Episode 303: Practical Everyday Use of AI with Drs. Stahl and Saddawi-Konefka
51:31 Share

For a limited time at Verizon, you can get our best price ever for a single line. Just $45 per month when you bring your phone, which is less than you spend on too-tired-to-cook takeout every week. Get one line on unlimited welcome for $45 per month with AutoPay Plus taxes and fees. Visit your local Verizon store by April 2nd to save.

$20 monthly promo credits applied over 36 months with a new line on unlimited welcome. In times of congestion, unlimited 5G and 4G LTE may be temporarily slower than other traffic. Domestic data roaming at 2G speeds. Additional terms apply.

The game is on the line, and I've only got a shot clock's worth of time to tell you about Subway's 699-foot-long deal. That's right. Order now on the Subway app and get any footlong on the menu for just $6.99. Fresh sliced deli meat, fresh crispy veggies. Ah, too much good stuff. Get it now on the Subway app. Any footlong for just $6.99. Use code 699FL now. Only here for a limited time. Redeemable at participating U.S. restaurants. Subway app. Online orders only. Add-ons. Additional plus tax. Additional fees apply on delivery. Excludes footlong snacks. Limited time.

♪♪♪

Hello, and welcome back to ACRAC. I'm Jed Wolpaw, and I am really excited for this show today because this is something I think anyone is going to be able to walk away from with some practical tools you can use in your life and your work, really no matter what you do. This has helped me a lot. We're going to talk about

not in the big kind of here's what the large language models do and how they work on a kind of very technical scale, but on a how can you use AI if you are pretty much a novice to it? How can you start using it in a way that will help your life? And where this came from is that I've been –

at several talks recently about AI, and they all come from a very educated, very, to me, high-tech level. And people often talk about kind of where this technology is, where it's going, how it works, where we may see it changing healthcare in the future, and all that is fabulous. But it doesn't leave me walking away thinking, I know what I'm going to do today to do something different with this technology that's going to help me

And at the same time, I was talking to two good friends and colleagues who are much more tech savvy than I am. They are not IT experts. They are not AI experts. But they clearly were farther ahead than me in starting to incorporate things like chat GPT into their lives and work and understand.

They taught me just little things, and I thought, this is great. This is so useful. We should do a podcast on this so that the rest of you can learn some stuff from them as well. And so I'm really excited to welcome Dr. Dan Sadawi-Konefka and Dr. Dave Stahl to the show, both fabulous anesthesiologists. Dave, former program director, current chief of critical care at Stanford. Dan, current program director at MGH. And guys, I'm really excited to have you here. Thanks for agreeing to do this. Thanks so much. It's going to be fun. Happy to be here.

Great. So let's start with just a, you know, why don't one of you or both of you take a stab at, you know, why I kind of, I think laid out what I think, but why is it worth talking about this at all in your mind? I think that, you know,

These tools have started to emerge, and I think that they're getting to a tipping point now where their utility outweighs the activation energy of using them. And where somebody who isn't necessarily tech savvy can start to get a lot out of them in ways that really tackle some of the most annoying parts of our jobs and lives.

and really can shave minutes and hours out of your day. And so I think that's where now makes sense to talk about this. I 100% agree, Dave. I think that I first heard about these a couple of years ago, and it sort of seemed maybe a little bit nebulous. How are we going to use these? And I think, Jed, you know this, Dave and I tend to be a little bit more tech-friendly. So we just

Both played around with it. And exactly as you've said, we've found ways to shave off minutes, if not hours, from our week to week. A lot of the routine stuff that doesn't take, I think, the advanced creativity, we sort of outsource it to. It's like having a really, really good assistant who's not going to get everything right, but they'll really help you move projects along.

Yeah, fabulous. All right. Well, let's talk about some everyday uses of AI for busy professionals. This certainly applies to us in medicine, but really, I think would apply to anybody. Give me some examples, guys, of what are some ways that this can be used in kind of everyday life?

I think one of the places we could start is by talking about how do you manage your email? Like, I don't know about you, but like email is something that that can easily consume hours and hours of your day. And I think it's a it's a space where it's also large language models are built for dealing with things like that. And so I think it can be a good place for us to start. Yeah. So what tell me about that? Like, how can we use something like ChatGPT or some of these others to help us with that?

So I'm a chat GPT user. Maybe at some point we should talk about what are the different options. I'll pin that for now, though. So if there's a particularly long email thread or there's just one long email and you want to make sure you're not missing any key points,

Uh, just putting it into it and saying, Hey, what are the key points? Or here's the email they sent. Here's my reply. Am I addressing everything that I need to address? That's, that's a real nice, especially for the busy folks who get long emails. That's a real nice way to, you're not make sure you're not missing stuff. I'll, I'll caveat that. And, and I think, uh, uh, anyone listening is probably thinking this too. Um,

Obviously, not for sensitive information, but if it's the type of thing where it's fine to have it out there, then I think it really helps with that. Dave, how do you use it? I'm usually... I'll copy-paste the stuff in there that I want it to think about. Yeah, I think we should also pin and come back to what data you're sharing and what data you're not sharing. But I think that...

I have seen it really help. And you've probably seen this on Apple intelligence commercials now that have come out where it can adjust the tone of your emails really well. Like, like sometimes I write an email and then I look it over and it's, I'm kind of waffly and I'm like, no, I want to be more direct, like more forceful and I'll ask it and it'll make it more forceful or, or I'll have a bad day and I'll write an email and I'm like, is this,

a rude email and I'll be like, can you make sure that this tone is not inappropriate for professional setting? It could do that. I had a faculty member that I used to work with who just

um, always came off kind of brusque in email, like was very direct. If you, if you read it without any context, you would just be like, yeah, it's just direct. But sometimes people thought that that person was, was rude or insensitive. And they started using, uh, one of the AI models to modulate their emails to be more sensitive and then got positive feedback. Like,

Like, wow, they've really turned a leaf on like how they communicate. And so I think it can really affect kind of our professional persona if you use it correctly.

