WEBVTT
00:00:27.559 --> 00:00:28.199
Hi, everybody.
00:00:28.359 --> 00:00:31.640
Welcome back to another episode of Between Sessions with Berries.
00:00:31.799 --> 00:00:33.159
We have the coolest guest today.
00:00:33.239 --> 00:00:34.119
I'm so excited.
00:00:34.359 --> 00:00:34.759
Dr.
00:00:34.839 --> 00:00:41.960
Steven Ferrara is a nurse practitioner, educator, and health policy leader with over 25 years of experience.
00:00:42.200 --> 00:00:47.479
And he's at the forefront of how AI is actually entering clinical practice.
00:00:47.719 --> 00:01:01.879
He's the founder of AI Nurse Academy, the immediate past president of the American Association of Nurse Practitioners, and the founding associate dean for artificial intelligence at Columbia University School of Nursing.
00:01:02.039 --> 00:01:14.920
He's here to talk about something every therapist is wrestling with right now: how to use AI responsibly, how to evaluate the tools flooding the market, and how to cut the documentation burden without cutting corners on care.
00:01:22.920 --> 00:01:24.200
Steven, welcome.
00:01:24.280 --> 00:01:25.640
I'm so excited to have you here.
00:01:25.959 --> 00:01:27.319
Thank you so much for having me.
00:01:27.560 --> 00:01:29.480
Yeah, well, let's just dig in.
00:01:29.560 --> 00:01:32.759
I always like to just start with like the real obvious questions.
00:01:33.000 --> 00:01:36.759
What do most clinicians get wrong when they start using AI tools?
00:01:36.920 --> 00:01:38.839
Like what are you seeing with all the clinicians you're working with?
00:01:38.920 --> 00:01:44.599
And when I say clinicians on your end, it's nurse practitioners that you're really talking with the most about these tools, right?
00:01:44.839 --> 00:01:45.079
Sure.
00:01:45.239 --> 00:01:45.479
Sure.
00:01:45.719 --> 00:01:58.039
And other clinicians as well, whether that be physicians and physician associates, therapists, what we have are a flood of AI tools in the market.
00:01:58.359 --> 00:02:02.199
And I do think that it's going to improve practice.
00:02:02.280 --> 00:02:12.199
I mean, that's why I am so passionate about AI, because I think at the end of the day, it can make me a better clinician and provide better care to my patients.
00:02:12.439 --> 00:02:16.120
I will say right from the outset, it's not going to replace us.
00:02:16.360 --> 00:02:19.640
And shame on us if we allow it to, right?
00:02:19.879 --> 00:02:27.319
Because it really is in our control collectively as a society to determine how we're going to use AI.
00:02:27.640 --> 00:02:32.599
And what I want us to do is really have those conversations.
00:02:32.840 --> 00:02:34.120
We may not have the answers.
00:02:34.280 --> 00:02:35.480
We don't have the answers.
00:02:35.960 --> 00:02:38.600
Things are moving incredibly fast.
00:02:39.000 --> 00:02:48.680
It's a full-time job to stay up to date on what is happening in the AI space, especially when it comes to health care and clinical care.
00:02:48.840 --> 00:02:49.080
Yeah.
00:02:49.320 --> 00:02:57.800
But I what I really, you know, want people to look at is, you know, it's not automatic, it's not fail-proof, and to use it critically.
00:02:58.040 --> 00:02:58.600
I love that.
00:02:58.680 --> 00:02:58.920
Yeah.
00:02:59.000 --> 00:03:04.120
So that really leans into that question like, what are you saying that people are doing wrong with AI or clinicians are doing?
00:03:04.280 --> 00:03:06.840
So are you seeing people not using it critically?
00:03:06.920 --> 00:03:10.280
Are you having to correct that behavior and the people that you're working with now?
00:03:10.520 --> 00:03:15.560
Just curious, like how that's coming up in the work you're doing with clinicians, what you're seeing like on the ground level.
00:03:15.800 --> 00:03:16.040
Yeah.
00:03:16.200 --> 00:03:17.960
It's a whole different workflow.
00:03:18.040 --> 00:03:23.719
And AI Scribes is the biggest example that we have in healthcare today.
00:03:23.960 --> 00:03:28.600
We're going from author of the note to reviewer of the note.
