Tomorrow's Medicine, Ep. 14: T1D Education Platform - John Pemberton

With John Pemberton, Diabetes Dietitian & Creator of The Glucose Never Lies. Hosted by Arseniy Arsentyev. Published May 13, 2026.

https://aarsentyev.com/podcast/ep14-john-pemberton

Machine-transcribed and lightly edited: filler words and false starts are removed, and the wording is otherwise as spoken. Where the two differ, the recording is the record. Times are minutes into the recording. Not medical advice.

[0:00] Arseniy Arsentyev: Hey everyone, it's Arseniy and it is the 14th episode of Tomorrow's Medicine podcast. Today I'm talking to John Pemberton, who is an expert in diabetes education, founder of The Glucose Never Lies and most importantly founder of Grace, which is the future AI system that will accelerate diabetes education. Let's go straight to our conversation and the first question. John, please introduce yourself, tell us more about the work you do at The Glucose Never Lies and Grace.

[0:33] John Pemberton: Yeah, so my name is John. I'm a diabetes dietitian from the Children's Hospital in Birmingham, but I guess more personally, I have type 1 diabetes myself. And it's interesting, when you go on a journey from, I was a dietitian before I got diagnosed, so I was a dietitian from the age of 21, 22 and I got diagnosed with type 1 at 27. And you realise very quickly in that enormous learning phase in the first month or two, that even what you learnt at university as a healthcare professional is the theory, but the practice, the science and the art are two very different things.

[1:09] So that was kind of like my first introduction to, what you're taught at school and university is great for a generic starting point, but these things require personal experience, they require trial and error, and they require an appetite to learn fast if you want to go fast. And that isn't necessarily always available in the standard healthcare system because you're limited by how many appointments you can have with the healthcare professional. Have you got a good healthcare professional who knows their stuff, or have you got one who's like got three kids, two dogs, and you know, is it the limit? And for them to learn the new stuff is just not possible. So there's a big sort of discrepancy. And I sort of picked that up a little bit and learnt myself.

[1:53] But then the real turning point for me came when my son, you can do pre-screening for the likelihood of getting type 1 diabetes. And at the beginning part, it's just a finger prick and a blood spot. And what happens there is it tells you basically there's a good chance that you've got these antibodies in your blood that will eventually tackle your pancreas and stop you producing insulin. And you've either got them or you haven't. So when this test I had my son and my daughter tested, my daughter was negative, and my son came up positive.

[2:23] And at that time, it wasn't known what the false positive rate is, which when a false positive rate is basically out of 100 people who have this test, how many of them go on to actually have the real thing? And at the time, I thought it was like 100, but it's actually only 95. So it's still a very high chance. So what you have to do then is from the blood spot to get rid of that 5% error. You've got to go and get a blood draw from your veins and actually double check that. And that's the statement of truth.

[2:53] But the gap between having that positive result, getting the venous test, and then getting the confirmation or the negative is about six weeks. And I remember sitting there for six weeks thinking, I was watching my son because I obviously had diabetes at this point. And I was like, wow, all this information and education that I give to the parents about you need to get your insulin before you eat and 15 minutes before, and you need to do some activity after eating and all these good things that you're supposed to do. You realize how difficult it is to try and put that into practice because you're seeing your son and he's not finished his meal. And you're like, wow, okay, what would I do here? What would I do here?

[3:30] And that really just made me think, I've got a lot of knowledge. And the education materials that we give to the families who we see is good, but it's static. It doesn't move. It's not dynamic. And also it doesn't learn at the cadence that certain individuals want to learn. It learns at a cadence of you come in once a week for six weeks and you get your education. Now, some people want to learn faster than that and some people want to learn slower than that, but it doesn't matter because that's the program. So I kind of sat there and thought, well, within me, my wife, my mum, my dad, my mother-in-law, my sister, all the people who are going to maybe look after Jude, they all learn at different speeds.

[4:11] So I need to create something that allows them to learn at the pace that they want to learn. So it started off with a website. I got a hold of my mate who's, he's into tech, which we'll get to later on. And he sort of created me a WordPress site and I started, knocking out what was, I believed, good quality education. And so we got the basics, like how to survive and how to thrive down over about six weeks. And I was like, right, done. We've done the sprint. And then the test came back negative. So it was like, oh, okay.

[4:41] But then I was already on a journey. I was like, if I know now that families and things learn at a different speed, and I've already kind of said the model that we're using, it's okay. But it only suits maybe 30% of the people, 30% want to go faster, 30% want to go slower. Do I carry on making this education? And it was about at that time that quite a few people reached out to me and said, we really love your stuff. You've written stuff like we're quite geeky. We know we want the detail. We like the detail. So give us more of it. Okay. So then that was kind of a point of like, what do I do with this now? So it was just a hobby. So that was back in 2019.

[5:16] So it was like a hobby for about five years where I would learn. It was my learning process. So I would go to a conference. I would learn about something new and I'd be like, okay, so how do I distill this to people other than where I work? So I'd go in there, I'd write a guide and stick it on the website. And then, I don't know, we probably got to about 180 pages with 20 different guides and all this like massive information portal.

[5:41] And that's kind of where it built to. And at that point, we were satisfied and happy. And we sort of heard, you know what, we'll start a bit of a podcast because we've got some materials and I'll have a bit of a laugh. So we did that and we ticked along and things were going well. And then my best mate, who's a senior AI developer, programmer, et cetera, said, that stuff on your website, like it's great. And a lot of people have said this, but it's just too much. It's like, it's overwhelming, 180 pages worth of content. It's just like, it's great, but it's like, it's great for people like you who want to consume that all, but like no one's going to read all that.

[6:16] And it's true, but then you're kind of thinking, well, how can you make it easier? And it's like, I can help you create a wicked agent that can be like you on your very best day, on the best day when you're kind of like absolutely on fire. You know everything, you connect the dots, you get all the information and you deliver it in the perfect way to the person who's asking the question. To a seven-year-old kid, to a 70-year-old grandma who's got slight, learnings challenges. You can tailor that information that you've got there to them and you'll get it perfect every time. I was like, what are you on about? And he was like, we'll create an agent.

[6:54] And so we sort of began on this journey of he was helping me essentially bed in a knowledge base and create this bounded model of the best type 1 diabetes references, education, thought process. And we spent so long building this kind of like bounded resource hub. And then it was like, what are we going to do with it? The possibilities become endless. So that's kind of, that is like a, from 2019, what we're in now, 2026, a seven year journey. The beginning five years were spent writing these guides without AI. And then probably in about 2021, Phil introduced me to ChatGPT.

[7:42] And then obviously over time you get ChatGPT, like you do projects and develop your own styles and you get it to remember and you put all like the prompts in and how you want it. So I kind of like learned how to coach these tools like ChatGPT and Claude by making them like me, essentially making their agent of me like me and then create guides online and things started to get better. And then you can create some cool graphics and all those sorts of things. So we've probably spent five years manually doing all this and like improving the quality and the podcast is good. But then, as I said, if you're a new person, newly diagnosed with diabetes, and you go to the site, you'll be like, holy crap. That's a lot of information.

[8:24] Whereas what you really want to do is you want to go and ask one specific question. You want to know one thing. You came to the website for one thing. What is that one thing? And that's where Phil said, we can make an agent that is you on your best day and people can come and ask simple questions and get really robust answers in a way that suits them at their cadence and pace. So that's kind of like the seven year build to what is kind of now known as Grace, which is this central knowledge hub. And I can tell you a little bit about how we built her and things, but does that give you a bit of a frame for how kind of the journey has got there?

[8:58] Arseniy Arsentyev: Absolutely, it did. Let me summarize your hero's journey, as we can say. So you were working at NHS, then you got your son, if I'm not mistaken, diagnosed on the first trial for diabetes. And it took six weeks before the second trial got results and you understood that your son is not diabetic. After this turning point, you began to assemble a knowledge base about diabetes that will help families all around the globe to summarize and to fasten the process of understanding and living with this disease. Because obviously there are no drugs that can fundamentally cure it. And at some point, your platform gets so huge. Because guys, everyone, check out the Glucose Never Lies website. It's so huge. John has so many sources, articles, testimonials, feedbacks, procedures, and so much more.

[10:16] And I can understand at that point, the AI agent was kind of an obvious idea to structure and summarize it. The next question that I want to ask you is kind of going to the first part of this journey of yours, of assembling this database. And let's just review some terms for the general audience to keep on track. What's the type 1 diabetes briefly? And what's the difference between type 2 diabetes?

