Will ChatGPT and LLMs Revolutionize Healthcare?| Persimmon Health

Chris : 

Hello and welcome to the digital health community podcast by persimmon health, the healthcare arm of leapfrog technology, where we will explore the groundbreaking advancements in artificial intelligence and their potential to revolutionize healthcare as we know it. And by this, we really mean generative, specifically the MLMs, and GPTs of the world. I'm Chris, your host. And I'm joined by my good friend Bimal, who has been our head of product going on eight years.

‍

Bimal:

Hey, Chris, happy to be here.

‍

Chris : 

Yes, so, Bimal. We've both been working with AI for quite some time now. But for our listeners, why don't you share a bit about your background and AI and health tech?

‍

Bimal : 

Sure. So I would like to say three things. Number one is I'm deeply interested in applied AI. So by that means I build a team around applying AI to different applications and use cases and health at LeapFrog. That's one and second and by the extension of that, we launched our AI startup called kaboom within the frog, which was a no code recommendation letter. And finally, the third one is, for last couple of years have been working with healthcare startups how to integrate AI into their applications and their software's and their products. And even I wrote a table called Lynn's AI playbook for shops. And since then, the fourth one, which is one more thing, which I didn't add it was playing with Chat GPT. since it came out primalist It came out in November 2003. And having have been using around using the pay playgrounds for different use cases have. I'm stunned by how good it is, and how great the potential to disrupt every piece of the industry is. It is a great defensive.

‍

Chris:

 Right, right. Yeah, and for me, I would say AI is more of my I would call it old stomping grounds. Back to my grad school research days, with GE and working on projects like we work together and like Iris Chatbot. That was a pre GPT chatbot that helped patients reschedule appointments and refill medications all by chatting with an agent. And like you I've collaborated with numerous startups around AI. And I know we're both passionate about AI and its potential for healthcare. So today, I think we'll really focus in on the role of AI and modern healthcare, but specifically focusing on the new, generative AI and LLM and ChatGPT. Yeah, no, I mean, it's, it's, it's an exciting time to be in this field. And I don't think we're the only ones that think so I know that many prominent figures are discussing this platform shift. Yeah.

‍

Bimal:

So if you listen to all in pod, or for our few eminent figures, like a Bijay Pandey from a16z, and who writes fiction, who wrote a couple who writes a lot about these things, they have been really, really thinking about, Okay, what's the impact is going to be an all of them are calling it a platform shift. And there is amazing article by SSG, which they say, right, okay, next biggest company is going to be a healthcare company. And they lay out the reasons why it's going to be and I think, this large language model, which, which actually sits on top of the paper come from and pretends and is all you need, which is transmitted by Google, and University of Toronto paper, actually, that's harder, and that all of those things are looking to like nails to be surprised to be solved. So this is a great type of SIFT. I think a lot of things change. And these potentials of it is not just off, of course, like everybody's talking about how they can integrate last managing models into their products.

‍

Chris:

Yeah, and they're using big language around this platform shift. They're saying, saying it's the biggest thing since or on par, even with mobile and social and things sacks who said even on par with the Internet. So yeah, it's something we have to pay attention to. Just in case they're 25%. Yeah.

‍

Bimal:

Absolutely can be like, which astonishes me is it's the fastest growing, fastest growing company or fastest growing product which has higher user adoption. Right. And I think that's not the only complete story to it. If you think about which are other fastest growing companies, Facebook, Google, and all of those companies were the fastest growing prior to open chatGPT. Yeah, but difference is that they though those were multimedia products. And what that means is there's an inherent network effects. And ChatGPT is a single user utility. And by that, even if when it was a single user, actually the per doc, it grew up $200 million, a few million users in a few months, right, which is extraordinarily ordinary. And imagine what would happen, they would start unlocking those multi user products. And when they're more, you know, they're going to use reinforcement learning and particular use cases that expand, they're going to expand to like billion users pretty quickly.

‍

Chris:

Yeah, no, I think that's incredible. I mean, Facebook's and the Googles and the like, they were basically the impetus for creating the term viral within tech. And yet, what took them years took CharGPT three months, like you said, 100 million users. And, gosh, I think it's been quite a ride. They say innovation is slow, and then all it wants. And I think that absolutely is true in AI and healthcare from recent advances, like deep learning and uses in life sciences, like automated drug discovery, and, of course, a monumental potential in the form of GPTs. I'd say personally, I've explored via chat GPT in the like, from navigating medical literature to managing minor health queries. And ChatGPT has been impressively helpful. And but I would say, beyond our personal anecdotes, this could shake up the entire healthcare startup industry, right?