Yeah, that's awesome. Go ahead, Dan. Oh, I was going to say 100%. I think the tone thing is huge. If one of my sort of triggers for thinking about using it is if I have any emotional valence as I'm writing the email, I'll paste it in there and just say, hey, can you check me here? And it's really nice. I can specifically remember one

when I said, give me feedback on this, what should I adjust? And it called out the one sentence that I, even when I wrote it, I knew like, I don't know if I should say this, this might be a little too direct. And it called it out and said, you might consider this alternative way. And it was still, it was just, it was, it was still getting the point across. It just took out all of that unneeded, unintended confrontational feel. Yeah, that's amazing. So

Could you let's say you get added on to an email chain that's been going on for days. Right. And it's like there's a million back and forth that you haven't seen yet. And now someone adds you and says, what do you think? In my life, that means I have to comb back through and figure out what's going on here. What was the initiating issue? Right. Could could I paste that whole thing in and then say, like, tell me what I need to focus on?

Yeah, you can even make it simpler. You can just say, give me the TLDR and then paste it in it. And they'll know that what that means. And it'll, it'll give you the bullet points. I'm going to, I'm sure we'll all come back to this a number of times, but I'll say you got to make sure that what you're pasting in there isn't sensitive though. Yeah. And sorry, you're going to have to tell me what TLDR means. Oh, too long. Didn't read.

Too long, didn't read. Okay. And it'll know, it needs to tell you what you need to know. Yeah. Yeah. Yeah. I think, uh, uh, forgive me. I don't know that or there's bluff, right? Bottom line up front, all these little things, but, um, Hey, uh, Jed, if you're, if you, if you like, uh, you can go to chat GPT, you can say, Hey, I just learned about TLDR and bluff. What are other things like this that I should know about? And it'll give you the whole list. Oh, I'm sure there's a long list, but that's awesome. So, uh,

So, okay. Let's... A couple things. One, it's come up a couple times, so let's just address this. You know, obviously, I'm guessing, but you guys tell me, like, you wouldn't put, you know, an email that had private patient information in there, right? Or, you know, things like that, right? Are there other things you think, like, draw, you know, where you draw the line of what you would and wouldn't upload to it? I think that you can, for almost all of the large language models, you can tell them stuff that you are okay sharing or not. And so, I think...

Probably for most of us, the best practice is to not share information, especially when you're using it for these personal type tasks, because it's not really going to help it build this large language model. And that's one checkpoint on you accidentally putting something in there that you didn't mean to. I think when it comes to the head vibration information, that's just easy. I don't think any of us use it for that purpose. And it's just...

not helpful. Now, I'll say there are certain places, so for example, at Stanford, we have a secure GPG that's approved for even patient care information in certain settings. And so there are institutional barriers that you may be able to use one or more of these tools in that space. But I think

Patient or private information is just a no. I think the in-between sensitive information is where the turning off the yes, feed this into your model gives you an extra layer of safety. And then random mundane stuff like what is TLDR, I think you'd be fine with on any model.

For sure. Yeah. I'll add, you know, depending on what you're using it for, there's some of the, obviously, this is not allowed either legally or policy, like patient stuff. And the policy stuff, though, say you're working on a manuscript or here's another one. I do a lot of peer review.

Um, a lot of publishers, a lot of journals will say, actually, we do not allow, uh, you to use, uh, AI for peer review because you'd be uploading content that's proprietary and so forth. So there's those sorts of policy things. And then I think there's also the, uh, sort of the gut check thing. Like if, if, if, uh,

If people knew that you were using AI for this, how would that come across? How would that feel? I can't remember the place, Dave. I don't know if you remember. It was Vanderbilt. Yeah. I think last year they sent an email in response to the cool shooting at Michigan State University. And at the bottom of it, it said that it was paraphrased from a GPT model. And they got just a ton of backlash because it was like,

Really, you kind of like outsource being sensitive to student needs at this time. And I think that's just a bad look.

Yeah. Yeah. So there's the, there's the strict don't use. And then there's also the, if other people knew I was using it for this, how would that. Yeah. One of my friends who's in this space, like said, if you feel like you would need to open an incognito browser to search for this or do this, then like, that's probably a, like, maybe I shouldn't use DPT for this. Interesting. I love that. Okay. So yeah. And one thing, Dan, I think you told me, uh, was that, you know, often maybe not all of us, but a

if I have to, you know, write like a two page thing, I would way rather edit it than write it from scratch. Now that may not be true. Everyone's definitely true for me. I know you said it was true of you. And so that is something I think, and, and I, you know, maybe it's going to be, we're going to get more guidance on this going forward, but if you, I wouldn't do it if you were writing like a journal editorial, right. But for an email, something like that, you know, a summary of something, it seems to me like you could do this,

Tell it what you want, get a two-page draft, and then edit it as extensively as you want. But at least you're doing editing instead of having to come up with it from scratch. 100%. There's that first – the first step of any sort of writing or thinking that you're doing is just putting down the stream of consciousness ideas of what you want to include.

All you have to do is just give that to chat GPT or cloud or like any of these other ones and say, I want you to turn this into an email, a policy, whatever it is. Here's an example. I don't know how many listeners will be familiar with it, but the ABA is changing its policies around the basic exam.

And the there's a program director society that just released a jet. I know you're very familiar with this, this recommended policy. So again,

What I think I'm likely to do is meet with my core team we're meeting this evening. I'm going to collect all of the notes, all of the thoughts. I'm going to put in both of the ABA and the program director group's policies and say, hey, based on all of our collective thoughts here, help me draft a policy that I can send around to my team for review.

