Uncategorized

Ep 121: Faster and More Accurate Results From ChatGPT with ScholarAI

By,
  • 12 Oct, 2023
  • 27 Views
  • 0 Comment

Resources

Join the discussion: Ask Damon and Jordan questions about ChatGPT

Upcoming Episodes:
Check out the upcoming Everyday AI Livestream lineup

Connect with Damon BurrowLinkedIn Profile

Related Episodes

Video Insights

Overview

In today’s fast-paced, data-driven world, businesses across industries are increasingly relying on artificial intelligence (AI) to make critical decisions. AI models, particularly large language models, have the power to generate creative content, provide insights, and assist professionals in their everyday work. However, ensuring trust, accuracy, and reliability in AI-generated information remains a pressing concern. In this article, we will delve into the subject matter covered in the “Everyday AI” podcast episode, exploring how business owners and decision makers can harness the potential of AI while maintaining the integrity of the information they receive.

The Challenge – Accuracy and Transparency in AI:

As demonstrated in the podcast, generative AI systems like large language models have the remarkable ability to produce content that spans various domains. While these systems excel in creative tasks like storytelling or songwriting, their accuracy and truthfulness become crucial when used in professional knowledge work such as legal analysis, medical diagnoses, or business decisions. Trusting AI-generated responses is essential, but so is ensuring transparency and understanding where the information originates.

Tethering AI to Reliable Sources:

One effective approach highlighted in the podcast is tethering AI-generated output to peer-reviewed scientific articles, publicly accessible databases, and authoritative texts. By doing so, the AI system draws information directly from reliable sources, reducing the likelihood of misinformation or hallucinations. This approach instills trust and confidence in the AI’s accuracy, as the generated responses are grounded in established knowledge and recognized expertise.

Real-Time Access to Up-to-Date Information:

One valuable tool discussed in the podcast is ScholarAI, which enables real-time access to the latest information through API endpoints. By harnessing these features, business professionals can stay abreast of the most recent peer-reviewed research, market trends, or legal opinions that directly impact their decision-making processes. ScholarAI updates its databases daily or multiple times a week, ensuring the information remains current and reliable.

Conclusion:

AI has the potential to revolutionize the way businesses operate and make decisions. By establishing trust, accuracy, and reliability in AI-generated information, business owners and decision makers can leverage the power of AI to drive growth and success. Integrating AI systems with tethered information sources, embracing cross-referencing, and utilizing visual representations can enhance the quality and reliability of AI-generated responses. As the technology continues to evolve, business professionals must remain vigilant in understanding and utilizing AI systems in a thoughtful and responsible manner. Together, we can unlock the transformative potential of AI while ensuring the integrity of the information we rely on.

Topics Covered in This Episode

1. Importance of Accuracy in AI Systems
2. ScholarAI Plugin for Reliable Information Retrieval
3. ScholarAI Plugin Demonstration
4. Addressing Hallucinations and Lack of Transparency in AI Models
5. Infusing Trust and Accuracy in AI Systems with ScholarAI

Podcast Transcript

Jordan Wilson [00:00:16]:

We talk a lot about ChatGPT plug ins on this show and How they can help you get more reliable and accurate information out of Chappy ChatGPT. Because here’s the reality all. You don’t know what you’re doing and sometimes even if you kinda do, ChatGPT can lie quite a bit. It can hallucinate just like any large language model. But There’s 1 plug in that we even recommend to people that really helps with that. And I’m excited because we’re gonna have the cofounder today of ScholarAI Talk about how to create more reliable and trustworthy AI specifically within ChatChippity. Not just that, but a lot more. So welcome to Everyday AI.

Jordan Wilson [00:00:54]:

My name is Jordan Wilson. I’m your host, and this is your daily livestream Podcast and free daily newsletter, helping all of us, everyday people, make sense of what’s going on in the world of AI because there’s so much, and it can be difficult To actually understand and apply it to grow our careers in our company. So we’re gonna be doing that today. And if you are joining us live, thank you. Make sure to get your question in, whether it’s about How to create more reliable and accurate, you know, conversations within ChatGPT or if you just wanna know more about the ScholarAI plug in, make sure to leave a comment. Leave a leave us a question. If you’re joining us on the podcast, make sure to check out the show notes. We’re gonna send a link, or leave a link so you can even come and join the conversation on LinkedIn.