‍

Bimal:

For sure, I think from let's say, let's look at this startup industry from the patient's point of view, right? For example, right? If I were a patient, what kind of products that I will be using? One product is first and foremost is gathering information. Which one, which one gathering information about particular disease, right? And which one is the most prominent one, Web MD, we go to whatever. For example, I have a, let's say, constipation, and I go to WebMD, and see what constipation, what does what the symptoms are, and try to do some from home remedy as recommended there, right. So that's what we do. But the, but the limiting factor of WebMD is that it's very general, in general sense that everybody looks at the same piece of content. And not only that is the same comprehensive, comprehensive level, it's not about like, even the PHD level sees the same thing, even the fifth grade or sixth is the same thing. That's where the Web MD came in, disrupt in such a way that so I can feed in my information. For example, I did this, I had a blood work done, I had a blood work done. And if I go and Google about it, I have blood work done, but I will not say my specific use case because it will be HIPAA violation. So I do feel my privacy intact. So I said the following thing one, but this is the this is what I just said the following prom. I had a blood work done for x, here's the report. Explain that to me. Right and explain it to me into my personal context. And I gave in my like, of course my age, my whatever the food I have been eating, what habits do I have when I fit into all of those things? ChatGPT  gave a really, really great answer, which is very personal to me. And again, it helped me in such a way that it humanizes the technical languages of medicine. So what that means to me was when I went to went to talk to my doctor, I was aware what it meant. And of course, when you when you go and talk to the doctor, they don't have enough time to explain to you that they only had eight minutes in a professional video and I think because of that ChatGPT explained to me what each of the parameters are what I see improvement decreasing, I could ask very specific questions to my doctor, right and this is what the WebMD cannot cannot do from the patient. Patient patient perspective. Now imagine this this is from the patient perspective gathering information. Now, if you think about so when I say in the healthcare space, right three types of people want who are achievers who want to, you know, like what athletes want to achieve things, they need to personalize things. The second category is people who want who are living what healthy life this is a lifers. So even for them, like they need to have a holistic picture of it, and thought about the categories of fighters who are who have chronic diseases. So it has in for them right managing their health conditions, understanding things, and, and even even growing business, they can be of different literacy level. And according to that particular particular literacy level, even healthcare providers can provide you with the patient education according to their level. I think all of those these things can was possible before but cost was immense. Now it can be done at the that 100th of a cost because the fundamental technology is there. Like I said, hammer is just there, and is need to be implemented. So in that regard, in that regard, from the patient perspective, I'm talking about the patient information and education perspective, it can be truly transformative, like WebMD example. And let's talk about this, then they say look for providers sense right diagnostic, and diagnostic. People have been working on like concurrency neural network for many, many years, and they have been really good at diagnosing for diagnostic diseases using like dapper mazes, and extra images and so on. For example, one of my friends work at Vidya health, and they what they use is they use dental images to diagnose all this, they're working on that point.

Chris:

Yeah, zebra medical vision AI doc, yeah.

‍

Bimal:

And all of those companies right. So, it has been there. Now, the interesting thing is that the limiting factor for them for those is those are not multimodal, you know, they can only take images, they cannot take structured data, they cannot take unstructured data, imagine like structured data, unstructured data, construct data means Congress in that the average a doctor shop today that means all the like all the all the this variable data, all sleep data, all the all health data that can be collected, and by comparison that like all X ray or DICOM images that can be done it created. And, and, and all using all this information, more holistic picture of patient can be painted. And and doctors can diagnose, diagnose it really well. And it scales. So So those are the those are these two examples right at the next would be the administrative efficiency about payers are talking about right now you know how much waste that happens, even in efficiency, doctors, taking notes, transcribing and all those things is a lot of work. And then all of those can be automated and augmented by AI. So therefore, from a Christian perspective, it's amazing the transmitted and no more people saying this a platinum served.