And that's going to – I'm sure it's going to be decent starting quality, and it's going to save a lot of time. Yeah, absolutely. And, of course, you're not going to send it out verbatim, I mean, unless you want to. But you can edit it as much as you want. It's just a good starting point. 100%. Let me ask you guys this. There are – at least ChapGPT for sure has both a free and a paid model. And I know that you guys convinced me it was worth – and I now completely buy into this – that it was worth paying for the paid model or the paid version. Is that –

At what level do you – is that just kind of across the board, like you should do it, period? Or is there sort of a, like, you're using it this much or for these things, then you should get the paid version? I think for any of the listeners who haven't tried them, I think trying the different ones out, whether that's barred with Google or Quad or ChatGPT, and seeing how it goes. I think the two biggest upsides you get – and I'm curious what Dan thinks. But for me, the two biggest upsides in the paid model are you get access to the newer –

uh ai models so they're just better they're better in their factor like they give you a more complete answer usually it's the it's better and then they usually have other features that come with them so for example with chat gpt you can you can paste data into it and it will give it back to you in a table and you can download the table file or you can download the powerpoint slides or other things that it'll create for you as as files as opposed to just giving you text and that can be really powerful when you're trying to get it to do more complicated things

Yeah, 100% agree. Dave, I heard you say that it's just smarter model, better reasoning. I think that's true. The extra features, like being able to make custom GPTs for things you do routinely or upload files or make images, those sorts of things. And then the third thing is that sometimes bandwidth is throttled. So I think, and I don't want to misspeak, but I believe that with...

cloud. Um, I think it's certain number of messages, certain amount that you can do for free every X hours. And that's a substantially higher amount that you can do. Um, if you do the paid model, uh, just for, for listeners, most of these seem to be in the ballpark of $20 a month, um, for the subscription. So I, I'll tell you what I did was I started with the free version. Um,

And then I thought, I'm using it enough that it'd be worth trying the other one. And at this point, the hours that I spend are well worth the $20 a month to me. But yeah, Dave, did you start? Did you just buy the advanced one right away or did you start just trying it out first? No, same thing. I tried the free model, a couple of different ones, and then settled on buying one of them. And that's worked. And then there are times where I use some of the other free ones for different tasks as well.

All right. Awesome. So let's talk about teaching. How can these tools help us with some teaching that we may be doing? One of the tools that's relatively new that I've been playing around with quite a bit is called Notebook LM. And it's a Google tool. It is a kind of pre-production tool. So you need a Google account to log in. You don't pay for it.

But I would say what it really excels at is it's a closed model. So it will only use what you feed into it. So you create a notebook and then you put content into it. So, for example, like I might throw a book chapter, like download a chapter from online access of a textbook like Miller, put it into there and then ask it to make me a podcast. And it'll make a two person podcast using just that book chapter.

No, it's not always perfect. Sometimes it gets a little perseverative. And it definitely does better with narrative text than it does with, you know, a manuscript where there's complicated analysis or things like that. But it can give you, it can give you a summary. It can give you a quiz. It can give you a podcast and it, and it uses only the material that you feed it. You don't worry so much about it coming up with strange things, but I I've,

done that with, hey, Rez, you listened to this podcast the night before, and then we're going to come talk about it. And I think there's higher uptake for that because they can do it while they're driving or working out or other things rather than trying to sit down and find 30 minutes to read a book chapter.

And I assume these are only topics not covered by ACRAC, right, Dave, that you're using? Obviously. Yeah. Okay, good. Yeah, and Jed, even if they were overlapping, they don't hold a candle. Thank you. Yeah. So one thing, a couple things I want to mention. So that Notebook LM by Google is a pretty neat –

a neat thing. I've used it as well. We had a patient at one point with neurofibromatosis. I uploaded an article that was neurofibromatosis and anesthetic implications. It made a eight minute podcast and I sent it to my residents that listen to this. And it also made a little study guide and do these short answer questions so that when we talk about stuff, you'll, we'll be able to go back and forth on it and sort of take you to the next level. So I think it's, I think it's really, really helpful for that. One thing though, that,

Dave was sort of hinting at with this closed model. A lot of them, ChatGPT, BARD, Copilot, Cloud, they borrow information from other places and they don't always get it right. In fact, they can get things very wrong. And so I don't think any of these are, especially if you're using it for

Stuff that you'd consider high stakes, to be honest, even for low stakes stuff. I feel like you should review it. A classic example. I just tried it. It looks like it's better now. Until very recently, if you would ask ChatGPT, how many R's are there in the word strawberry? It would say two.

And it's just not right. And then you point it out and say, oh, yeah, yeah, there are three. But I've definitely seen it do things like that with clinical care, clinical scenario. The other thing that I think it still takes some massaging, I've made a custom GPT to sort of make mock oral stems to teach residents. And so I'll put in a basic made up here. Let me...

I'll go to it right now. Is it okay if I do it live in front of us? Yeah, let's do it. Let's show it. So I'm going to say 55-year-old M, so it says 55YM with COPD for left upper lobe ectomy. That's all I'm going to write. And I just hit enter. And I could say, like, I want you to give these complications, all these other things. And just based on that, this is a custom GPT.

This is what it says. This is what it's output. Patient profile, age, 55-year-old male, medical history, COPD, longstanding, moderate severity, history of chronic tobacco use with cessation one year ago.

Surgical procedure, left upper lobectomy for long adenocarcinoma. Clinical context, the patient's scheduled for an elective left over lobectomy. He's been experiencing dyspnea on exertion but denies breast dyspnea. PFTs show an FAV1 of 55% predicted and a DLCO of 50% predicted. The patient's six-minute walk test shows a distance of 400 meters. And it goes on and on and on. And then it has these questions that it divides. It's got some risk stratification questions, some optimization questions, questions around the anesthetic plan.