Daily AI news

Jordan Wilson [00:01:34]:

Alright. So before we get to that, let’s talk about the AI news like we do every single day. So big news out of ChatGPT, speaking of. Right? So, it can talk to photos now. So this new, feature being able to upload photos natively is now being rolled, is now being rolled out to many more users, that have the ChatGPT plus subscription. So, this is part of a larger group of updates that are kind of Unofficially being called, GPT 4 v with the v standing for vision. I don’t think that’s an official name. It’s just what a lot of people online are calling it.

Jordan Wilson [00:02:10]:

So early users have had access to this for a few weeks now, but general access, is happening now. Even, myself, got access to it, late, late last night. Actually, shout out, doctor Harvey Castro, a former guest on the show who said, hey. Do you have access? I have access. So we’re gonna have more on this in AI in five. Next piece of news. Google’s Bard is coming to new places. So, Google’s newer smartphones like the Pixel 8 and Samsung Galaxy S 24, we’ll feature an assistant with Bard.

Jordan Wilson [00:02:40]:

That’s not new news, but, there is some new news because the publication 9 to 5 Google, recently found some references in the source code, claiming that you could, as an example, tell the BART assistant to Draft an email to my recruiter to accept the social media manager job offer and negotiate a later start date. So we’re gonna have more about that in the newsletter, But some, some some new details leaking on how this new assistant with Bard, will work natively inside of some of these new, you know, Google, smartphones such as the Pixel 8 and Samsung Galaxy S24. Alright. Our last piece of news to go over. AI is soon going to be consuming more energy than an entire country. Yeah. That’s right. So all of these data centers that that are helping to produce generated AI.

Jordan Wilson [00:03:30]:

They take up a lot of electricity and a lot of water to, you you know, a lot of times cool these systems down. So a recent report from the journal, Jolley, I I I believe that’s how it’s pronounced, but, researchers in Netherlands showed that by the year, 2027, That all of these different, you know, servers, server farms, and, plants, essentially, where they help to create these generative AI systems Could use anywhere between 85 to a 134 terawatts of energy per year. I don’t speak Terawatts in energy, but, apparently, that is about as much power as Argentina uses in a year. It’s actually very fascinating to talk about, Electricity and power in Gen AI. We actually had, you know, someone from NVIDIA talk about that this week. So, we’ll we’ll drop the link to that in the comments. But You probably showed up here not to talk about energy and, how to cool generative AI systems. You probably wanna know.

About Damon and ScholarAI

Jordan Wilson [00:04:25]:

You probably wanna hear a little bit more, about ScholarAI. And and full disclosure, we have people hitting us up all the time and saying, hey. I have this, You know, this plug in, I wanna come on your show. I have this product, and we normally say no. But with ScholarAI, this is something we’ve been using and recommending for months. So I’m extremely Excited to bring on to the show and welcome on as a guest we have today. And help me welcome on everyday AI. We have Damon Burrow, The cofounder of ScholarAI.

Jordan Wilson [00:04:53]:

Damon, thank you for joining us.

Damon Burrow [00:04:55]:

Hey, Jordan. Happy to be here. Thanks so much.

Jordan Wilson [00:04:57]:

Alright. So hey. Just like Brian right here who said he’s excited to learn in today’s episode. If you are excited, if you have a specific question and are joining this live, Please please drop please drop your question for for Damon if you wanna know a little bit more about ScholarAI. But, Damon, let’s let’s start high level. What is ScholarAI, what do you all do and and kind of how does this thing work?

Damon Burrow [00:05:17]:

Yeah. Excellent excellent question. So at at the very core, ScholarAI is building systems that infuse trust Into large language models. And right now, that large language model is the GPT 4 transformer that powers, ChatGPT, And we do that through some very specific applications. Very happy to dive into the specifics of any of that, but, again, kind of remaining at the highest level. We essentially, tether The AI generated output that comes out of chat gbt to, peer reviewed and publicly accessible, scientific articles, databases, textbooks, Etcetera. What that does is it significantly reduces, almost virtually eliminates, hallucinations in general. It pulls information directly from those sources, And it provides hyperlinks to those sources such that anybody using those generative AI systems can immediately follow-up.

