There is a talk I give about how I think clinicians should look at diversification of career and do have side things they do to keep them engaged. This is one of my side excitements and we have incredible people here. So this is what M&I do every day with a lot of other people in the room trying to figure out how we can push the limits with this. James, who is leading the video and audio here, is also a physician. He's still practicing and he has a side hustle. So it's a fun thing to do. It keeps you alive. So with that, these are two other people that do side things. And actually, before I asked M, they do this every year. This one is more focused on everyday AI stuff that is not necessarily related to patient care. Things that we think everyone needs to know about if you're somewhat close to being current. Okay, so Carlos, starting. We have a bunch of videos in the presentations. But with Todd's request, we'll do some workshops over the next year. So some of them are just going to be like glimpses of what you might see in the future. Yeah, and to explain that. So we will start giving workshops in depth on each of these things we've done one. We're going to do one at Steve's Vale course. We're going to do them throughout the year because there's been enough requests. We're going to give just high level here if you want more info, contact us and we can show you more. Perfect. So let's start with the tools that will make your life easier. So we already know open evidence. How many people here use open evidence? How many know? Okay. So we had like four or five hands recent. How many people here have used open evidence with their advanced capabilities that only people in the US have access to? We got. Okay. So let's talk about it. So we know that open evidence has had lots of partnerships with any GM, GM with JAMA, et cetera. So it's been becoming a tool that has partnerships with Baluat using it as a medical provider, as a medical practitioner. Next please. So now we can give you a CME. Now you can use it for studying. Now you can use it to further Baluat what you're working on in a more... Yes. All right. I'm going to cut you off. Hold on. Does anyone here only like two people knew what open evidence was? So let's back up a little bit and explain to people. This is one of the most important sites that has come out in the last year. You 100% have to know this. This is not a maybe. This is probably the most important site as physicians we need to have. The guy who came out with us became a multi-billionaire in about two months. It is absolutely critical. Can you explain what it is? So open evidence is a large language model which we already know and we've been hearing so much about them. Cloud, chat GPT, Lama, et cetera. But the thing that we have to know about large language models is that they're really big, fancy, very well-funded text completers. So they depend on what text you train them on. Open evidence has been making sure that they've been training these models with medical text. These partnerships that they've been creating ensures to us that the text that they're checking out is medical text. So that's why in recent papers that they've been publishing, they give better answers against other open elements. Any questions about open evidence up until now? I don't think we can add that the review of all the resources were... Ask Mike, Mike, Mike, they give you also where they come in the information from. Yes. It doesn't hallucinate like any other large language models out there. It gives you more than resources. If you bring up a case scenario which is a part of the HIPAA compliant. If you get registered by your business administrator from your hospital, you can even put your actual patient information and ask for a case solution and it diagnoses and management options. Alright, seven minutes. Perfect. So what's new? Now since it's HIPAA compliant, it's been enabled only in the US to receive attachments from your guidelines. So now you can upload the guidelines from your hospital and it will use actual up-to-date information comparing your guidelines to what's happening up to date. So it's really, really useful. It sites how it's comparing your information against actual information but always with a little asterisk because it still sometimes hallucinates. But we always have to check what it's producing. So here's a little example of the attachments. Here's a little example also of what's called a deep consult. Normally you ask a question, it gives you a quick answer but here it's using something called reasoning where it could, we could say think about the question. It analyzes the papers that more better respond to your inquiry and you can see here in this video that it not only citing the papers but it's citing important graphics, important tables, etc. And at the end it gives you the actual papers that you can check out. What kind of question did you ask to get that result? The question was, what kind of question was asked to get that result? You can see it actually. The question was in children with uncomplicated appendicitis. How does primary, not operative, oh my god, this again? Nonoperative management with antibiotics compared their appendectomy to at one year of treatment success. 