So I could take this as a starting point and say, oh, this doesn't feel realistic. I want you to ask more about this or that. And I mean, this took us...

whatever, 30 seconds in five minutes, I can have something pretty robust that I can take into the OR and go back and forth with a resident about. Yeah, that's awesome. And when you say custom GPT, basically this is you go into chat GPT, you're basically creating a GPT of your own or essentially a thing of your own that when you type in that thing, it knows what you want. You've trained it to say, to give you like a thing that looks like an oral boards.

done. Yeah. And that may make me seem like just this IT guru. No. Basically, the way that you train these custom GPTs is you say, I want you to do this with this type of information that you get. And it asks you questions like, oh, do you want a professional tone or an informal tone? And then you try it out and you say, well, no, I want a little more depth than that. And so there's no coding. This is

for this sort of thing. It's like putting a conversation to come back to. Like, it's like you've started a conversation with the robot and it's, now I'm going to be able to come back to that exact point in the conversation and pick it up whenever I want. Yep. Yep.

Yeah, and I will vouch. So, Dan, you showed me how to do this. And again, like there's not a lot of people who are less tech savvy than me, but I was able to do it. So this is as complicated as it sounds. It's quite easy. All right. So lots of cool teaching examples. What about kind of organization? Soundboarding, Dave, I think you use it for this, right? Talk to me about that. How do you use it for soundboarding?

Yeah, I think where you're just trying to get an idea off the ground and you're not really sure –

where to go. It's something that you can just, like I would pick up the phone and call somebody and be like, hey, what do you think about this? Recently, for example, putting together a proposal for an ASA talk, I was like, hey, I need a creative name for this talk. Here's what I'm going to talk about. Throw out some creative names. And sometimes it gets too crazy. You can be like, no, less crazy than that. Or you can be like more catchy than that. And I think that's a way where you can kind of just go back and forth with it and help it generate things. And it can come up with some really cool stuff.

Dan, what do you think? Yeah, 100%. I actually, it rarely is creating something like, oh, completely, I didn't think about this at all. But it will have some use cases or some ways of approaching it or thinking about it or framing it that I haven't thought of. And it's a little bit of that, you do the thinking on the inside of your head, this makes it on the outside. And I feel like it really accelerates the process.

Yeah, that's amazing. All right. What about visual creation? And I will tell you a story. I think it struggles with this. At least it can, because I asked it. My family is going on a trip to the Galapagos Islands, and my brother thought it'd be fun to have a T-shirt, you know, to commemorate this trip. And so I asked it, you know, make it make a T-shirt. I told it like what I wanted on it. And it could not spell visual.

Galapagos or our name correctly, no matter how many times I corrected it. And it would get one, but not the other. Then it would, then I would say, okay, that's right. All I want you to change is spell our last name this different way. And then it would change the whole thing, get the last name, right. But misspell what was hard previously. Like it just has a really hard time with this, I think. But you tell me like, do you, am I, am I just not doing it right? Or what's going on here? No, I think you're fine. I've at least for now I've bailed on using it for,

for, uh, uh, sophisticated picture production. It can make simple things, little logos, that sort of stuff. Um, another thing that I find is that the more you go back and forth with it, if it's, if it's taking you in the wrong direction, you just need to stop that session, start a new session from, from scratch. But pictures are not something where, um, at least for me where I'm using it on the regular yet. Dave? Yeah, a hundred percent. Like I think I,

Both, once it goes down the wrong track, it just seems to steamroll down into the worst track. So definitely killing it. And it's terrible with words. And I was asking one of my friends about this, and he said it's because it's not writing words, it's drawing pictures. So it's drawing a picture of words, and therefore it...

riffs on what words look like and so um i'm showing them a logo that i made but i had to redo the text because it couldn't get the text right but the pictures are actually pretty cool that it would come up with so i asked you to make a stanford critical care division logo um and so some of the stuff i think it does okay i think low stake things for this it can it can do a neat job but it definitely does not do tech as well yeah how did you how did you add your own text dave

So I added it to PowerPoint, and then I tried to map the text. And it's not perfect, and I've only played around with this for a little while. But then this is just like an overlaid text on top of the image. And Dave, even there, it looks like – is that a stethoscope on the left? Yep. Like it's a little funny looking and like – Yeah. Yeah, yeah. Jed, I'm with you. Yeah. Yeah.

Yeah, but it's neat. It's fun to play with. I don't think she'll ever listen to it. My 10-year-old daughter, this was a couple years ago, she really needed to brush her hair. And I told her that she looked as though she was ready to house a family of badgers in her hair. And so I asked ChatGPT to make a little blonde girl with a family of badgers living in her hair, and she looks on Kempton. It was great for that, but not so much for...

But I really need this to look a particular way. Yeah, that's awesome. All right. So I'm sure it will be improving on that in the years to come. How about data analysis? You guys touched on this a little bit. Like you said, you could, Dave, you said you could give it some data and tell it to give you back the data in a spreadsheet format. But how can this be used for folks who maybe just want to analyze some data? Stay with us. We'll be right back.

Hey, ACK Rackers, we have a fabulous new sponsor, and this one is near and dear to my heart, TrueLearn SmartBanks. I've used TrueLearn for my residence for years, and it is hands down the best smart bank out there. The questions are excellent, and the answers have clear explanations that break down why each choice is correct or incorrect. I love that it's customizable, so you can get target weak areas or take longer or shorter practice exams anytime.

Their metrics are great, and their mobile app lets you study on the go. We use their ITE, basic, and advanced smart banks for our residents, each with 1,000-plus questions, realistic visuals and graphs, and even a calculator that mimics what you have on the exam. Our residents love it and get great results. TrueLearn even has the CME smart bank where you can get AMA, PRA, Category 1 credits after you finish training, and an adult cardiac anesthesiology smart bank as well.

