#1839 AI and Diabetes Care with Dr. Sarah Gebauer

Summary

Juicebox Podcast #1839: AI and Diabetes Care with Dr. Sarah Gebauer

Date: May 1, 2026
Host: Scott Benner
Guest: Dr. Sarah Gebauer, anesthesiologist, AI healthcare researcher, and parent of a child with type 1 diabetes

---

Episode Overview

This episode dives deep into the intersection of artificial intelligence (AI) and diabetes management, bringing expert insights from Dr. Sarah Gebauer. As both a practicing anesthesiologist and the parent of a type 1 diabetic child, Dr. Gebauer brings unique expertise—blending hands-on family experience, international travel, medical practice, and a professional focus in AI models for healthcare and national security. The discussion covers the current uses of AI in diabetes tech, the promise of coming innovations like digital twins, barriers in regulation, patient empowerment, and how everyday people can harness AI for diabetes and beyond.

---

Key Discussion Points & Insights

1. Sarah Gebauer’s Background & Book on Traveling with Kids and Diabetes

• [02:49–05:35] Dr. Gebauer introduces herself: anesthesiologist, AI researcher, parent to a T1D child, and globe-trotting family (over 50 countries in 4 years).
• She’s authored a self-published book on long-term travel with children (including a section on diabetes), aiming to empower families to travel and manage diabetes confidently.

“You're a parent. You do complicated things all the time. This is something you can figure out and help people not be left with that kind of sense of, 'Oh, shoulda, woulda, coulda.’”
— Dr. Sarah Gebauer [04:30]

---

2. How Dr. Gebauer Became an AI Expert

• [06:08–08:05] During extended travel, Sarah self-taught herself AI (Stanford/MIT materials), leading to research at RAND on AI risk in biosecurity, and then a shift to AI evaluation in healthcare.

“I taught myself all about it…I watched videos…read computer science textbooks…then started writing a substack on healthcare AI…”
— Dr. Sarah Gebauer [06:25]

---

3. Making AI Understandable: What It Is and Why People React

• [09:45–12:35]
  • AI is already ubiquitous and used in diabetes tech for years.
  • AI = Big circle; machine learning (ML) is a smaller circle inside.
  • Neural networks are the newer tech driving today’s AI leaps.
  • Examples: AI enables protein folding in minutes (vs. months); meanwhile, chatbots can still fumble simple tasks (“Can’t book a flight!”).

“AI is great at some things and terrible at other things…”
— Dr. Gebauer [09:45]

---

4. Agents: The Future of AI-Driven Support

• [15:29–17:12]
  • Definition of agent: "A brain" that accesses different tools (internet, PowerPoint instruction, booking abilities) and decides how to combine them for a task.
  • Agents underpin much of the emerging AI landscape and will become central to how we interact with tech—including healthcare.

“An agent is basically a brain…that has access to different tools…it can decide for any given question or task which of these tools should I use…”
— Dr. Gebauer [16:21]

---

5. Practical Use of AI for Everyday Tasks & in the Podcast

• [17:11–25:37]
  • Scott shares how agents streamline laborious tasks (e.g., scraping and collating Facebook responses, managing documents, website upgrades) and amplify solo productivity.
  • Discussion of technology delivering on childhood promises (“this is what was promised to me when I was a kid” – Benner), and how AI may change job markets and democratize what an individual can accomplish.

“All the things that I accomplish in the course of a day are weeks worth of work…”
— Scott Benner [19:05]

---

6. Where AI Is Used in Diabetes — Current State

• [25:37–27:06, 42:04–45:25]
  • Closed looping, predictive technology, pump algorithms: widespread, but usually “old-school” machine learning, not “next-gen” neural networks.
  • Next wave will see:
    • Increased personalization, e.g., exercise prediction, better dosing.
    • Digital twins: virtual versions of yourself constructed from your data—testing treatments on your digital twin before trying them for real.

Notable Explanation: Digital Twins

“It’s a digital version of all the data we have about you…so you can try stuff on the digital twin before you try stuff on the real person.”
— Dr. Gebauer [27:09]

---

7. Barriers to Progress: FDA, Regulation, and Personalization Limits

• [29:30–34:41, 42:56–45:25]
  • Regulatory frameworks (FDA) are not set up for continually self-learning, personalized devices—hard to approve because the device “never stays the same.”
  • Recent FDA relaxations on advice from chatbots (e.g., not enforcing against healthcare info) may open new possibilities.
• [43:45–45:25]: Explanation of the regulatory tension between “static” devices for many and “dynamic” devices for each individual.

“It’s a real struggle to move from devices that are meant to work the same in…millions of patients, to devices that are meant to work differently in every single person.”
— Dr. Gebauer [43:45]

---

8. Democratization and Empowerment — From DIY to Patient-Powered Innovation

• [37:15–39:35, 46:08–47:54]
  • New AI tools allow even non-coders to build practical apps.
  • Loop, Android APS, other open-source platforms as precedents: “Is there going to be another generation of those people…or are some of those people going to have their thoughts reignited?”
  • AI = “Pattern recognition”; ideal for algorithmic diabetes tech (forecasting glucose, adjusting insulin, etc.)
  • The real next leap: Truly adaptive tools that respond to the individual's needs in real time.

---

9. Interactive, Personalized Diabetes Tools — The Near Future

• [34:41–37:18, 54:30–66:08]
• Discussion of calculators, “estimators,” and agents that can:
  • Let users see potential starting settings (carb ratio, basal, sensitivity) by entering their weight.
  • Tools for meal bolusing personalized to the user—“I think I could build that app right now,” says Benner [37:19].
  • Dreams for an app that combines community, daily affirmations, and all-in-one diabetes support: “I would just like an app that you open up…that says something…valuable or supportive about diabetes…”

“At some point they might have their own agent who knows them and knows their personality…might even be able to just engage on its own with your tool and then bring that information back to them without them having to search it out.”
— Dr. Gebauer [35:53]

---

10. AI as a Productivity Supercharger, Especially for Advocacy

• [53:54–63:38]
  • AI empowers creation of content, advocacy, communications (writing posts, structuring arguments, prepping for community engagement) that would normally be impossible to fit into a day.
  • AI used not just for management, but shaping community messages and support.

“It really is about creating those opportunities to do something where nothing would have been done that are the biggest yield.”
— Dr. Gebauer [59:24]

---

11. Limits and Real-World Usage: Encouragement for Listeners

• [52:33–54:30]
  • Even if coding and “code blocks” feel intimidating, just using AI tools regularly will yield intuition and empower users.
  • Most leading LLM providers offer classes for upskilling—experimenting is key.

“My experience has been just using them on a regular basis for things you actually need to do…by far the best kind of learning.”
— Dr. Gebauer [53:54]

---

12. Why AI Sometimes Fails — And Why That’s Not a Stop Sign

• [47:54–50:15, 69:16–72:13]
  • AI is probabilistic, not deterministic (“it doesn’t always give the same answer given the same information”).
  • This makes evaluation and regulation hard—but shouldn’t stop innovation or foundational progress.
  • AI is improving at unprecedented speed; listeners should “be excited that it’s moving forward so quickly.”

---

13. Broader Implications: The Future of Medical AI and Advice

• [72:18–74:16]
  • Apply self-driving car progress to medical research: AI could soon monitor labwork and suggest medication tweaks that even doctors may not achieve.
  • Call to action: “Pick one thing you’re great at and try to make it better”—everyone can participate.

---

Notable Quotes & Memorable Moments

• On Travel and Diabetes:
  “There is a window that you can do…these kinds of experiences and adventures with your kids more easily.”
  — Dr. Gebauer [04:30]

• On Public Fears About AI:
  “I think everyone thinks the Terminator is going to come and step on your skull…All I can say is maybe. But if we can get through this and make it work for people, I think it’s going to be magical.”
  — Scott Benner [18:25]

• On AI’s Real Power:
  “Computers were supposed to speed us up…and a lot of times we ended up adapting to the computers instead of the computers truly adapting to us.”
  — Dr. Gebauer [21:09]

• On Individualized Diabetes Tech:
  “If we can have a digital twin…the thought is you can try stuff on the digital twin before you try stuff on the real person…”
  — Dr. Gebauer [27:09]

• On AI’s Shortcomings:
  “Until very recently, chatbots were not able to count the number of Rs in strawberry…”
  — Dr. Gebauer [09:45]

• On Enabling Non-Coders:
  “Now coding is conversational…most people, even coders, give the agent context in mostly natural language…”
  — Dr. Gebauer [40:07]

• On App and Tool Creation:
  “I think I could build that app.”
  — Scott Benner [37:19]

• On Productivity Supercharge:
  “It really is about creating those opportunities to do something where nothing would have been done that are the biggest yield…”
  — Dr. Gebauer [59:24]

• On Agency:
  “You also have agency and you could read it and decide if what it told you makes any sense or not.”
  — Scott Benner [64:36]

• On Regulatory Hurdles:
  “AI is probabilistic and not deterministic…how do you trust that? That’s one of the main things that regulatory bodies have really struggled with.”
  — Dr. Gebauer [47:54]