‍

Chris:

No, yeah, absolutely. Yeah. In fact, on the LeapFrog side of things, what we've been using chat GPT for is summarizing 360 degree reviews, performance reviews, where people get multiple assessments from their peers and their managers and their customers, etc. And GPT summarizes, what are the strengths? What are the weaknesses, where do you go from here? And yeah, it's incredible, even on an untrained data set, what it can do, and get you mentioned, a lot of great use cases. And I think what you nailed is that the and I think we should circle back to this, this paradigm shift of taking in context, and truly personalizing and humanizing responses even with webMD, the user has to synthesize it themselves, you know, without something that's personalized or takes an input from them. And then you look at big recent industries, let's say since COVID, like telemedicine, we've seen companies like TellaDoc and Doctor On Demand flourish. Now imagine integrating something like ChatGPT, it could provide real time information, assist doctors with potential treatment options or break down complex medical terms for patients in context. While they're curious about their treatments, and I'm sure there's more mental health, maybe you could talk about that.

‍

Bimal:

For sure, I think mental health such as what this big information such as the EU in the consumer health space, right, that's about let's say like, even comm app or the headspace app, or ginger or any I think Live Health is one which is more b2b. All of those, like more policy information about what what human preferences are, for example, like, people talk a lot about, let's say meditation for ranges and for for like, mental health. And, you know, learning meditation is extremely hard you and I know right I need both practice and meditate and is meditation is really hard. But sometimes like things, other things such as hypnotherapy, or like yoga nidra, and breathing, breathing exercises are those could work. And in that scenario, I think what is really helpful is based on the feedback, I think that kind of personalized, let's say, play, this can be made, and they can be, you know, people can start inputting those things, right, longer period of time. . So it's because more long period of time and then feedback is inbuilt. So that what works for them can be reinforced, and what doesn't work cannot be informed and, and can really personalize, I think it's very transformative. 

‍

Chris:

Yeah, that personalization is not just at the long term, playlist level, it can be at the immediate level, for example, remote patient monitoring, or physiological monitoring. And you think about companies like Withings and bio forests, ChatGPT could really enhance these platforms by providing real time personalized advice based on the data they've collected, basically, right now.

‍

Bimal:

And also, it's, I think, we were working with one of the startups and I won't name it, but one of the startups from that star was trying to solve was comprehensive, right? When you even talk to your coaches, you want to when you come to say, see a psychologist, when you talk to your therapist, right? One of the huge challenges is the language, right? So me not having the right words at the right time to express myself and elucidate my problem. And, and what it was trying to do was augment that, and standardize the conversation between the patient and the lesson therapist or provider, and therefore reducing the MIS communication, right. And that's where this scalable form of this language model language model is extremely useful in such a way previously, it's not right, it's not scalable at all. Now, it's like one there is an assistant who is going to translate and transcribe and standardize all the conversation that happened between patient and the provider. And that's going to be transformative and delivery of care, be it mental or be other or not. 

‍

Chris:

Yeah, and so I think the scope is indeed massive. And as these MLMs mature, the potential for startups and even existing healthcare players to leverage AI is only going to increase so it's really an exciting time to be in this field.

‍

Bimal:

So I think we should also exciting in such a way that I was watching this, Google IO and devil is made palm two wishes, the train it for in, in the context of the medicine. And I think the the med pump two can pass US ml, mle. at the expert level, I think 87% or something like that. The average is around 60 People get 60%. So okay, so it's already on better than most, you know, t test takers. So in that scenario, we were actually talking with someone who has a vast scope of knowledge. And what another technical thing about last language model is, did not forget, that's what it is attendance, all detail is that they do not forget, and they can synthesize multiple levels of it. Right. And, and that's amazing. When you're talking to someone, or their doctor is using it, you're actually having assistance of somebody who has the app, let's say, access to a portal, which has who's knows everything and anything. Right. And that can be you know, at the, its disposal to make a decision while while we're giving care providing care to the patients.

‍

Chris:

Absolutely. Yeah, they can't forget. And we all know that. The customer support AUTOMATED VOICE respondent version of forgetting and you know, saying which one to do next, or dialing in, and oftentimes you end up back at square zero. But yeah, I think conversational AI has come a long way. And maybe we should review some of those evolutions. And let's try to keep it as interesting. But take a step back and examine the timeline of AI.

‍

Bimal:

I think your your encyclopedia in that.

‍

Chris:

I'm a student. Yeah, not an encyclopedia. But I'll regurgitate what I've learned.