Hey, folks, absolutely no joke. Last night we were eating our factor meals and my daughter said, how do they make it taste so good? It's like we're at a restaurant.

Even my two younger daughters, who are very picky eaters, are loving every meal we get from Factor. Some favorites are the chicken tikka masala and the chicken taco bowl, but they love everything. In addition to 40 different meal options across eight dietary preferences, every week you can also choose from smoothies, add-ons, breakfasts, and more to keep you going all day.

We added on some breakfast options and the kids love those too. The convenience is amazing. Two minutes and the food is ready to go. Honestly, I'd still eat them for the convenience even if they weren't so delicious. But the amazing thing is that it's super fast and incredibly tasty. I wouldn't have believed it until I tried it and they're super flexible. You can change your order up anytime, pause or reschedule. Eat smart with Factor. Go to factormeals.com slash factor podcast for 50% off plus free shipping.

Use code FACTORPODCAST, that's F-A-C-T-O-R-P-O-D-C-A-S-T, Factor Podcast at factormeals.com slash factorpodcast.

For a limited time at Verizon, you can get our best price ever for a single line. Just $45 per month when you bring your phone, which is less than you spend on too-tired-to-cook takeout every week. Get one line on unlimited welcome for $45 per month with AutoPay Plus taxes and fees. Visit your local Verizon store by April 2nd to save.

$20 monthly promo credits applied over 36 months with a new line on unlimited welcome. In times of congestion, unlimited 5G and 4G LTE may be temporarily slower than other traffic. Domestic data roaming at 2G speeds. Additional terms apply. All right, and we're back.

Yeah, I think you could do a couple of different things. And I think one of the things that I was really excited about doing it recently is I was giving a talk. I wanted to present some of the data from a paper. I didn't want to use the table or the chart from the paper because I wanted to respect copyright. But I also didn't want to like

Didn't have time to do a whole big analysis on my own, but I pasted the paper and I said, hey, can you give me a boxed and whisker plot for just this row of data? And it did. And then I could refine it from there. And it actually came up. It was actually a much more powerful add to my presentation than if I just shown the table or even shown the table and highlighted that row or anything like that. And I think.

That works. It works pretty well with stuff like that. And also it can take tabular data and make it into a new table or take a graph and estimate what those numbers look like into a table or vice versa. I think that can do well. So you you pasted the article in and you said, look at figure, you know, or table or whatever it was, you know, look at table three. And you said I just screenshotted the table like I just I just get a screenshot of the table and said, here's the table.

I want for this, it was pulmonary vascular resistance. I want you to make a box and whisker pot for just this pulmonary vascular resistance. Nice. Okay. Awesome. Dan, how about you? I only learned about that capability from Dave actually last month. So I haven't gotten into it a whole lot, but it looks incredibly powerful. I will be using it.

It's cool. You can also ask it, like, if you want to do real analysis, like, you want to look at resident true learn scores and then how they do on an exam. And you could ask it to do, hey, multivariate analysis, you throw in these other variables or, and then you can ask it, how did you do that? And it'll show you what...

what usually uses R, but it'll show you how it did it, how the stats was done. So even if you want to kind of peek under the hood, it can be really helpful. I do think a caveat of like, it can make mistakes like anything else. So I think being thoughtful about that and giving it just at least the sniff test of like, wait, does this make sense or not? But for some of those things where you're like me, like I can't remember the code for the T test that I want, but I can ask it and it will come up pretty quickly with the right stats test for simple things.

Yeah, I actually just to echo some of that, just in general, I use it when I use Excel a lot. Sometimes I'm like, I don't want to think about, do I clip the left part of the string? I'll just say I have this is what the data looks like, and I would like a formula to make it look like this.

So just earlier today, I wanted if you have things in Excel that are in date format, you can't use them as titles of column headers. And so I remember there's a way to make that a text. I just didn't remember how. So I just give me the formula and it gives you the formula. You can have to do really complicated formulas, too, and then just paste them in.

Wow. That's amazing. All right. Let's talk about we've kind of touched on a variety of or that there is the existence of a variety of tools. I think the ones we've mentioned so far are obviously ChatGPT, Cloud and BARD. But let's talk about what what ones do you think people should know about so they can kind of play around and decide what they like?

I think those are the big players. I think Copilot for, especially if you're an institution that is Microsoft heavy, Copilot's the Microsoft version.

I'll add, we talked about Notebook LM is a nice one. And all of these, by the way, you just type them into whatever your favorite search engine is and it'll be the first link. Another one, I don't know, Dave, if you've seen this yet. I'm new to it, so I don't want to give it two thumbs way up, but it seems initially pretty promising. It's called Open Evidence. It's specifically for healthcare workers, specifically.

Uh, and it, it seems to draw, it only allows you to ask like clinical stuff or stuff that would be published in papers. Um, and it seems to draw, and then it will reference like, here's the PubMed paper with the link to the paper. So you can ask for like, what's the data for tranexamic acid for postpartum hemorrhage. Um, and it'll give you a little summary and,

here are the three trials that are related to that. So far with the fairly simple questions I'm putting in, that's been pretty impressive. The other one that I started playing around with, but not gotten a ton into is called Research Rabbit. And it's researchrabbit.ai. And it allows you... I think it does a neat job of showing you, if you give it one paper, what papers are related to that paper and how tightly they're related and what

concepts they're related by. And so if you're writing a review article or you're just trying to get a sense of the literature in a certain area, I can see it. It was a little buggy the first couple of times I've used it, but I think it's getting better and it'll be interesting to see if that's helpful for people who are more research heavy.