Damon Burrow [00:06:02]:

They can ask more questions, those those kinds of things. Longer term, we we plan, more applications outside of ChatTBT. So like Google BARD, like you mentioned earlier, We’ve got systems coming for that. We’ve got a browser plug in, browser extension coming for, Google Chrome and others, and then we’re also gonna have some dedicated web apps insight of very specific domains in which, some of our core user bases, have are are going to be perceiving some some outsized value there.

Jordan Wilson [00:06:29]:

Yeah. I love it. Yeah. And, I think people who use ChatGPT pretty frequently have probably either heard of ScholarAI or have used it before. So So from a very high level, you know, perspective, Damon. And and and we are gonna, explain this here in a couple of minutes, but, you know, maybe let’s answer the question, Why? Like, why do we need to use, you know, a plug in like ScholarAI to get better results?

Damon Burrow [00:06:54]:

Yeah. So I think I I think in some cases, you don’t. I think the reality is so where some of the, you know, truly generative AI systems really shine through are for creative tasks. So if you’re trying to write a new and Short story or you’re trying to write a new song or, you know, design some sort of, theatrical play or something like that. You don’t really care If the output in that is is grounded in what would be considered truth across the board. Right? If you transition though into professional knowledge work, so think about you’re a doctor trying to save somebody’s life, Think about your CEO of business trying to use, the last bit of runway you have to either save your business or to expand into new domains. Right. Think about, a lawyer who is doing a patent search for a client of theirs, being able to express whether they are free to operate in those space.

Damon Burrow [00:07:36]:

In in those kinds of things, you you, must be accurate essentially with with your use of AI. And so it’s those times in in In which the systems that ScholarAI is building, become essential. And again, the way that that works is we essentially, distill, the the creativity, out of these generative AI systems? A little bit. Not completely, but just a little bit. Just enough such that all the output that comes out of there is grounded in Truth as is, kind of supported by the domain expertise. Right? So basically, the way that this works in science, I can speak most, thoughtfully towards that because I am a PhD student. I I kinda live in this world every single day. Right, the scientific literature, that has gone through the peer review process is kind of the gold standard For the most up to date knowledge of, yeah, across scientific literature.

Damon Burrow [00:08:21]:

And so Scholar Ag gives people, who want to use the large language models to interact with these and Sources in a much more thoughtful and much more rich way, access to that to that material, and so that their, responses that come out Of the AIs that they’re using are, again, more accurate, more reliable, and then ultimately more transparent.

ScholarAI allows for up-to-date info

Jordan Wilson [00:08:40]:

Yeah. And, you know, I think it’s worth noting and and maybe even hitting, like, Rewind here a little bit, Damon, because if you are brand new, if if if you’re listening and maybe you haven’t really used ChatGPT a lot or Or you’re just a casual user. You know, there’s a couple important things to keep in mind. You know? It has a knowledge cutoff. So if you’re using the free version of ChatGPT, It has a knowledge cutoff of September 2021. If you’re using the paid version, it has a knowledge cutoff of January 2022. But Regardless. You you know? So the I guess there’s 2 things that I kinda wanted to talk about here, Damon, and maybe to ask a question.

Jordan Wilson [00:09:12]:

So, yeah, 1, can ScholarAI give us, You know, more more recent knowledge, you know, past that kind of cutoff date. And then 2, can skull ScholarAI also just give us more, more specific and more fact based things, you know, even included in that knowledge, you know, cut off just just for more accurate and And more specific citations.

Damon Burrow [00:09:35]:

Yeah. So all all excellent questions, and and, yes, you basically nailed it. So basically, you know, the the the cutoff dates are September 21 Or January 2022, depending on exactly which model that you’re using. ScholarAI provides access to up to date information seemingly in real time. So as soon as the information is published, Scholar has access to it through our various API endpoints, depending on which database you’re you’re pulling from. Those usually get updated Daily. Sometimes it can be, you know, twice a week, those kind of things. So in in very short, yes, and then to your larger point of how we maintain accuracy across the sources, we basically Go beyond not just the cutoff window, but also what is called the the context, window not not the knowledge cutoff in the context window rather.