• On Patient Empowerment:
  “Pick one thing you’re great at and try to make it better.”
  — Scott Benner [73:19]

---

Timely Highlights

| Timestamp | Topic | |---|---| | 02:49 | Sarah introduces herself & book on travel with diabetes | | 06:25 | Sarah’s self-taught journey in AI and application to healthcare | | 09:45 | AI vs ML explained simply and its history in diabetes tech | | 15:29 | Definition and significance of AI agents | | 17:11 | Real-world use: Facebook scraping/collation and automation | | 25:37 | How AI already powers closed-loop systems and prediction | | 27:09 | Digital twins: definition, purpose, and their future potential | | 42:04 | Challenges with FDA approval and why AI tools in diabetes remain slow to change | | 43:45 | Regulatory struggles: “every device different for every person” | | 47:54 | The probabilistic nature of AI and regulatory evaluation | | 53:54 | Encouragement for users: “play around and use for things you need” | | 64:36 | Agency—people’s power to assess and use AI outcomes | | 69:16 | Acceleration of AI’s evolution—why speed is a good thing for patients | | 73:19 | Practical call-to-action—everyone can help improve a “piece” of the world |

---

Concluding Thoughts

The episode delivers an energetic, candid, and pragmatic look at how AI already shapes—and will increasingly transform—diabetes management for patients and families. From practical, actionable tips (how to use AI now for your management or advocacy), to visionary futures (digital twins, real-time personalization), Scott and Sarah urge listeners to be boldly curious and proactive.
They demystify both the technical jargon and fears; this is not about robots taking over, but about making life easier and better for people with diabetes through tools more accessible than ever before.

Sarah’s final encouragement:

“…this is something now that has gotten to the point that normal people without any coding experience can use these tools and create really cool things.”
— Dr. Gebauer [53:54]

Scott’s closing message:

“Pick one thing you’re great at and try to make it better.”
— Scott Benner [73:19]

---

A highly recommended listen for anyone interested in diabetes, AI, or patient empowerment.

Transcript

[A]

Here we are back together again, friends, for another episode of the Juice Box Podcast.

[B]

Hi, I'm Sarah Jabauer. I'm an anesthesiologist and also the mom of a type 1 diabetic kid. I get to do all kinds of cool stuff with AI and I'm thrilled to be back here today talking about AI stuff, which is what I do when I'm not in the operating room.

[A]

If this is your first time listening to the Juice Box Podcast and you'd like to hear more, download Apple Podcasts or Spotify. Really any audio app at all. Look for the Juice Box Podcast and follow or subscribe. We put out new content every day that you'll enjoy. Want to learn more about your diabetes management? Go to juicebox podcast.com up in the menu and look for Bold Beginnings, the Diabetes Pro Tip series and much more. This podcast is full of collections and series of information that will help you to live better with insulin. If you're looking for community around type 1 diabetes, check out the Juice Box Podcast. Private Facebook Group juice box podcast type 1 diabetes but everybody is welcome. Type 1 type 2 gestational loved ones it doesn't matter to me. If you're impacted by diabetes and you're looking for support, comfort or community, check out Juice Box podcast type 1 diabetes on Facebook. Nothing you hear on the Juice Box Podcast should be considered advice, medical or otherwise. Always consult a physician before making any changes to your healthcare plan or becoming bold with insulin. This episode of the Juice Box Podcast is sponsored by the Contour Next Gen Blood Glucose Meter. Learn more and get started today@contour next.com Juicebox Today's episode is also sponsored by Cozy Earth. You can use my offer code Juicebox at checkout to save 20% off of your entire order@cozy earth.com everything from the joggers that I'm actually wearing right now to the sheets I sleep on, the towels I use to dry myself with, and whatever else is available@cozy cozyearth.com just use the offer code juicebox at checkout. The podcast is also sponsored today by US MED usmed.com juicebox or call 888-721-1514. US MED is where my daughter gets her diabetes supplies from and you could too use the link or number to get your free benefits. Check and get started today with usmed.

[B]

Hi, I'm Sarah Jabauer. I'm an anesthesiologist and also the mom of a type 1 diabetic kid. I get to do all kinds of cool stuff with AI and I've worked in hospitals, clinics, all kinds of settings professionally, and then also gotten to interact with the healthcare system. We've traveled as a family to more than 50 countries now around the world over the last four years. And I'm thrilled to be back here today talking about AI stuff, which is what I do when I'm not in the operating room.

[A]

You and I recorded together already. I'm trying to decide if your episode came out or not yet.

[B]

I don't know if it did, actually. Well, exciting news with that also is that I have written a book on traveling long term with kids, and there is. There are some sections in there on traveling with diabetes specifically. So, yeah, I just thought it would be great to help empower some of the families that I meet to. To travel long term. You know, as I mentioned, it's. We say that it's for our kids, but it's really for my husband and me because we love forcing them to spend time with us. And I talk to a lot of other parents who are interested in doing something similar, but it just seems too huge and unmanageable. So the book is really trying to break that down and to help people feel like it's something that they can do easily.

[A]

I don't know, Sarah. I feel like you're just on here to make me feel bad. Like you're.

[B]

She.

[A]

She said. I said before, before we started, I said, sir, do you have exactly an hour? Do we have extra time? She says, well, I have to be in surgery later. Then five seconds later, you're like, oh, I wrote a book. You just wrote a book? Why? All right, let's start with that real quickly. Why did you, like, how does that happen? How do you say to yourself, I'm gonna write a book, and then you actually accomplish it. Is it published or did it self published or what is it?

[B]

Yeah, it's self published and it'll be coming out formally at the end of this month. So we'll do a launch then and I'll let you know when that happens. But, you know, we talk to a lot of people who, who have kids and always say, oh, I would love to do something like that, but we just never quite got around to it. And what really stuck with me was one. One surgeon that I talked to said, you know, we always said, oh, that would be so great, we should do that. And then we just never did it. And now my kids are too old, they're in college and never going to get the chance. And I just wanted to help People feel like it doesn't have to be as complicated as it might seem. This is something you can do. It's totally manageable. You're a parent. You do complicated things all the time. This is something you can figure out and help people not be left with that kind of sense of, oh, shoulda, woulda, coulda. And then at some point, it is, you know, your kids are old, they have their own lives, their own things that are happening, and there is a window that you can do this kind of travel and these kinds of experiences and adventures with your kids more easily. So.

[A]

And even though Sarah doesn't know it because she's too busy to listen to my silly podcast, her episode is 1,617. It's called 50 Countries with Diabetes.

[B]

Wonderful. I will check it out.

[A]

Yeah, I love that you didn't listen to it.

[B]

All right, I'm sorry. I. I honestly. I usually run in silence. I almost never listen to podcasts these days. I do listen to Buddhist meditation while I run.

[A]

Hey, listen, don't give those people, the people listening, that idea you, you have. Sorry, no. You have to be listening to podcasts when you're doing stuff that you can' in silence. Never silence.

[B]

Right.

[A]

But when. When you and I were talking last time, it kind of came up that you had an understanding about how AI was working. And so why don't you explain to people first, like, how it is you have that understanding, and then we're going to move forward and talk about some things specific to AI and diabetes.

[B]

Yeah. So actually, during one of our big family trips, it was the first time that I hadn't worked full time since I was, you know, been in school since I was, you know, tiny. And I really got interested in AI and just mostly how you can teach a computer how to understand language. I just found that really fascinating. And this was kind of in 2022, before the big leap with ChatGPT and the Neural networks really started. So I taught myself all about it, everything I could. I watched videos because I'm a huge nerd from Stanford and MIT and read computer science textbooks. I already knew how to code from some previous work I'd done, and then just started talking to people about what are they doing and what are they interested in, and then started writing a substack on health care AI and since then have written that steadily formed a group of physicians interested in health care AI and then a few years ago started working at rand, which is a large think tank in the US doing AI model evaluations for national security Risk. So really trying to look at how would we know if some of these frontier models created new kinds of risks for biosecurity and if they do create them, what kind of mitigations? Different kind of mitigations might be needed to help decrease those risks. So. And then transition to doing more of a healthcare focused AI evaluation and governance role in a new company that I started at the beginning of this of last year. So I help hospitals.

[A]

While you're being an anesthesiologist too, right?

[B]

Yes, while I'm prepping and traveling all the time. Yeah, I like to. I like variety, it turns out.

[A]

I think I'm. It's possible I'm the only podcast host who sits holding in a laugh while someone's explaining something that impressive. Because I just want to laugh. I'm like, why? How are you doing all of this? I'm still stunned, but we're gonna get past that part because I want to get to the AI thing. I like how you went from, like, I found it interesting how you could teach a computer to blah, blah, blah, and then you're like, and then I did this and that and started a business and I worked for the government. I'm like, hol, like, you feel like a spy. You're not a spy, right, Sarah?

[B]

I am not a spy. Although sometimes when we traveled to, you know, very far flung pit places, I got a little nervous when, you know, they were, you know, looking through my laptop and such just because, you know, you don't want people to get the wrong impression of what is actually happening.