00:03:28.680 --> 00:03:29.960
And that's not a bad thing.
00:03:30.120 --> 00:03:30.360
Yes.
00:03:30.600 --> 00:03:31.879
But it's a different workflow.
00:03:32.040 --> 00:03:37.320
I mean, when I went to NP school 26 years ago, like that wasn't a thing.
00:03:37.640 --> 00:03:43.960
So, you know, for the vast majority of people out there, you know, they're using these tools for the first time.
00:03:44.280 --> 00:03:51.640
And we have to look twice when we're reviewing a note that we didn't write ourselves.
00:03:51.879 --> 00:04:03.879
What people shouldn't do is assume that everything in the transcript is correct and look at it, you know, critically and you know, make any necessary corrections and and then sign off on the note.
00:04:04.200 --> 00:04:04.680
Totally.
00:04:04.760 --> 00:04:14.280
I mean, even in my own practice, I catch myself doing that a little bit because you get so comfortable in a way with the AI like taking over that burden that it's almost like, oh, the notes are written.
00:04:14.360 --> 00:04:19.000
I can go like take a walk real quick or go to the bathroom for once, you know, like that kind of thing.
00:04:19.240 --> 00:04:22.680
You know, it's easy to get in that pattern of like, you know, it's good.
00:04:22.920 --> 00:04:26.279
Sign off, go to the next client kind of thing where I feel you on that.
00:04:26.360 --> 00:04:28.040
Like, I hope everybody hears what you're saying.
00:04:28.120 --> 00:04:32.519
It's like AI likes to fill in the blanks, it likes to hallucinate a little bit.
00:04:32.680 --> 00:04:38.040
We have to be super careful and read over it and make sure, like, yes, this is in a court of law.
00:04:38.120 --> 00:04:43.400
If somebody read this note, I want it to be accurately written what I'm saying here, yeah, or what they're saying.
00:04:43.639 --> 00:04:43.960
100%.
00:04:44.199 --> 00:04:48.360
And yeah, and I fear of the the the idea of good enough, right?
00:04:48.519 --> 00:04:57.879
Yeah, this looks good enough, and because we're busy, we're busy clinicians, and you know, you finish with one patient and then you're already thinking about the next one.
00:04:58.040 --> 00:05:07.560
So it really requires us to stop, reflect, make any necessary corrections, and then finalize the note afterwards.
00:05:07.879 --> 00:05:14.600
Because we do a lot of scanning, you know, with our eyes and with our brains, and we need to do better than just scan.
00:05:14.840 --> 00:05:15.400
I love it.
00:05:15.639 --> 00:05:23.480
Any other mistakes or things that you're seeing people getting wrong off the bat right now when they're just hopping into AI without education, like the type of education you offer?
00:05:23.720 --> 00:05:35.079
Yeah, I I think one of the hallmarks is making sure that patients are consented to use, to agree to use an AI scribe in the office.
00:05:35.240 --> 00:05:37.240
I can't assume anything.
00:05:37.400 --> 00:05:46.360
And, you know, I I think it's incumbent upon the clinician that's using, you know, the technology to describe it in plain language to the patient.
00:05:46.519 --> 00:05:53.160
You know, there's a lot of misinformation out there, you know, does it mean it's gonna be discoverable for everyone?
00:05:53.319 --> 00:06:14.920
You know, I I think we have to do a better job collectively on the clinical side to describe what this is, what it's not, and how ultimately it's gonna help, you know, your provider, it's gonna, it's gonna help, you know, the clinician to be able to work more efficiently and more effectively and hopefully provide some some more eye contact during the visit.
00:06:15.240 --> 00:06:16.360
Yes, absolutely.
00:06:16.519 --> 00:06:21.480
You could actually pay attention and not have to like have that, like, how am I gonna write this up like in the back of your brain later?
00:06:21.720 --> 00:06:22.120
Exactly.
00:06:22.280 --> 00:06:22.680
Exactly.
00:06:22.920 --> 00:06:23.240
Yeah.
00:06:23.400 --> 00:06:26.759
Is there like language like in your teaching that you recommend?
00:06:26.840 --> 00:06:28.120
Like, this is what I say to my clients.
00:06:28.199 --> 00:06:30.680
Like, do you have a script or is it different for everybody?