[10:48] John Pemberton: Yeah, sure. The easiest way to explain it is type 1 diabetes. Well, they both come down to the insulin that's produced from the pancreas. And basically, it has lots of jobs. But the job we're going to talk about it with diabetes specifically is it releases something called insulin. And it releases it into the blood. And the job of insulin when it's released from the pancreas into the blood is it comes along. It grabs a molecule of glucose. And it goes to either the muscle cells or the fat cells and opens it like a lock and a key. The insulin is like the key and the lock is on the cell. Opens it up and allows the glucose to go in. And then that glucose gets burned for energy.

[11:23] So you can use that to run, to sleep, whatever you need to do. So if you have a situation where the pancreas cannot produce insulin, then you cannot move the glucose out of your blood into the cells. Therefore, you have no energy in your cells and the glucose level rises very high. So with type 1 diabetes, what happens, as I was maybe alluded to with my son, is if you have antibodies and antibodies are essentially the white cells in your blood, the defense mechanism, the things that fights off colds and viruses. So when you get those, your body goes, OK, we're going nuclear on this horrible thing that's come in. We'll produce loads of white cells and kill off these invaders.

[12:05] But unfortunately, for type 1 diabetes and any autoimmune condition, it's basically like friendly fire. So you come along and you get a virus and the body kicks out these white cells and these white cells kill off the virus. But unfortunately, it also detects the insulin producing cells in the pancreas are also, for some reason, a virus cell and then begin to destroy them. So the type 1 diabetes is an autoimmune condition. It's friendly fire. Your own body turns on itself to basically destroy the pancreas cells from insulin. Therefore, very quickly, not quite overnight, but over a month or two, you go from being able to produce as much insulin as you would like to virtually none. The blood sugar goes extremely high.

[12:47] The energy production in the cells doesn't happen from glucose and can only go from fat. So you end up burning fat at such a high rate, you lose weight. But that burning fat at that level causes acidic products. And that's what's basically called diabetic ketoacidosis. The acidic part is your pH drops so low that your body can't function. And if you don't take insulin, you die. So 100 years ago before Banting and Best, it was a death sentence. If you had type 1 diabetes, you died because you could not move glucose into your cells. So the first amazing thing for people with type 1 was the synthesis and the finding of insulin so that actually people can survive this condition, which is amazing. So it's all about replacement of insulin for type 1.

[13:31] Now, type 2 is slightly different. You do not have the autoimmune destruction. What you have is generally, not always, but generally, people with type 2 diabetes carry extra fat around the midriff and around their organs. And what happens there is if you can think of a cell that has that lock that the key is meant to open, if you have lots of fat around your central organs, that lock mechanism gets stuck. What they call insulin resistance. The fat basically stops the insulin from working very well. So what has to happen then is the pancreas has to push insulin out at double time, over time, triple time, three time. And then basically you end up with such a high level of insulin in the blood that you then end up storing lots of extra fat.

[14:14] And the pancreas just becomes tired out. It's like you can imagine working a 90-hour week for 10 years. At some point, you're just going to go, hell no, that's too much for me. So the difference with type 2, it can be if you catch it early enough and you lose the fat around the organs, you can actually take the stress off the pancreas, get it back down to a 40-hour week and actually reverse the condition. If you catch it early enough, with type 1, you can't reverse the condition because the destruction is autoimmune. So hopefully that explains the difference between the two.

[14:46] Arseniy Arsentyev: Whenever anyone asks me what diabetes is, I'll clip this out and send it to them for you to describe. And I'm sure everyone understood what diabetes is. And the next stage is whenever you acquired such a vast amount of data, I'm sure you understood that everything is so messy. I am a psoriasis patient myself and I've got so many controversial advice in my life and I tried so many things. And it was one of the reasons why I started the show, to unpack these ideas and find some kind of ground truth and to acknowledge that there is a mess in the data. Could you walk us through the sources?

[15:34] Is there evidence in this data?

[15:36] John Pemberton: Yeah, so I think it's very interesting. So the way that I view this is, I'm thinking back to when I was at university. So the information that I learned at university was in the textbooks, typically not so much in journals. So your typical textbook is on average 10 years behind the current evidence base, on average 10 years. That's crazy to start with. So 10 years at best case, you're behind the eight ball. Which means if there's been a paradigm shift in the understanding of certain key elements, which that's what good science is. Good science is about disproving, trying to go hardest to disprove what the current narrative is or the current model is. And if you can't disprove it, despite your best attempts, it's the closest version to the truth you can imagine.

[16:20] But the purpose of trying to disprove it is to move knowledge along. Which means if you've got good scientists, if you've got new techniques, new nanotechnologies, new diagnosis things, these paradigms should be changing all the time. They should be changing by the month, by the week, certainly by the year, not by 10 years. Things happen faster than that. So inevitably, when healthcare professionals and education models that are evidence-based because they had a randomized control trial five years ago saying that that was the thing to do, it's already gone. It's disappeared. It may have been the source of truth at the time. And it might be good, but it might not be the best.

[17:03] And I think once you acknowledge that as a healthcare professional, as a patient, as a person living with a condition, I don't think it's right to be sort of down on the current healthcare system. It's just an acknowledgement that what you want is you want things to move fast. You want knowledge to progress. You want therapies to go fast. So if that's the case, the price you pay is that typically the healthcare professionals that you see, if they're still based off textbook learning, are going to be on maybe outdated models, maybe outdated mechanisms. And it's just natural. That's the way it is. And you'll get some healthcare professionals who are on the ball, who are on the money, who know their stuff.

[17:41] And they're the typical ones that people pay a lot of money to go to see because they're the ones who are there. So I think once you've kind of got that frame of understanding, what you have to do is kind of think, okay, if we know this, where is the best place to start, of what's close enough to the truth, that gets updated regularly enough, and there's a consensus across a broad range of healthcare professionals that that's a good place to start for everybody?

[18:09] And the answer within diabetes is the International Society for Pediatric and Adolescent Diabetes, ISPAD. They update their guidelines every two or four years with the very best people in the world. Now, I've sat on a couple of those guidelines, not saying that I'm the best in the world, but when you sit on those guidelines, you realize the top, top dogs who are there, like they know their stuff. Like they are on the five-minute science. So that's their life. That's what, the professors, they dedicated their life to it. So when you kind of go on a consensus guideline, it takes you six months to write. And by the time you've published it, you've already been disproven on a couple of things because the evidence is already gone.

[18:46] But that's fine because if you're only two years out or at maximum four years out, you know that those guidelines are the best effort of truth by a set of individuals who really know the stuff. So ISPAD for Pediatrics, the ADA, American Diabetes Association, and the European Association for Diabetes, EASD, are the adult equivalent of the best version of the truth that's two years old, is what I would say. It's a really good place to start. It's what everyone agrees is a very good place for now. So that's the place to start. And with Grace, that's what we did. We took all the chapters from the current ISPAD guidelines for Pediatrics, and that's your base layer of truth. We took all the ADA and EASD. That's the adult source of truth.

[19:30] And a lot of it is very similar, very similar, but there are some specific differences. So that allows you with your model, the base layer of truth when you're thinking about kind of algorithms and where you're going to retrieve your information from. If you're an adult, you're going to go over here. If you're pediatric, you're going to go over here. So that's kind of our base layer of information. But one thing, the beauty of me spending seven years writing guides out is I was translating these guides, which sometimes are very dry.

[19:59] 15 pages of recommendations with ABCD and no clinical delivery whatsoever in language that is like horrible to understand. And I spent seven years changing those guidelines into understandable, hopefully a bit of fun, delivered in a way that is understandable by someone who's 10 years old. And typically on my guides, it'll be like, here's the TLDR at the top. You can stop here with this 20%. Now, if you want to get nerdy and get into the weeds, I'll take you deep into the weeds for the next 80%. But you can stop here and you can get most of things that you need to know.

[20:33] So we kind of then layered the second layer was all the guides that I've developed over time, which included stuff that people don't talk about, how to take drugs, how to drink and go and get wrecked on a festival for three days because I've gone and done all that stuff. And I've been out with people with type 1 who've been out there and lived it, kind of. So you've got the lived experience element, but you've also got the practicality of understanding that the guidelines will get you 60 to 70% of the way there.

[21:01] A good clinician will add an extra 20% on top with questions and framing. The final 20% is on you because your individual physiology will determine that you will not meet the average person. So you will have a good place to start. But that 20% of self-discovery for any condition is on you. So that's kind of how it's built up. You've got the source of truth of the guidelines on the bottom. You've then got the framing from seven years of learning how to deliver this stuff with lived experience in real life and clinical delivery to over 300 children, young people and families for 10 years. I worked in adults for 20 years.