[A]

Oh, my gosh, I love you. I swear. I think you'd be disgusted with me inside of 35 minutes if we were in the same room together, but I think you're fantastic. So. So explain this to me. You sent me a little list that I'm thrilled to have gotten for our conversation today, and you kind of broke it down to bullets. I want to just follow your bullet points like, so let's lay it out for people and explain to them where it's already being used and how it might be used in the future. I'm going to talk a little bit about how I use it interspersed in inside of the conversation. But I think mainly the general public has, as far as I can tell, either a really, like, kind of harsh reaction to the words, you know, when somebody says artificial intelligence, or they're just like, too Pollyanna about it when they talk about it. I don't really hear anybody talk about it. I think thoughtfully in common conversation is my point. Does that make sense?

[B]

Totally. And I think, I think there are a lot of reasons for that. And one of them is that I don't think the AI community has done a good job of explaining what AI is because we all have been using AI for a long time. AI is in everything now, but for diabetes, for example, it has been for a long time. Anyways, AI is kind of a big circle and within that is a small circle, which machine learning is in there too. Machine learning is old and that's, you know, kind of looking at data and predicting patterns and that kind of technology that is actually encompassed within the, the umbrella of AI. Generally, what is new is neural networks and those most ML to find a pattern, you would kind of say, these are the things that we think AI. These are the things that we think might predict a pattern. So, you know, kind of look in the data, say, okay, this seems more related, this seems less related. This is how we can kind of group these things together, what neural networks can do. And what's really exciting is it can look at a huge, huge amount of data and find patterns and synthesize that information. And it doesn't have to be told this is where the connections might be. It can find those connections on its own. And so, you know, it is superhuman in that way. And I think there's this, you know, there's this tension between, as we know, AI is great at some things and terrible at other things. So AI is already superhuman at doing a lot of things, like finding patterns, as I mentioned. But also, when I was an undergrad, I did chemistry research on protein folding and it would take months and months to figure out how to, how a protein actually folded in real life. And this was a, this was a huge task. And AI now with the help of John Jumper, who won the Nobel Prize a few years ago in chemistry, and the team at Google DeepMind, found a way to figure that out within minutes. So that used to be this huge problem in chemistry and now it's not a problem. Now it's completely defined, it's completely solved. I mean, obviously there are a few outliers, but that kind of ability I think makes people both excited and nervous. But then in other ways, you know, until very recently with AI, couldn't the chatbots were not able to count the number of Rs and strawberry. People might have, might have seen some of, some of those online as well. You know, it just, if you asked it how many Rs Strawberry had in it, it would just give it wrong answer over and over. And over. So, and it's still, you know, to me, it still can't book a flight. So, you know, if you can't book a flight, it's, you know, it's not that superhuman, is it?

[A]

Not going to pivot again though? Because that, listen, I'm going to say a lot of names. I don't, I don't know this guy's name. So there's a guy who's been huge in this space, took a break, came back and then coded a Claude bot or something like, or Clawbot or something. And then didn't chatgpt just hire him? Like, aren't we getting towards agents that work for you? Friends, I just placed my [e-mail] they're today's sponsor and I'm here to tell you about them. Use my offer code juicebox at checkout when you buy and you'll save 20% off of your entire order. That's everything in your [e-mail] save 20% with the offer code juicebox. Now. Why am I excited? Well, I just ordered the cozy Earth blanket. It's the viscose bamboo blanket. I'm super excited about it. It looks comfy as can be and it's going to go so well with the sheets that we already have from Cozy Earth now. Yeah, I'm a bit of a cozy Earth convert, I guess. I'm sitting here in my joggers. I used my towels coming out of the shower this morning. I slept on my sheets last night. Slept like a baby. By the way, cozyearth.com they pretty much have everything you want. Use the offer code juicebox to save 20% at checkout on skin care, women's and men's clothing, bath and sleeping accessories. And don't forget, Valentine's Day is coming up quickly. Get those pajamas. Cozyearth.com use the offer code juicebox at checkout to save 20% off of your entire order. I used to hate ordering my daughter's diabetes supplies. I never had a good experience and it was frustrating. But it hasn't been that way for a while. Actually for about three years now because that's how long we've been using usmed usmed.com juicebox or call 888-721-1514. US Med is the number one distributor for Freestyle Libre systems nationwide. They are the number one specialty distributor for Omnipod Dash, the number one fastest growing tandem distributor nationwide. The number one rated distributor in Dexcom customer satisfaction surveys. They have served over 1 million people with diabetes since 1996. And they always provide 90 days worth of supplies and fast and free shipping. USMED carries everything from insulin pumps and diabetes testing supplies to the latest CGMs like the Libre 3 and Dexcom G7. They accept Medicare nationwide and over 800 private insurers. Find out why USMED has an A rating with a Better business [e-mail] juicebox or just call them at 888-721-1514. Get started right now and you'll be getting your supplies the same way we do.

[B]

Yes, we're getting a lot closer. So the technology is improving so quickly. And that's one of the issues that I think is important to talk about today too is what is out there and what can be done is not the same as what's being done in the healthcare field. Because healthcare is so understandably conservative and risk averse. A lot of what is possible. It takes years to translate that. Just because of the systems we have with the FDA and devices and all those concerns which are there for a very good reason, it is making it hard for them to regulate AI in a meaningful way. As you mentioned, agents are the future. Already most of the AI that you use is an agent, meaning that it has. And do you think people. Should I define the term agent?

[A]

Go ahead. Yeah, please.

[B]

Okay, so an agent is basically a brain. So think of it as a little brain that has access to different tools. And tools might be something like the Internet. The Internet could be a tool. The instructions on how to create a PowerPoint might be a tool. Instructions on how to book a flight in the future might be a tool. So has kind of all these different contexts and tools that it has access to. So it can decide for any given question or task, which of these tools should I use to do that, either alone or together. And then how can I put those together to make a good output? So when people talk about agents, it's often a compilation of different AI tools that are being controlled by a central AI tool.

[A]

Okay.

[B]

Does that make sense?

[A]

It does, actually. I'm. I guess somewhat unironically, I have an agent scraping a Facebook post for me right now. Like, so I put up questions to explain to people one of the ways I use it. I will put up a question that I'm trying to crowdsource how everybody feels about something. I've been doing this for years and years and years. Right. What are your. I have a, an exhaustive list, for example, of like what people's struggles are with type 1 diabetes. Diabetes. We created the entire Grand Round series off of a 90 page document that asked people the question simply, what do you wish someone would have said to you? A diagnosis. What do you wish someone would not have said to you? A diagnosis? And wow. We used to just put up the post and then get all the responses back, and then Isabel would take all of those responses, read through them, collate them, say, oh, this one and this one are the same. She'd kind of put them together. She did that all for me in the background. Now I send an agent to a post, it scrapes it out, and then I have it do that. It takes about like 10 minutes maybe.

[B]

And isn't it amazing?

[A]

Yeah, yeah, no, it's. It's absolutely fantastic. And I'm 54. I don't know how old you are. I'm sorry.

[B]

40s? Yeah.

[A]

Oh, in your 40s. Okay. When I look at computers right now and I look at all this, I go, this is what was promised to me when I was a kid. And when I look up and I see people scared about it, I'm like, all right, I get that everybody thinks the Terminator is going to come and, like, step on your skull and everything. Like, And. And that might. And all I could say to that is, is like, maybe. But we could get there a lot of different ways. If we can get through this and make it work for people. I think it's going to be magical in what it does. Is it going to change the job market? I'm sure it will. Like, I mean, because, listen, I don't really talk about a lot, but this podcast is huge. I run it completely by myself.

[B]

That's amazing.

[A]

I don't have a marketing team, I don't have a writer, I don't. I rob edits the audio. But I mean, like, the rest of it, like. And I used to do that too, by the way. It just. I didn't sleep much, so. And so, like, you know, all the things that I accomplish in the course of a day are weeks worth of work. Or you say, well, you could have hired somebody. But no, I could not have hired somebody. I don't have that kind of money. I couldn't have done that. So it just would not have gotten done. And it's. I don't know, it's just really fantastic. And when people think about it in their diabetes technology. You said something that I meant to get back to. I'm sorry, I'm pivoting. But you were like, health care is risk averse, but there's something specific about it. Right. Like, I forget, I'm. I'm a little messed up here because I don't have all my words that I need. But in health care right now, give people examples of where AI is being used right now in their diabetes technology, then I'll ask my question. I'm sorry, go ahead.

[B]

Well, first I want to say how amazing it is you're able to do all of this on your own. I can't imagine how much work that is.

[A]

It's just every moment I'm awake, that's all.

[B]

And that's all it is, is all your time. And that is. I mean, I hope that as the agents get better and you can offload even more work to them, you know, which I find myself doing. You know, every few months there. There seems to be a meaningful improvement and in what the agents can do, and I find myself offloading more work to them on a regular basis. So I.

[A]

My goal is to have an agent who's thinking about the podcast the way I am right now, and telling other agents what to do.

[B]

Yeah, I think that is actually possible now.

