Event Replay: From Terminal to Turnaround: How GitLab’s Co-Founder Leveraged ChatGPT in His Cancer Fight

[00:02:00] Good afternoon, everyone, and welcome to the Open AI Forum. [00:02:05] I'm Chris Nicholson, and I'm really glad to have all of you joining us today. [00:02:11] For those who are new, the Open AI Forum is a global community where we bring together people across industries to share how AI is being used in the real world, what we're learning, and how we can shape its impact together. [00:02:23] Today's conversation is a powerful example of that impact. [00:02:27] We're joined by Sid Sijbrandij, co-founder and executive chair of GitLab, and Jacob Stern, a geneticist, working closely with Sid. [00:02:36] Together, they're going to share a deeply personal and deeply technical story about using AI in the context of a serious illness. [00:02:46] I've known Sid for more than a decade now, and I'm very grateful that all of you are here today. [00:02:52] So with that, I'll hand it over to Scott McKinney, researcher at Open AI, to more fully introduce Sid and Jacob.

[00:02:59] Thanks, Chris. [00:03:00] I'm Scott McKinney, I'm a researcher at Open AI, and I'm thrilled to have Sid and Jacob here today. [00:03:05] As Chris mentioned, Sid has been battling osteosarcoma, it's a rare and tenacious form of bone cancer. [00:03:10] It affects fewer than 1000 Americans a year, and if it recurs after the first line of treatment, survival rates are typically measured in months. [00:03:17] I myself have been fighting the same fight since 2020, and I've seen firsthand how sparse, blunt, and frankly gruesome the treatment options are, and I've seen how quickly they can be exhausted. [00:03:29] After Sid's cancer came back, faced with the unfaceable, he and Jacob went into founder mode, tackling the problem with intensity and audacity. [00:03:36] When the traditional healthcare system brought them to the end of the road, they decided to pave their own. [00:03:42] As you'll hear in their talk, they're chaining together bleeding-edge tools in conjunction with AI to push the boundaries of personalized medicine. [00:03:49] And what I love about this story is that it's not just about resourcefulness in response to one's own medical crisis, but rather, as entrepreneurs, they're interested in blazing a trail for others to follow. [00:04:00] I was first connected to Jacob and Sid last year in the context of my own diagnosis, and they've been enormously generous in sharing their playbook, and it's totally changed how I've navigated the situation. [00:04:10] Thank you. [00:04:11] I like to think that their approach offers a glimpse into the future of medicine that could one day be accessible to all. [00:04:17] It's given me hope for my own family and for every family afflicted with cancer. [00:04:22] I've been inspired to watch Sid leverage his energy, ambition, and resources to chase his own cure. [00:04:27] In my own small way, I feel like I get to do that here at Open AI. [00:04:31] AGI may be the tide that lifts all boats, but many of us are particularly excited about how it can transform medicine, and we're working hard to make that vision a reality. [00:04:40] The team has created public benchmarks for evaluating models in the healthcare domain, deployed clinical co-pilots in primary care settings, and we're working to democratize access to medical expertise in Cat GPT health. [00:04:52] In this regard, I think the future is very bright. [00:04:55] With that, I'm honored to give you Sid and Jacob.

[00:04:58] Well, thank you so much for that wonderful introduction, Scott and Chris. Really excited to be here today and to walk you through my cancer journey and emphasize along the way how AIS helped us work through enormous amounts of data and navigate new treatments. I came to the US more than a decade ago in 2015 with a startup called GitLab. Six years later we took that public.

[00:05:26] But the next year after going public, a lot happened. You all at OpenAI released ChatGPT. There was a Nobel Prize awarded to a very interesting technology. But also on the downside, I got some weird feeling when I was doing a bench press. I was doing a bench press and I felt a pain around my heart area and I thought, oh, I know this is not going to be a heart attack because I've had this before. But unlike the few other times, it didn't go away and after two weeks, I couldn't sleep in the night. At 4 AM, I went to the hospital to the emergency room and they said, well, there's no such thing as a two-week heart attack, so that's good.

[00:06:13] And they took an X-ray, sent me home, there was nothing going on, according to them. But a few hours later, I got a phone call and it was my GP. He said, do you know how to meditate? I thought that was a very Bay Area question of him and I said, yes, actually I do know how to meditate. He's like, well, you better start meditating right now because you have super high blood pressure and you might have an aneurysm and the pain you're feeling might be your aorta starting to burst. So chill out and also get to the emergency room right now or call ahead.

[00:06:50] So I did that and they had good news. My aorta was fine, but they also had bad news. There was a six centimeter, what turned out to be a tumor, growing from my vertebrae. And so I had to do surgery really, really urgently to remove that. They put in a frame, they did a spine fusion, we did radiation, we did very extensive chemo. And as Scott mentioned earlier, it's kind of medieval how rough that is. It was hard for me to even get to the restroom. I needed four blood transfusions to even keep alive.