Damon Burrow [00:10:16]:

So we actually dissect, that source material through some proprietary and trade secret algorithms that we developed such that we can actually package more dense information into large large language models Such that it can actually hold, a larger breadth of information. Think about, you know, thousands of sources at a time in its kind of working memory, so to speak, such that it pulls and Directly from those sources rather than having to rely so much on its predictive capabilities. Right? So we’re playing less of a probabilistic game, much more of a of a of a information, Gathering game.

Jordan Wilson [00:10:47]:

And just that that seems like a very timely response because yesterday, on the show and and we’ll make sure to leave a link to that. We talked like tokens and in memory. So so you’re also saying even the way that ScholarAI processes information, it kind of takes up less of of that token memory so So it can retain knowledge longer.

Damon Burrow [00:11:04]:

Yeah. The way that I would classify it is we so based on the semantic search functions that we’ve built, we can identify the most and Important aspects of each source such that we can kind of disregard the, like, less important tokens, if you will, so to speak. And so what that does is it it doesn’t actually truly in the context window, but it puts more of the information you care about in the context window. So that’s everything that you’re pulling from is the most relevant, the most accurate, the most important pieces of all those articles. And so what, again, what you get is reliable, accurate information from the most relevant of even parts of sources. So, again, we’re trying to synthesize information Across sources, not just pull information from a single source, and so that’s when it really becomes effective to kinda break those into their most meaningful chunks, if you will.

Jordan Wilson [00:11:46]:

Yeah. Absolutely. Hey. And, you you know, for those we have a lot of people here in the live stream. Let me know. Have you used ScholarAI? What What do you like about it? What are things that you wanna know? Because, I’m gonna ask Damon here in just just a minute, to kinda share his screen so we can take a look at this. But before we do that, Damon, like, talk let’s let’s talk a little bit more because you said, hey. In the future, won some of these other large language models, you know, Officially release, you know, plug in capability or third party, you know, developers to to be able to, work within their, models.

Jordan Wilson [00:12:18]:

But can you access ScholarAI right now without ChatGPT? Can you just log in and and use it, or what’s is is are there other use cases, I guess?

Damon Burrow [00:12:26]:

Yeah. So the the public cannot. We have a few, kind of applications that are in their alpha or beta forms for specific, you know, what I would consider professional copilots. So think about a lawyer doing the patent search, like that earlier as as an example. The the short answer to your question is no. Those those Things are coming down the pipeline. Still TBD on when exactly those will be released. For the public, what I can tell you is people should, Actually, avid users of Scholar actually be excited because in the relatively near future, we’re gonna have things like, browser extensions for their Google Chrome, Chrome, etcetera.

Damon Burrow [00:13:03]:

And we’re also gonna have some dedicated, UX UIs that give us some increased, capabilities that things that we’re just simply limited to insight Inside ChatGPT. So all those things are coming down the pipeline. There is no publicly available link, right now, but if you are a, ScholarAI user And, you feel that there’s, an element of ScholarAd that’s missing, then we would love to hear from you. We’ve got a community of about, 180,000 Right now, users, and a lot of us, you know, we we we would like to engage, with them and and see if there’s a gap that we can we can fill. So.

ScholarAI plugin walkthrough

Jordan Wilson [00:13:35]:

Yeah. Love it. Well, hey. Let’s let’s actually give, give everyone a chance to kinda see it side by side. So, Damon’s gonna pull up his, His screen here, and we’re gonna walk you through. So we’re gonna show, Damon’s gonna show an example of what you might see when you’re using ChatGPT by default. Because like we talked about, the beginning of the show, and we talked all about it yesterday, is if you really don’t give, a large language model access to the information and the resources that it needs, you run the risk of it kinda Sometimes making things up or just, you know, giving you, poor responses. So, Damon’s now gonna kinda walk us through, you you know, what kind of Happens kinda before and after so we can see the difference that a plug in like ScholarAI makes.