[A]

Yeah, that's. That's where I'd like to be, because I have a plan. I know how I run my day and my week. If something else was like, overseeing that, that would be a big deal. Then I could actually sit down and, like, you know, read my emails. Not once a month or once every two months. I could actually do it, you know, every couple of days and have, like. I could do more human things, I guess, is what I'm talking about.

[B]

Exactly. And I think that's the promise. I think that we are all so used to, you know, the minutiae of using computers. You know, computers were supposed to speed us up, and I think a lot of times we ended up adapting to the computers instead of the computers truly adapting to us.

[A]

Oh, 100%. Computers just cause different busy work.

[B]

Yes. That's all.

[A]

Yeah.

[B]

I mean, how many. I mean, I've created many, many PowerPoint presentations in my life. Moving a text square from one side of a PowerPoint to the other side is just not a meaningful use of my time in any situation. And the fact that now AI can produce beautiful PowerPoints in, you know, 30 seconds that are. Are very nice and actually makes sense. I mean, to me, that's a meaningful improvement.

[A]

Sarah, I recoded my entire website over the weekend.

[B]

Yes.

[A]

I don't know anything about coding.

[B]

Right.

[A]

Yeah. My website is so much better than it was on Friday. I completely changed the search right now on the front page, there's the last four episodes of the podcast are right in front of you. You can arrow through and go back, I think through like the last 30. If you want to listen online. Most people don't listen online. There's a search audio. If I just type in 1617, your episode is in front of me now, I can click on it, go listen to it in Apple or in Spotify or, you know, right here, if I wanted to type it into a different search box. 16, 17, now it searches the website. It takes me right to the webpage that I created for your episode. There's now a beautiful menu on the side that lists out the guides and the estimators, the series, the collections, different links in the site. Like, I completely remade juiceboxdocs.com, which is a website where you guys can send in like great doctors that you use. It's now searchable. It now tells you if the. If the doctor has type one, you can search by that. You can submit your own doctor, which used to go into my inbox. Then I had to sit down and then go in and make a text box on the webpage and recreate that. Now it goes into a Google Doc somewhere where someone looks over it with human eyes and then slides it into the other page of the Google Doc and it appears on the website. Right.

[B]

It's amazing, right?

[A]

Not only that, but you can click on a phone number when you're in there, call the doctor, go to their website, launch a Google map for it. Have you ever heard me talk on the podcast about. I don't really understand what the podcast does for people. I make it and I know it helps them because they tell me, but I'm trying to figure out functionally, what does that mean? Like, if I. If I told you to sit down and be me, what is it I'm doing? I know that maybe is sort of existential, but I realized I was never going to figure it out. Exactly. So I just loaded in all of my transcripts and I asked AI and it explained to me why people interact well with me. The Contour Next Gen blood Glucose meter is sponsoring this episode of the Juice Box podcast. And it's entirely possible that it is less expensive in cash than you're paying right now for your meter through your insurance company. That's right. If you go to my link contornext.com juicebox, you're going to find links to Walmart, Amazon, Walgreens, CVS, Rite Aid, Kroger and Meijer. You could be paying more right now through your insurance for your test strips and meter than you would pay through my link for the Contour Next Gen and Contour Next test strips in cash. What am I saying? MyLink may be cheaper out of your pocket than you're paying right now, even with your insurance. And I don't know what meter you have right now. I can't say that. But what I can say for sure is that the Contour Next Gen meter is accurate, it is reliable, and it is the meter that we've been using for years. Contornext.com juicebox and if you already have a contour meter and you're buying test strips, doing so through the Juicebox podcast link will help to support the show.

[B]

What did it say? I'm so curious. I mean, I have some ideas, but I'm curious what the AI thought.

[A]

I'll pull that up and we can talk about it at the end. Okay?

[B]

Okay. Okay, great. And then I just wanted to pull out one other thing that you said. You said it allows you to do human things. It allows you to do human things and it does things that you wouldn't have been able to do otherwise.

[A]

Absolutely.

[B]

And I think that is really what we're trying to get to with AI and I think it really directly applies to a lot of the diabetes pieces as well. Because really what we are trying to do, I think what we're moving towards with diabetes is that we're able to analyze data in a way we never were before. We're able to do precision medicine and individualized medicine in a way that was never possible previously. And then we're able to figure out how well things work in a way that no human would have been able to. So I think that's the hope. So kind of big picture what AI is doing now, I think we all know the closed loop predictions, the predictive technology, that's all AI. Technically it is older AI. For the most part, it's mostly ML, which is the older kind of AI technology. We are going to see more personalization, more things like exercise, prediction, better dosing. And then pretty soon we're going to start seeing digital twins and AI that can really be more close to you. And then I think also looking at larger population health and trying to figure out better ways to predict diabetes as a whole and predict things that influence care and improve care.

[A]

Talk a little more about what digital twins means.

[B]

Yeah, great question. It sounds really bizarre and scary, I think. But what it really is is it's a digital version of all the data we have about you. So that would be things like, you know, diabetics have so much data about them that most people don't have. You know, that just even if you only look at the glucose monitors, you know, you can. You can guess at what was happening during multiple points of the day. And then if you add, you know, test results and other pieces of data in there, you basically have a version of yourself that is just a whole bunch of data, and that is a digital twin. The advantage of that is you maybe now, but in the future, the thought is that you can try stuff on the digital twin before you try stuff on the real person. And that hopefully the digital twin has enough data to be a realistic representation of you and how your body will respond. And that way, you know, really what we've been doing for a long time is more or less experimenting and being like, well, here's. I mean, as a doctor, I could say this. We give somebody medicine, we say, all right, well, you know, it works for a lot of people. It doesn't work for some people. You know, I hope it works for you. And the hope is that with more digital twins and more data about people, we'll be able to make much better predictions about what kind of treatment, what kind of therapies will be most efficacious for different subsets of people and even for different specific people, which are really. Is the change. Already I'm getting alerts in my. My medical inbox when I prescribe something saying this person has. Has had testing for a specific enzyme, for example, that speeds up the metabolism of certain kinds of medicines. And so if you prescribe this, you know, either you want to avoid prescribing it depending on what the medicine is, or if you prescribe it, it may not work as well, or it may take longer to get out of the system. So already we're seeing a little bit of that. But that's just one data point, really. That's just that one lab test. What I'm talking about is having the whole set of all the data points of you and being able to test things on you before it actually gets to the person themselves to make sure it actually will work.

[A]

Because we have all the data already. It just doesn't do anything. See, I'm overwhelmed with that idea right now that I've recorded 1800 plus episodes and that if you kind of colloquially talk about it, people say, oh, I listen to the podcast and my A1C goes down. Which means that the answers to your issue are in there. Somewhere. And so if they're in there, but they have to come out conversationally, isn't there a way to like pick through them and distill it even more? Right, like, so the podcast is great for people who enjoy conversationally listening to something that they want. Long form talking. Right. But some people just don't want that. And some people will tell me, I've listened over and over again and nothing's happened to my A1C.

[B]

Interesting.

[A]

They just don't learn the same way. So when, you know, it feels like there's a big dark room and all the answers are in it, but I don't have a light, I can't turn it on. And even if I could turn it on, what I would find is millions and millions and tens of millions of words that have to be gone through to figure out what is valuable and, you know, and what isn't. And I've been thinking about that for years and now all of a sudden it's like, it's right here. I tried to service the other day. I don't think it's quite ready for primetime where you load all of them in. You can just talk to just like you basically create a large language model of just the transcripts. So it's only going to that it's close. It didn't do a bad job, but the engine was like GPT4 and it just wasn't quite right.

[B]

Right.

[A]

I thought, okay, this company, like, if they keep doing this, hopefully they'll stay in business or somebody else will figure it out and maybe, you know, a couple of years later. But then you immediately run into the problem of you have to give somebody a prompt and then they have to ask it the right question to get the answer out of it, which is unlikely. Like, that's probably not going to happen.

[B]

Right.

[A]

So then the, the model needs to be able to already know your questions, even if you don't know them, so that it can serve you the information. But I'm telling you that before I'm done, there is going to be a prompt. Juiceboxpodcast.com is going to be a prompt when I leave. It's going to have questions that are, that you don't even know to ask. You're going to click on them and it's going to tell you the answers and that's going to be that. But the problem becomes there is like, what if someone types into the prompt? My insulin to carb ratio is this. My sensitivity is that I'm about to eat 50 carbs. Blah blah, blah. What's the. What should I bolus? There's enough conversation inside of the podcast to answer that question, and then that becomes a Class 2 medical device.

[B]

Well, the FDA just came. The FDA just loosened the rules.

[A]

You thought they loosened them?

[B]

Well, they. They said they weren't going to take any action with. With chat or Claude on healthcare advice.

[A]

Perfect. Because I have a. An estimator, I have to call it an estimator on my website, where you put in just your weight and it gives you starting settings for everything. Because what I figured out one day is I was watching, I went into the office, I was talking to the practitioner, and I said, you know, I think Ardent settings are messed up. Like, I don't really, you know, I'm not, I'm not sure, like, kind of like where to, like, reset the meds back way before I knew what I was doing. And she just said, how much does she weigh? And I told her. And then she pulls out a piece of paper and she's writing and she's scribbling and scribbling and writing and talking about. And then I, you know, over conversations with Jenny, I realized there's, you know, there are prescribed calculations they do off of your weight to give you starting settings for everything. Carb ratio, basal sensitivity, the whole thing.

