An LLM that fits on your phone, works offline and has passed the US Medical Licensing Exam. Oh, and by the way, it's open source.
Today in health, it Dr. GPT. I know what you're thinking, but let's just talk about it for a little bit. Cause I it's very interesting to me for a couple of reasons. My name is Phil Russell. I'm a former CIO for a 16 hospital system. And creative this week health, a set of channels dedicated to keeping health it staff current and engaged.
We want to thank our show sponsors who are investing in developing the next generation of health leaders, short tests, artist, site, parlance, and service. Now check them out at this week. health.com/today. Having a child with cancer is one of the most painful and difficult situations. A family could face in 2023 to celebrate five years at this week. Health.
We are working to get back. We will be partnering with Alex's lemonade stand all year long. We have a goal to raise $50,000 from our community. We are up over that number for the year. And we thank you for that. , we asked you to join us though. We are going to plow through this number and see how far we can get in our website and the top right-hand column. You're going to see a logo for the lemonade. Stand, click on that to give today.
We believe in the generosity of our community. And we thank you in advance. All right, Dr. GPT, that was like, is this really a thing I'm sure is what you're thinking. And the answer is yes, it is. In fact, it's open source, Dr. GPT. PT is an large language model that can pass the us medical licensing exam.
It works offline it's cross-platform and your health data stays private. And by the way, I'm looking at it on GitHub. So it is absolutely open source. You can come out and get this thing and take a look at it. It's based on a Meadows Le , oh gosh. Metta is a Lama to platform. I just blanked on it for a minute. , and they released a couple of different, , versions of the, , Lama to model. There's a 7 billion parameter model, and then it goes up from there.
I think there's a 24, there's a 70 something. , model. , the reason this one's based on the smallest version is I think one of the most compelling things about this model. And it is that the w once they get done training, it turned into a, a. Tensor virtual machine and putting it on an iPhone. That it is, , only three gigabytes in size. It fits on any local device. So there's no need to pay for an API to use it. It's free. It's made for online, offline usage.
Which preserves patient confidential confidentiality. And it's available in iOS, Android, and web. , It's really fascinating to me. So let me tell you why I'm talking about this. One is, , passes the U S , medical licensing exam. Now, first of all, it passes it at the lowest level possible to pass it.
But to be honest with you, we have no idea if the doctor we're seeing today. Past it at the lowest level or the highest level, we have no idea. But that's it. You're not going to use this for that kind of a model that, that high level kind of medical advice anyway. Here's what this is really valuable for fits on the phone.
It is standalone on Google and Android. It, , Which I think is fascinating. In and of itself. Right. So we can now put this thing on a phone. It does not need to get to the internet to get the answer. The model is standalone. It is in the, , the application itself. Okay. So that again is very interesting, but it's not dynamic in that. It's not going to continue to grow and that kind of stuff. You're gonna have to recompile and, and, , continue to train it.
, elsewhere. I think the other thing that's interesting to me is the, , the, , I watched the half-hour video video on this. If you get a chance to see it, It's a meet Dr. GPT. It's by the gentlemen who put it together and he walks you through every aspect of it. , when you, you get the Lama to model it.
, we'll get like a 20% on the us medical licensing exam. And then you do reinforcement training. You do, , , you do a lot of different training models. You pull down these different things and he tells you exactly how he built it and what he did. , so the, the, , video itself is fascinating. If you're really geeking out on this stuff, you really under one understand.
How they train these models and the different, , reward systems and that kind of stuff that they're using, how they're actually tapping into. Chat GPT to, , to validate, , some of the answers. And D to validate, , some of the models it's again, really worth, , worth watching. Meet Jack, Dr. GPT, offline AI that ACE the medical tests. It didn't ACE, the medical tests. That's that's a little, ,
, overstating, but essentially. , thinking about thinking about it this way. There are large. Areas in our country that do not have access to doctors and yes we can have, , or around the world is probably a better example around the world. We have places where it's hard to get, to see a doctor.