Damon Burrow [00:14:15]:

Yeah. Thanks, Jordan. To so it’s incredibly well said. The one thing I will say is OpenAI is making, chat gbt, both the models 3.5 and 4, better all the time. And so it is getting better at just simply telling you when it can’t do things other than just making it up, but people still do need to be careful because, they the answers the the hallucinations, still do happen, especially with a 3.5, model. But, I’m kinda jumping into this demo. What you’re seeing here on screen is a prompt that I’ve used several times, which I’m asking, the standard, chat gbt running the transformer 3.5 To show me 3 new papers about recent advances in artificial intelligence. And, unfortunately, it says, it cannot do that.

Damon Burrow [00:14:53]:

Some of that’s because it doesn’t have access to the Internet in real time, and also it has a knowledge cutoff that is in September 2021, so it can’t actually see anything that’s happened beyond that. It tells you, You can use other academic search engines like Google Scholar, at PubMed, etcetera. The problem there is that you can’t actually then use the large language model to interact with those sources. Right. So we don’t want to just use this as a reference finder. We actually want to engage with that content. We want to read PDFs. We want to ask questions from that concept.

Damon Burrow [00:15:18]:

We want we really want to, you know, get the knowledge Out of that source material, we don’t just want to simply find the material, and that’s what those LLMs are so great at. So swapping over to a use case after, ScholarAI, the The plug in has been installed on, the GPT 4 transformer. Here, you can see a much different response. Staying prompt, this is this is exactly copied and pasted one into the other, and here what we’re seeing is, 3 papers. I asked specifically for 3. You can ask for much more. I just wanted to do 3 just just for the sake of time here in this kind of conversation. But what you’re seeing here is papers with direct link to papers, and you can see here, directly to your question earlier, Jordan, about, the the dollars cut off.

Damon Burrow [00:15:55]:

So this paper came out in December of 2022, Obviously, well, after both knowledge cutoffs, but given access, to CheckCVT via, the ScholarAI plug in. We’re also seeing an abstract here, How many times been cited if the PDF is or is not available, and then just notice that in this, this title was actually hyperlinked. So if anyone wants to kinda go up and then follow-up with the sources, they can do that. One interesting thing is, what I can do is I can say to this gallery I plug in, as we said, we don’t just care about the information finding, we actually wanted to provide a summary of paper 1. And then kind of in real time, I might have to zoom out just to show you that, but what it’s going to do is it’s going to actually, give you a decent understanding of what the summary is. So, in the cases where it can find the full text, it will be explicit, and it will tell you That it has found the full text, it has parsed the full text, and has created the summary from the full text. If, because of publisher restrictions, ScholarAI does not have access to the full text, It will tell you it doesn’t have access to full text, and it will still create a summary from the bits of information it has access to. So it’s not just gonna guess and extrapolate the information.

Damon Burrow [00:16:58]:

It’s just gonna say, I’m only limited to seeing this much. Here’s what I can do for you. Here’s what I can give you that is accurate and reliable. There’s kind of your information. If you want more information, please follow-up through the hyperlink.

Jordan Wilson [00:17:09]:

Yeah. I love it. And, you you know, I think David just did a great job of explaining, and and, you know, if you’re listening on the podcast, again, come in. We’ll leave a a link so you can watch this. But, you know, even just showing, in real time, it’s fast. It’s responsive. Something that I love that I wish all plug ins would do is always, providing a link to read more to the original resource because, when you talk about transparency, that’s something that we always kind of worry about with large language models, you know, because there’s always that little, seed of thought where you’re like, alright. Is this real? Is this a and So being able to have the link is great.

Other use cases for ScholarAI

Jordan Wilson [00:17:45]:

Damon, like, right away, I I thought of a couple great use cases. Right? So if you’re trying to learn a new subject, this seems to really cut down the time. Even, you you know, using ChatGPT cuts down the time, but using ChatGPT with scholar, Amazing. Right? So if if if you’re a reese maybe if you’re a student, you you know, working on a topic, working on a paper, this seems great. Maybe what are some other and Kind of everyday use cases or very, wide ranging applications of the ScholarAI plug, plug in.