[00:07:33] As I hinted earlier, 22, something else happened. There was a Nobel Prize for click chemistry. You might not have heard of it, but it's a way to combine two compounds in a human without any side effects. It doesn't do anything else. It always happens. It's a very specific reaction. It's genius. That's why it won a Nobel Prize. And I was lucky enough during YC to meet someone, Jose is his name, and he was starting a company around it. He was first in patients with click chemistry and over time he struggled to fundraise. Now he's OK. But at that time before the Nobel Prize, it wasn't trendy. Biotech VCs didn't believe in it. And over time I became his biggest investor. It became my biggest investment.

[00:08:20] And we ended up becoming best friends. And now that I was in trouble, he said, you know what, let's try this treatment for you. And that opened up my eyes that it was possible to get a treatment just for a single patient. I always thought it had to be a big trial. But if you're in a really bad spot, there's an option called a single patient IND. The FDA approves ninety-nine to seven percent. They even say now that they are approving 100 percent. And it's a great pathway if you're with a disease and there are very few options.

[00:08:54] Despite all of our efforts, two years later, we had a remission. It started growing again. It had a local progression and there were no more standard of care medicines. My oncologist couldn't recommend something that he thought was going to make a difference. He said, hey, go look for trials. But it's a rare disease. There were no trials. So this became life or death for me.

[00:09:22] And I quit my day job and started going founder mode on my cancer. And I went all out, I did all the diagnostics I could get my hands on. Typically, people say only do the diagnostics if you know like what you're going to do with it. We didn't do that at all. We did every single technology we could find and we collected a lot of data.

[00:09:47] We started making treatments. If there's no treatment left, you need to make them. And we started doing as much as possible in parallel. We were out of time and most people died because.

[00:09:56] of time and most people die because the disease progresses. [00:10:01] And we're trying to scale it for other people. [00:10:04] In the maximal diagnostics, we've done everything under the sun, and if you want to see some of that, there's 25 terabytes of data available on osteosarc.com. [00:10:14] One of the things we did was single cell sequencing. It's a really cool technology where you individually sample thousands of cells. [00:10:22] It allowed us to isolate my cancer cells and look at the properties of the cancer cells. We found out they had a lot of FAP, it's kind of a fibrous tissue. [00:10:33] And that was great because we found in Germany a doctor that had an experimental treatment where you combine FAP with radioactive substances. [00:10:44] So I went there, did it twice. I was in the isolation ward afterwards because you had a glow up. And it was really, really successful. [00:10:52] The two treatments led to 60% necrosis and 20% shrinkage and the cancer detached from my dura and we were able to go in and scoop it out of my body. [00:11:05] And when we looked at the sequencing we've done along the way, the TCR sequencing in particular, we saw that we've been able to, all the immunotherapy treatments, get a lot of angry T cells in me. [00:11:18] So I did everything under the sun. I did dual checkpoint inhibitors, NK cells, super antagonists, and oncolytic virus. [00:11:26] So we had my immune system riled up, but the tumor micro environment was so, was suppressive. It didn't allow the immune system to go in. [00:11:33] And we suspect what happened is that because we disabled the FAP cells, they stopped signaling to the neutrophils. And we turned the cold tumor hot. [00:11:45] Suddenly the immune cells were able to go in. We don't know this for certain, but that's what we think happened. [00:11:52] We also looked at MDM2 as a therapy for me. I am really, really high on the MDM2 expression. It's just some protein, but I'm off the scale. [00:12:00] I'm the index tumor over here. I'm in the .3%. And we started looking, like, are there any drugs? [00:12:09] And there were drugs. People made them. But then they stopped all the development. [00:12:14] Because drugs only get to market if it's a blockbuster, if the majority of patients are served with it. That wasn't the case for MDM2. [00:12:21] And so they stopped it, and they were about to turn off the freezers. So we're now paying to keep the freezers on. [00:12:29] And I'm looking for a pathway to bring this medicine to market because it might help other people beside me, not the majority of patients, but if there's 10% or 20%, that'd be really awesome outcome as well. [00:12:43] One of the best things that happened with Maximal Diagnostics is that the single cell sequencing got me in touch with Jacob, who's now running the enterprise of my kit.

[00:12:52] Jacob?