Now we have a lot of different models that we're working on. We have, obviously we have a telehealth models. We have a. There's some. I'm sure there's some other models I haven't really thought about at this point. But in this case, you could actually. You don't have a situation it's pretty urgent. You don't have access to a doctor, but you have access to your phone. You've downloaded this app.
And it's available. That's one of the things I think is very fascinating to me. The second is. , Lama too is open source, completely open source. In fact, the only limitation, if you're going to build an application on this is once you cross 700 million. Users. That's a lot of users. , you actually have to have a conversation with Metta about it. And one of the reasons they wanted to do that is they didn't want their competitors, their largest competitors to utilize.
The Lama to platform, right? You don't want Google picking it up. I know it's kind of a server, but you don't want Google picking up or Microsoft picking up starting to build applications on top of the model. , you want, , you want the entrepreneurial community to see this. And start to build on it. So it's, it's truly open source model.
, it's a model that you watch this video, you get a lot of different ideas for how to train these models with reinforced learning and that kind of stuff. And, , I think there's an opportunity here. Let me tell you what my, so what on this is, as I'm sort of. Doing this on the fly. Cause I don't really have a script that I'm going from here.
I think it's interesting and it might be worth considering. Depending on the size of your organization, whether you need to start developing a muscle around development projects that have these kinds of capabilities, if you could imagine your digital front door, whatever that term means for you. Having the ability to, , ask these basic questions.
Of a doctor. A again. , a doctor. That has passed the us medical licensing exam. I don't know if there could be use cases where you want that to be your organization. I'm still looking for the healthcare organization that really adopts digital transformation of healthcare. And my concern is if we don't adopt it ourselves, if we don't reinvent ourselves, that there are going to be.
Other organizations that come in, roll these things out in our markets. And then we end up with a fragmented. , method for our markets to receive care. I keep coming back to this model. If I were a C E O of a health system, I want to be the trusted source for health information at every, every aspect of health information.
For my community. If that means rolling out a tool like Dr. GPT with my branding on it. I would want to do that. If that means. , you know, tele-health first, I would want to do that now. Clearly we have a business model. That is tied to reimbursements and tied to a lot of those things. And we'd have to keep that in mind. But at the end of the day, I think my organizing concept would be.
The the source of truth for health information in our community. You can go to a lot of different places, but at the end of the day, when you want a trusted source for information, you're going to come back to us. You're going to come back to us. The health system. And so when I see a tool like this, I think open source available to any health it organization around the country today.
, you could incorporate this into your digital front door. You could actually roll out an application that is, you know, branded with your brand and those kinds of things. I think there's an opportunity, at least I believe there's a discussion to have. And I can hear the, the, , rolling of the eyes as I even talk about this. But I think you find your early adopters, you find your champions and you say,
All right. Let's discuss. , you know, let's have this kind of conversation where we put this out there and just discuss, is there a use case where this works for us? Is there a use case where we have an underserved market that we could serve better with a tool like this? , and, and then, you know, if there is a use case, then you can play around with it.
See how it answers the questions, have your doctors put it through the tests and see if it gives you the kind of, , kind of responses and the kind of answers that. , you've deemed of enough quality to be utilized by your health system. But again, these kinds of tools are coming. They're coming fast and furious. The more I dig into this.
, the, the, the more tools that are available, the more stand-alone tools that are available, the more, , cloud-based tools that are available around large language models. , I read a bunch of stuff over the last couple of days about, , Elon Musk's, , endeavor. And if you haven't seen the team that they have put together for this thing, it is.
It's a. Just a drop dead, amazing team. Of 20 of the best AI scientists you can imagine. So. , I expect interesting things out of there. And then, you know, this is Metta medicine Lama to. Is, , is looking to be the platform. That. , democratizes access to these large language models that. That you can train.
On whatever data source you want to trade it on. So. Anyway. That's all for today. Just something to consider. If you know someone that might benefit from our channel, please forward them a note. They can subscribe on our website this week. Dot com or wherever you listen to podcasts. We want to thank our channel sponsors who are investing in our mission to develop the next generation of health leaders, short test artists, site, 📍 parlance, and service. Now check them out at this week. health.com/today.
Thanks for listening. That's all for now.