‍

Bimal:

Let's go, go through a briefing of conversational AI.

‍

Chris:

Yeah, I mean, the first generation of conversation AI was like, let's say the early stages of a child learning to speak. And this is where we let's say first encountered things like Siri, which debatably has not come that far, or had those frustrating chats with rule based conversation trees, driven chatbots. And they were capable of handling specific commands, but struggled with context.

‍

Bimal:

Right. So even now, when you reuse, say, I use Google assistants and hey, Google, do this and do that. And if I speak in a string of languages is cause it's because Bruce activism standard thing.

‍

Chris:

Exactly. And they did their best, but they were quite limited. And still, they paved the way for the next generation. And as we moved into kind of a iteration generation to machine learning came into play. And the AI systems could learn from data and improve over time. And they were more accurate in terms of assessing say, a patient's intent, but still lacked context and struggled with complex interactions. And then generation three, built upon generation two, start integrating a lot of data. So integrating deep learning neural networks, and this allowed them to really better understand language and recognize patterns. But they still struggled with providing consistently accurate and contextually relevant responses or generating on canned responses that made any sense in context.

‍

Bimal:

Yeah, I think one of the drawbacks you will have the third generation was they use somebody called Rico recurrent neural network. To give an example a translation. So if you're translating from language to language, be what it did was say my name is Bimal. If you want to translate that to like Nepali, Mero Naam Bimal Ho is like what it did was, so it trusted my internet believers and more. And named into another word just did Word and Word instead of taking the entire in a sentence to sentence translation, right? And that's where the it breaks down. Sometimes like writing in different languages, subject, verb object, sometimes in other languages object comes first. And that's where the entire the, it reads out there, because it doesn't have that memory off. Like, what did what was the context of it?

‍

Chris:

The memory is fascinating. And I think that's where the story gets really exciting, is with this fourth generation like ChatGPT, that we're seeing a real game changer. But with generation four, we're witnessing this tipping point, in conversational user experience. They can understand context, generate relevant responses, creating a more natural and conversational and inviting user experience. And this is a massive leap forward from that toddler stage we started at and it's opening up the world of possibilities in healthcare and beyond.

‍

Bimal:

Awesome. That's a great summary of four generations.

‍

Chris:

Well, thank you. Now. So that's where we're at. I think Bimal, it's time to put on our futurist hats. So let's imagine we've entered this utopian scenario with ChatGPT, and it's completely transformed healthcare. Can you give me a hypothetical snapshot of that future?

‍

Bimal:

Sure, I think, as I said before, so if you look at from a patient's perspective, the patient education, the die, you know, state, let the administrative efficiency and the pay in the payer side of it, right, all of those things. So what I'm actually let's say not interested me, many people have predicted this, like, the biggest company is going to be a consumer healthcare company. When you say consumer is end user is the consumer, show a few examples, like the grid, Article restricted by a16z, called the title called the biggest company is going to be the healthcare company. If you're familiar, if you talk about like, for example, if there is a $4 trillion spent in the US alone, and the US alone profits, think about like all these big companies such as Apple, Google, Microsoft, and they are interlinked to us and healthcare space because they want to add the next trillion dollar valuation into their, into their, in their companies, right. So my prediction is that there is going to be a full stack for an offense. Full Stack company, for example. What's the full stack company in another physical space is full stack company is Tesla. So if you think about it, Tesla is a car company by You can? Is it a battery company? Is it a car dealership company? Is it a software company? Or is it a car manufacturing company is all it's all of those, right? That's the reason why there's a race to uh, and it's possible because technology has reached to a certain point, like many things can be automated and automated manufacturing can be automated. And you know, there are a lot of batteries can automate it, the software can automate because of that it made it possible. I'm not discounting it is hard. But if you want to do that, this 30 years, let's say 30 years ago, pre Internet era, it would have been impossible to do so. Right. So therefore. So my prediction in the consumer interest sense is that because most of the most of the patient information, the diagnostic links and decision support of the industry efficiency and all those automated What 50 People used to do one isn't can do it right. Therefore there is going to be Ai first healthcare company that's going to do differently than the what the the incumbents are doing. And therefore what is going to is boss is capital of crane is vertically integrated.

‍

Chris:

No, absolutely. And you nailed one thing that is the hard part of a full stack company to scale, which is the people. And what I would challenge is can we think of ChatGPT as your personal healthcare assistant, and your your 24/7 virtual Doctor, what do you think about that?