[00:12:53] Cool. Thanks, Sid. And thanks so much for having us. I'm not a doctor. [00:12:57] I met Sid through Jose, the mutual friend, who was working on the Click chemistry-based drugs. [00:13:05] And at the time, I was working at 10X Genomics, a company that makes the equipment that enables the single cell sequencing. I was there for six years. [00:13:14] And when I met Sid, I started talking about what we're doing. I'm at 10X, we do single cell. Sid explained how he and his team were using single cell to actually inform his care. [00:13:23] And at 10X, the mission statement is to master biology to advance human health. This is the first time I'd actually met someone in person who was doing this, mastering biology to advance his own human health. [00:13:34] And I found this to be super compelling. And as we got to know each other, and I got to know the story, I got to understand the extent to which he was living in the future and had ambitions to do even more. [00:13:45] And so what you can see here is the vast array of technologies we're using to analyze Sid's cancer. [00:13:51] We're doing stuff that I'm more familiar with, like single cell sequencing, or some of the bulk DNA and RNA sequencing, but we're also stretching way outside of my comfort zone. [00:14:01] We're doing targeted radio diagnostics, extensive pathology staining, using organoid models to test the effect of drugs directly on Sid's cancer. [00:14:10] And so for me, I've become an extensive user of AI to basically bring myself up to par and enable me to talk with experts about these different domains and have intelligent conversations about how we can drive this forward for Sid and actually make use of this data. [00:14:27] So I'm actually going to take you through a little bit of my historical GPT history so you can see a few examples of how I actually did this in practice. [00:14:36] So this is a screenshot from last summer where we wanted to just run an experiment. Let's take the output of one of these bulk RNA sequencing experiments, where we're basically counting the number of times that each gene is being detected at the RNA level in the sample from Sid's tumor. [00:14:54] And this is expressed in a CSV file.

[00:14:54] And this is expressed in a CSV file. It's genes and then counts.

[00:14:58] And we fed this to, at this point it was 4.0 on the pro plan. And said, what do you think of it?

[00:15:04] And I wanted to see what came out. And frankly, even then it was remarkable. You'll see here if you zoom in, it flagged B7H3, which you'll recognize later in the presentation.

[00:15:14] And then it also recognized some of the immune dynamics that Sid talked about earlier, which we've studied in much more detail since at the single cell level.

[00:15:24] And now what we're doing is much more advanced. I mean, these models are progressing at an absolutely insane rate.

[00:15:30] And we built some harnessing for ourselves, where we can ask a question in natural language and set up a series of agents that can go do literature search, formulate its own hypothesis, basically structure a bioinformatic analysis.

[00:15:47] And then execute that and bring back a summary. So what I have here is a recent example, where I'm asking the system that we built about CHIP, Chlonohematopoiesis of Indeterminate Potential.

[00:15:59] This is something that many people get as they age. It's basically the blood cells losing polyclinality as we age. And it can, if things go wrong, lead to leukemia.

[00:16:10] And as we got sort of a hint that this might be happening in Sid from our team, it spooked us because this can be a side effect of the chemotherapies that Sid took a number of years ago.

[00:16:20] And so one of the first things we did is I went to the system and just asked, you can see the question that I posted here. It's pretty natural language, and it went out.

[00:16:28] It was about $20 in API costs. It talked for 30 minutes, did a series of tool calls. It did its own literature review, formulated the set of markers it wanted to look for.

[00:16:39] This is hooked up to about 600,000 single cells from a number of time points that we've taken from Sid's blood. So it can actually run the analysis directly in Sid's blood and come back to me with a report of, here's what I think, here's my conclusions, here's the interactive plots, and then here's my whole history, here's the code it wrote in Python, etc.

[00:17:01] And I'm not gonna trust this out of the box, it can certainly make mistakes. But what it does give me is, it's a rapid way for me to get up to speed and start to understand the sort of circumstances around this disease.

[00:17:14] Again, I'm not a doctor, I didn't train and ship. But here, I can rapidly come up to speed on the disease more generally and also on Sid's biology specifically.

[00:17:23] And it's let me be a good counterpart to the informaticians who have since looked into this in much more depth, and thankfully we've put this risk to bed.

[00:17:33] So on top of doing the maximum diagnostics, we're also making a number of treatments from scratch, specifically for Sid.

[00:17:41] And this is even further outside of my comfort zone than the analysis is, and it's frankly been radicalizing for me to find that you can just make your own drug.

[00:17:49] And so before Sid goes through and talks about some specific vignettes here, I just wanna sort of give an overview of some of the modalities just to level set.

[00:17:59] So the first one he'll talk about is a cancer vaccine. In this case, an mRNA vaccine specifically.

[00:18:05] So you can think about this similar to the vaccine that you've taken for COVID, or for the flu, whatever it is. Where the idea here is to educate the immune system to prepare itself to fight something that's foreign.

[00:18:18] In the case of the COVID vaccine, you're exposing the body to the spike protein of COVID itself. In this case, we've encoded a number of mutations that are present in Sid's cancer, which basically differentiate the foreign cancer from Sid's normal tissue into the vaccine.

[00:18:35] And we put that in, or priming his immune cells to be ready to patrol and try and fight the cancer if it sees it.