So those two that you just mentioned, obviously, Open Evidence and Research Rabbit are going to be obviously more for research. ChatGPT I think of as like basically anything, right? And are there other of these? Notebook LM we already discussed and how that's closed. So whatever you put in and it can give you a podcast or a quiz or something like that. Are there other things you think, oh, for this kind of thing, I would go to that tool?

I think Claude tends to be focused a little bit more on natural language processing. I'm only starting to use it a little bit more, but it seems to get tone a little bit less robotic than chat GPT and a little more consistent. So I like that one for if I'm thinking about writing prose or maybe it'll even be what I switch to for emails. Yeah.

Yeah, I mean, if you're really into the Google infrastructure, BARD is increasingly integrated. And so you can use it within Google Docs or Google Calendar to do certain things. It still has some limitations. But if you want to take something from your Google life and move it to another part of your Google life, it's, I think, increasingly good for that. And I would say Copilot similarly for Microsoft does some of that. Yeah, totally. Yeah.

Fabulous. Okay. So lots of interesting stuff and I'm sure more coming out. I know with chat GPT, there's now a functionality where you can talk instead of type. I've played around with that. My kids think it's a blast to like ask it some question and have a conversation with it. But from a practical standpoint, I mean, I guess in terms of like putting the input in whatever you prefer, right? You can talk it in, you can type it in. In terms of getting the input back out, I'm not sure how useful it is to have it talk to you because you can't then like cut and paste it anywhere. But have you guys played around with that at all?

Yeah, I have. We're talking about AI for work. I tend to use the voice stuff for AI for work.

not work and there's a lot of use cases for that so i mean you you can say hey i want you to play a game of 20 questions with me and my kids um and so you think of an animal and we're gonna ask you questions or um uh i wasn't paying attention to this tv show it's episode three and so i missed what happened here can you tell me about these two characters like what did i miss there's all sorts of uh

I'd say non-work stuff that's just like things that I would have typically or historically Googled and then try to find something to do it all. If it's simple enough, I'll do that. But for me, it's the voice. I use it mostly for that sort of thing or just as like a cute thing. Like, oh, isn't this neat? Right.

Yeah, similarly, I haven't used it a ton. I've used it when driving to be like, hey, is there a taco place around here? And it does really well with that. I would say often better than just Google searching or Google Maps searching or Yelp searching. It'll find something because if you tell it, it'll tell you where you are and then it can search a number of different sources.

Nice. Is there anything for kind of making life easier in academic work that we haven't touched on that you think people would find useful?

I don't know how useful it would be, and this is academic but also personal. If there are long policy documents or long – like everything from like what is required for me to submit this article to this journal, and it's just this really long thing to – I don't know. You want to apply for a certain credit card, and here's the 20-page contract that nobody reads. So you can just plug it in there and say, hey, what do I need to not miss or –

For the contract, like, is there anything I'm signing it for that I should really be aware of? You know, like those sorts of things. And it'll read it out and like, oh, just you're agreeing to arbitration here. Like that's the only thing to pay attention to. Like those sorts of things. So I do like it a lot for really anything where this isn't a high level thing that I need to do. It's just going to take a while and I have to slog through it. It's really nice for simplifying those things.

Yeah, I think similarly for repetitive tasks. I had the other day this messy... It was a list of names and emails, but they weren't all formatted correctly. And I used to sit there and pick out the emails and copy and paste them over. And now I just throw the whole thing into ChatTPC and say, hey, give me a list of the emails, separate them by a space and a semicolon. And it does. And I copy that right into an email bar and it's ready to go. And that just saved me five minutes of annoying repetitive work. Nice. Yeah.

Dan, I know you had showed me a version of something where you can put some information and it'll make you slides. That is another way to kind of use this. What was that called? I'm completely blanking on it right now, but it won't take me long to find it. So if we talk about something else for a minute, I'll look it up. We'll come back to that. I will say about it while you're looking for the name of it that you had showed me. You were trying to do a presentation. It was just short. You needed like six slides.

You put stuff in there. I tried using it for the same thing, found it very useful. I think it's limited. You can't do more than a certain number of slides. So I think you can't like tell it, make me a 75 slide, you know, hour long talk. But you can use it for little things you want to present. And it does a pretty nice job of making the graphics, the text. And you can, of course, edit it just like you could anything else to make it what you want.

That's exactly right. And hopefully I'll find it. And if not, are there show notes? Yes, we'll put it in the show notes. Yeah, for sure. All right. So let's talk about limitations of this. We've talked about some. We've mentioned that, you know, it obviously isn't always correct. But when you guys think of cautionary, you know, things to be aware of in terms of limitations and potential concerns, anything we haven't covered that, you know, you think we should cover? Sure.

I think all large language models are,

They're built from something. So whatever they're built on, they're going to propagate that for it. And I think there's increasing understanding of this in terms of things like bias. So if it got fed a whole bunch of stuff that has some underlying misogyny or gender bias, it's going to propagate that for it in suggestions that it makes. And so being thoughtful about that and about sensitive topics, I think, is really important. Yeah.

um i also uh i also just reiterate the like the ai is set the way it's creative is because it's hallucinated so sometimes it will tell you things and that they are just not true and so um being cautious and using it as your starting point and not your finished project that it could be helpful yeah yeah i i couldn't couldn't agree more i think the the issues of of bias are real i think the issues of uh

of, uh, you really, it can't be your finished product. If this product, if this is anything that's important at all. Um, and then the only other thing just to be aware of, um, the muscles we don't use, we lose, uh, in general. So, um, I wouldn't be surprised if, if my creative writing, uh, ability decreases for emails, uh, over time, that's a skill I'm happy to lose it. I'm okay with that, but, uh,

It's just something to keep in mind. Yeah. Yeah. I wonder if our kids are going to grow up just like unable to do this without this kind of technology, you know? Yeah. And I think we did find, Dan, the name of that slide software. What is that called? It's Gamma. G-A-M-M-A. Gamma. Yeah. One other, sorry, one other way that I've gotten burned early, early in my chat, GPT use,

I looked at the email. It looked fine. I sent it. But it was not my voice at all. And so it struck the recipient very, very strangely, which was appropriate because it's just not something I would have written. So I think we talked about reliability, accuracy. We talked about confidentiality and regulatory. We talked about dependence or over-reliance. And then that last one I would add is just voice.