00:06:30.759 --> 00:06:35.160
And just whatever you have, because probably people listening are like, yeah, but I don't know what to say.
00:06:35.319 --> 00:06:36.920
Like half the world hates AI.
00:06:37.079 --> 00:06:38.759
Like, how do I breach this topic?
00:06:38.840 --> 00:06:41.879
You know, do you have like a script or a way to go about it?
00:06:42.120 --> 00:06:42.360
Yeah.
00:06:42.600 --> 00:06:55.160
I try and break it down with with AI literacy, and I know we're gonna get to a little bit of that, but you know, the idea is that it's not an audio recording that is verbatim of the entire conversation, right?
00:06:55.240 --> 00:07:14.759
These are large language models, LLMs, if you will, that have been trained on data to be able to learn, and I use quotes because I am cautious of referring to AI models that when they think or when they reason, right?
00:07:15.240 --> 00:07:17.160
They don't they don't do those things.
00:07:17.400 --> 00:07:19.160
We do those things as human beings.
00:07:19.319 --> 00:07:33.560
What they're doing is a lot of statistical calculations behind the scenes to be able to predict what the next token or word for you and I is going to be in their output.
00:07:33.800 --> 00:07:43.240
So using the the proper language, it's a large language model that's been trained on you know massive amounts of written word and spoken word.
00:07:43.400 --> 00:07:59.000
And what it's going to do is listen to the audio transcript and then give me the output in the format that I've asked it to, you know, whether that be a soap note format or progress note format, this is what it's been trained to do.
00:07:59.480 --> 00:08:12.120
And making sure that there's assurances to the patient that this audio recording is going to be deleted, or doesn't it doesn't exist beyond, you know, this session.
00:08:12.439 --> 00:08:22.199
Again, you know, those clarifying things I think are really important because especially in the mental health space, we're talking about very sensitive and confidential things, right?
00:08:22.360 --> 00:08:34.679
All of our care is confidential, but here particularly, and I think we have to make sure that we're telling our patients exactly to the best of our knowledge what is happening and what's not happening.
00:08:35.000 --> 00:08:35.639
Exactly.
00:08:35.799 --> 00:08:39.240
And that leads into the literacy piece that you were just talking about before.
00:08:39.399 --> 00:08:47.319
Like as a clinician, we're responsible, I think, and you I'm sure you do too, to know what these tools are we're using, what they actually do.
00:08:47.399 --> 00:08:49.799
Like talking about that, yes, it does record the session.
00:08:49.879 --> 00:08:52.279
If you choose that option, a lot of them have lots of ways to do it.
00:08:52.439 --> 00:08:52.679
Correct.
00:08:52.840 --> 00:09:00.200
But if you choose that option and it's generate a recording, like we need to know that it's actually generating a recording and then what happens with the recording, right?
00:09:00.440 --> 00:09:00.919
So absolutely.
00:09:01.240 --> 00:09:03.000
What do you teach around the literacy piece?
00:09:03.080 --> 00:09:10.360
Maybe we could educate the listeners right now, like head AI Literacy 101, just they're like, I haven't even used Describe before, I don't even know anything about it.
00:09:10.519 --> 00:09:12.680
Like, where do we start with the education on that?
00:09:12.919 --> 00:09:24.680
Yeah, and I'll I'll pick up from my last response, you know, large language models or these models, when we talk about AI, for the vast majority, I'm talking about generative AI.
00:09:24.759 --> 00:09:36.279
So this is AI that's generating something, whether that be new text or audio, this is what it is because AI is very, very broad as a term and has existed for a long time.
00:09:36.440 --> 00:09:38.759
So we're talking about generative AI.
00:09:39.159 --> 00:09:41.159
We're talking about tokens, right?
00:09:41.320 --> 00:09:52.680
Tokens are what is being produced behind the scenes in those statistical calculations to make sure that the output that the model is giving you is most correct.
00:09:52.919 --> 00:09:55.639
And you know, those are the main principles.
00:09:55.879 --> 00:09:59.399
There's the idea of weighting and temperature.
00:09:59.560 --> 00:10:08.200
And I go into this very briefly because, again, I think it's important to understand how a token or a word is weighted, right?
00:10:08.519 --> 00:10:16.360
So in the training data, the words that appear more often are going to be weighted more heavily.