[00:18:42] The next vignette will be around a TCR T cell therapy. TCR stands for T cell receptor, and T cells are one of the business ends of the immune system.

[00:18:49] T cells are very killy. They use a T cell receptor to detect something very specific, and then basically they shoot out proteins that rip holes in cells that are close to them.

[00:19:04] And so, what we can do here is something that's a little bit more direct than what we're doing in vaccines. Where through some of the work we've done with single cell sequencing, and some very capable partners have done very thorough work as you'll see, we've identified TCRs, T cell receptors that are likely very specific for Sid's cancer.

[00:19:22] And we can engineer those directly into healthy T cells, grow a bunch of those up, and then inject them directly into Sid. Hopefully we never have to do that, but we're preparing in case we need to.

[00:19:33] The last one I'll talk about briefly is the CAR T cell therapy. In this case, CAR stands for chimeric androgen receptor.

[00:19:41] This is an extension of that same concept of the TCR T, where we're taking a T cell, the killer immune cell, as a chassis. And souping it up, and then engineering in the receptor, basically what it's gonna use to.

[00:19:52] The receptor, basically what it's gonna use to recognize. And we've picked a target that through the maximum diagnostics we saw is present on most if not all of SIDS cancer cells.

[00:20:03] And through this, this is the really killy one where if we really needed a Nuke to deploy, we're putting this on the shelf as an insurance policy, a sort of super killy CAR T that's specific and very potent.

[00:20:17] Yeah, so we did the personalized mRNA vaccine. It's super exciting. I was the first patient of an investigator initiated trial. And while doing that, it was awesome to learn what goes into making that.

[00:20:30] There's all these antigens that you can encode in it. You have bits and bytes basically to encode, base pairs to encode in the virus. You have a limited set. So what you select matters.

[00:20:44] And we used tons of different ways to determine that. Like what was in my tumor samples? What is scoring high in different models? And right now this is more of an art than a science.

[00:20:55] But we've already seen companies that are starting to use AI to do this automatically. And that is very, very exciting because that means you can do it for millions of patients.

[00:21:06] I can see a future where you get a personalized vaccine against your cancer. And that will have to come from AI because we don't have enough doctors to do this. Also, typically these engineering AI approaches yield better results in our opinion.

[00:21:24] So very, very exciting what AI can do here. We were very fortunate that we got this from project start to injection in only six months. And we see those times coming down even further in the future.

[00:21:40] Jacob mentioned the TCRT. It's a very interesting thing to make as well. It's very complex to make. But a lot of decision making goes into it and a lot of data goes into it.

[00:21:52] And what we've seen is that the diagnostics, if you have good kind of tumor samples and you do a lot of sampling, it really helps inform how you design it. Here again, there were lots of decision variables like, do we do that? Yes, no, why?

[00:22:06] A lot of reasoning, a lot of logic. And to do this at scale, you'll need AI as well. The last example I'll give is, I think, a really powerful one of how the diagnostics are influencing the medicine development.

[00:22:20] We selected B7H3 as a target for my CAR-T. The CAR-T, remember, that's that super killy nuclear bomb that you're gonna set off. We wanted to know where is that present?

[00:22:31] And you have genetic data about me, but there's nothing as good as doing an actual scan. And we were able to access one. It was in China, in Beijing, so I traveled there in October.

[00:22:45] And we did the scan and the doctors came back and they said, we have really good news and really bad news. Well, what's the good news? The good news is they didn't see any cancer, so yay, still clean.

[00:23:02] The bad news is that my liver was showing up three and a half times as much as the 20 Chinese people they scanned before me. So now we were very worried. If I ever were to take this medicine, I might lose my liver.

[00:23:14] So we went back to Cole Robo, who's making it, and he said, can you do something about it? He says, well, actually, I was the first one to invent a logic gate for CAR T's, an end gate.

[00:23:32] Two things have to be present. So we went back and looked at what's really not present in the liver. It's FAP, you know that from the first one, the radioligand therapy.

[00:23:42] That has very low liver expression, and by needing those, both things to be present, the B7H3 and the FAP, we're gonna make sure that if I use the nuclear bomb, it doesn't go off in my liver.

[00:23:57] And it was amazing to see that feedback loop, how a better scan informed that. Last anecdote is something we saw, it's called PENX3, and it, under genetic data, is like, sprung at us.

[00:24:08] It was represented 10,000 times as much in my cancer as in my healthy tissue. That's an incredible difference, and we were eager, like, okay, let's see what medicines are available.

[00:24:25] But when we scoured the literature, there was nothing. There were a few mentions of it. No one's really done the research. And most protein targets, people will have published on it.

[00:24:38] They just go look for what's expressed on the outside of a cell. So this was really strange. Why wasn't there anything? And we think we figured it out, it's a hydrophobic thing.