[21:38] And then on top of that, we then put philosophy. So you've got so much information in here. You're at risk of someone asking a question and this thing just spits out at you everything it knows. And it's a horrible experience because you're kind of like, I don't want all that information. I just kind of want the key stuff. So for me, like I said, my best mate, Phil, who's also like an AI developer, is obviously helping me do this along the way. He said, I don't know if you know Nassim Nicholas Taleb.

[22:08] Arseniy Arsentyev: I know. Oh, huge fan.

[22:10] John Pemberton: So me and Phil are huge fans. We've read the book so many times and just agree with so much of it. So a lot of things. So, for example, I will say that type 1 diabetes is a black swan condition, which basically means things can go perfectly for 10 years. You only need one serious slip up and that could be it for you. You forget to take your insulin or something happens to your insulin for two or three days. You're gone. If you go out and drink loads of alcohol and take loads of insulin by accident, you might not wake up.

[22:40] So you might think things are going okay for a long period of time, but you've always got to have in the back of your mind, because the risk level is so high, you have to be cautious about the quality and the accuracy of new technologies. You have to be very happy about the stuff you're using, that there's a serious evidence base behind it before you'll stick it on yourself or someone who you are supporting. So we built that within our model.

[23:10] Grace will only give you education on devices where we're happy with the data. If you asked it about certain devices which are available, she'll go, I don't know about those because they don't meet the data situation. So we know that type 1 diabetes is a black swan condition. So we apply that precautionary principle in terms of what devices and what information gets in there. So that's kind of like the first filter. The second filter is what we call via negativa. Like you said before, there's so much information out there, but there's only a few levers for most things that really account for most of the difference. So what are the three or four levers for each of these things that look after the majority of the risk?

[23:50] So for exercise, for example, if you would say to someone with type 1 diabetes, tell me about what things might influence the glucose level during exercise. And they'll be able to reel off. They'll say, oh, the temperature. Oh, it's how much insulin you've taken. It's the time of day. It's this. And they'll come up with 15. I was like, OK, so how are you going to ask somebody about all of those 15 and help them make a plan? Because to me, that sounds like a lot of information and you're going to be like, you can't make that decision. And realistically, there's three things that make all the difference. Number one, what's your glucose level when you're starting? Number two, is it moving up or down?

[24:25] And then number three, how much insulin have you got circulating in your body from the previous insulin dose you've given the last eight hours? If you know the answers to those three things, your risk, you can give a plan very simply to moderate your risk of going low, extremely simply. But unless you've lived it, unless you know the data and unless you kind of have a frame for less is more, i.e. via negativa, focus on the things that make the biggest difference.

[24:49] You're going to end up giving output saying, oh, well, there's these 15 things that you can do and you need to take account of all of them. You don't. You just need to take account of three things, and you're golden. So we wanted to make sure that when people were putting questions in, it was only going to give education around the things that really moved the needle. And then the final bit, which is Nassim's last book, which for me is so important with the framing of this is called Skin in the Game. So essentially, you should not be giving information and prescriptions and exact advice to people who you don't know. So we absolutely were so keen and it's all over the consent forms and everything.

[25:33] It's an educational advisor that for 20% of the information will get you 80% of the way there. The other 20% is for a human to human conversation, for self-discovery learning, for an interaction with someone who knows your context, who knows your family, who knows your social history, who knows what your tendencies are. AI cannot do that, but it can get you 80% of the way there like that. The other 20% is that's what human beings are for. Like get AI to help you get 80% of the way there. And then as a clinician, that's amazing. It means I can sit there and talk to you about what's going on in your life, about what other things other than diabetes are affecting, what's going on.

[26:14] But if you can't get 80% of the way there, you spend the whole consultation trying to get to 80% and you miss the most important part, which is speaking to the human being in front of you. So we wanted to be clear from a skin in the game perspective that no one should take any of the things for prescriptive advice. They should take it as an educational advisor that will give you what the average person with your insulin or your weight would suggest from the consensus guidelines to a starting point. But do not take it as prescription advice and go and learn with someone who knows you. And similarly from a healthcare professional perspective, I could imagine people will look at Grace and they'll go, it's going to take my job.

[26:57] It's got all that stuff in there. My perspective is completely different. It's going to be like, this should make you amazing at your job because you don't have to nerd out on the intricacies of each algorithm for these hybrid closed loop systems, because this thing will just tell you what the primary levers are, give you a directional nudge of what you can consider. And then you can speak to the human being in front of you. So the people who are good human to human communicators will love it. The people who are not good human-to-human communicators, they may feel like, oh my God, I've been hiding behind this mask of knowing all this detail for so long.

[27:30] Now I've got to speak to the human in front of me. I'm a bit uncomfortable with that. So there may be a bit of that. I will, we'll wait and see. But hopefully that gives you a framing. The base layer is all about the grounded truth from the experts, from the consensus guidelines. The delivery layer is years of lived experience and how to translate that. And then on top of that, we've got a philosophy of type 1 diabetes is a black swan condition. So precautionary principle for anything that comes out.

[27:58] You've got to have a framing and an understanding that you need to look after your own backyard, and that it cannot be giving advice is very important as well. And then most importantly, what are the two to three things for each question that will move the needle the most, not give 15 because then no one can process 15 and they don't need to process 15. So, it's kind of almost been, me and Phil had a bit of a laugh and sort of said, it's like, it's the 20 year overnight agent, that's kind of just appeared.

[28:32] You've got my clinical experience of 20 years and you've got Phil. I'm talking all this like I know what I'm talking about, but Phil is the person who's developed these structures, these wiki-on-wiki structures, and developed this database RAG system that ensures that when the questions come in, they get to the right bit of Grace's brain quickly, it's connected to the right places, and then the output that you get, number one, is quality, but number two, is affordable moving forward, because I've got a few stories about how much money I spent unnecessarily for the first few months.

[29:06] But if I didn't have Phil with Phil guiding me, it would have never made it because it'd be too expensive.

[29:14] Arseniy Arsentyev: We'll definitely touch on you putting everything, putting your skin in the game as Nicholas Taleb says and three layers of Grace sources, your distillation philosophy that you put under it.

[29:28] John Pemberton: Well, I guess the final thing is we have a filter. So we have an agent that goes out to Google Scholar and all the areas that Grace covers, pulls the papers in for the week and there's typically about a thousand, like there's way too many. There can't be that much good research. So then we have a filter of Grace critical appraisal who then appraises these papers and filters it down to 10 and then I choose the five that make it in. So there's only five papers a week that make it in because again, from that principle, we could stick a thousand in, but a thousand load of crap is not worth anything.

[30:01] If you can filter it down to the five nuggets that add something of value to an already valuable source, you're going to incrementally compound a very, very strong and robust system. If you just throw everything in there, which is the temptation, you're going to end up with such a weak signal coming out and such a messy situation. You've got to protect and bound your model if you believe in it and you should be the filter, not any, not an AI shouldn't be. It should get you like the 10 best and then you pick the five best. Don't try and wade your way through the thousand. Let it do that work. Let it get you 80% of the way there.

[30:38] And that final 20% of you, it's on you to make sure that you get the right stuff in your model.

[30:44] Arseniy Arsentyev: Yes. If you pour all thousands of papers in, Grace would be like, no, just fight against diabetes. You can do it. You can do it with your hypnosis.

[30:58] John Pemberton: Yeah, exactly.

[31:00] Arseniy Arsentyev: And extract the cells. Speaking of sources, I've seen that you did this classification of sources, which is such a great idea. I worked with it in finance. For every financial folks, you know that bonds are classified like AAA, which is the most riskless type of bonds. And down to the D, if I'm not mistaken, which is just a junk bonds. And you have the same for the sources. So as you said, this regularly updated every two to four year sources are an A level. And can you tell us more? What kind of sources on this grade scale?

[31:43] John Pemberton: Yeah. Within research, you have A to D. So A is either a systematic review or, very simply, a randomized control trial is the best that we have, which is essentially you get two groups. They're ideally as close as possible as you can be. And then you give one group the intervention. And then the other group is exactly the same apart from the intervention. So you follow them over a number of years and if they end up with different outcomes, ideally, if they're the same, you can basically attribute the difference to that one intervention. So that is kind of like the gold standard, but that's expensive and sometimes messy, especially when you've got a human condition going on, like matching. It can be quite difficult, but that is the gold standard randomized control trial.

[32:33] So randomized control trial counts, but what you really want is replication within science. So anyone can do a randomized control trial. And if you look at Nassim Nicholas Taleb Fooled by Randomness, you can get a random randomized control trial that just pumps out a result, which is unfortunately gives you a positive finding when it shouldn't have done. But if you've got 10 randomized control trials on one topic and they've all got pretty much the same signal, that gives you a high level of confidence that whatever it is working consistently and robustly. So that's what a systematic review is. It looks at all the randomized control trials on a particular topic. It puts them together and it says on the balance of things, does this thing really work?