‍

Bimal:

I think it's truly possible because there's, there's a core corollary to this 30 years ago, if you people ask right? You are going to walk with a supercomputer in your pocket, supercomputer in your pocket, if in 30 years ago, people said that they will be very crazy. Right. So now today is fortunately, 4 billion people have supercomputer in your pocket. Similarly, similarly, like, every human being is going to have a super doctor in their pocket. Because it's super doctoring. Like, because the transplant, we made the transformative thing about the mobile, even a PC was graphical user interface, right? Because of that, you don't have to have to write a command, there's a GUI change everything, therefore, it was accessible to common people. And that's the reason why and that, again, exponentially expanded and that category and releasing the smartphone era era. Similarly, in the super, super doctor in your pocket is because how do people engage in a physical environment through conversation, and the same user experience is being delivered through a technology? Right? Therefore, I think the conversational UX is that transformative, transformative bit of technology, who is going to empower you to have a super doctor in your pocket?

‍

Chris:

Yeah, let's talk more about doctors. Because we've talked a lot about this utopian potential for patients. What about doctors? What does this potential Utopia look like for them?

‍

Bimal:

So for them, I think what we hear mostly is from the doctors is that they hate they became doctors to treat patients not being swamped into administrative documentations, right? I think the lowest hanging fruit can be lowest hanging fruit can be this documentation, right? This augmenting them document this one. And of course, on our lowest hanging fruit is the clinical decision support with the various sorts of data and saying, okay, giving the right piece of information about our patient at the right time making it ready, right there, reduce their mistakes, right. So all of those, I think that's low, lowest hanging field. And another thing is we've flipped the business models completely. So we will complete this such as right now this because fee for service business models is because of that people have to document a lot, right? What if it can be transformed into a let's say, subscription based model in which like, documentation is nostrud. For example, today, if you go to any Sass companies, does any Sass company need to document what the user is doing? Not that is practice embedded into the product, the tracking is embedded in the product, right? It's just the is happening in the it happens in the background, there was similar kind of thing can can enable English the doctors can focus on focus in a, a, I think focusing on the patient and also more and more, they can focus on the they can give more time to the exceptional cases, whereas the majority of the cases can be You know, done by that? 

‍

Chris:

Absolutely, yeah an absolute reason, I think, I think a16z was painting a similar future, maybe a year ago, they had this canonical podcast on health care, and completely agree. And the waste that you're talking about is a real, I think there's the 2016 study by the Annals of Internal Medicine that for every doctor, every hour a doctor spends with the patient, they spent two hours on paperwork, and you're talking about this huge opportunity to improve the efficiency of the health care system. So we talked about efficiency, we're talking about experience. But the big thing is access, like how would all of this help democratize health care.

‍

Bimal:

So all of these health care is again, so when you are moving from medicine, 2.0 to 3.0, which is 2.0 is about treating disease, whereas 3.0 is about focusing life healthspan, focusing on preventing disease, and so on, so forth, right, because of access of data, more patient we are hoping is more patient are aware of your and they can make it take care of their own health, so they be grind less sick in the first place. And in that, that I think there's personal access to health information, which is much more holistic is is a huge enabler for patients to build their health and take charge of their own health, right, that's a huge Democrat isn't in itself. And second thing is democratic in the sense that it's cheaper and faster to build, say, AI diagnostic tools, right? Clinician support tools, those are going to enable the the applications are going to immerse much faster than it used to be and the developing developing those applications will be much cheaper. That's going to democratize it. One is direct impact and other is enablement.

‍

Chris:

Right. Yeah. And absolutely agree about that information aspect. And right, I will, we talked about the fee for service is sick care, right? We only care for people when they're sick, instead of preventing them from getting sick or keeping people healthy. And I think a lot of that barrier for many people, myself included, is just social and psychological. It's hard for me to talk proactively to someone about my health. And that could be an uncomfortable subject for me, but it's in some ways, it's easier to talk to a bot. Right? If I'm curious about how to take care of myself, I'm curious about symptoms. No, I'm not gonna go to a clinic and wait for an hour and talk to a stranger and open up. And I think that does lead to people coming in only when they're sick only when they have clearly chronic conditions. So I think the potential reward.