[B]

Oh, wow.

[A]

And so I was like, oh, okay, so I'll like, find out what that math is and then I'll just put it all together in one place. And I put it together and I was like, oh, okay, now this is a tool. I can't put a tool up there. You can't type your weight in because then it's a diagnostic tool. But if I put a slider up there and call it an educational tool and you get to pick a weight just to see what happens to the settings, it's not my fault if you pick your own weight, but that kind of stuff is ridiculous. Like, do you know what I mean? Because everyone should have access to being able to reimagine their settings like that. That shouldn't be a big deal. I don't think. It doesn't say that the settings are perfect if your basil set at 1.5 an hour because you don't bolus for your food. Well, and your doctor just keeps pushing your basil up and over basil's you. And you could, like, learn one day, like, oh, gosh, you know, it seems like my basil should be more like one an hour. Huh? And my carb ratio is, wait, one unit covers 15. I've had it as One unit covers, you know, the wrong thing the whole time. Like, it would give you a place to kind of start over again. And I just think that, that if you could then spend a little time getting your settings together, then go to the other estimator where you can put your settings in the carbs, the fat, the protein of what you're eating, and it breaks out exactly how a bolus would look. Is that not what we want for people? Like, do you know what I mean? Like, that's so. I'm glad to hear that you feel like they loosened it up because.

[B]

Yeah, yeah, they basically said. And what a. First of all, what an amazing tool. And I can't imagine how helpful that is, is and will be for, for so many people. I mean, just having these kind of resources.

[A]

Yeah.

[B]

That are on your website especially. You know, I remember it wasn't so long ago, I guess four, four or five years ago, that I was just starting out as a type 1 diabetes mom. And even with all the medical education and knowledge that I had, it was still completely overwhelming to figure out, you know, what you should be dosing for different, at different levels. And I definitely relied on your podcast. I definitely, I definitely was a person who learned from your conversational approach and have appreciated it. But I think the more, like you said, the more tools and the more ways you allow people to interact with the information, the, the better experience people are going to be able to have for themselves. And at some point they might have their own agent who knows them and knows their personality and knows their, you know, where they usually struggle. Might even be able to just engage on its own with your, with your tool and then bring that information back to them without them having to search it out. Because that, because their own chatbot will, will proactively know. Oh, look, hey, there's this problem coming up. I'm going to bring this information to the person.

[A]

Yeah. Or leave me out of it. I basically just vibe coded it and I just, I just vibe explained it to you. At this point, you could go to a window and say, what are all the implications that, you know, are taken, you know, into account? When I'm bolusing for food, it'll just tell you. ChatGPT will tell you about the Warsaw Method. It doesn't need me to tell you about it anymore to make your point about the agent in the food. Like, if you know your sense, your sensitivity and your carb ratio, and that's pretty much what you need to know. So if you know your sensitivity, your carb ratio and the impacts of what's coming from the food. Fat, protein, carbohydrates, right? Boom. Here's the. You know, it's a 4.6 unit bolus, and then you need another 1.6 units over three hours to cover the fat. Like, something like that. Right? And then you said to it, well, you know what? This is a. I don't know. This is a Cheeseburger Happy Meal. Remember that? Remember that I'm having a cheeseburger, that these are the. The carbs, the protein and the fat for a Cheeseburger Happy Meal. And by the way, here it is when I do it with a milkshake and then just build a library behind your. You could have an app on your phone in two seconds that you could literally just pull it up and hit a search bar and type in Cheeseburger Happy Meal, and it'll tell you how to bolus.

[B]

Exactly.

[A]

Based on your settings.

[B]

Based on you. Yes.

[A]

Yeah. Right. And so, like, that's. That's not just, like, futuristic, but I'm telling you that me sitting here right now, I think I could build that app.

[B]

Oh, you definitely could. I mean, my kids have been experimenting with all the tools and building apps pretty frequently. I mean, it's amazing how much it's democratized the ability to create a website and an app and different tools, because these are tools and these are tools for people to use, and they're for the people. But there have always been people with great ideas who just didn't have access to a programmer and resources and money to build the thing that would actually help them and help other people. So, yeah, I think it's amazing. That has really been made available to everyone now.

[A]

I swear to you, on Sunday afternoon, I hate the menu at the top of my website. And I finally just was like, I went to. Let me start by saying I'm not using the free version of one of these things, okay? So I'm paying a fair. You know, some of them are, you know, a couple hundred bucks a month, but you get deep research, you get unlimited, you know, tokens. Like, you can. You pretty much go as much as you want. I've been doing a lot of it in Gemini, and that's been working really well for me. But I'm talking about Gemini Pro, their ultra plan. Like, I think it's like $250 a month or something like that. Don't just go to the window, like, to the free version and be like, tell me how to take, because it's going to make more mistakes. Right? So anyway, I go to the window and I just say, look, go to juice box podcast.com and look at the menu at the top of the page. And then it comes back and I go, I hate that menu. Can you write a better one? And it just did.

[B]

Yeah.

[A]

And that. And that's it. And now I. And then I looked at it, I went, oh, I don't like that. I'm like, put this here. Can we put. On the right side of the page. Can we do this? You know, when you mouse over something, I'd like it to light up a little bit. And then it was a little, like, too much. I was like, not that much. And then it dialed it back a little bit. And then I was like, here's all the links. I want to be in there. Put them in there. Make them alphabetical. Except I want the pro tips, the bold beginning and this one to be the top three. And then it can be alphabetical. There's no coding involved in it. Like, I just literally spoke to it what I wanted to happen. It made me feel like the typing was slowing me down and I should get a headset.

[B]

You probably should. Those work quite well with these, these AI tools now. But I think what. What's interesting to me is that now that's what the coders will say too. You know, engineering used to be coders. You know, people think of the hackers, you know, typing away with. You know, there are a lot of semicolons and the, you know, the. The whole screen and everything. And now they're even. They are doing a lot more conversational coding. I mean, obviously it's easier for them to see the whole architecture. And that's, you know, I think just like with any.

[A]

Where value still is. Yeah.

[B]

So, yeah, I think when people talk about coder think about coders, they think about, you know, people actually typing code. And I think if you talk to programmers now, you'll find that the job, even that job itself has changed so much. And now a lot of them are interacting in a mostly natural language with tools that help them. You know, they say some similar to what you are saying. You know, you have this desire to do something different and, you know, how can you change it? For me, what would the code look like? It spits out the code. Of course, they're able to change that code more easily and to modify it in a more sophisticated way, but that's kind of how the job is progressing very quickly. And to me, that is interesting too. Basically, they are. The programmers are becoming supervisors for the Agents that are going out there and coding things for them, and they're basically managing all these agents, you know, giving them context, giving them information like you would for any employee. It's neat for them, I think, too, because it used to take them days and hours and weeks and months to create one thing. You know, you're typing, typing, typing. Finally, you get something that, like, kind of works, and then you would, you know, then you would debug it for another several months. And now you can do that in. And like you said, days, hours, minutes in some cases.

[A]

When I was, like, 12 or 13, I saved money for, like, two and a half years, and I went to Radio Shack and I bought a computer and I went home and I had this book of it, just codes in it. And I spent an entire day, like, typing the code from the book into the computer. And I remember hitting the, you know, enter, and it just failed, right? And so I went back and spent hours reading through the book, and then I found the. Like, literally, the one place I put a comma in the wrong place or something. Wrong, right? And I hit enter, and a stick figure popped up on my television because the computer was attached to the television. It did one jumping jack and it stopped.

[B]

Yes.

[A]

And I put that computer back in a box and returned it and got my money back. And I was like. I was like, this ain't ready for me yet. And now today, I'm telling you, like, you have to kind of, like, listen to what I'm saying. Listen to what Sarah's saying. But then imagine it in the hands of the company making your insulin pump. We already got to see it with loop and Trio and all the. The Android ap, like, all the, you know, the people online coding, like, you know, algorithms for insulin delivery. If you really stop and listen to what's being said, what this stuff is good at, as is pattern recognition, right? Like, that kind of stuff. And forecasting glucose ahead, adjusting your basal insulin, like, delivering correction doses, like, this is all, like, kind of forecastable stuff. Stuff. And it's moved into all the other, you know, all the pump companies have a version of it now.

[B]

Yeah, absolutely.

[A]

What we're trying to wonder is, like, will a company ever get to the point where they're going to be comfortable making something that is so personal to you? And then back to the thing that I thumpered through before, that I couldn't really talk about because I couldn't find the right words for, is this stuff gets through the FDA because it's simple. It's pattern recognition. If this happens, then do that. If this happens, then do this. But it doesn't change because if it learned while it was going, then the FDA would need to approve the next thing it was going to do. And that can't happen because that'll happen ad nauseam, over and over again. Like, so that's where the rules have to catch up to the technology. And God knows how long that's going to take. Can you explain that better than I just did? But you know what I mean, right?