00:10:16.600 --> 00:10:23.000
So you're more likely to get a response that has a certain word because of how it appeared in the training data.
00:10:23.159 --> 00:10:27.399
And I think that's important, you know, especially if we're using these tools clinically.
00:10:27.639 --> 00:10:34.919
It's those principles of foundational literacy behind the way that large language models work.
00:10:35.240 --> 00:10:44.680
And I believe that when we understand what they can do and what they can't do, then we're better able to use them in the right application.
00:10:44.840 --> 00:10:58.200
I mean, we're still early at this in generative AI in healthcare, and there is no one AI solution that will solve all the problems out there, right?
00:10:59.000 --> 00:11:08.279
We are now using scribes, we're using large language model, like evidence bases, you know, for clinical lookup.
00:11:08.600 --> 00:11:24.200
There's a number of tools that are out there, but again, they're better than your off-the-shelf like ChatGPTs and Claude's because they have been trained specifically on medical, most times peer-reviewed data.
00:11:24.279 --> 00:11:33.960
So you're getting a better sampling of the training as compared to some of the commercial language models, large language models that have been trained on everything.
00:11:34.200 --> 00:12:00.039
So, you know, these are the nuances that get a little bit boring, gets a little bit, but I think at the same time, will help us better to assess and use these tools and hopefully have us as the clinicians help our patients and be the ones to enable our patients to use AI responsibly and ethically as well.
00:12:00.360 --> 00:12:03.720
Yeah, I mean, it it is really showing up in the therapy room, I tell you.
00:12:03.799 --> 00:12:15.159
I mean, people are coming in with but, well, this is what ChatGPT said, you know, to be able to understand, you know, just technically as a clinician what the LLM is, and that it's not, we don't, we're not gonna humanize it.
00:12:15.240 --> 00:12:17.720
Like I call Claude he and I catch myself all the time.
00:12:17.960 --> 00:12:18.519
I do as well.
00:12:18.919 --> 00:12:20.360
He's not a he, he's a computer.
00:12:20.519 --> 00:12:23.000
It's hard not to because they're trained that way.
00:12:23.240 --> 00:12:26.519
You think they're your best friend, you know, and it's a problem with therapy too.
00:12:26.600 --> 00:12:32.440
I mean, I know we could go down that rabbit hole too, just like, you know, clients turning to that for support and that sort of thing.
00:12:32.519 --> 00:12:45.960
But but yes, back to like AI literacy as the clinician and tool literacy, and then being able to kind of break it down to layman's terms to explain to our clients, this is what's happening when it's involved in the session and how you're protected and all of that.
00:12:46.200 --> 00:12:46.440
Exactly.
00:12:46.600 --> 00:12:47.240
Yeah, exactly.
00:12:47.320 --> 00:12:50.279
And we're using AI all generative AI all over, right?
00:12:50.440 --> 00:12:56.519
So, you know, people are using it at home and they're redesigning their backyards with generative AI.
00:12:56.600 --> 00:12:58.039
And I think that's amazing, right?
00:12:58.120 --> 00:13:06.039
Like, you know, it's not so amazing for the landscape architects that may not be getting, you know, the business that they want.
00:13:06.200 --> 00:13:23.080
But I think when we use it thoughtfully and intentionally, right, it's not to replace those skilled individuals, it's to use it to be more informed as a coexistence and being able to have better ideas and a better idea of what it is you want.
00:13:23.399 --> 00:13:24.200
That's amazing.
00:13:24.360 --> 00:13:24.759
Yeah.
00:13:25.000 --> 00:13:29.080
And something you talk about is full practice authority.
00:13:29.159 --> 00:13:39.240
And we're kind of talking about that a little bit, but can you expand upon that concept and how it applies to us as clinicians using AI tools, whether it's Claude or, you know, a note scribe that you're using?
00:13:39.399 --> 00:14:00.519
What does that mean exactly when you're talking about So full practice authority is the idea of clinicians being able to practice at the top of their licenses to be able to use the education and preparation that you have garnered in school to be a clinician that could be effective in your patient care.
00:14:00.759 --> 00:14:04.120
So it's mostly used for nurse practitioners.
00:14:04.680 --> 00:14:04.919
Okay.