Damon Burrow [00:18:13]:

Yeah. Excellent question. So roughly 60 to 70% of our use cases come from what I would consider graduate students, so master’s, PhD students, and or researchers, Many people in academia doing kind of hardcore, you know, research. The other, you know, roughly 30 to 40% come from, you know, Other business professionals, specifically people doing due diligence. Again, trying to assess whether, some businesses technology is Above board, what the maybe kind of isolated advantages of 1 technology are over a competing technology, that kind of thing. Also, as I said, at the c level, executives can use this kind of to do, maybe maybe mergers and acquisition due diligence or maybe just market research and seeing, you know, which kind of, path Might be most advantageous for their kind of business in its current state. We see a lot of journalists using this platform when they’re doing a a story and their core expertise may lie slightly insight of the story that they’re writing about. Right? You can think of, without diving into the specific of anything, you you can think about, you know, how kind of, groundbreaking and how kind of earth shattering the the COVID nineteen pandemic was and how much, you know, kind of information was swirling and people were having a hard time deciphering what was real, what was not real, that kind of thing.

Damon Burrow [00:19:22]:

So you can you can imagine, some very important use cases there. And like you said, students of all ages just kinda coming in saying, hey. I don’t understand this. Can you please explain this to me at a level that I do understand? Can you please provide me links for this if I want to keep reading? That kind of thing. So We we do see a a very wide range of uses, and we’re getting better and better at serving each one of those use cases. Like I said, we kinda started With the science and medicine focus, and then we realized what we were building was, much more applicable, to everybody else. I’m also seeing the the the chat, the comment coming through in the chat. Yes.

Damon Burrow [00:19:55]:

A lot of clinician scientists are using us both to propel their research forward, but also help them treat patients. You can imagine, how important it would be to be helping a a cancer patient who has failed the standard of care treatments and to, you know, their their, Physicians who have whomever helping the nurses, etcetera, need to be parsing, not just published data, but also clinical trial information to say, How can we help this person? You know, how how can we give them improved quality of life? How can we help them live longer, live happier, etcetera, etcetera. So

Jordan Wilson [00:20:25]:

Yeah. No. It’s So many so many great use cases, and, you you know, the one that you really, mentioned there. And I have to call students out. You know, students, If you’re using ChatGPT, don’t use the free plan. Like, it’s it’s worth the $20 a month to be able to tap into plug ins Like what Damon is showing us here, ScholarAI, and there’s so many other great plug ins that if you are just trying to learn or help write papers, this right here is going to save you so much time, Not only improving the quality of what you’re trying to write, but it’s also gonna help you learn. It’s gonna help you learn faster, learn, learn a little bit better. So, wanna get into a couple questions here, David.

Plugins to pair with ScholarAI

Jordan Wilson [00:21:00]:

So speaking of like, I love this question. So Val is asking, you know, what other plugins maybe work well with ScholarAI because, you know, ScholarAI serves a very specific purpose, and it does it well. Right? Like, I I’ve used it a lot of our, you know, listeners, use it all the time, but maybe what other, plugins complement ScholarAI?

Damon Burrow [00:21:20]:

Yeah. The one that I would call attention to is is called show me diagrams, and there are other Similar program, plug ins that do kinda diagram work. But, like, for me, I I’m somewhat of a a visual learner, so I wanna go beyond the text. Right? It’s it’s very good to get that summary, but I actually maybe Want to work, maybe make a a mental framework or some sort of mental model, you know, diagram explaining some sort of kind of arcane or very specific concept to me that maybe I wouldn’t understand otherwise. So the one that’s that’s very top of mind, the one that we see most frequently is is show me diagrams. But, basically, anything that’s gonna allow you to take, information, especially dense information, And to simplify in some sort of visual visual graphic, is is, you know, really powerful.

Jordan Wilson [00:21:56]:

Yeah. Absolutely. It’s it’s it’s actually funny you say that, Damon. Like, we, we have our kind of free, prime prompt polished PPP course. We do it twice a week, doing it later this afternoon. So if anyone wants access. Just, you know, drop PPP. I’ll send it to you.