[33:12] So you've got systematic reviews and randomized control trials sitting at A and then B is typically not randomized control trials, but control trials. So you haven't randomly put them in. You've kind of split two groups and you give one the intervention and one not. And if you've got quite a lot of signal, like a strong signal there, you can be quite happy with the B evidence. But if the signal's a bit weak, you're a bit like maybe not. There could be a bit of randomness going in.

[33:37] C is observational trial. So large registries, big data sets, which I actually really like a real world data for one particular reason. As long as the data source is big enough to pick out a generic signal that then can be tested in a randomized control trial, because randomized control trials in diabetes are typically white middle class people who are very switched on, very health conscious. So you're bound to get a good result of them following whatever it is you're supposed to do. So it's good for showing efficacy that something will work, but effectiveness is very different. Efficacy is will it work? Effectiveness is if you give it to everyone, will it still work on average for everyone? They're two very different questions. So randomized control trials are good for efficacy to prove.

[34:21] But if you want effectiveness, the real world data kind of tells you when you give it to everybody, do they get the benefit? And an easy way to explain this is if you look at all the randomized control trials for these new hybrid closed loop pumps that adjust the insulin without you thinking about it: in the trials, the metric that we use is time in range, the time in the glucose range that you would like. They're typically in 75, 80 percent in trials. You chuck this thing out on a real world across the board data set, which we have done in the UK with children. The average time in range is 65 percent. It's 10 percent lower.

[34:55] Now, that's still better than the 50 percent you get on the other therapies, but it ain't quite 75 percent. So the real world observational data gives you a better frame for, on average, for the people who you're looking at, what result you're going to get. So it's understanding the levels. And then grade D is we don't have any randomized control trials. We don't really have any open trials. We don't have much real world data. What do the experts think? So that's down to anecdotal experience of individuals. Now I've been on some of these groups and you have some very powerful noisy people on there who've got a strong opinion about something.

[35:31] So you have to be a little bit careful about grade D evidence because it's open season for people who are very bullish about certain particular topics and they may be right, but they won't necessarily have the data to back it up. So we've graded the educational recommendations based on A to D and so that people can get a frame for, okay, Grace is suggesting this, but it's a grade D. And she's already said it's grade C, grade D. It's the best we've got. Kind of go and speak to your team. Or if it's grade A and it's been proven time and time again, like hybrid closed loop system, she'd be like, look, this is kind of the evidence. She's very, very strong on this topic. You want to go speak to your team.

[36:13] This is probably what the evidence would suggest is the best approach to consider moving forward. So again, when you're trying to distill a lot of information, which keeps on growing and growing and growing, you need a way to be able to discern what is kind of like closer to the truth. A bit closer to the truth and a bit more like D. I'm not going to base a strong decision on a D, that's for sure. So, yeah, the grading system, like it sounds like similar to the previous work that you've been involved in. You need some way to allow people to understand. I think A, B, C, D makes sense, doesn't it?

[36:51] Because like A is obviously the best and D is a bit like like if you've got a D at school in maths or something like you wouldn't want that person to be your accountant.

[37:03] Arseniy Arsentyev: Absolutely. Let's keep your distillation as a second layer of Grace because it involves a lot of technical computer science details and proceed with the philosophy. You said that you positioned Grace not as a medical tool and not with exact numbers. Take this pill, go to swim 20 kilometers, go to an ice bath and you positioned it more as educational colleague, I may say, educational mentor in the space. Could you tell me why you decided to proceed in such a way?

[37:42] John Pemberton: Yeah, well, first of all, from a simple legality and cost perspective. So if you want a medical device that gives prescriptions, that in CE marking from Europe is a class IIa or a class IIb device. And to get that through, it's like half a million pounds. So on a simple level, that is just like, it's impossible. It's impossible. But also for me, from a skin in the game perspective, very, very important that I don't want people using Grace to think that Grace knows them. And we've purposely done it so she cannot, there's no memory. She can't, you can't go in and tell her about you and she'll remember you.

[38:24] That's on purpose because it's impossible for an AI agent to have a connection with you that is genuine and real, that understands and empathetically can see in your eyes. You might be saying one thing, but actually what you're thinking is something different. I've sat in enough consultations where someone's telling me something and I'm looking in their eyes and I can see their face and I know that it's not true. So I will then change my approach to that individual with the questions that I ask from the frame of the discussion. And that can only be done by a human who can see and pick up those signals. That's not, it's not a tangible thing you can program. You cannot program that.

[39:05] It's like, you're looking at me now and I'm doing all these weird gesticulation things. You know that I'm into it. An AI doesn't know that you're into it. Like kind of, they can't get that realistically. So it was important to me that the message is clear. It's something that will get you 80% of the way there. It cannot get you 100%. You must have a human to human network. However, that is for you, whether it's other peers with type one, whether it's your healthcare professional, whether it's your mum, whether it's your dad, someone who knows you, who can see you to help guide you on that self-discovery of 20%. So there's kind of like a cost implication, but there's a more important framing of if you've got a good healthcare professional, they will know.

[39:44] And that 20% is the most important bit and that can only be done by guidance, support or self-discovery. So it's just so important to me that that's the case. And also from like a Nassim Nicholas Taleb perspective, if it was prescribing something and then someone dies because they followed that or someone gets seriously, seriously injured, I don't want to be the person who goes, oh, well, I've got insurance, so it's okay. That for me is such a cop-out and I think it's absolute bollocks. Like people who kind of hide behind a limited company and directors insurance or maybe even go online as a person, an influencer or whatever who's saying, use this, use that, I think it's great for everybody. It's like you don't know them. You don't know their individual circumstances.

[40:30] How can you be suggesting to do that? You're like, oh, I'm covered because I'm covered by public liability insurance or whatever. It's like, what? Don't give a shit about that. It's like, are you a human being? There's another human being who's going to be following your information. Are you happy that human being? Do you really feel comfortable about that? That for me, I am not happy with whatsoever. So it's important for me that that always will remain the same. That's why we set her up. She will not have a memory of you. She will not allow you to create a project and kind of dig into your personal things.

[41:01] She might allow you to set up a project for informational purposes only, but you won't be like, I've, got this and that and get the perfect plan for me. It doesn't allow it to do that. And that's very, very purposeful. And I know a lot of people said, oh, it'd be great to have like, it's given your prescriptions. They'll be like, you might think it would be, but it doesn't know you. It's never going to know you. So take the 80% and be grateful and then go and do the 20% with a human being. Because like for me in the future, I know we'll talk a little bit about at the end.

[41:30] AI for me should be about making human beings, allow them to get 80% of the way there for amazing things. And then sparkle with your 20%. Whatever makes you, whatever makes your amazing things about you or human beings, like what a chance to kind of like really show that off without having to do all the grunt work to kind of get 80% of the way there. That for me is such an important piece of Grace. And I'd hope that healthcare professionals, some won't, some will think it's kind of like too far, but I would hope that most would see it as an opportunity: that they don't have to learn the nuts and bolts of every single system, of every single guideline, to keep up with the nerds like me, for whom it is their life.

[42:14] They can actually be better than me because they can get that 80% and they're probably a better communicator than me. So they can actually be better for the people who they see. For me, that's a win-win. So yeah, a real, listening to Nassim Nicholas Taleb, and I think he tells the story about the 2008 financial crisis of how people were selling these like mortgages upon mortgages upon mortgages and kind of like going to the hill. And the people at the top knew what was going on. And then when the crash came, they were the people who'd been claiming bonuses of 15, 20, 100 million pounds. And then when the crash came, did they pay any of that back? No, it was the US government who paid it back.

[42:55] They were still off in their Hawaii and whatever things claiming, hey, we couldn't see it coming. Well, it came and you've still got 100 million quid. Like Johnny down the road has lost his mortgage and his pension. Are you going to give him any 100 million? Oh, no, no, I've earned that. Well, you've earned that by putting them at risk and you're now hiding behind. I've got a limited company and we're kind of like behind this insurance policy. That's just bad human behavior. That's just kind of like you wouldn't do that to your neighbor. So don't do it to someone over the other side of the world just because you don't know who they are. Like that for me is just like poor human beings.

[43:27] So it's very important to me that people will never approach Grace and be like, it's going to tell me what to do. No, it's going to get 80% of the way there. 20% is on you and your network. So yeah, I think if I wouldn't have read Nassim Nicholas Taleb, there's a chance that I would be like, we'll push it more towards prescription and we'll get people really into it. And we'll like, we'll charge them loads of money. But then when you kind of like you read the stories about 2008 and what it meant to people's lives that it destroyed.

