar AI copilot in places where scientists already were, maybe on their favorite machines, that kind of thing, would have been something that would have been very intriguing. Will we be able to pull it off? I think was a different question, but I think that's something that I certainly would have tried.
- [22:23]
Dr. Tamara Nall
Oh, I love that. And that's great advice for our listeners who actually probably have some business ideas like go to the user, go to the customer, if you will, to kind of get those insights. So thanks for that truthfulness and that directive. That's, that's amazing. So let's do our bonus rapid fire. So I'll ask a question quickly, list the answer. Most overrated tech trend.
- [22:48]
Damon Burrow
Oh, crypto. Crypto.
- [22:51]
Dr. Tamara Nall
Okay. All right. It's interesting. I actually had dinner last night with someone. It was like, crypto, Crypto.
- [22:56]
Damon Burrow
Yeah.
- [22:57]
Dr. Tamara Nall
Okay. Yeah, I guess in and out though.
- [23:00]
Damon Burrow
Yeah, that's right. That's right.
- [23:02]
Dr. Tamara Nall
Most under hyped AI breakthrough.
- [23:06]
Damon Burrow
Video models. Right now, I think the progress that video models are making are going to take people by surprise for good and for bad. I think we are going to see an explosion of AI generated content, both in the forms of images and videos. And I think the world is going to have to react to that in ways that are, that are thoughtful. But I don't think enough people are talking enough about how quickly that is getting very good and becoming almost indistinguishable from human generated videos.
- [23:35]
Dr. Tamara Nall
You're right again. Having conversations and somebody say, you know what, if anything happens, just say, oh, that's AI generated. They won't know, but then somebody will come up with a product that will rate whether or not it's AI generated. You know, those are already out there. So I'm sure you'll do that. One book everyone should read in general or about the future.
- [23:55]
Damon Burrow
That's a great question. I'm going to give a fairly typical answer here, but the way that I think about a lot of things comes from Daniel Kahneman's work in writing. Thinking fast and slow. So trying to break down a problem of, you know, is this a problem that requires my absolute attention and this is something that I should think thoughtfully and deeply about, or is this a question that may ultimately, you know, be okay with an answer that is, is Kind of somewhat knee jerk. Right? Keep it, keep it moving and kind of allowing my bandwidth to be devoted to kind of a higher thing, I think, I think that's a really great, great read.
- [24:28]
Dr. Tamara Nall
Awesome. Okay, the boldest AI prediction you believe in.
- [24:34]
Damon Burrow
I think that AI will help inspire the work in the early days and then probably be able to do a lot of the work that will lead to breakthroughs in disease treatment that we are yet to predict. So I think a lot of people are looking at cures for cancer or Alzheimer's and kind of the most deadly disease. And I think that will happen, but I think there will be overlooked disease that don't command so much of the public zeitgeists, don't command so much of the research dollars that AI will actually probably help us solve maybe more quickly because maybe they are less difficult to treat, less difficult to overcome, but that humans just haven't committed the bandwidth to and AI will sufficiently fill that gap.
- [25:21]
Dr. Tamara Nall
Yeah, yeah, No, I love that. I love that. Yeah, Cancer gets a lot of attention. I'm going to be honest. One area that scares the living daylights out of me are blood clots, because by the time you find those, it's too late. So that one, that one really, really scares me over. Covid. I had a number of friends, unfortunately, that passed away from it.
- [25:41]
Damon Burrow
Yeah, sorry.
- [25:42]
Dr. Tamara Nall
Yeah. And so. And then recently a couple of people that I was two, two people removed from, but still it's quite scary. So I can't wait for investments for that, like you get a notification early about it, so.
- [25:56]
Damon Burrow
That's right. And I think, I think to that point we are early days at understanding how we can best use data with AI. And right now a lot of that data is tethered to the digital world, the electronic world. I think, you know, I'm somewhat biased here because my PhD is in hardware for human health. But I do think as that technology sufficiently advances, right. Things like the Apple Watch or similar to where it's kind of 247 or at least with some relative periodicity, monitoring your blood, catching things like blood clots, that wouldn't have been possible because you can't have humans reviewing that continuous stream of data, but you can have an AI doing. I think those possibilities open up in this kind of new world that we're, you know, we're frankly kind of being, you know, that we're ushering in.
- [26:40]
Dr. Tamara Nall
Yeah. Oh, I might have given a secret sauce. I'm like, maybe I should have just done it. Done it. I use Scholar AI to help me with my methodology.
- [26:48]
Damon Burrow
That's right.
- [26:48]
Dr. Tamara Nall
And all that. So I don't know if I come out with a new company. I'll give you credit though. I'll give you credit. Well, Damon, I so enjoyed this conversation. Thank you so much. Thank you for what you're doing with scientific breakthroughs. Thank you for your, what you're doing for researchers and scientists and those like me that are like these, these geek out research folk. And I'm pretty sure that all of our guests are going to absolutely, absolutely love the product if they're not already using it since they're in chat GPT. So again, if people want to reach out, go to Scholarai IO or go into ChatGPT to use their app there or you can connect with him on LinkedIn. That's last name B U R R O W. So thank you so much. Do keep us up to date about your studies and I mean I can't wait to hear what other contributions you're going to make to this world. Thank you so much, much.
- [27:47]
Damon Burrow
I appreciate it. Thank you very much, Dr. T for having me.
- [27:50]
Dr. Tamara Nall
Absolutely. Until next time everyone. Bye bye.
- [27:54]
Damon Burrow
Thanks for tuning in to lead with AI. I'll see you next time as we continue exploring the cutting edge innovations shaping AI across the public and private sectors. Until then, keep leading with AI.