Now, Mike, my next question is really to ask you, how did you get involved in this? But before I do, we're recording this on March 16th and chat GPT-4 was just released several days ago. What is in brief chat GPT-4? How is that different than chat GPT-3 or 3.5?

Chat GPT is based on GPT-3 3.5 has in it close to 200 billion parameters. GPT 4 was just released, and I've not had an opportunity to explore it or determine how many parameters are in it, but there have been estimates that the number of parameters number in the trillions, and chat GPT is presumably going to become based on version four of GPT. But to my knowledge, it's not yet released to the public. I think it's been released. The company has made some examples of what GPT-4 can do, but there's not an interface to which the public yet can explore that. What I think we're going to see, though, is GPT-4 is going to be much, much better than GPT-3 And I think subsequently we will see versions 4.5 and 5 and so forth. And we will also see similar language models or transformers from Google and from other companies that are developing these things. There are a number of companies have been developing these language transformer models and have hesitated to release them because of the potential that people will believe things that might not be true.

And Mike, obviously, I'm hopeful that you can take us to how you see this being used in medicine and pediatric critical care in particular. But before we get there, just briefly, how did you become absorbed in this? Was it based on all of your experience being the PI of all of these consortia and data resource centers? Tell us a little bit about that.

Well, actually, Jeff, it's interesting that my exposure to this technology really arose by accident because of a hobby interest that I have in medical computer simulation. I've had an interest in medical simulation since the 1970s when I was a resident, and I encountered a computer program that was published in 1972 called McPuff. And McPuff was a program that simulated human respiration and oxygen delivery. I translated this program into numerous languages over the course of 30 years, and I corresponded with the authors. This program was actually used to predict ventilator changes and other changes in real human patients in the early 1980s. That effort was not continued forward, and I last corresponded with one of the authors of those efforts about 10 years ago, and they never moved forward with it, and currently they're all dead. So the simulation idea that I had was that there would be simulation in our monitors, in our ventilators, in all of our devices, and that people would be able to predict what would happen with a patient before they actually implemented an intervention. Now, this program has fascinated me, and I've become fascinated with the ubiquitous nature of iPhones. And so I've been programming a version that would run on the iPhone. And in the course of developing the software, I heard about ChatGPT and then I went on ChatGPT to ask it to write part of a subroutine that I needed for this program. It immediately wrote the subroutine in one computer language. I asked it to write it in another computer language and got an immediate response. And then I finally said, I need this to run in Swift, which is the language used on Apple iOS. And it immediately produced a version of the subroutine and explained how to use it and also gave an example and demonstrated that it worked. So that's how I became aware of this product. I also asked it to write some other things. I asked it to explain the routine to me in Czech. I asked it to explain it to me in Chinese, and it was immediately able to spew out what might have been correct. I don't speak Czech, nor can I read Chinese, but the capability to do that was absolutely fascinating and compelling. I will say that the routine that it gave me looked like it should work, but when I actually tested it in the Apple iOS environment, it did not work. And so this is one of the worries and risks of this program, which is that it can give an answer to you that seems plausible, but it can be completely wrong. And that's one of the issues that's going to come up, I think, when we get into what happens in medicine when people incorporate this tool into their lives.