[B]

No, that was a. I think that was a great explanation. I am hopeful that we will get there. And I think that is the future. And I think everyone kind of realizes that that is where we need to get. The FDA did make an allowance for AI to have some kind of planned updating, more or less where you kind of say when it gets this much information, it will do this kind of within a certain range. So it's still bounded and not just like it's gonna, you know, kind of do whatever it feels like. So they're kind of inching that way. But I think it's a real struggle to, to move from devices that are meant to work the same in thousands or hundreds of thousands or millions of patients, to devices that are meant to really work differently, and possibly very differently in every single person. And figuring out how to manage that in a way that is still safe with all the humans are often the weak point in a lot of these technologies. We do things that we're not supposed to, or we drop them, or we accidentally put an extra zero when we're typing something in. And so figuring out how to guard against some of the possibly very bad things while still delivering the benefit is something that. I think you're right that the regulatory agencies, not just in medicine, but I think in all highly regulated industries. So things like defense, education, and those fields everyone's really struggling with because it's such a different paradigm than we've been used to.

[A]

It's different than how we think too, right? Yeah. Like, we think very literally as well. Like, and it's, it's hard for people to jump ahead and have, like, fanciful ideas about what could happen. Like, I'm, I'm telling you, like, sitting here thinking, is it possible it's. I could task something with understanding the value of how I conversate. That's not a thing I was going to get done otherwise. I get people's reviews back. Oh, Scott's this or he's approachable. Like, they use words like that. But there's actual reasons why it works. And I don't know what it is because I'm not doing it on purpose. And they don't know what it is because it's just working for them. And nobody's going to spend the time figuring it out. But if I could push a couple of buttons and come back a week later and read a report that explains a little bit about that, I don't know that it would do anything for me, but I don't know that it wouldn't do something for me. Like, I just would like a deeper understanding of how it works. And I want to, I guess I want a deeper understanding of what, how conversation helps people or why it works for some people but not others. Because then if I know why it works for some but not others, I might be able to find a way to make it work for somebody else that's not touched by it as well, or just to open up my own mind to understand. Because I don't think we're going to get to what all this can actually do if somebody doesn't kind of run forward with their hands up and go, like, hey, what does this do? I'm sitting here right now having the conversation with you for the first time, thinking, the DIY community for diabetes is amazing. Like, each and every one of those people is wonderful, right? Anybody who put time into sitting down and banging out code to make loop or something similar to that, no one will ever be able to thank them well enough. But is there going to be another generation of those people, or are some of those people going to have their thoughts reignited? Like, are you gonna wake up a couple years from now while the industry is struggling to figure out what to do? Like, is, you know, are four guys, you know, connected in, you know, all over the world, and, and, you know, some wonderful lady who sits down and, and writes out the whole, like, instruction manual for how you put it together? Like, are all those people going to come back together again or reform, like, a new version of the Justice League or whatever, and make, and make a version of this that just that you pick your phone up and go, hey, I'm going to McDonald's and I'm buying this. And, and is that it? Like, do you know what I mean? Like, is that gonna. Because it's not, not doable.

[B]

I, I, I'm so glad you mentioned that, because I do think that the diabetes community is so lucky to have so many truly dedicated and interested participants and people who are very active, who have a range of expertise that is really the best possible environment for AI, because AI is so multidimensional and so multimodal. You can get somewhere with a coder, but you can't get as far as you would get with if you had a coder who understands AI. And you also have people nearby or involved who understand some of the social aspects, some of the medical aspects, some of the, yeah, the user hardware, the user aspects. I mean, all these pieces to it, you're going to create something much more meaningful and amazing. Especially if all those people are able to use AI to speed them up and to refine their ideas and to get better products out in the hands of people faster, which is really what the industry is trying to do overall and get feedback more quickly. All this can just be sped up so much. I do want to say one thing with the evaluation of these models and why it is harder to do than it was in the traditional machine learning models. And that is because AI is by nature probabilistic and not deterministic. And what that means is it chooses the most likely answer from a range of answers. It doesn't always give the same answer, given the same information. So because of that it's hard to test if it's working because say, maybe even 17 out of 20 times it'll give maybe not exactly the same, but a very similar answer, for example. But then three of those 20 times, maybe it gives a very different answer or kind of a strange answer that's not quite as understandable or it's just off enough that you don't really feel like you can be like, ah, yeah, that's a good one. How do you trust that? And how do you say, well, it's did a good job most of the time. Is most of the time going to be sufficient for the users and then if you multiply that by the thousands and hundreds of thousands of millions of people who use these tools, you can see how those evaluation challenges would be very difficult. And that's one of the main thing that things that regulatory bodies have really struggled with.

[A]

Well, I agree with you, but that shouldn't be the end of the conversation. That's all, that's all I'm saying. Yeah, you don't hit a road bump and they go, oh, see, it's it, you don't do what you see online. It's not always right. I asked the asked the same question three times. It said three different things. Well, okay, well I guess this doesn't work anymore. No, no, no. I have transcripts on my website. Right. This is not Obviously delivering insulin. But I have transcripts on my website. They're AI generated, but they're ugly. And because they take up so much space, I have to put them behind kind of an accordion, like a collapse thing, which makes them not searchable by S for SEO. And. And. And that's problematic for me. I would like my site to be. I would like the transcripts to be searchable. So I finally had time to sit down and I said to my. My prompt, I was like, here's my problem. What can I do? And it said, oh, you can give me the transcript and I'll turn it into code. You can put it into a code block, and then the code block can stay open partially, and you can click on it to open the full thing. And that way Google will be able to see it when it scrapes your site. And I was like, oh, awesome. Go ahead and do that. So it. It did that. And I was like, all right. I wish it was a little more like this. I wish you should pull out some key takeaways, put them at the top. I want the formatting to be more like this. I need it to be more readable that, you know. And then I got it exactly where I wanted it. I was like, awesome. Now, I've been using AI long enough to know that if I just start dropping a new text file and saying, do it again, do it again, do it again by the third one, it's going to mess it up somehow.

[B]

Yep.

[A]

So it gets to the third one, and all of a sudden it starts, like, leaving. It says, site start at, like, at every. And I just go back in. I'm like, do not put the site start language in the final product. And it takes it all out, but then it gives you a abridged version of the transcript. I said, you took out site start, but then made the transcript abridged. I need you to rewrite this so that you don't do that again. And so then it. It does it. And I finally got it to a point where I realized that what I need to do is I. I got it to write me a prompt. I take the prompt and I drop it into the window with a text box. I hit return, it gives me back code, I drop the code in the code block, and then you get to see the transcript on the website. It's very readable and lovely. But what I need to do is, every time I drop it in, I need to drop it in with the prompt. I can't just say, do the next one. Do the next One to your point, like, I, for the life of me don't understand why after the third or fourth one, it just starts to mess up.

[B]

It loses context.

[A]

Yeah, yeah, yeah, yeah. But it. But do you understand why that happens?

[B]

Yeah, yeah, it loses. It loses context. And it so basically most easily sees the most recent things. But if we think about our brains as having all these different memories in them and basically being kind of like employee handbooks that are within our brain for different things that we do or information we have, the AI doesn't have nearly as much of that, especially for a specific task. So it has kind of broader tasks like write a file or develop code, but for these very specific tasks, it doesn't have that kind of memory to draw from. The further away it gets from what it was asked to do initially, it just kind of starts the edges, more or less. But I also want to say you really have gotten deep into using these, and I think it can be really intimidating for people to hear the words like, oh, it made a code block, and those kinds of things. And just want to really emphasize, especially for this diabetes community that's so innovative and so, so dedicated that it really just involves playing around with it for a while. And then you get this intuition, like, you have about, like, well, after the, you know, takes a few times, and after that, it really goes off the rails. But understanding the nuances and the complexities of it is really from just using it. And you can take classes. There's online classes for most of the major, you know, frontier models have classes on how to use the different tools and how to upskill in them. But my experience has been just using them on a regular basis for things you actually need to do and get taken care of is by far the best kind of learning. And I think I would really just want to leave your audience, too, with that message of, this is something now that has gotten to the point that normal people without any coding experience can use these tools and create really cool things. Really cool things that previously would have needed a serious programmer to do.

[A]

If you don't like the way that sounds in your mind, think of it as construction.

[B]

Yeah.