[44:01] And there's human beings on the end of that and it's just like, this is not okay. So yeah, hopefully that gives you a bit of a frame of, I'm really strong on that because I would rather make one pound and help loads of people and get good educational advice than become a millionaire doing this prescription thing and then someone eventually has a problem. But by that point, I've already made loads of millions of pounds and sod them, they've had the problem, that for me, I just couldn't live with.

[44:28] Arseniy Arsentyev: And you work with six, if I'm not mistaken, six major companies that are in diabetes space and you have this ranking of sources. How is input from such huge pharmaceutical concerns influence the Grace's behaviors? These are philosophically challenging questions that I want to ask you.

[44:54] John Pemberton: Yeah, no, it's a really good question. So we don't take sponsorship. We only take educational grants. So when you sign an educational grant, it says on it quite clearly, they're giving you the money to create education. So they're not allowed in theory to influence. So they just give us the money and I will go to them and say, this is what we're planning to put into the content for decisions, et cetera. I'm up for a debate. I'm up for a debate because you know your product really well. You know your data really well. But don't think because you've given me money that you're going to influence my decision because at the bottom line, we've signed a piece of paper.

[45:27] So you've got to put trust in me that you believe that your product is good and it's good for people with diabetes. In that case, I'm happy to have a discussion. So we won't take any money. So for example, on a CGM perspective, there's five simple questions about when this device was tested. There should be five simple questions about the study design, which will tell you whether it's been tested effectively. And of the ones in Europe, of the 14 available, only five meet that criteria.

[46:03] So they're the only five that we talk to. We have other people who come to us and go, we'd really like to be in Grace or on your website because a lot of people go in to see which the CGMs are. How much does it cost to kind of sponsor a podcast? You haven't got enough money. You haven't got enough money for how much it would cost you because it's priceless because I cannot have someone come in to my website or the app and they go in and they get recommendations to try this thing. And now I don't know what the risk level is. It may be an amazing product. It may not be. But if you can't provide me with the data so I can understand the risk, then you don't get in.

[46:42] It's as simple as that. So the data determines whether you have a conversation. And then if you have a conversation, then we'll have a conversation. If you want to put educational grant money in, then great. And if you don't, your products will still make it in because it's good. So there's products that don't do an education grant with us and they get the same input as the others. Now, the speed of the input might be different. So, for example, if you have a company who's given us an education grant and we've got a communication pathway and they've just released a trial, which is really important and really impressive and they tell me about it, then, of course, I'm going to feed that information through. Because why wouldn't I?

[47:21] But if I don't have a communication portal with someone and you come to me cold about your product and you haven't given an educational grant, it's probably going to get in, but it might be a bit slower. And that's probably the level of influence is what I would say is. The good stuff from the best companies gets in quicker because it's a relationship. The good stuff from the other ones, it will get in there. There may just be a slower cadence of how quickly it gets in. So, of course, there's going to be some bias and that's just business. But I would like to think that that's a reasonable way of doing it, as opposed to taking sponsorship, which then has a problem if you rely on that sponsorship for how your model works.

[48:00] So that's kind of hopefully gives you a frame. There's of course going to be some influence. And there's also relationships like there's certain individuals within the companies that I really get along with who are a great laugh and I'll go and have a beer with them and you know, be brilliant. But I used to work in industry. I used to work for Medtronic. I've worked in the private industry. I know how it works. I know when people are trying to butter you up. I don't care. I'll take your free drinks, but I'll still be there when you come back like just because you bought me a few free drinks doesn't mean that your stuff is going to get any faster because I've worked on that side. I know how it works.

[48:29] So I think that maybe people who have had interactions with me and seen the research that we've written and the papers that we've done and the level of honesty that we've done will realize that I'm not up for sale basically, but I'm up for collaboration with good products and I'm up for collaboration with products that will really help people. And of course, I'll be happy to take your money and fast track your stuff through if I think it's good stuff. But if I don't think it is, I'm not interested. So like that's kind of the philosophy behind it. So it's a little bit based on trust from their perspective that I'm going to not misrepresent what they believe, but that's where a conversation and relationship comes up with.

[49:09] And I think I've maybe earned that right over 10 or 15 years of being involved in research and guidelines to have these conversations with them and be frank with them. I don't care whether you're the vice president of global affairs for Medtronic. If I think you're talking shit, I think you're talking shit and I'll tell you. I don't care how much money you earn. I've got nothing to lose. And I think that type of attitude is like, I don't need your money, but I'll take your money. And if I don't think you're actually got the right information, I'll tell you I don't think you got it. I don't care how many PhDs or how many whatever you've got behind you.

[49:41] I've been on guidelines and had the same conversations with scientists, kind of who believe they're right about certain things. And often I'm happy to be wrong too. So I'm also happy to change my opinion. Strong convictions loosely held is the way to approach it.

[49:57] Arseniy Arsentyev: Great mindset. And that is what people do not understand about, let me say, personal projects. I have tomorrow's medicine. That's a purely my intention to do this. I just I'm a computer scientist. I have a personal incentive to make it through and to meet new fascinating people to bring their perspectives to the world. And as I know, you have crazy stuff going on because this educational grants has not yet reached you and you're paying out of your pocket. And it's not 100 bucks. Let's be honest, it's not 200 bucks. AI now is expensive. Sources are expensive. Computation is expensive. And let's dive into this topic. It's a little bit off tomorrow's medicine side, but it's a very important human side. And I want to hear you explain it to the audience.

[50:54] What is the mess going on with?

[50:56] John Pemberton: Yeah, so I'll give you a bit of a journey. So if you think of just a basic WordPress website for a year, I don't know what 200 quid, maybe something like that is kind of low profile, not expensive to run a simple website. But then as the pages go up, and then as your ability to want to allow people to have a positive experience, i.e. not take ages for the information to cache, etc., the cost starts going up and then you're looking at like a grand or something like that.

[51:26] And then you start to do a podcast and you're like, I want to do a podcast and I want to have all these cool widgets that come in and make it nice and cool. And then you've got your Riverside subscription and then you want to get someone who's helping you to come in and do all that as well. And all of a sudden the cost is like creeping up and then you've got a microphone which costs like 700 quid and you've got a camera which costs a grand. And then you've got the cost of the podcast episodes and then you've got editing things and you've got your own personal time. How much do you charge for that?

[51:56] Which obviously you don't, because you do it because you love it, but ultimately that's a charge as well. So without the Grace stuff, probably to run The Glucose Never Lies and the podcast on an annual basis, I want to say is probably in a region of £10,000, something like that, maybe around there, something like that. And you think of that over like five, six, seven years. So I've been able to finance that because I do talks for companies and different entities that pay you to go and do their talks. So I would probably earn and break even on that. So maybe 10, like I've got my job at the NHS, which is just pays for the bills at home.

[52:33] And then you kind of got your side project, which is paid for by going doing these talks. I won't make any more for these talks. That is just to finance the platform behind it. And then you get to AI. So obviously you kind of build this thing. And then you realise when you built this kind of bounded model, which you think is great. And then you start doing queries on it. And then like the cost per query comes out way more expensive than you're anticipating. And also as you're building it in the background, the cost of building it is also expensive depending on which structure. So when we've started making it and Phil was like, okay, get yourself an Anthropic account and get yourself like an Anthropic Max.

[53:12] And then we'll kind of start building these things in the background. And then you're like, you've got all your structures that you're building. And then every time that you kind of get a new idea, you've got to change the structure of it and you've got to change how it's presented. And the bigger your kind of model becomes, all of a sudden, I was like, oh my God, I've just burnt through a thousand quid in a week. Oh crap. Now that's £2,000 in two weeks. And I haven't got, I'm not at the model that I want yet. So now I've got to go with Phil and I was like, Phil, this is going to bankrupt me personally within about three weeks. He was like, okay, now you're in my game. You're in my game.

[53:49] So he works in basically what we call token management. So essentially, I had what we call a flat structure. So I had all this information that's just flat across the way. And Grace is trying, Anthropic is trying, Claude is trying to mash all this together to come up with answers. And it's so expensive because he doesn't know where to look. So Phil was like, okay, there's two approaches. There's kind of like, if you think of like a wiki file, which is kind of got like structures and tree structures, we can stick things on there. But then once you get to a point where that becomes expensive, you can then do database approaches with RAG style.

[54:21] So we've kind of now gone from this really flat structure to a wiki with RAG and these kind of cost per quotas that go into things. So it's like, I'm talking about like, I know about it. I kind of got this. I would say I'm about like a 60% the way I understand it. Phil understands it to a T. But essentially we put this in place and then we reran the validation of how much it would have cost me. How much of that two grand have cost me if we'd have done this from the beginning? And the answer would have been 50 quid. It would have gone to 5% of the cost.