[00:24:49] So if you do your tests in water, like most.

[00:24:50] your tests in water, like most people do, you're not gonna find this thing. The only reason we found it is because we went back to the data and did a painstaking analysis. And this is again something that AI is incredible for. It has much more patience than the human. It will go through all those terabytes of data and find the proteins that matter. A human would not have the patience for this. And we were lucky that we were able to find this. And now we're running a binder campaign trying to see if we can engineer something against this. So that's some of the anecdotes of how we make the personalized drugs and diagnostics. Back to you, Jacob.

[00:25:31] Yeah. So once again, we'll dive back into my ChachiPT history. I would say this is an area, again, where I'm even more out of my depth than I am with some of the more adventurous analysis modalities. And so what I'm doing absolutely would not have been possible two years ago. Because if I were trying to navigate to the expert who's at some biotech company, you know, advancing some program that's going to be used for millions of people, the knowledge is very sparse. But what AI has made possible, particularly I think I'm a super user of deep research, is the democratization of that specialized knowledge. And it doesn't make me a specialist, but it makes me someone who is competent enough to talk to specialists and push some of these programs forward while owning the objective of let's keep SID alive.

[00:26:21] And so here, you can see some of the stuff we were looking at as we were thinking about the mRNA vaccine. Last summer, as we were considering how are we going to do this, we were looking at options ranging from working with one of these, you know, big professional contract development manufacturing operations that makes drugs professionally, makes the COVID vaccine, actually, down to can we actually just have some academic group make it in their lab as if they were putting it in a mouse, and would we put that in SID? And so for that, we actually thought from first principles around what are the quality control metrics we would want to see from something in order to feel comfortable that what we're putting in SID is going to work, it's going to be what we expect, it's going to be clean, it's not going to give them sepsis, et cetera. And of course, all that work was done with AI.

[00:27:09] And that meant that when we were introduced to the academic team that ended up making the mRNA vaccine, which has been an amazing collaboration, it did two things. One, we were prepared to ask reasonable questions and understand what they were saying and what they were planning to do. And perhaps even more importantly, it made us good partners to them because we were sophisticated enough to understand what they were doing. And that built a relationship of trust. And we had to fight through a lot of things to actually get this done, and we did that together. And that was built on sort of this mutual understanding and mutual ability to talk about the nitty gritty details here.

[00:27:47] Here, this is on Penexin-3. As Sid explained, this is a really understudied target. When we're thinking from first principles about Sid, if you go back to that plot, we don't have to go back there, but on that plot, the Penexin-3 was way off the diagonal. It was clearly the best target for Sid, but it's not the best target for almost anybody else. And so, it's a target that's understudied. And so, as we're trying to do a binder discovery campaign against Penexin-3, we're sort of MacGyvering through it, we're inventing new reagents, we're thinking through new assays. We're trying to triage different data types to see if we have a good binder or not. Is it a false positive or a false negative? And as we're thinking through that, AI does feel like an Iron Man suit to allow us to assess all these different specialty things and think about these really arcane, nuanced points in this arcane process and it makes it accessible to somebody like me.

[00:28:47] If we think about the treatments in parallel, where did that come from? I think, something you have to be aware of as a patient is that the incentives for the doctors are very, very different from the incentives for you as a patient. The doctors want to minimize their liability. Their liability is when they prescribe a treatment and there are severe side effects, including death from those effects. As a patient, you want to maximize survivability. In my case, because it's a nasty cancer with medieval medicine, I'd rather die from a treatment than from the cancer. Dying from cancer is a really miserable way to go. That is not what you will get with doctors. They have very different incentives than you. And most patients, they get too few medicines and they die because the medicines that they got didn't work.

[00:29:48] The medicines that they got didn't work, they didn't have time to try the other things, they get metastatic cancer, and they pass away.

[00:29:56] AI is amazing at helping you suggest things to discuss with your oncologists to combine things. And the standard, I think, pushback you get from an oncologist is, this has never been tested in the randomized control trial. That is true. The randomized control trial is $100 million. We can't test every combination of medicines that way. But also, it's not needed. From first principles, you can just see, do the side effects, do the medicines have the same side effect profile, do they add up? You almost, you kind of need to look at this organ by organ. If one drug is hitting your kidneys, you want the other drug to hit your liver. You don't want to take two kidney drugs at the same time.

[00:30:45] So, AI is incredibly useful to help you through that, to have the dialect with your doctor, but to come well informed and to push them to get closer to maximum survivability. Because that will not be their default mode. If you want to see what treatments I've combined, you can see so on osseosarc.com slash timeline where we have all my different treatments and all the modalities listed up. Some of the things I'm doing are incredibly expensive. Some of the things I'm doing are not incredibly expensive. You can get bulk RNA sequencing these days for 50 bucks. Whole genome sequencing started at $500. AI is amazing. For $20, you get super capable tools. Treatments in parallel will consume more drugs, but if those drugs are generics, the pricing is super affordable. So, there are things that should be accessible to people if they can convince their doctor to prescribe.