[A]

Imagine if you had green lanterns ring and you could just sit here and go, like, make wood, Put it there, do this, make that taller, like that kind of thing. It feels like that to me. And it's. Obviously, it's all digital, but our lives are digital at this point, so it's. Right. It's not a stick figure doing a jumping jack. It's an actual thing that can impact, I don't know, your lives. Like, I don't know if everybody will, you know, can you use it around your home? You know, I don't know. You could get it to, you know, and there's arguments, by the way, like, I don't let it answer my email because I start thinking like, well, if I let AI answer my email, then why are we emailing each other? Are AIs just talking to each other? That's not really, really valuable. I answer my email all the time, like by hand, by myself. But the other day, something happened online and I needed to make a response to it. And it was Sunday morning. I guess I have a weird job, right? Like, so people are kind of are talking about something and I need to get involved and I need to really, like, thoughtfully give my, my ideas around it. Okay, right. So if that's going to happen, then I've got to wrap my, my mind around what's going on and then I've got to read what people are saying. Then I have to make sure that I feel about it the way I think I feel about it. Then I have to think about how to talk to them about it. Then I've got to write it out, Then I've got to edit it and do it again and make sure it doesn't. Make sure it covers all the bases. I'm not trying to be offensive. Blah, blah, blah, blah, blah. It takes me about, like, when you see one of those posts from me online, you're like, oh, Scott's such a well thought out guy. He must. Blah, blah. It took like three hours to do that because I'm also not a classically trained typist. My brain's the right person for it, but the rest of me is not the right person for it. I was able to explain to a window what the problem was and how I felt about it, and then say, I want to talk about this. And I want to say this and this and this and this and this. And it structured it for me in, like, no time. And then I was able to read that back and go, that I don't agree with. That is exactly what I meant there. I would say this differently and then basically rewrite it. And instead of me taking three hours, it took me 45 minutes. And. But moreover, what you don't know is that if it was gonna take me three hours, I just wouldn't have done it right. I would have looked at it and said, okay, I can't get that done today because it's Sunday morning. And my family's getting up and we're doing stuff and I don't have time for this. So that thing would have just sat there untouched. Now, that was just a Facebook, you know, conversation. But I thought it was a big deal. And then after I read it, I thought, oh, I'm gonna make a podcast episode about this too. And I think it's going to help people. And to give it more context is very simply. I don't, I'm not going to get in a soapbox here and waste your time. But I. At the moment, there's a lot of conversation around the Eladon trial out of Chicago and the, you know, and people are getting islet cells put in their liver. They're taking this new immune suppressant called Tego and they're not having a lot of any side effects of. Most of them are saying, and so, you know, these people have a functional cure. And this, you know, there's a, this is a trial going on. It's not FDA approved. It's like, you know, it's a trial and it's exciting. But because of how social media is set up now, everybody online is like, they cured diabetes and blah, blah, blah. Like, and I don't like that. I don't like giving people the idea that this, it's almost over because it's not. Even if they got through the FDA today, there's still a ton of reasons why it's not going to get to all, like 1.8 million of you probably ever or, you know, cost or. I just think we should talk about that, like, adults not use it as fodder for Instagram and TikTok to get likes and posts and retweets and stuff like that. I just kind of contextualized how long I've been in the space that I don't like talking about it this way. I do think it's very important to talk about. I've got somebody coming on the podcast to explain their situation, but you need to understand, blah, blah, it's a more thoughtful thing. And I just never would have done it. I just, I would have run out of time and not done it. But instead, that post gets 20, 30,000 views. People are, you know, hundreds of likes and hearts and you make people feel comfortable. It's a good thing. And then I'll probably sit down and do like a talking head episode about 15 minutes long explaining this because, you know, I understand everybody trying to be an influencer nowadays, but, like, come on, like, don't jerk people around about them Getting cured about their diabetes. Like, it's, it's okay to explain to them what's going on. It's not okay to make it sound like it's imminent, in my opinion. Like. Right. So then that's going to be my perspective on it. And, and, but anyway, without AI, like, I would not have had time to put my thoughts together and put them down like that.

[B]

Well, it's so important to have, first of all, I think, you know, to have a, A seasoned voice out there and a voice of that can really, like you said, context is so important both for people and for AI to understand what's actually happening and where we actually are in this. And like you said, I think it really is about creating those opportunities to do something where nothing would have been done that are the biggest yield. And to have you talk about these issues and to give that kind of very reasoned and helpful picture to people. I mean, that's a huge benefit to the conversation. And kind of a little ironically, it actually, you know, you having put that out there and then it being engaged with so many times, thousands of times, because these models are scraping from the Internet, that actually helps give these foundation models better information over time and a more reasoned viewpoint. Just by you using AI to put your thoughts together more quickly and put that viewpoint out there.

[A]

Yeah, well, maybe one day it won't tell Mark Zuckerberg to value people arguing over people talking, and maybe then some of these posts will get seen by other people.

[B]

There is that.

[A]

There is that. Huh? So I'm not going to tell you what it said, but I will tell you that my prompt for trying to figure out why the podcast is valuable people says you are analyzing a long form podcast transcript. Your task is not to summarize. Your task is to extract moments where Scott gives directive advice, expresses a belief about how diabetes should be managed, challenges a common mindset, Reframes fear into agency, pushes back against conventional thinking, describes what works or what doesn't work. Return direct quoted statements, one to two sentences of context for each quote. Label each as tactical instruction. Mindset principle, philosophical belief, behavioral pattern. Do not invent ideas, only extract what is clearly present in the transcript. Ignore guest only statements unless Scott affirms or reinforces them. But I didn't write that prompt. AI wrote that prompt with me explaining to it what I wanted it to do.

[B]

Ah, I love that.

[A]

Yeah. Because what I've learned is, is that my dummy brain can't talk to it as well as it needs to. So instead of jumping right into the task, I Pre bolus the task with another task. I go in and I go. Instead of just saying, like, go into these episodes and find out why I'm so great. Like, you know, which I'm sure is how some people deserve that. Instead of doing that, right, I say, here's my goal. Here's what I think might be happening, but I first need you to read a couple of transcripts and tell me if I'm wrong. And then it comes back and says, well, I think your impact might be this, this, and this. I think we should look for these things. And I go, okay, write me a prompt for you that will help you do that the best you can. Like, that kind of stuff. Like, I talk to it in, like, cleaner language, like, or, you know, more colloquial language like that. And then it comes back and it gives me the prompt and I go, okay. Like, is there anything about this prompt that will lead us to basically tell it? Like, I don't. I'm not looking for you to glaze me. I'm not. I'm not asking you to kiss my ass. Like, I'm. I want real, actual. I want you to really think about human psychology and why things impact people. And, you know, and. And then I end up with this prompt. And then the prompt does a really good job of pulling out ideas. Now, where could I use that in the short term? Probably social media, right? Like, there's quotes in here as it's going through that are all, like, they're really valuable things for people living with diabetes. I see each and every one of them. But if you ask me to go, like, remember what I said and make a piece of social media about it, I can't do that. And even if you asked me to go back and listen to the whole episode and jot down takeaways, like, I don't have the time for that either. I would never get that accomplished. I'm taking something that I already know helps people, and I'm finding a way to repurpose it to help different people. And that with that, without this, that doesn't happen.

[B]

That has a lot of implications for diabetes in terms of. Also, there's the medical side and the, you know, the medical. The device, the really in the weeds side. But then there's also the side of advocacy and communication about, you know, what. What is this to a broad audience and then within the schools and within different settings. So I think using AI for those circumstances is also probably underutilized right now in terms of people saying, oh, I'm doing A fundraiser. And I think this is important and I can maximize it much more easily if I use AI or I don't really have the words to describe why I'm having, you know, to describe a certain issue to the school nurses. Can you help me put it into a way that they might understand better? So I think a lot of those kind of communication pieces are great use cases for AI, too, and especially for the diabetes community.

[A]

Yeah, I just. In general, I think some of you are just not thinking about this the right way. That's all. Like, there's real ways to use this for yourself right now. You just have to kind of. You have to just step back and see how it works and how it thinks and how you talk and how you can do those things together to lead it to do the thing you want it to do. Like, it's not just. That's what happens when people say, like, I asked it something that got it wrong. I, like, I would love to see what you wrote into that because I bet you didn't have a chance in hell. And it does get stuff wrong like we talked about like that, that I understand as well. But you also have agency and you could read it and decide if that's. If. If what it told you makes any sense or not. I just think, like, using this as an example in my heart, like, I don't know how to do this yet. I haven't been able to teach myself the whole thing. What I would like is an app where you can listen to the podcast, but we're also like, daily affirmations might pop up like that are just from context, from the podcast. I would love it if one day that app had. You know, I know you can't do this because Facebook won't let the API out, but I would love it if the Facebook group just lived, if it all lived together. I don't imagine that's going to happen. I think the code would get crazy and, and, and it wouldn't work. But I would just, I would just like an app that you open up that I don't know, when it opens, it says something to you that you might find valuable or supportive about diabetes. You swipe up and there's the little app where you figure out your bolus for your day or something like that. That.

[B]

Right.

[A]

I just think that might be nice for people and, you know, a way for them to take a break or to be reminded of something, because I hear that all the time from people. Like, one of the things that somebody will say is, like, I Already really know how to take care of my diabetes. But listening to the podcast keeps me the way they tell me is like focused on it without being too focused on it. So it's not front of mind and back. And they're not always like, God, I'm always thinking about my diabetes. But it's around just enough that they find themselves making good decisions. And I wonder if like just having something pop up in front of you that says like, you know, you get what you expect and, you know, if you expect a 130, you're probably going to get a 130 setting. You know, if your high alarm set at 150 right now, try moving it down, right. And maybe you won't do it right then, but maybe it'll be stuck in your head the next time something happens and you know, like that kind of thing. So I don't know, like, I'm going to do my best with it and see what I can figure out. I'm hoping that the companies are doing their best with it. I imagine that they're going to go incredibly slow compared to my desire and I understand all the reasons why they would do that. And I really do hope that those brilliant people who already came up with Loop and Trio and all that other stuff, I hope they're out here like, wondering how to like, judge it up a little bit. So I don't technically know what that means, but I think we all understand what I'm saying. Like, you know, we'd like to lift some burden and, and make better decisions and do so in a way that is as blended into your day as possible so that you don't find yourself always interacting with, you know, numbers or the thing or whatever. It can, it can feel more natural, I guess, is the. Should be the long term goal, in my opinion.