[54:50] And it was like, if I didn't know that, I would be gone and bankrupt in three weeks time because I couldn't afford to do it. So that now is now we've built the model. So we kind of just stopped there and I was like, I can't afford to carry on like this. But then we've kind of done it. So that level of work now costs significantly less so we can carry on. But if I didn't have that technical expertise by the side of me, I would be gone already. Like Grace would be disintegrated because it'd be too expensive.

[55:18] So, yeah, I would say to anyone who's thinking about building a model, if you've got the ideas and you've got the knowledge base, but you don't have the architecture, the best money you can spend is finding an AI developer, number one, who's interested in your project because they personally like you, i.e. it's my best mate. He was the best man at my wedding. I was the best man at his wedding. So he's doing it for me. He's like, like I paid him, I think I paid him six grand or maybe 10 grand to help me. That should have cost 150 grand.

[55:50] He's done it for me on mates rates at like 5% of the cost. I wouldn't have been able to build it. Otherwise, it's like simple as that. But he's doing it because obviously now he's part of the company and he's got skin in the game. He's part of The Glucose Never Lies. Of course, that was going to be part of like, I'll pay you a tenth of what you would have got paid. But I'll also give you like 20% of the company. So are you in or are you not in? So he's got skin in the game with me.

[56:13] And I've obviously got skin in the game with him. But without that, it wouldn't have been happening for sure. So we've kind of got there through me. Probably Grace now owes me 20 grand, I reckon. And I reckon over the years through contributions from getting from the companies and education, well, the personal earnings from the companies about 50 grand and then about 30 grand. So it's probably about 100 grand's worth of money, plus Phil should have cost another 100 grand. And then we had a very fortunate interaction where a member of the type 1 diabetes community who's a VC, he read my drugs and alcohol guide and was like, that's amazing.

[57:01] There'll be people who want to go out and get smashed who are going to read that and will be able to do it in a much safer way. And I see you've got this model. I'm going to give you 24 hours with this thing that I've got, which is a 10 year data set of people's CGM insulin data. And you can test your models, you can test your algorithms on it and see what you get. And he allowed us to do like 33 tests and of those 33 tests, 29 came out suggesting that the model that we had was very good with some slight adjustments. And that's just another bit of fortune that would have cost like 150 grand for me to access that database.

[57:37] But it was like, because I like what you're doing, I'll let you do it for free. It's like, okay, you've got 24 hours. So that was caffeine, whatever stimulants I could get to kind of like hammer down these things. I was broken by the end of it, but I was like, I'm going to make the most of this opportunity to get this validation work done. So it's a bit, it's just like, it's pure fortune, like work for seven years, pure fortune of my best mate being an AI coder, pure fortune of having this person who read my blog and kind of like lent a hand. And then for me, it's about now trying to get a sustainable model moving forward to give back for people with type one diabetes.

[58:13] So it doesn't matter whether you live in Tanzania, it doesn't matter whether you live in New Zealand, doesn't matter whether you live in China. Within your pocket, you will be able to ask questions about type one diabetes that will give you 80% of the way there very quickly, very efficiently and delivered in a way with the warmth that Grace offers.

[58:31] Arseniy Arsentyev: Okay, and well, you just named a crazy amounts of money. And do you have a plan in mind how to get any return on the investment? And does the educational grants from the companies cover the work on Grace or is it a separate realm?

[58:49] John Pemberton: Yeah, so the personal timing, I reckon, oh my God, it depends on how much price in my hours. Because if I price my hours at £10 an hour, God, it'd probably owe me about 200, 300 grand. If I price my hours at like £30 an hour, which would be cheap still, it'd be about a million quid. And if I priced it at market rate, it'd be about like £4 million. So like if you ever get involved in a personal project, you can just forget your own time. You're doing it because you believe it's going to be helpful for people. But I think we've got to a model now where it is good, well, very good.

[59:24] It is cheap enough to be sustainable. And then the question comes, okay, so if you want to get this for free for people with type 1 diabetes, how are you going to do it? So there's quite a few ways. The first one will be, it will be free for people with type 1. But if there are people who access it and they really like it and they want to pay it forward, they can give a charitable, not charitable, a donation, which is not for the subscription. You can have it for free. But if you donate, then we'll give another license for free to somebody else. So it'll pay for yours and somebody else's from the Global South. So that's one way.

[1:00:02] If people in the community really like it, if they go to the Buy Me A Coffee thing, that £5 will be registered. Then another one will be freed up for the next person who's on the waiting list. So that will be, so if you're in and you then pay a fiver, then someone else, the door opens for someone else. So the community can like self-fund that way if they want to, if they find it valuable. There's no pressure. It's completely their choice. And at the moment that they retract that, then that license disappears. So that is like a revolving door policy.

[1:00:30] The second will be if companies or charities think that it's really useful, they can sponsor however many for a fiver at a time. And for every fiver that they chuck in, we'll chuck in one for free. So if they buy a hundred, we'll chuck in another hundred for free. So if you've got charities, if you've got medical device companies, if you've got individuals who've got quite a lot of money with type 1 diabetes who would like to sponsor like a thousand, then of course we'll just match it. And obviously that'll be a bit of a downer for the people when they no longer have access to it because that sponsorship sort of gets reduced. But if they've really valued it for five pounds a month, they can probably carry it on themselves.

[1:01:12] So it'd be up to them from there. And then the other one is we've approached kind of like big, big organizations who've got a lot of money, who are funding cures for diabetes and various different things and saying, look, you're pumping all this money into hoping that there's a cure. That's great. How about you give us a portion of that money to support people who've currently got the condition? Number one, understand about it, but we've also got a Grace Max as well. So Grace Max is kind of like the off the chart, research assistant, manuscript writer, ideas generation, content creator. So if people want to purchase an expensive per month account, they can do and that will, they will be able to do like amazing things with it.

[1:01:58] But researchers, for example, me, I've got Grace to write me my last two papers and they are so much better than like what I would do. But I have to be the person who filters in. This is the idea. This is the structure. These are the evidence base that we're going to look at. These are the diagrams that I would really like and the figures. And this is the framing I would like. I would really like this kind of framing and to come from this angle and really set up this nice prompt. And then all of a sudden it spits out this 4000 word thing. I would like it in Diabetes Care.

[1:02:32] The author guidelines are in there. So I want it in Diabetes Care and I want it delivered. And I actually, this is my style. I like to write like this. This is how I want it to come across. Short, sharp sentences, punchy, with evidence base kicked in. And then what it spits out gets you like 80 to 90 percent of the way there. And then it's on you to make it sing with how good you are. Having written papers before, it is a painful experience. It's like you get the feedback from your mentor and you get hammered. So my mentor, so before I sent her the one that Grace did, I didn't tell her that I've got Grace to do 80 percent of the work. And then I did the final 20.

[1:03:07] But I sent it to her and she was like, what's this? You usually serve up like dog crap for the first one. And I've got to kind of like coach you into getting it there. But this is better than I could have written. How have you written this? I was like, well, let me tell you. So do you want now me to create a Grace persona for you? Here's your account. Off you go. So if people like really want to do some amazing stuff. And if you're a researcher, if you're a marketer, if you're someone who really wants diabetes stuff, you can have the Max account and pay obviously quite a bit more for that.

[1:03:41] But so that's potentially some way to fund it. But what's actually happened recently, which has kind of blown our minds a little bit, me and Phil, is we had a conversation with a couple of manufacturers and they were like, you have got a grounded source of truth there, which is amazing. And we'd really like to tap into that to help educate our company. So we kind of like we sat there on it and chewed on it for a little bit. And at first I felt uncomfortable because I felt like I don't want the purity of Grace as an educational advisor to be tainted by doing dirty things for commercial companies. That was my first thought and immediately it just made me feel a bit sick.

[1:04:26] And I was like, I don't think I'm down for that at all. But then I reflected on my experience at Medtronic and I sat there in the educational meetings. I sat there in the marketing meetings and I sat there in the end of year reports and the shit that I heard people talking and the terrible marketing and the incorrect information around from the from the medical affairs people that was passed down. I actually thought, do you know what? Ethically, if we put the Nassim Nicholas Taleb filters of Black Swan first, Via Negativa, Skin in the Game, we can create agents for these companies that will be a marketing agent, will be a compliance agent, will be a medical affairs agent, will be a education agent, will be a ethics agent.