00:14:05.159 --> 00:14:15.320
And you know, the idea is that uh we have 50 states and when you count DC, we'll count them as as an extra as a 51.
00:14:15.639 --> 00:14:28.120
And for NP practice, there's a lot of variability across the states, and that's because the states are the ones who regulate and legislate clinical practice.
00:14:28.279 --> 00:14:34.519
There is no one like law that any healthcare provider follows, right?
00:14:34.600 --> 00:14:38.440
You have to follow a law in the state that you are licensed.
00:14:38.600 --> 00:14:42.039
And if you're not licensed in that state, then you can't practice in that state.
00:14:42.279 --> 00:15:04.120
So the idea is that you know, the board of nursing in the state that regulates the practice of nursing and nurse practitioners has the sole authority to set professional standards and discipline and everything that goes along without the requirement of another discipline being involved in the provision of patient care.
00:15:04.279 --> 00:15:06.440
So that's the idea of full practice authority.
00:15:06.519 --> 00:15:24.039
Right now in the US, there are 27 states that are considered full practice authority, and all of those other states that are not, it is legislation that must be introduced and passed in order for those other states to be considered full practice authority.
00:15:24.440 --> 00:15:25.320
So interesting.
00:15:25.480 --> 00:15:27.080
Yeah, I mean, I'm on the other side of it.
00:15:27.159 --> 00:15:39.080
I I have a whole nother business called the traveling therapist, and it is such I live in Airbnb's, I'm fully nomadic, and it is such an issue, like that we're not just under one regulatory board across the entire United States.
00:15:39.159 --> 00:15:39.799
It's so complicated.
00:15:40.200 --> 00:15:41.799
It's very, very complicated.
00:15:42.360 --> 00:15:50.120
You have different states and different sets of rules, and it's up to you as the licensed individual to make sure that you're following the rules of that state.
00:15:50.200 --> 00:15:54.519
And what makes things so much harder is telehealth, right?
00:15:54.680 --> 00:16:09.639
So yeah, you can be literally seeing a patient in any one of the 50 states, and you have to follow the law, and you are in your state, but you have to follow the law and be licensed in that state that you're providing that care to where that patient is.
00:16:09.720 --> 00:16:12.680
And it's very, very difficult and very confusing.
00:16:12.840 --> 00:16:17.720
And again, it's incumbent upon yourself to know what those laws are and follow them.
00:16:17.960 --> 00:16:26.759
Otherwise, you don't want to get a knock at the door from the board of whatever profession you're practicing and saying, you know, you're not practicing, you know, correctly.
00:16:27.080 --> 00:16:27.480
Yeah.
00:16:27.720 --> 00:16:28.840
Yeah, it's so interesting.
00:16:28.919 --> 00:16:32.440
And I'm seeing so much AI legislation come down in different states.
00:16:32.600 --> 00:16:34.039
I'm just so curious.
00:16:34.120 --> 00:16:42.440
I mean, this is probably a future conversation to see like how is all this going to play out for different states and professions and boards, and how are they gonna interpret these rules?
00:16:42.600 --> 00:16:48.039
And it's just kind of like spins your mind up a little bit, just trying to keep track of all of it.
00:16:48.120 --> 00:16:49.639
It's really an interesting time right now.
00:16:49.879 --> 00:16:52.200
It's nearly impossible to keep track of that as well.
00:16:52.360 --> 00:16:53.480
The legislative side.
00:16:53.560 --> 00:16:58.840
I mean, there are states that are saying no use of chatbots for therapy, right?
00:16:59.159 --> 00:17:03.480
I I mean, even states have passed legislation.
00:17:03.720 --> 00:17:14.359
Washington state is one of the first states in Oregon passed a law that said you can't call an AI bot, chatbot a nurse, right?
00:17:14.440 --> 00:17:24.119
You yes, the term nurse is a protected term, and you need a license and you need to be a human being that can be called a nurse.
00:17:24.279 --> 00:17:34.759
So there are states that have passed these laws that you know make it illegal for an AI company to use the term nurse so-and-so as a chat box.
00:17:35.000 --> 00:17:39.000
But my bigger question is where does this end, right?
00:17:39.079 --> 00:17:44.440
Do we need a law for every profession in every state to do this?
00:17:44.599 --> 00:17:51.240
Now, I believe that you know, if you're a therapist, if you if you're a nurse practitioner, that should be a protected title.