[B]

Absolutely. Well, I hope for that too. And I do think that there's a lot of pressure on these foundation model developers to move quickly from a competition standpoint. So I, I think they are trying their best to, to juice it up to, to pun. To have a very bad pun there as quickly as possible. I also want to say that I loved when you said I'm. I'm going to give it a bolus of instructions.

[A]

Yeah, I pre. Bolus the. The task with another. Yeah. With another task so that we. My dumb brain isn't the one putting the marching orders together like we do it together. I, I think of it all the time. Is like it's a thing that can order my thoughts better than I can.

[B]

Yep.

[A]

I don't have the recall and I don't have the mathematical ability to put things in order because I'm a person. And it does. It takes away a lot of my. A lot of my frustration when I'm thinking about things. I used to tell people, like, one of my favorite exercises was like, think about a thing up to the end of my understanding and then wonder what's on the other side of my understanding. I don't really have to do that anymore.

[B]

Interesting.

[A]

I can tell it what I know and it can fill in the blanks about the parts I don't understand better than I can. So it's sort of how it feels to me. But anyway, I have no idea. I had another thought. I completely lost it, which is upsetting. Oh, wait, no, I. Here it is. Here it is.

[B]

Great.

[A]

I can hear people saying it's going too fast. But what I would tell you is imagine if we made one step forward. Imagine we go back, I don't know, three years ago, and it's whatever, CHAT GPT, whatever the first one was and how bad that was in the old world. You'd live with that for 10 or 15 years until people could figure out, oh, you know, if we turn this knob, this will do that. Because you have to live to. People have to go to work. They have families, they have lives. Their kids are sick, their wives broke up with them, they can't afford their. Like, these people all have a life. They can't be thinking about this 24, 7. Right? So it takes 10, 15 years to get to the next version of Chat GPT. And then like, look at how technology works. Like, it's so goddamn upsettingly slow. I saw a car 20 years ago, had a push button transmission and I was like, wait, you made that? Now why don't we have them in our cars? Like, oh, it just takes too long and we can't do that. We gotta use up the parts. Or it's the way people think is just ridiculous.

[B]

Right?

[A]

You should be excited that it's moving forward so quickly because your whole life won't get wasted figuring out ChatGPT 3 and 4. You'll actually be alive for whatever ChatGPT 8 does. And like, that should be exciting to you. The way I think about it over and over again is that five years ago Tesla said, hey, we have self driving and it wasn't good and now it's awesome. And they did that by building their own computers and getting their own data and telling their computers to specifically crunch this. And then once it got so technically good. They said, you know what, it's really great, but it doesn't feel natural. So they just gave it video of humans driving. And we're like, here, make it more human. And then it did that. Like, are you crazy? Why are you complaining about that? That's amazing. That's one small idea. I don't care if you want your car to drive itself or not. Apply that idea to everything. Like, imagine you might actually live long enough to see something cool now, instead of it just being like, oh, when I was a kid, the Internet wasn't here and now phones have glass on them. That's nice, but I want to see what else is coming. I'm getting older. You know what I mean?

[B]

Exactly. I mean, as I'm sure you are aware, for Claude, Claude actually built Claude code. So Claude built a tool for itself to use to speed it up, speed itself up to improve its capabilities. So I think we're going to just see more and more of that where we as humans don't have to be the bottleneck in some of these innovations. Like you said, whether it's thinking or just pure capability or resources, we don't have to be the bottleneck. We can move past some of those more quickly with some of these AI tools. And that, that to me is the exciting part. That's, that's what I hope to. To really that the society capitalizes on and that the diabetes community in particular really feels like they can, they can use.

[A]

Right. Use the self driving as an analogy for medical research.

[B]

Yep.

[A]

I don't know. Something as simple as it watching your blood work and telling you to turn up or down your Synthroid. You think that's crazy. Like your doctor is spitballing. Okay, but totally. But if you have a unmanaged thyroid, it is impacting your life in a myriad of terrible ways. Just imagine if you went for blood work every six months and then the blood work told you, hey, you're taking 0.88 right now. You should really be taking 0.88 and then skipping a day. Or take 0.88 and then take two on the seventh day, and that'll really help you. Your doctor's never going to figure that. You got to get a great doctor to figure that out. And it's still variable because that doctor still had to go to work today and they're tired and their husband was yelling at them when they left the house and their kids are on crack and like, they've got, like, human problems. There's so much Here. If people would just focus. If people would just do what I'm doing here. Ready? I'll get on a soapbox. Now pick a thing that you're good at and try to make it better. Like that. Like, instead of.

[B]

I love that.

[A]

Yeah. Instead of just doing. I'm going to fix the whole world. You're not going to fix the whole world. Pick one thing you're great at that you really understand, and see if you can't open your mind up and make it better for somebody or just for yourself, even. You don't have to help other people if you don't want to. Like, pay 20 bucks a month and go talk to the damn thing and see if you can figure out something. You know, how come me and my husband are arguing all the time? I bet you it knows there's a draft in my. In my. My electric bill is too high. What should I be doing? I bet you it knows. It knows what's on the Internet. Is it always going to get it right? It's not. But, like, I don't think. I do think that there'll be a mechanism at some point in my lifetime that will minimize mistakes to the point where an average person will feel good about this. Somebody should be working on that, by the way, like, one. I need a. Four dorks in a room figuring that out right now. Or they. Geeks or dorks. Sarah, which are you?

[B]

You know, I don't know. I always thought of myself as a nerd, so I.

[A]

Okay, fine, Whatever. Listen, Sarah, I'm taking up your time. Don't you have to go to surgery now?

[B]

I do, actually. I have to run into the O. R. Now, but it was such a pleasure talking to you, and I am just thrilled that you're using AI So actively. I think it's a great example to the diabetes community about what people can do who have so much knowledge and so much expertise and can get messages out there and get information out there to the rest of the community more quickly with less time and with less effort. And I am so happy that you are in the space and embracing parts of this technology so quickly and so skillfully.

[A]

Well, thank you. And let me say this because I don't think you're busy enough. You have the link to get on my schedule. You are free to get on it whenever you think you have something to say.

[B]

Okay, wonderful. Well, be careful what you offer.

[A]

No, no, no. I'm happy to say that I. I would. I'd be happy if two years were 15 episodes of Sarah talking about technology. So thank you very much. I really appreciate it. Good luck in that. What kind of surgery is it? What are they doing today?

[B]

Hip replacement.

[A]

Oh, wow. Good luck to everybody.

[B]

Yeah, people feel better? Yeah.

[A]

Yeah. Excellent. All right, thanks so much.

[B]

All right, thanks again.

[A]

Take care.

[B]

All right, bye.

[A]

This episode of the Juice Box Podcast podcast was sponsored by US MED usmed.com juicebox or call 888-721-1514 get started today with US MED. Links in the show notes links@juicebox podcast.com I'd like to thank the blood glucose meter that my daughter carries, the Contour Next Gen Blood Gluc Glucose meter. Learn more and get started [e-mail] juicebox and don't forget, you may be paying more through your insurance right now for the meter you have than you would pay for the Contour Next gen in cash. There are links in the show notes of the audio app you're listening in right now and [e-mail] to contour and all of the sponsors. A huge thank you to Cozy Earth, a long time sponsor. Cozyearth.com use the offer code Juice Box at checkout. You will save 20 off of your entire order when you use that code. Don't let me down, kids. Head over there now. Get yourself some joggers, some towels, some sheets. Save yourself some money. Support the podcast. Make your life beautiful and comfortable all at the same time. Cozyearth.com use the offer code Juicebox at checkout I can't thank you enough for listening. Please make sure you're subscribed or following in your audio app. I'll be back tomorrow with another episode of the Juice Box Podcast. My Diabetes Pro Tip series is about cutting through the clutter of diabetes management to give you the straightforward, practical insights that truly make a difference. This series is all about mastering the fundamentals, whether it's the basics of insulin dosing adjustments or everyday management strategies that will empower you to take control. I'm joined by Jenny Smith, who is a diabetes educator with over 35 years of personal experience and we break down complex concepts into simple, actionable tips. The Diabetes Pro Tip series runs between episode 1000 and 1025 in your podcast player or you can listen to [e-mail] by going up into the menu. If you have a podcast and you need a fantastic editor, you want Rob from wrong way recording listen. Truth be told, I'm like 20% smarter. When Rob edits me, he takes out all the like gaps of time. And when I go and stuff like that, and it just. I don't know, man. Like, I listen back and I'm like, why? Do I sound smarter? And then I remember because I did one smart thing. I hired [e-mail].