30%. So one of the news that was very exciting, this is the first AI in the Luzge language model history that actually scored a perfect 100% on USMLE. And now you can use it even for USMLE study sessions for step one, step two and step three. You picked the topics and stuff topics and it creates a scenario for you. And then you do have your quiz questions and if you answer correctly it changes the level next time with a little more difficult. And if you do wrong, you can use the explanations, references for you to learn more and if you can connect with a third party apps to make flashcards. And now they have templates to make patient notes that you can either create your own templates that you can use in your hospital or you can use their existing templates that would aim to decrease the amount of time you spent on the chart. And the next thoughts most excited part about chat GPT. We hear a lot about the operator projects GPT-5, the codex, agent, research mode, deep thinking. They have a lot of capabilities that you can use in chat GPT, they keep adding more and more and more stuff. And we'll talk about the operator projects GPT-5 and custom GPT today. The operator is a human and chat GPT co-pilot, something together. And AI does most of the job but as a human you can interact interacting, where you jump in any time to steer and change the direction of the topic going on. And Todd will show us how to do it. He said make me reservations for two people between 6 and 7 pm near this area at a stake restaurant. And go ahead see what you can find. So here's what it did. All right so I typed in my request, I put in my name and email info number so it could fill in and you don't have to do that. And then it searched on open table, it found a stake restaurant near me and then it found the right time, it removed the ad there and then it started filling in my information. So when it was done it basically said here you go, it's done. Do you want me to go ahead and do it or not? And you say yes or no. It's amazing. Go ahead. So the next one is the projects. We want to expand it so it'll be more Todd, the projects part. Yeah, you can do it. Oh, this is just projects is the idea that you can put, who uses projects? All right so it's classifying you create a smart person in your team. So you have a whole team of people to help you. So I have a bunch of different people on my team and all this stuff related to that that I uploaded all the stuff I need them to know. I go to that person for a specific request. So I have my team. This kind of shows it right or is this not a big deal? Did you know that you could literally build an entire staff that works for you in chat, you be team. So all you do is go into chat, you can see and go to projects on the left. And just make a bunch of projects that are I hear a mind social media sand executive coach Eddie financial Frank. I have travel agents. I have my stoic philosopher. I have everything you can video editor Vicki. I have all the ones and I just keep adding them to it to it. And then under each one you have all the different projects. So if I'm doing social media, I said you are an absolute ultimate expert in social media. I don't know anything. Here's what I'm doing blah blah blah and it remembers everything you can upload documents and then every time you want to do something different, make a new project underneath that expert. So that staff handles it for you. It's sick. It's insane. Now I just got chat GPT pro to try it for a month. I'll let you know if it's that much different. But go ahead, give it a try and tell me what you're using it for. So I can copy from you like Tom. Okay. So this one's one of talks, cost them GPT's one of his passion projects. All right. I believe the best way that we learn in residency is by screwing up. And I want to figure out how we learn better in the future. Now part of me is wondering, do we still need to learn because you're going to get answers when you need them as we just saw from open evidence. I still think we need to learn certain things and figure out how to do it. I think making mistakes is how we learn best. So we designed this GPT. It's not perfect yet. I'm happy to send it to anyone to test it. It is designed to take you through case scenarios, but not tell you if you make a mistake. If you make a mistake, the consequence of curves now your patient is coding, but it doesn't tell you why your patient's coding because you forgot to resuscitate a pyloric stenosis and you went straight to the operating room. You're not they're not going to tell you that. It's just going to say now your patients in the hospital. So I have it almost for every specialty. We're trying to make it a hundred times better with images and video, but it's basically a board review. But it's it's the best. Now we're a senior clinician being tested on a scenario involving an advanced stab wound to the abdomen. Remember this is all for educational purposes, not real medical advice. Now would you please confirm that you're ready to begin the scenario? I'm ready. All right. Let's get started. You have a young adult patient brought in with a single stab wound to the abdomen just below the amylicus. The patient is hemodynamically stable for now, but