00:17:51.319 --> 00:17:54.440
And and not not anyone should be able to call themselves that.
00:17:54.680 --> 00:18:02.759
But I believe we need a comprehensive strategy that isn't individual, state by state, for every profession because it's it's messy.
00:18:02.920 --> 00:18:05.720
It makes it that much harder to track.
00:18:05.960 --> 00:18:12.839
Yeah, and I mean, don't get me wrong, I'm grateful for the legislation because I don't want AI therapy bots taking over the world, you know.
00:18:13.079 --> 00:18:21.400
We said earlier you're not gonna be replaced, but like trying or attempting to replace our professions, any of our professions, you know, it's it's a super interesting time right now.
00:18:21.720 --> 00:18:25.880
What's great about a chatbot, right, is that it's available 24-7.
00:18:26.440 --> 00:18:26.759
Right.
00:18:26.920 --> 00:18:30.200
And for the most part, a lot of them are free, right?
00:18:30.440 --> 00:18:32.599
Yes, and we know nothing is free.
00:18:32.920 --> 00:18:33.079
Right.
00:18:33.400 --> 00:18:48.839
So when you're using an AI chatbot, and if you don't turn off the setting that says don't train on my data, your data and your conversations are being used to train future models.
00:18:49.000 --> 00:18:51.559
And again, is that a bad thing?
00:18:51.880 --> 00:18:58.599
Maybe, maybe not, but you know, maybe if it's getting better, it'll be more useful.
00:18:58.759 --> 00:18:59.640
But at what cost?
00:18:59.799 --> 00:19:09.000
Again, and I think that's where I take the position that we as a society are the ones that need to decide how how to use this or not.
00:19:09.400 --> 00:19:09.960
Yeah.
00:19:10.279 --> 00:19:11.640
And you brought that up at the beginning.
00:19:11.720 --> 00:19:13.079
I think it's such an important topic.
00:19:13.240 --> 00:19:20.839
Do you have sort of your own guardrails now or guardrails that you teach around how to make sure or ensure that we're not crossing that line?
00:19:21.000 --> 00:19:34.599
And I mean, you're talking about like turn off the settings so it's private, they're not using your data in the large LLMs and that sort of thing, but I'm just wondering, are there other things that maybe somebody hasn't thought about that we can keep our autonomy and not like, you know, totally feed the machine kind of thing?
00:19:34.839 --> 00:19:35.160
Yeah.
00:19:35.400 --> 00:19:38.359
And I think, I mean, the biggest is the privacy settings.
00:19:38.599 --> 00:19:39.240
Yeah, okay.
00:19:39.400 --> 00:19:39.640
Yeah.
00:19:39.880 --> 00:19:41.240
That's first and foremost.
00:19:41.559 --> 00:19:46.359
And then it's, you know, it's making sure that you're verifying outputs.
00:19:46.599 --> 00:19:55.240
It's making sure that when you see a reference link that you actually click on the reference link and see what it goes to.
00:19:55.559 --> 00:20:02.120
Because we know that LLMs are very good at we call them hallucinations.
00:20:02.279 --> 00:20:04.839
I also like to call them confabulations.
00:20:05.240 --> 00:20:05.880
Oh, nice.
00:20:06.120 --> 00:20:10.680
Because these have been programmed to give you an answer, a response.
00:20:10.759 --> 00:20:11.000
Yeah.
00:20:11.240 --> 00:20:18.519
Because what would happen if you put in your prompt in your favorite LLM and it said, I don't know.
00:20:18.759 --> 00:20:19.000
Right?
00:20:19.160 --> 00:20:19.319
Yeah.
00:20:19.480 --> 00:20:20.519
I don't know the answer to that.
00:20:20.759 --> 00:20:21.799
You wouldn't use it.
00:20:21.880 --> 00:20:24.200
You'd stop reading and say that that didn't help me.
00:20:24.279 --> 00:20:25.240
So unhelpful.
00:20:25.400 --> 00:20:25.480
Yeah.
00:20:25.720 --> 00:20:26.039
Exactly.
00:20:26.359 --> 00:20:26.839
Unhelpful.
00:20:27.400 --> 00:20:34.039
So they are trained to give you a response, and sometimes we don't see what's happening behind the scenes.