Yeah. Yeah. And let me ask you this. Can you say, here's, you know, a thousand of my emails. I want you to write, you know, an email in my voice. Will it do that? Yeah. Yeah, you totally can. You can actually even, even for fun, here's a bunch of my emails. Describe the way that like, describe my tone. Like what are, how would you describe how I write emails? It can be kind of neat for some, some self to, to bolster some self-assessment.

Now, that's awesome. So let's say that I have a paid account for ChatGPT and I put a bunch of my emails and I say, you know, I want you to remember my voice. Do I have to do a custom GPT in order for it to remember it or will it always remember it? I think for the most part, there are some things that you can apply across all of your sessions. But for something like that, I think custom GPT is probably the way to go. A lot of times you start a new session and it's as if it just doesn't know you at all.

All right. Dave, would you agree with that? Yeah, I agree. I think if you want it to turn, like if you want to make an email GPT, and again, this is not something that's very hard. Like it would take you probably 15 minutes to set up or to get it to like a decent place.

Yeah. Awesome. Okay. So if someone is listening and they're like, I haven't done nothing, I have never tried this at all, but I want to get started. What would you tell them how to start? What's the process to make this, you know, something they could sit down today and do?

Pick a model. If you're not sure what to pick, ChatGPT was like the earliest one and it's what a lot of people use. So just go make an account or sign up through your Google address and just play around. Just pick something low stakes like, hey, I'm

make this email shorter or, um, tell me the history of baseball in this country or whatever comes to your mind and just play around with it. Um, and then, uh, uh, hopefully that decreases the, the activation energy. And then you start saying, well, maybe I could use it for this or this. And, um, at least that's, that's, and that's not, that's good advice for tech in general. Just, just start playing with it a little bit. And I think it definitely works well for, for this space as well. What do you, what else do you think Dave?

Yeah, I mean, if you're like me and you have always too many Chrome browsers open, just like leave a browser with it open separate and then you'll see it on your screen. You'll be like it'll be a reminder to like ask it something or the next time you're stuck on a task to see if it can help you. And I think that's a good place to start.

One thing that I found pretty funny when I started playing around with this is I would ask it for something. It would give back, you know, whatever it gave. And I would kind of find myself thinking it's not exactly what I wanted, but it's probably good enough. And I feel bad asking it to redo it. Right. As if it were a person. And then I'd have to, like, say to myself, wait a minute, wait a minute, wait a minute. Like, that's not it. It doesn't care how many times you ask it to do it.

Uh, 100%. So when I, my first thing in using, uh, uh, chat GPT, I was actually, uh, interested in like doing a little coding in Python. And I was like,

I tried to learn, I tried to do some stuff and I would have some issues with my code. And so I would paste it in there and say, can you tell me what the issue is? And they're like, oh, you have an extra space here. You need a semicolon here, whatever it was. And then I would go back and work a little bit more and I would feel so bad coming back to it. And now I just don't even care. I'm just like, nope, do more, do more. But it does, it does feel, uh,

there's a little bit of a human feel to it, you know? And so you feel bad about taking advantage of it. But I don't know, give it a couple months of regular use, you get over that, I think. Yeah, yeah. I just found that to be really funny. Like it's programmed into us to like not ask to, not inconvenience other quote unquote people too much, you know? And it's just hard to get that out of it. But maybe our kids won't have that same, right? Totally. My four-year-old doesn't mind screaming at Alexa to play his favorite song over and over again. Absolutely. Absolutely.

Any thoughts just kind of briefly on the broader impact of this stuff? We've focused a lot on, you know, the very practical day-to-day, but just a little bit of, you know, how we see this impacting things on a broader scale. You know, I think anesthesiology, we have so much data that is coming at us. If you think about most of our monitors are getting vital signs at 60 hertz. And so we're

We have a really cool data science initiative here at Stanford, and they're working on all sorts of things. But one of them is like, can you predict vital signs into the future by using some of this data? Like we're, I think, only scratching the surface of what you can do when you have an exponential increase in computing power to deal with large data. I think like...

Dan's example of like throwing the long email chain in there and getting a summary is, is just like a microcosm of what we could do in terms of patient information and, and being able to summarize and analyze. But that stuff is really exciting to think about what it could look like. And then,

I think also then what that frees you up to do, like Dan was saying, maybe your creative writing for email skill gets worse, but maybe that opens up your skill to do something that's much more important and powerful. And I think that's really exciting to think about.

Yeah, I couldn't agree more. The idea that maybe on the airway team walking to a code, I could just open up my EMR and get, Jed, this is just for spaced repetition for you, the TLDR on this patient. I don't want to read through all the notes. I just need, what are they here for? What are their comorbidities, recent labs? Maybe there's even a button that I just hit that just that they've arrested. And it says like, here's what you need to know. And here's

Top five things that you should probably consider for causes of arrest because that's what it's really good at, taking all the data and just very, very quickly organizing, summarizing, and helping us interpret. There's a quote, Dave, I don't want to get it wrong. I got it from your session that you did at the SAAAPM meeting. It was something like, AI won't replace humans forever.

But does this sound familiar? Yeah. But humans with AI will replace humans without. Yeah. AI is not going to replace humans, but humans with AI are going to replace humans without AI. Yeah.