[1:05:20] And we can create these agents that communicate with Grace. So they can't influence Grace, but they can be a set of prompts, basically a really good employee who can go to Grace and pull out the information. And yes, we'll work with companies to create these agents in the way that, let's just say, I'm going to call it company Omnia. So how, Omnia is a new diabetes company, a CGM company, a medical device company. We'll go and speak with them and we'll say, OK, tell me about your compliance. Where are you going to go? You're going to go CE marking, FDA. Where are you going? OK, right, great. We know that. We already know that. But how do you want your particular compliance? And some, compliance in some companies are different.

[1:06:05] OK, we'll feed that through our ethics and see whether that's OK. Great. Marketing. How do you want to market your product? Are you someone who likes to market to the very strong emotional component, to the individual with diabetes, or are you more of a like towards a health care professional? OK, brilliant. We'll put your flavor in there. OK, education. How do you want your education to be? Do you want your education to your patients and to be like this? OK, sales. What's your sales philosophy? Do you work on the challenger mindset? Do you work on the insight selling or are you more of a relational seller? How do you want your sales reps to be? Or do you want them to choose that?

[1:06:38] Do you want them to be able to plug in and choose that? So we can put all that in for the companies and then say, right, do you need all those marketers? Do you need all those sales trainers? Do you need all those compliance people? Or do you need 20 percent of amazing human beings who dictate and adjust these agents and therefore you get their 80 percent done brilliantly? And that's the way the future is going. Now, people don't like that. Like some people are going to go, you're going to take away jobs. But the bottom line is it's coming, whether you like it or not. Like this is going to happen for all marketing jobs or sales jobs or compliance jobs. These agents are better than you on your best day.

[1:07:24] But what they're not better than is defining the parameters, defining the model, being that influencer who kind of gets the best agent. They're the 20 percent of best people whose jobs are absolutely safe. They're going to dictate these agents. But the people, the 80 percent of people are not very good at their job or who are just followers in those particular professions. They're gone. They're going to disappear. So I'm not saying like people might think that that's a very crude mindset from me to say that I'm happy for those jobs to go. But whether you like it or whether you like it or not, it's coming and it's going to happen.

[1:08:02] So why don't I be a person with a diabetes perspective who helps that happen whilst making money so that people with diabetes get Grace for free? The money that we make, I'm not taking, I'm going to take a portion of it. Like I'm not saying that I'm going to live a squalid life. I will take some money for it. Yes, but I'll take a fair portion, but I will push through those licenses for free so that people can get educated. And if it means that a few people in private industry lose their job a bit sooner than they would have done anyway, well, I would imagine that people with type 1 diabetes don't give a shit about that because they have got a tough life and they would like some quality education.

[1:08:42] So I'm happy to take that ethical position and say, if there's a few markets and salespeople lose their job, well, if it means that people with type 1 diabetes get great education, I'm willing to sit there and take that on the chin.

[1:08:55] Arseniy Arsentyev: John, and I mean, if we do this agency in a Nicholas Taleb way versus the usual big corporation Armageddon way, like break everything, they will still fire these people. And freak everything up.

[1:09:16] John Pemberton: Yeah, we've already had three very good conversations. One with a huge, huge 25 billion turnover company. And we've had this conversation. I can see their eyes. They're like this. We're going to save loads of money on these people. I'm like, okay, that's fine. But if you're going to use us, we're going to do it the Taleb way. And second of all, we're only going to do it for products that we believe in. So, for example, I would not do continuous ketone monitoring through this process because I don't believe on the balance of things, it's number one, a good product. I don't think it'll help enough people. And I think on the balance of things for most people, it will be a signal that will be more annoying than helpful.

[1:09:59] And the same for continuous glucose monitoring for people in the wellness space. The signal-to-noise ratio is terrible. So you'll end up with people who are making decisions on these glucose numbers that are actually not accurate enough to be making those decisions. However, continuous lactate monitoring is the canary in the coal mine 10 to 15 years before you're going to get metabolic disease. That is a signal that is worth paying attention to. If you want to, use Grace to explore that. To get marketing agents, compliance agents, education, etc. I'll help you because I believe that that is going to be good for people. So I'll help you do that. But if you come to me with an idea to try and make money and use Grace to kind of supercharge that, piss off. I'm not interested.

[1:10:46] Not interested in your money at all. If it ain't going to help people, and it ain't going to go through a Taleb process, of we'll help you, but in the right way and in a fair way, we're not going to help you if you're just trying to make money off a product that ain't going to help people. Not interested.

[1:11:04] Arseniy Arsentyev: Last but not least. Let's talk about tomorrow's medicine. And let's forget diabetes, glucose, Grace for a moment. And look at the bigger picture all throughout the world. Throughout the US, the UK, other countries. And what do you envision? I'll let you paint two pictures. The first picture is what do you think will happen in the 10 years? Let's say 10 years. And the second picture is what do you want to happen to the medical system at all? Can you speculate on such topics?

[1:11:45] John Pemberton: I guess from a pure Taleb perspective is like I have got no clue.

[1:11:52] Arseniy Arsentyev: Nobody does.

[1:11:53] John Pemberton: What I'm about to say now, take with a pinch of salt and laugh because speculation on this sort of stuff. What I knew about how to create an AI agent two weeks ago, literally two weeks ago, is a joke to what I know now, because I've already built stuff which I'm blown away with, which might mean in three weeks' time, what I can build will be so much better, and so ridiculously a joke to what I thought two weeks ago that it's like embarrassing.

[1:12:26] So what I will start by saying is I have got no clue. But what I do know is it's going to happen fast. It's going to happen extremely quick and it's going to catch most people completely unaware. They're going to have this job in marketing and sales and AI programming. And in three weeks they're gone. So as an example, there's one major, major online retailer who got their best AI programmers to put everything into this AI agent. And they all enjoyed this project and it was completely brilliant. And then they sacked them all because this AI agent took over all their jobs. And all that remains is the 20% of people who are dictating these AI agents.

[1:13:09] So as I described to you with what Grace might be able to support with these agents for these medical device companies and the jobs in sales and marketing, that's what I would foresee: the best 20% of people are going to be safe, because they're going to work with companies like us, with companies who are better than us, who know miles more than us. And they're going to be able to create these agents which do those jobs so much better, so much efficiently, and so much more in the way that that company wants them to do it. And those jobs are going to disappear. So that would be my short-term view. Not like 10 years, that's like three years.

[1:13:48] For the companies that really are kind of forward-thinking, their employee structure will be changing rapidly, especially in the resource generation, dossier creation, coming to market with compliance stuff, like all those jobs disappear. What won't disappear, in my opinion, is the good human communicators, the relational people who can look in your eyes and know when you're talking shit, know how to press your buttons and know how to structure a conversation that allows what AI has generated to sing. So the communicators of this world, the relational people, that is the currency.

[1:14:28] That's the educators. That is people like you. People can have a communication, they have a conversation, can ask questions, can press someone's buttons, can kind of find out what is the secret sauce to human being communication. Like that is the currency that is going to thrive. And that's teachers. That's great nurses, great doctors who can communicate, we're going to thrive. Information jockeys, you're gone. SOP-driven people, you're disappearing. Your job's, well, you're not disappearing, your job's disappearing. The question is, can you adapt to that? It's not, don't sulk about it. It's just the requirement of human beings to adapt. And the people who can adapt the fastest are the ones who are going to thrive. The people who are going to sit there and go, do you know what? It used to be like this.

[1:15:24] It used to be so good. We had it so good. It's like, well, go back to the farm. Go back to the cave. Go back to that point in time. Because those days are gone. What's going to happen is the people who can adapt fast, the people who can discuss, the people who can talk, the people who can have a relationship are the people who are going to be okay with this. The people who are going to whine and moan about the past of when they used to be there. People used to come for me because I used to know my stuff on the technical bits.

[1:15:53] That's not going to be needed anymore. So that would be my initial short-term speculation.

[1:16:04] Arseniy Arsentyev: The first picture. And what's about the second picture, what you want to see? Is this utopian stuff?

[1:16:13] John Pemberton: Whether it's good or bad, what's coming is irrelevant. It's coming. So I don't really have a valence to whether it's positive or negative, because it's irrelevant. It's coming. Like I'll hear people online going, AI is going to come and it's going to destroy the human race. I'm not going to sit there and whine and moan. I've thought about it long and hard. And I've thought about it. I've been in that mindset for quite a while. I remember reading a book. If you want a read that's going to scare you, this is it.

[1:16:50] Yuval Noah Harari, Nexus. Like I read this when I was in a depressed state. So I've been in a depressed state a couple of times in a really, really bad way. Not to do with diabetes or anything because I had a functional motor disorder, which meant that I was overthinking moving because I'd had physical injuries, which meant that I was, my brain was trying to understand how to move. And when you're in a position where you're locked into your brain about moving and you can't get out and you can't get present, it's a pretty terrible place and you need a lot of stuff to distract you. And that book, it was a good distraction because it's a deep and philosophical and scary read in a lot of ways. It's about how information networks progress.