[00:31:55] And then, aside from that, for some of the more adventurous things that we're doing, part of the way we're trying to scale this for others is we've started a series of companies to try and put this on rails, because for Syd and a few other families who are pursuing this at the absolute maximum, has been coming up the learning curve and starting from scratch and developing relationships and developing competencies, et cetera. But what we want to do is pave the road behind us so that it's easier for the next person, the next person, the next person. If Syd and a few others like him are Roadster model 001, how do we do the Roadster that somebody can buy off the shelf and work our way from the model S to the model three? And of course, AI is deeply embedded in all of these things because it's a deflationary technology. It's how we bend the cost curve and how we get things to scale.

[00:32:44] And so I'll give a couple of examples for the portfolio. One of the companies is Thalus. They're doing the maximum diagnostics from a gene expression perspective, running things like single-cell sequencing and running through all the raw bulk RNA sequencing data to find the right target for somebody's cancer regardless of the tissue of origin. And so here you can see the plot looking at a specific patient and the expression levels, et cetera. But once the target is identified, in this case DLL3, using AI to pull in all sorts of context about this target. Active clinical trials, biological data, pharmaceutical companies to work with, et cetera.

[00:33:27] And then for another one, Arden. Arden is taking a similar approach to complex unresolved immune disorders by doing maximum profiling of blood to try and get into a treat-measure-analyze-repeat cycle with a combination of targeted immune modulators that are tailored to whatever the person's blood is actually saying. And the CEO of Arden, when he starts working with somebody, often what comes back is a gigantic Google Drive folder, or in one case, a 9,000 page document documenting somebody's medical history. And so he's actually used AI to build an AI tool that helps him parcel this information and get up to speed with the patient and help them parse the history to make a better plan going forward.

[00:34:14] So I've been very, very lucky that what we did worked and there's no evidence of disease since we did the radioactive treatment and surgery. It's not for a lack of time. A couple of weeks ago, I did a bunch of experimental scans. These scans combine a protein binder, for example, for B7H3, but also IFAR-2 FAP with a scan. And I'm super excited that there's more and more of these proteins kind of becoming available so you can do a scan. I can see a future.

[00:34:46] do a scan. I can see a future where, instead of just trying a drug, even a generic drug, if there's a protein it binds to, you want to make sure that is expressed in your cancer, and you want to be aware where else it's expressed so you can view for those side effects. That differs per person. If we can do the scan first, that's great. With some of these, you can do the scan first, and you can do the scan in the morning, and you can do the treatment in the afternoon where they bind something radioactive to it. I think this field is poised to take off really, really rapidly. By searching and scouring the world, my therapeutic ladder, the options available to me, have gone from zero treatments to 30 treatments that I don't want to use. I'd rather not use, but it's great to have options. I want to thank you so much for watching us, and we super look forward to your questions. Thank you.

[00:35:41] Hey, guys. That was amazing. Thank you so much. One of my takeaways that people maybe who haven't suffered cancer or haven't gone through cancer like you, they may be surprised is how dynamic and proactive you have been with cancer. You've learned from it so quickly, and it feels like the speed of your learning process is the key to getting this far. Is that how you feel?

[00:36:12] Yeah. We've learned a lot from other people, other patients, concierge medicine groups that are super, super good, but also a lot of AI and a lot of first principles. I think medicine has lost its way in a little bit that they only work with randomized controlled trials, but some first principles thinking would really help in a lot of cases.

[00:36:29] Yeah. People throw around the term a cure for cancer a lot and lament that AI has not given us the promise to cure for cancer, but I feel like what I'm learning from you is that curing cancer may happen one patient at a time, right? So you're really painting a future of personalized medicine where there's not a single cure.

[00:37:03] Yeah. We're trying to bring the, I think some of the stuff we presented today is going to be the standard of care 30 years from now. We're trying to bring it to the present. Some of these to the present for older people and all of this to the present for some people.

[00:37:21] Yeah. I heard you making some recommendations. I know there's a lot of families out there who have had brushes with cancer or who are going through cancer now, and what I hear you saying is there's affordable ways to gather data and there are affordable tools to help you analyze, understand that data so that you may ask for better treatments and eventually get to better outcomes. Is that the gist of your message?

[00:37:41] Especially if you have a rare cancer, it's super important to be a good advocate and even if you have a known rare cancer, we've met many survivors who said that stepping up their advocacy has really paid off for them.