Yeah, that's really profound. You know, I'm really looking forward to, I'll in a couple of days be interviewing Christian Mayhoff and a colleague of his. They're going to do a pro-con debate about whether we should be doing continuous monitoring of all surgical patients post-op. And

You know, that might seem like a no-brainer, except I think one of the big things is exactly what we were just talking about, is what do you do with all that data? And we are not capable of dealing with all that data, right, as human brains. But the better AI gets at this, I think, we'll see what they say, but I think

This is going to be the answer is when AI gets good enough to be able to know with extreme reliability what is and is not important and tell us so that we're not trying to comb through, you know, seven trillion pieces of data to pick out what we think is important. That may be when the tipping point comes.

Let's wrap it up. I think, you know, we've talked about a lot of really interesting things. And the really, I think to me, the summary here is that you can make your life easier. You can save time. Even if it's just a few minutes here and there, it's going to add up by using these tools. And it does not take a lot of expertise or time to learn how to do them and to start incorporating them. You guys have taught me, and it's not like just a listener's note, why

We're not in the same place. We're not on the phone every day. I mean, all that I learned from you guys, I learned in about 10 minutes of sitting next to each other at the SAAAPM conference. So it's very easy to learn, pick up, and start using. And I think if people take that away and just start playing around, you're going to find it's really useful. But what do you guys want to add as a wrap-up?

I think that was beautifully summarized. Give it a try. Have some fun with it, too. You don't all you don't have to do all the I use it a lot, honestly, for like just fun, silly things with with my family. That's all I had that. Awesome. All right. Let's turn to the part of our show where we make random recommendations, something fun to share with the audience for them to check out. Dave, let's start with you. What would you recommend?

So my new California lifestyle, Thanksgiving weekend, went hiking with our kids and some friends in Rodeo Beach in Marin, which I had actually never been to, but is a really cool place. If you're on the West Coast, near San Francisco, it's like a quick drive over the Golden Gate Bridge and an easy kind of up and down hike. You can make it into a big thing or not a big thing. The views are incredible and just something different to try.

for me getting to live where it's 65 degrees in December. That sounds awesome. Dan, how about you? That's very awesome. So I got a book as a gift a few years ago from someone. It's like a kid's book, but it was pretty deep, pretty profound. And given the topic, I asked my friend AI here to write a very short summary about it. So it's called The Boy, The Mole, The Fox, and The Horse. Okay.

It's by Charlie Maxey. It's a beautifully illustrated tale about friendship, kindness, and self-discovery. Through their journey, the four characters share gentle wisdom and reflections on life's challenges, offering an uplifting message for all ages. Dave, didn't you just recommend this to me? I did. Did you really? Yeah. It's a good book. It's not very expensive either. It's like maybe $10, $15. Yeah. You can read it in like two hours probably. It's a good one to read to the kids too. For sure. Nice.

Nice. That's awesome. Well, thank you. I just finished a book. I can't remember where I heard about this, but I've gotten into these really long fantasy series, which are great, but man, you commit and then you've got like 10 books in front of you. But this is just a standalone one book. It's not fantasy. It's kind of fantasy.

I guess it's a little bit of a mystery and maybe thriller, but not too over the top. But it's called Conclave by Robert Harris. And it's just a fiction. It's a novel about a new pope being chosen. I am not Catholic. I don't really know anything about it, but it was still a really enjoyable, really fast, easy read. And kind of really interesting and well done. So that is called Conclave by Robert Harris.

All right, gentlemen, thank you so much for coming on the show. Thanks, Jed. Thank you. All right. Hopefully you got as much out of that as I did. That was really fantastic. Let us know what you thought. Go to the website, acrac.com, where you can leave a comment. Others can learn from what you have to say.

If you are a fan of the show, you can follow us. We're on Twitter. We are on Facebook. We are on Reddit. And we are on Instagram. I'm at jwolpa on Twitter. And we're at Akrak Podcast. And you can find us on all those other platforms as well.

If you are a fan of the show, please consider going to Apple Podcasts or wherever you get your podcasts and leaving a comment and a rating. It really helps others find the show. If you'd like to support the making of the show, please consider going to patreon.com slash ACRAC. That's P-A-T-R-E-O-N dot com slash A-C-C-R-A-C, where you can become a patron of the show.

Even if it's just a dollar or two that you pledge, it makes a big difference and we really appreciate it. You can also make donations anytime by going to paypal.me slash akrak or looking up J. Wolpaw on Venmo. Thank you so much to those who have already made donations and become patrons. We really appreciate it. Thanks as always to our fantastic Akrak crew. Sonia Amanat is our tech lead and Sophia Wu is our social media manager. William Mao is our production assistant.

Thank you so much for the great work that you do. Our original ACRAC music is by Dr. Dennis Kuo. You can check out his website at studymusicproject.com. All right. That is it for today. For the ACRAC podcast, I'm Jed Wolpaw. Thanks for listening. Remember, what you're doing out there every day is really important and valued.

For a limited time at Verizon, you can get our best price ever for a single line. Just $45 per month when you bring your phone, which is less than you spend on too-tired-to-cook takeout every week. Get one line on unlimited welcome for $45 per month with AutoPay Plus taxes and fees. Visit your local Verizon store by April 2nd to save.

$20 monthly promo credits applied over 36 months with a new line on unlimited welcome. In times of congestion, unlimited 5G and 4G LTE may be temporarily slower than other traffic. Domestic data roaming at 2G speeds. Additional terms apply.

Geico makes it easy to get affordable renter's insurance in as little as 90 seconds. And then you'll start to wonder why everything else isn't so fast. Like, why does it take so long to do the dishes? A meal that's gone in three requires 45 minutes worth of dishes. Doesn't make sense. But it does make sense to get renter's insurance through Geico. That way, that pile of dirty dishes and the other things you love are covered. Get more with Geico.