‍

Bimal:

Not only that, but also see like you when you have a mental health problem, or when you have a sick disease, right? One of the reasons why you don't go visit providers as a provider or doctor is that it's costly, it's expensive, right? For example, you want to expensive, therefore you wait till the end, when you till, you really have to go I think what democratic thing is that I have this problem right now. I have an expert, the Super Doctor who I can talk to by paying a few bucks. Right, relatively 1000s of dollars or hundreds of dollars with Machar. Yeah, I think that itself is going to Kenny back at us, you know, she used in healthcare.

‍

Chris:

Yeah, Bimal. So there's two more issues I want to cover and maybe a few minutes each. And we've talked about the utopian scenarios, but there's this double edged sword where we've been talking about the sunny side of the street and AI driven healthcare utopia. But let's take a detour down the dark alley, and we're talking about potentially Black Mirror levels of dystopia here. What are some of the risks and concerns that would keep you up at night?

‍

Bimal:

For me, I think one of the things which has already again, I love to give the correlates, right. So let's talk about internet. Like when you talk about internet, it was all open Internet, everybody had access to Democrats that what really happened when I talked about 4 billion people accidentally trading before 4 billion people having a supercomputer in there. So what's the portal into the into the wall is this Facebook, TikTok, YouTube, YouTube and let's a few other Microsoft products. So four or five handful of companies monopolizing. That's the internet for them, right? I think in the same in a similar vein. What worries me most is is when these technologies enables happened when the AI right now isn't, isn't have not been become right that tool is prolific, but when AI isn't become prolific and even close AI become AI loops become prolific. So in that scenario of small company, a handful a company can have a monopoly over healthcare delivery system. And then and human beings are leaving as a subject to those few handful of companies. Right. And that's scary. Terrifying, right? That's a black mirror.

‍

Chris:

Yeah, no, that's a black mirror. And then, you know, I mean, I think another one is, you know, there's this haunting question of accuracy, right? So imagine a patient gets a diagnosis from an AI system, they follow the recommended treatment, but it turns out the diagnosis was, was wrong. And then I mean, the internet is full of instances where ChatGPT is wildly assertive, and its opinions but completely factually incorrect. And the consequences could be catastrophic or even fatal. Even today, when you ask a question to chat GPT is it says, prefaces, I'm not a certified health professional and do not take this as medical advice. I am not a lawyer type thing. But the challenge is, a lot of people are going to ignore that. And, you know, I think we've been paint a picture of this black mirror episode now. That's a chilling thought. It raises a big question, which is if something does go wrong, who's on the fault?

‍

Bimal:

You're right, I think that's exactly the reason why ChatGPT, really open AI objectivity has that disclaimer that I am not a healthcare professional. So that's the disclaimer in watch. Free fees them from the all the, you know, repercussions that come from that. Having said that right thing. I think that's a technological challenge. What that means is the corpus of data that ChatGPT is trained on is an open Internet, right after just strain on the healthcare specific data. And, and again, the the model is tuned, after this reinforcement learning to human feedback and the human feedback is more doctors more to ends, right one is the corpus of data is get trained, or is the medical literature. That's one and second is the model is tune to the human feedback. And the human feedback is given by the more more doctors and the accuracy will be much more higher. Plus, I think another layer is the human in the loop, right? When somebody is happening that can ultimately will flag on somebody else needing it and the multiple AI? Isn't cross verifying each other further, this is what I do. I told you before I haven't used they're called God's pair, and I'll be curating content about gut microbiome. And this is exactly what I do. So I know what is right and what is wrong. But for me to scale faster, I used to memorials for example, I use ChatGPT for to create a content. And sometimes I need to fact check that what I do is I copy paste and go into a bar and say fact check each other sentences. Right. So yeah, so when I'm mixing to the model.

‍

Chris:

Right, right, but you're you're you're like uniquely positioned to know that you need to do that. Right, exactly. Yeah. So I think there is this delicate balancing act, and we need to push for democratization of healthcare, making it more accessible and affordable and efficient through AI. But at the same time, we need to guard against those risks you were talking about with monopolization, the potential for AI mistakes, and you know, of course, we we've hardly touched on privacy of data and who owns what and you know, all of the light that comes with compliance. But yeah, go ahead.