[00:37:57] Yeah. Amazing. Okay. I'm going to open this up to questions from the community. So we got Jason DeLuca who asks, did you ever hit a point where things felt absolutely hopeless? And what was your breakthrough moment with the tech?

[00:38:10] I think it wasn't a tech thing, but I felt very, very hopeless for 15 minutes when I got a radiology result. And it was a classic. I was in a meeting, but all of my reports, it was during an off-site and I got a message from my GP. He says, it's positive, not subtle. Okay, great. Positive. I'm like, oh no, no, wait, it's the reverse in medicine? What's going on?

[00:38:43] So I opened it. My lungs lit up like a Christmas tree. And with bone cancer, you're really afraid of spreading to the lungs. And it was a single diagnosis from the radiology that the cancer has spread. And there were so many, it was inoperable. And so that was, I walked out of the room. You process for 10 minutes, you call your wife, you call your CFO and CLO, and then you go from there.

[00:39:16] And the next day, the board came over, that was fun. And you kind of like, well, this has been a ride. Thanks, everyone. All the doctors signed off, like, oh, I'm so sorry to hear about the news. And then off the seven doctors we had in the loop, one guy, one guy, it's like, I don't like how this spread by the lymph nodes. This is not how osteosarcoma spreads through the lymph node. And if I get 60%

[00:39:44] And if I guess 60% this isn't cancer. [00:39:48] And it turns out it was the remnants of COVID. [00:39:52] But that was certainly a low point. [00:39:54] Yeah. [00:39:55] But yeah, you also learn 10 minutes plus five minutes. [00:39:59] Yeah. [00:40:00] And you're back and you're gonna, you're making the most of what's left. [00:40:05] Yeah. [00:40:08] That's a wild ride. [00:40:09] That was wild. [00:40:10] Yeah. [00:40:12] Okay, here's another question. [00:40:15] Let's see. [00:40:17] And like for an AI angle, it would have been nice to have AI look at it because it might've given a differential diagnosis. [00:40:24] This could be cancer or this could be COVID. [00:40:26] That would have been super, super helpful in that situation. [00:40:29] Yeah. [00:40:30] And a lot of folks don't have seven doctors with one at the end. [00:40:34] No. [00:40:35] That's six. [00:40:36] That one doctor that has seen 10,000 sarcoma cases, there's probably only one Sanchala in the world. [00:40:44] Yeah. [00:40:47] How... So it sounds like this journey has been personally transformative for you. [00:40:54] For sure. What has changed in you, in your approach to life? [00:41:01] Like how has this changed you as a person? Not your body, but your spirit? [00:41:07] Yeah. [00:41:09] I think it makes you realize that kind of the things you learn as kind of, when I was a software entrepreneur, I didn't think those skills would translate to cancer. [00:41:22] And I thought in the beginning, I'm this delusional tech guy doing this. [00:41:28] And no way that without a PhD, I can have a positive impact. [00:41:35] But we learned along the way, there's lots of ways in which the system is misguided. [00:41:41] Look on my website, you'll find over 10 things, I think that you change in the system to be more patient first. [00:41:48] And it's given me the confidence to like try it on other things outside of cancer. [00:41:54] And yeah, might not always work. Maybe I got lucky, but it's worth trying. [00:41:58] And as long as people try and AI has helped us find the courage to question the status quo. [00:42:06] It sounds to me that you kind of have a movement in mind here, that movement could come out of this experience where many people whose lives are at stake, seek a greater change.

[00:42:20] So, in addition to reading your site, and I want you to repeat your site so everybody knows where to go. [00:42:28] But in addition to reading your site, where do people, what can people do? Not just for themselves, but also for everyone. [00:42:35] I think take agency, ask your chatbot. Don't take it as the truth, but help it inform your conversations with your oncologist. [00:42:48] We've talked to tons of patients, and Scott wasn't the only one. We'll continue to do so. [00:42:53] Today is the day I hired someone full-time to help us, to enable us to help more patients. [00:43:00] So, shout out to Purnima. She's gonna have a busy first day, I think, because cancer at siteseek.com might get some emails today. [00:43:09] We're looking forward to that, and there's amazing people in the industry, like Oxundra, who's trying to drive clinical trial abundance. [00:43:17] AI is gonna enable us to kind of discover and make so many more medicines, but we still have to run these trials to prove that they work, and she's doing amazing work to make those trials more accessible and affordable, enabling many more medicines to come to market, because we need both. [00:43:37] We need the AI, but we also need clinical trial abundance.