How ScholarAI came to be

Jordan Wilson [00:22:09]:

But we actually have, you know, ScholarAI in in that exact plug in, you know, diagram on the same page, you know, for our recommend, our Our recommended plug in, so I I I love that you brought that one up as well. Question question here for Monica saying the episode’s blowing her mind, but saying, she loves learning from leaders in the AI space and asking, How did you even come up with the idea for ScholarAI?

Damon Burrow [00:22:30]:

That’s a that’s a fantastic question. Thank you, Monica, for asking that. So I’ve been doing AI and what would be considered more ML research since roughly 2015. So back when it was kinda called gradient descent, I was doing it in undergrad, and there I was helping, Optimized radiation therapy for, cancer patients. So basically, just trying to get the radiation to go to the cancer tumors and leaving the healthy tissue alone. You know, roughly a year ago, we saw this kinda momentum of these large language models gaining power. They were becoming, you know, more useful. People were kind of interacting with them in their everyday lives, and we were using them both in in the research setting and also kind of just commercially, Whatever was available, and we were seeing some of these shortcomings.

Damon Burrow [00:23:13]:

Right? The hallucinations, just the lack of transparency. In the very beginning, you would ask these large language models for sources. And rather than telling you they couldn’t give you sources, they would just predict the text that was supposed to come next, and they would just make them up. So they’d just be they would be real author names, not real people. They’d be real titles, not an actual title that’s linked to a source, fake BLIs, etcetera, etcetera. And so we said people are going to be using this, in their professional lives and they’re going to be making, You know, important decisions based on this, and these systems are going to need, guardrails or structural supports, if you will, that are going to make them ultimately useful for the things that we care most about. And, again, Scholar Eye emphasizes, any use case in which the loss of life, the loss of revenue, or the loss of reputation, it it is kind of occurring, and so you can think about fields like medicine, science and research, law, business, etcetera. But that’s that’s kind of how we Came up with this week.

Damon Burrow [00:24:03]:

We saw we saw the momentum of large language models, becoming more useful, becoming more powerful, but but we’re still kind of, you know, being, handicapped, if you will, by these, shortcomings, and and we saw a need to overcome these shortcomings. And, you know, we think it’s gonna be An evergreen problem because even as some of these AI systems improve, there’s always going to be this superabundance of sources, in which case you don’t really want to be just pulling things randomly from the Internet. You want these things to be context aware. You want the most relevant sources being pulled into your into your chats or into your AI agents, in general. And so that’s That’s kind of, you know, where we where we hang our hat at Steller Eye.

Audience questions

Jordan Wilson [00:24:43]:

Yeah. Yeah. Love it. Love it. Alright. We have a couple more questions. We’re gonna try to go rapid fire here so we can try to fit Fit more in here. And, again, if you’re joining us joining us a little late, we have, Damon Burrow, the cofounder of ScholarAI joining the Everyday AI Show going through some questions On, the very popular ScholarAI ChatGPT plugin.

Jordan Wilson [00:25:01]:

So a question here from Gabriel asking, do you include international sources in ScholarAI?

Damon Burrow [00:25:07]:

Yep. Answer is yes. So we just recently, as of this week, have access to over 200, 1,000,000 different articles, databases across the world. So globally inside of the United States and outside, we have access to, virtually all, published scientific, literature, even the preprints.

Jordan Wilson [00:25:23]:

Wow. And, like, As someone that uses y’all, like, a large language model a lot, the ability to tap into all of those sources in 1 plug in is Extremely valuable. Right? Especially when you are kind of kind of limited, you know, because you can only have 3 plug ins at once, when you start a new chat. So you wanna be a little judicious about what those are. Alright. Another another great question here from, Douglas. Douglas, thanks for the question saying, would this be something you could look for drug interactions? Obviously, consult with your doctor, but, you know, When it comes to drugs and how they work, is that something you can also, ask ScholarAI?

Damon Burrow [00:25:56]:

Yeah. Yeah. Short answer, yes. All those clinical guidelines come from research that is conducted in the industry academia, all that normally is published in the academic research, all of which ScholarAI has access to. So, yes, obviously, consult with your, physician and doctor. Don’t trust these things blindly, but but direct answer is yes.