[1:17:34] And it's like how human beings are expected to be in a tribe of 150 people. You're meant to know everybody. You're meant to know everyone. And you're meant to have a brilliant time with 150 people. You know who's talking shit to you and you know who's honest. You don't have that with the internet. Like we kind of know each other, but you could be an avatar for me. Like you could just be this avatar that's an amazing piece of Riverside tech that is now like the latest podcast. And that could be true.

[1:18:00] But what I do know is I go downstairs and look my kid in the eye and they're a human being in front of me. So that's what we know. But that's not the world we live in. The world we live in is information. And now it's moving at such a pace that you have presidents of the United States who've won elections on knowing how the information networks work. And that's what AI is going to do. So whether you like or whether you do not like that is irrelevant. It is coming. So it's about leveling up in understanding it, number one. And then number two, adapting to it. And it will take you some time.

[1:18:34] It will take you some time that if you used to be this finance guru who people would come to you for financial advice because you were the person who could predict what's going to happen. And you took some very fortunate risks and things and you were the man. You are gone. Like AI is going to be able to synthesize and process and pick patterns that you could never believe were even there. And your job has disappeared. But can you adapt to that? And can you adapt that your skill set might need to change? You might need to be a better communicator in educating about stuff, about being a better dad, about being a better husband, about being a better wife, about being a better communicator with someone.

[1:19:16] Those are the skills of tomorrow that are maybe skills humans have kind of lost, skills that are unvalued before. They're going to become the currency because your ability to do all these fancy knowledge things, that's what AI does. It does it better than you ever will do and can ever dream of doing. And it will get better and it will get faster and it will get even more clever. But what it can't do is have a conversation with your kid and look him in the eye and go, do you know what? You're a good one. You are. You're a good human.

[1:19:48] Arseniy Arsentyev: I'm sure that AI will change and will replace many individuals that have some applied jobs, some applied in a technical skill set. However, to be honest, I'm skeptical about it, replacing like mathematicians and...

[1:20:04] John Pemberton: Proper thinkers. Like proper, proper thinkers who've got ideas, who'll sit there and connect the dots in a way that only humans can do. Psychology is an interesting profession. Probably people are going to hammer me for this, but psychological research never replicates. You look at the replication of randomized control trials or evidence for psychology, it doesn't replicate. And there's one reason for that is nobody knows what's going on inside of here. The connections of neurons and millions and billions of connections that happen for a thought, for whatever, you cannot lay that out. It's impossible. But what you can lay out is a randomized control trial, whether a drug works or not. You can figure that out. But you can't work out from one day to the next how this brain works.

[1:20:58] And that's the beauty is kind of the uncertainty of that means that human beings will just generate the next idea that can't be done.

[1:21:09] Arseniy Arsentyev: And may I expand a little bit my thinking? Because I work in ML a little bit. And the modern LLMs and other models are statistical engines that are trained on some data that are cleaned, that's filtered. Because if you put junk data when pre-training an LLM, it will be some kind of a junk. But what am I getting at? These models are statistical and based on the past data. And I'm not sure that it is capable of going into these dark spaces in the brain where top scientists go and produce something that is uncorrelated. So LLMs is basically correlation engines. And these groundbreaking ideas, these questions that transform the world, they're 90% of the time are outside of the box.

[1:22:08] John Pemberton: I keep harping back to the man, but Nassim Nicholas Taleb would absolutely be high-fiving you at this point. Because the whole point of a black swan is, if you could predict the future from everything that's happened in the past, you would be a billionaire. AI can't do that. So you cannot say to AI, and it cannot predict. And this is why I can't predict. Okay, ChatGPT, Claude Opus Max, version 15.3. Tell me what lottery numbers are going to be next week. Tell me what stock in five years is going to be the thing to buy. Tell me in two years what it's going to be to buy. Anyway, the level of craziness in the world of all these connections of things that happen for things to be the way that they are are not predictable.

[1:23:04] People pretend that they can do. You can't even predict the weather. You cannot even predict the weather five days out with the level of certainty that you would need to know as to whether you're going to buy a t-shirt or whether you're going to buy a raincoat. So in the UK, certainly, maybe if the levels of certainty are a lot higher in certain ways with the weather. But even in like Africa, you couldn't say for certain in a year's time it's going to rain on the 5th of April. Like you couldn't say for certain. Like there's just not enough certainty in the world. And that's the risk of AI is that you believe that the model is going to get you 100% the way there.

[1:23:42] And I think Grace in some ways, 80%, it may be higher than that. But certainly from an education and a diabetes education perspective, I would never say it could be any more than 80%. So I can't tell you what to do. It's impossible. And if anyone tells you that they can, they're talking shit. And that's the way I would say with AI is I presume that the evidence that the answer it's going to give me is going to be 80% of the way and 20% of it's going to be shit. And it's up to me to work out what that 20% is. And if people are trying to tell you it's more than that, you best have your guard up. You best have your guard up.

[1:24:15] If it's going to give you the financial advice for the rest of your life and you're going to swallow it and you're going to stick all your money on black because of it, then you need to have a word for yourself.

[1:24:26] Arseniy Arsentyev: Thank you, John. The main summary, the main lesson that I got out of our conversation that I need to have you for the second time on my show because we've got so much topics to discuss and you got me so many ideas. We'll probably talk to you again after the Grace launches.

[1:24:48] John Pemberton: I'll probably be bankrupt. Yeah, I'll talk to you, but I'll probably be bankrupt. One thing I will have done is I will have learned something. Regardless of what happens, I will have learned something and it may be good or it may be bad and I'll try not to take it too much on the chin. But it's like I say with AI, it's coming fast. Let's just learn. Let's just learn and see where we get and correlate the dots backwards and see if we can learn.

[1:25:10] Arseniy Arsentyev: When is Grace to be released?

[1:25:13] John Pemberton: Yeah, it's a good question. So at the moment, if you actually went right now and you clicked on, I'm interested in the thing, you'd actually get an account straight away. So we basically have got enough money to allow a certain number of people in and we will extend the letting in as the money comes in. So effectively, the launch has kind of happened, but it's kind of guarded saying it's the beginning of June. But the reality is if you go and click on, I'm interested in an account, we'll give you an account and we're just gating that so that the number that come in can be paid for comfortably. And then as people, if they decide to put a fiver in to cover their own, five pounds, which is about what, seven dollars?

[1:25:59] That means it will open up the gate for them. Plus, we will chuck the money in for one more. And then if sponsors, sponsors of people to use Grace, not to sponsor the content, for example, a manufacturer goes, John, I will give you five thousand pounds. That will open ten thousand accounts because the five thousand for the individuals plus the five thousand that we will chuck in as well. So we'll just do it like that and we'll see what happens and see what the feedback comes. And they may think it's terrible and they may think, you know what, it's not worth seven dollars a month or whatever.

[1:26:36] That will be what it will be. And if it's only ever good and we can only ever have 100 people who really love it, that for me is enough. I don't need to make money. But if we can make loads of money by selling these agents who use Grace to make said CGM manufacturer upscale the speed that they can get to market with high quality education resources and high quality stuff for the CGMs that have got good data, that is. And for the AID devices that have got good data, we'll help them because we believe they've got a good product.

[1:27:13] If a company comes to me and goes, we think what you've said about lactate is the way, we'll help you get to market quick because we genuinely believe that continuous lactate monitors in what we're in, 2026? If they move fast, 2028, 2029, certainly 2030, continuous lactate monitors will be the thing that will determine 10 to 15 years before metabolic problems happen. Then insurance companies will be bending over backwards to help people get on one and bring their lactate down. And that will mean that they don't have to go on diabetes meds, have leg amputations, lose their sight. 10, 15 years before, they typically would find out with glucose.

[1:27:59] That will be the next thing. Now, will we be making money off that? I do not care. If someone listens to this and they go, that's the thing. That's the thing I want to pump my money into. And they go and find somebody else to do it with and they make loads of money off it and people get helped. Brilliant. I don't care. But if it's us, it means that more people with type 1 diabetes get education for free. So that's, if you're interested in that, we can do the two together.

[1:28:26] Arseniy Arsentyev: I'll reach out to you in a couple of months. Everyone, audience, thank you for listening. It was the 14th episode of Tomorrow's Medicine. I hope you learned a load. It was John Pemberton with me. Thank you, John. And yeah, see you soon, hopefully.

[1:28:42] John Pemberton: Absolutely. Absolutely.