[00:43:41] Thank you. I think, Chris, something that's pretty striking just in doing this every day is people in biotech and medicine almost uniformly are in it for the right reasons. [00:43:47] So it's like people want to help people. For sure. [00:43:54] People under the hood in the pharmaceutical industry, they want to make medicines that advance human health. [00:43:59] Same with anybody who's going into medicine. These are hard roads, and there's easier ways to make money, frankly. [00:44:06] But as we've gone out to talk to people, to scour the world for collaborations, whatever it is, and sort of put the energy of what we're trying to do out there, the response has been really encouraging, because people want to help, and people want to find a better way, and people just want to help patients. [00:44:26] And so I think there is a positive sum energy out there, at least from what we've encountered. [00:44:33] And so that gives me a lot of optimism. [00:44:37] It feels, I think a lot of folks outside the business may not be aware of how much of a bottleneck clinical trials are.

[00:44:42] Speaker 1: Bottleneck clinical trials are? The cost, finding candidates, it sounds like you two are really thinking about how to alleviate that bottleneck. What are some of the problems there? Please talk to me about how you've understood clinical trials and how you think they might be improved.

[00:45:00] Speaker 2: Yeah, for sure. Roxandra has done an amazing job publishing on this. And for example, one of the things that makes clinical trials expensive is that you have to get approval from the FDA before you can start it. And that takes time. And time kills everything. People run out of money. So in Australia, they've just been using a notification. That is so much better, and you can see in the safety data that it's just as safe. So that's kind of a no-brainer to kind of copy that from Australia. The costs for these trials are getting out of hand. It's more than a billion dollars to successfully get a drug to market. That is insane. That's limiting the drugs that get to market. There should be 100 times more medicine if we can get the cost down.

[00:45:49] Another thing that we've really struggled with was the IRBs at hospitals. These are so-called ethical review boards. They are incredibly hard to work with because there's no incentive to kind of go in the IRB. What would be really great is if there's IRB freedom, where you're not forced to use the IRB of the hospital, but you can go to independent parties as well, who are proven more affordable, more efficient, and just as safe.

[00:46:26] Speaker 1: How does it currently work? You've got an IRB in a hospital, and if you want patients associated with that hospital to participate in a clinical trial, you have to work through the IRB?

[00:46:31] Speaker 2: You have to use the hospital IRB, and what happens is as a patient, this ethical review board, I get wheeled in to pre-op at 7 a.m., and they're discussing kind of commas and formulations that really don't materially matter until the point where I got rolled into the operating theater at 3 p.m. without an IRB sign-up still. It's insane. That's protecting me, like waiting the entire day to do some dots and commas on a file that no one's ever going to look at again, and it's not material to any of it. It's absolutely bonkers, and the best way to solve it is to give researchers freedom to pick.

[00:47:25] Speaker 1: And it sounds like you're saying sometimes the delay in care actually has much graver consequences for patients.

[00:47:30] Speaker 2: Of course. With all my resources and all my agency, if I'm running into that, think about how many people never get treated at all. And how much energy and enthusiasm gets sucked out of the investigators when we get these trials.

[00:47:45] Speaker 1: And how about the matching problem? How are people developing potential treatments supposed to find the folks whose electronic health records qualify them for the trial? It seems like an unsolved problem.

[00:47:51] Speaker 2: Yeah, for sure. So like navigating clinical trials, there's good websites, but AI could really help, especially if you have more diagnostic data and using that to find a trial that really fits you.

[00:48:13] Speaker 1: Yeah. This has been deeply inspiring. Let's see. We've got, okay, I think we're at the end of these questions, so we're going to move to wrap this up. I want to say before we wrap it up, I've learned a lot. I think our community has learned a lot. We're going to keep learning from you guys after this talk is over. So thank you for coming in and being partners with us.

[00:48:39] Speaker 2: Yeah, thanks for having us.

[00:48:41] Speaker 1: Yeah. So I think this gave us all a better understanding of what's possible with AI, and that's one of the purposes of this series is to inspire people by highlighting what innovators like you are doing so that other people might find their own path forward. And hopefully, AI can also lower the frictions for them in their search for their own way of navigating this complexity.

[00:49:06] And as you said, Jacob, a lot of people want this to work better, and what we're fighting often is not each other but the complexity of a system, a legacy of previous years and decisions. So thank you, everybody, who joined us today. The forum is shaped by you and your questions and your curiosity. We're grateful for you here in the conversations that we get to have with you.

[00:49:30] As a look ahead, we have some more forum events coming up. That includes conversations like...

[00:49:40] Like reimagining cultural heritage with OpenAI, that's going to happen with the Sanskriti Foundation in Asmona next Tuesday, March 24th at 8 a.m. Pacific time.

[00:49:53] That session's a replay of a recorded event, OpenAI forum event. It was originally held in India. Brought together by OpenAI, Asmona, the Sanskriti Foundation, which is one of India's leading cultural institutions.

[00:50:07] So we'll be sharing more details soon. Stay tuned, keep an eye on your inbox and the forum platform, and thank you for being here, and we'll see you at the next event.