Jordan Wilson [00:26:12]:

Yeah. I love that because I always tell people, like, don’t trust me just because I recommend something. Right? People always ask me on I say no. Don’t don’t, you know, trust some random person on the Internet or a random plug in. Go go try it for yourself. You you know? Make sure fact check it first. Make sure it works for your use cases before you just blindly start, you know, trusting anything, really. I I think that’s a great, a great, point there.

Jordan Wilson [00:26:34]:

Mike asking, how is this monetized?

Damon Burrow [00:26:36]:

Yeah. Excellent question, Mike. So we we have 3 tiers. ScholarAI is available to, everybody and For free so long as they subscribe to, ChatGPT Plus, which does cost $20 a month. That is limited up to 25 per requests per week. So it requests anything like Search for a paper, show me the full text, give me a summary, those kinds of things. The ScholarAI basic package is available for 4.99 per month or $50 Per year, and that basically just comes with, unlimited use. We also have a premium plan that’s 8.99 a month or, $85 For the year, and that comes with, unlimited use, but also some advanced features.

Damon Burrow [00:27:11]:

Like, one of the things that we’re most excited about is, figure and table extraction. So, in the sources that you find or the PDFs that you upload to ScholarAI, you can actually ask questions not only of the text of that material, but also from the figures, and the tables itself, for premium, subscribers. And you can see at the very bottom there, we have a meta analysis, coming soon. A meta analysis is, Probably a much longer discussion than needs to be had here, but basically, is information synthesis across a bright a wide range of sources on a on a singular topic. So think about, condensing the information insights that come from, 10,000 sources on, infectious disease, as an example. So

Jordan Wilson [00:27:50]:

Great great overview. And I I I love the ability in that, highest plan to be able to read tables, charts. That’s so important. Right? Like, we opened up the show talking about how We can kinda do that now with ChatGPT Right. With the uploading the photo, but the downside of that, y’all, is right now, that only works in default mode. So you can’t combine that With any other plug ins, so, that’s that’s a huge, huge benefit to ScholarAI. So, alright. Well, Doug, like like we, David, we went through a little bit of everything, in today’s episode.

Final takeaway on ScholarAI

Jordan Wilson [00:28:18]:

You you know, you took us on, like, why this started, you know, what in ScholarAI does, the benefits of it, how it can help us get more reliable and accurate information. But maybe what is the one, takeaway that you hope people get, on the benefit of ScholarAI and how it’s going to help us, you know, just put out and create more reliable information.

Damon Burrow [00:28:38]:

Yeah. I I think that, like I said, these large language models are going to continue to become better and better over time. I think that we’re going to see, especially, professionals in in the knowledge world, science, medicine, law, business, journalism, etcetera, are are going to have to learn to use these systems more thoughtfully, And and we want to be the system that helps people do that. Right? So, they can get to doing the things in their daily lives that make a maximum impact and don’t have to worry about the information that’s coming out of their things being accurate. Right? We want them to be able to trust the responses coming AI out of their AI generated, either chats or kind of interfaces regardless of whether those whether that is, and we also want to create the level of transparency, like we said, to where we’re not necessarily saying trust us To kinda kind of, you know, make your impact on the world doing kinda whatever you, are best at.

Jordan Wilson [00:29:33]:

Oh, love that. Yeah. You always gotta show your sources. An old saying, I believe I I learned in journalism school as someone said, hey. If if your mom says she loves you, get it in writing. Right? So, hey. I I I love that ScholarAI, plug in does this. It’s one of our favorites.

Jordan Wilson [00:29:47]:

It’s one we recommend. Damon, thank you so much for coming on the show to join us.

Damon Burrow [00:29:52]:

Yeah. Thank you, Jordan. Thanks so much.

Jordan Wilson [00:29:53]:

Alright. Hey. Just as a quick reminder, we covered a lot. If you didn’t if you can’t type as fast as we can, take notes. Don’t worry. Make sure to go to your everyday AI .com. Sign up for the free daily newsletter. We’re gonna have a lot more about ScholarAI, maybe even some things that we didn’t get to that the plug in can do.

Jordan Wilson [00:30:08]:

So make sure you go do that, and make sure you join us Back tomorrow and every day for more everyday AI. Thanks, y’all.

Leave a comment

Your email address will not be published. Required fields are marked *