‍

Bimal:

We are still so early that this innovation is moving so fast that regulations are far behind for example, right even the regulations such as when you talk about in the earlier we talked about it can be multimodal multimodal sensor input of data, and also in the output of let's insurances, right? One model can perhaps like it is it is for like for cancer or not to it is there limp or not? It doesn't have all of those things that can multiply both in training and both in the inference. Whereas, whereas the current in FDA and all like it's very, okay, very specifically the one use case, right? Does this model does x y x or not? Right. So the FDA approval process is that so what happens to those the FDA approval process when the even input is multimodal? The inference is multiple model, right? We don't know. Right?

‍

Chris:

Right. Yeah, yeah, yeah. And then there's this the opera Ready to operate Jason analyzation production realisation of these things like is ChatGPT ever going to be HIPAA compliant? Are they like with non health related conversations going to use private and regular date regulated data to train their models? And is there an ethical and legal quagmire there that we're going to have to cross and this whole data ownership issue?

‍

Bimal:

So I look at from two perspective, right, from a technical perspective and the human doctor perspective, from a technical perspective, what I think about this platform about open AI or Google is that they are more at the platform level, which is more at the infrastructure level. So on top of that, there will be application layers, such as, like, simply whose application layers, right, for example, like there will be let's say, I'm making this up. Health, HealthGPT, right which will be like for HIPAA compliant, HIPAA, HIPAA compliant, and, and for example, data is stored in separate, separate, separate and they're more anonymized happening all of those things like GPT compliant based on the any any compliance regulation, I am guessing that's going to happen. That's the reason why even open chargeability wants to prove to the world that this is possible, therefore, they have gotten the chance to be part of it. But the healthcare specific, I think there'll be application layers on top of the platform, and it will design about the HIPAA and all those things. And there will be multiple, I hope there are multiple players on that part there is there is competition, all that parcel that there is there is no monopoly on that part, right on that part. And still, like we get the benefit of the Democratic progression of it. But from more for me, that's more of a technical challenges. And the rate is the rate, the money that's flowing, the capital that flowing into the human capital that's going to build those things they'll extract. But more importantly, what is really challenging part is the human adoption, right? For example, adoption part. Technology, easy, right? For just scale bits is a scale, right with atoms versus physical goods is really the hardest skill, because we're manufacturing laws takes transport and all of those things, but human emotions are much more complex, complex complex in that regard. So in this healthcare space, right, so we can give you a brief example of April 7, right with operability. It came out in I think, is and the FBI came out in like, late 2000s, 2009 the end. And still we're talking about interoperability, and there are huge challenges. And why is that happening is because like, if to make a date to make a decision in the health system, there are multiple stakeholders, right? There's doctors that editors drive that nursery that patients right, and they already have this system, there are a lot of things to worry about. So the foot further reason, I think it's going to take a little harder, longer to longer to

‍

Chris:

Yeah, healthcare is always slow. And then you now the human factor is there, because AI is competing with that human factor, not just IT. So I think Bimal it's clear that the journey towards AI driven healthcare is complex. I think we both believe it's a journey worth taking, because the potential benefits are huge. And we're excited to see where it leads.

‍

Bimal:

For sure, man, I think I'm very excited about this thing. My hands like I'm excited about like intersection of this health, as well as an AI and building paradox. And again, that and we have the opportunity and the platform to make more holistic and more patient centric. I think that's the reason why I concur with the likes of David Sachs. And every reason it's a platform series like the internet is because first even for a small because we know from that right even for building a personalized system, building a recommender system when we built kaboom. Right? It was a lot of heavy lifting. Now, you now you can go simple, you can do it for yourself, I give three movies that you like, hey, and prompt like this, Hey, ChatGPT, hey, like, I like these three movies. Can you give me recommend another movie? And most likely, like we figured that prom category will give you a very accurate prediction for you. Right? And that's absolutely operationalize. And that's very easy to do. And hopefully this thing is transferable to healthcare space as well.

‍

Chris:

I think we're going to find out because of that money in startups and everything that's behind this. And I think we're going to need to revisit the state of learning and trial and error just as consistently as this evolves. Bimal, I look forward to our next podcast, where we explore maybe the current state, maybe share some of our own health related LLM conversations we've been alluding to, and maybe invite one of the startups working with the ChatGPTs of the world to share their progress. We want to thank you for your time and insights.

‍

Bimal:

My pleasure, I'll see you next time.

‍

Chris:

See you next time, thank you.

‍