there are signs of pertinatal irritation. As the examiner, I'm going to ask you what do you think is happening internally? And what is your initial management plan? What is your next step? Well, if you've got a pair of new signs, I'm worried about a visceral or bleeding injury, but the first thing I want to do is assess his airway, his breathing and make sure his vitals are okay and see if he's still hemodynamically stable. And I get to two large 4IVs in place. So you decide to focus on the ABCs first, check his vitals and establish large 4IV access. After doing that, you find that the patient remains stable, but he's starting to have more dominole tenderness, been guarding. What do you think is happening and why and what do you want to do next? So as he's going on like that, but I keep picking mistakes on purpose to try to get it. I keep trying to, you know, oh, I want to do a cranny out of me. I mean, you can say whatever you want to try to see, oh, really? You have to say that it's a boarder. You have to say it's an exam or chat with people not like you share it. You have to say, unless you do it in a private tenant. And every time it's a different case. Oh, yeah, you're not uploading anything. You're just saying like I need to study about this and it has old knowledge and just creates this once. And I believe it took a lot of time for you to ask it not to hallucinate. Yeah, and then it gives you a summary at the end of what you did right and wrong and it can send it to whoever you want if you use an agent. It's a time zone. Two more minutes. Go ahead. Okay, so here's an example of something on the same pad way, which is a version of chat GPT that's called study mode. So it's a little like what Todd has mentioned where the end point is not the answer, which normally you want the answer that you're asking for, but you change the end point so that chat GPT knows that you wanted to help you learn. So the end point is it explaining to you what you asked in a digestible manner step by step and it adapts to how you answer. So it makes it easier. It explains more. It goes to the next topic that you're learning on. Yeah, by coding. Okay, I just want to mention one more thing with AI coding agents. We are able to make our own apps in our own terms with the every capability that we have. It's like when the computer is in it and I first started everyone would get a Gmail address and hotmail address and they would drop on online messaging to each other. So the idea of wipe coding there are different websites and different difficulty levels going on and you there's like fully automated ones. You just say like, oh, I want to make an app that plays to effect. So with me all the time and it's just going to make it itself, but we're working on some apps or websites. One of them was toss it by one and make a website that will help me hire interns. And we were able to make this website in three minutes with AI. We just said we need an app that we need to have said that we need to upload some CVs and have some information about interns and then you can find a domain name put it online and it's there. You don't need any help to build those things anymore. So this is what I think is probably transformational maybe a little early, but everyone needs to know about this. This is like the new internet. There's going to be precise customization of apps and solutions for each individual and everyone can make their own. You don't need massive developers. So we are trying to do this at our organization. We've convinced our leadership that there is a top down in a bottom up approach to AI. Yes, we can buy technologies. I want the entire workforce to be enabled to empowered to vibe coding means you just type it out. I want an app that does blah blah blah blah blah and there's your app. So we think about it a year or it's still pretty hard, but we think within a year it'll be democratized. Everyone will be making their own hyper solutions each day. Adding to what I thought saying, these things are so effective that it's making the question is software as a service that as a business being asked. We don't know the answer to it, but the fact that it's so effective that it's making us ask that question should enlighten to how effective these vibe coding is. Workflow automation, we're going division by division, asking what they do. We believe we can improve the efficiency of the entire hospital and say hundreds of millions of dollars over the next three to four years by allowing full automation of everyone's workflow. Because if you go division by division, we started with my wife's practice to see we can completely cut down half of the time that they're spending money they're spending and we can do it for every single division individually. So that's workflow automation. I'm telling you, these are just touch points. We'll do a whole workshop. You guys must know about it or someone in your organization must know about it. You have to have people on your team that meet three times a week to make sure you're up to date. Is that it? No. Okay. And Carlos can do a workshop so we can do more questions, but that was awesome. Thank you. Yeah. Thank you.
Click "Show Transcript" to view the full transcription (15795 characters)
Comments