00:20:34.200 --> 00:20:39.160
Sometimes it's not very confident in that response and it's going to just spit out something.
00:20:39.319 --> 00:20:49.559
I've clicked on links that have that DOI number that you know looks like a legit, you know, like it goes to PubMed, and then you click it and it doesn't exist.
00:20:50.039 --> 00:20:53.319
Or for some different topic, the paper.
00:20:53.480 --> 00:21:03.000
So you have to always, always verify the outputs and not assume that, you know, because it looks good, that it actually is good.
00:21:03.400 --> 00:21:05.079
Yes, I I love that.
00:21:05.240 --> 00:21:06.120
Such good advice.
00:21:06.359 --> 00:21:20.279
And I also think, I mean, what you're doing, like with your education, that sort of thing, like people like us showing up and trying to educate, you know, the people in our field around all of these topics can just help everyone have a seat at the table when the stuff's being developed.
00:21:20.359 --> 00:21:23.480
Like Barries, being the clinical consultant as a therapist, you know.
00:21:23.559 --> 00:21:28.519
It's I just think that stuff's so important for us to show up to the rooms too, where the stuff's being developed.
00:21:28.599 --> 00:21:49.400
If we can't do we have to partner with the smart engineers and the software developers, they're good at what they do, we're good at what we do, and you know, we need to work together on the solutions to make sure that, you know, again, we're using it ethically, responsibly, in the best way possible, based on, you know, the current limitations.
00:21:49.480 --> 00:22:04.839
And, you know, a limitation today, I'm borrowing this or I'm stealing this from very other prolific AI people who've said, you know, this is the worst that AI is gonna be today is the worst it's ever going to be, right?
00:22:04.920 --> 00:22:06.920
It's only gonna get better from here.
00:22:07.160 --> 00:22:09.240
And you know, again, what what does that mean?
00:22:09.319 --> 00:22:10.519
Uh, how do we use it?
00:22:10.680 --> 00:22:14.599
And just because we know knew something six months ago, it could have changed.
00:22:14.759 --> 00:22:16.359
You know, could have changed uh already.
00:22:16.440 --> 00:22:18.519
So we have to be up uh up to date on that.
00:22:19.079 --> 00:22:19.480
Wow.
00:22:19.640 --> 00:22:21.720
Yeah, there's so much to think about and all that.
00:22:22.039 --> 00:22:27.000
We kind of jumped around a little bit, but I would love to go back to the Columbia piece and how you introduced.
00:22:27.480 --> 00:22:30.279
The founding AI program with the nursing program there.
00:22:30.359 --> 00:22:33.559
I know that you're not doing that anymore, but you started it, right?
00:22:33.720 --> 00:22:34.039
Right.
00:22:34.200 --> 00:22:38.440
I'm just curious, like, like, where do you see AI in the role of education?
00:22:38.599 --> 00:22:42.920
It's I mean, it's so cool that you're starting that conversation in Columbia, no less.
00:22:43.079 --> 00:22:43.960
I mean, that is amazing.
00:22:44.120 --> 00:22:47.480
And I'm just curious, like, where do you see that going?
00:22:47.720 --> 00:22:49.960
Are all the programs just going to be AI forward now?
00:22:50.120 --> 00:22:54.759
We're just all going to like enter into the workforce with AI forward thinking, or what do you think about that?
00:22:55.000 --> 00:23:01.640
Well, I think we need to have the conversation to determine how we use AI in the classroom.
00:23:02.200 --> 00:23:11.640
And the challenge that exists, like you have your universities, and then you know, in the universities, you have your schools of you know the different health professions.
00:23:11.960 --> 00:23:18.440
There's very little discussion that's happening other than like, you know, a policy statement.
00:23:18.599 --> 00:23:20.440
Now, a policy statement is important.
00:23:20.599 --> 00:23:23.400
And at heart, I am a policy wonk, right?
00:23:23.559 --> 00:23:37.160
I love I love health policy, but these this technology requires more than what has been written on a piece of paper in trying to anticipate something that we really don't even have a good grip on, right?
00:23:37.319 --> 00:23:48.279
So what I believe needs to happen across every health profession in every school that teaches the health professions is that there needs to be a deep dive by the faculty.
00:23:48.680 --> 00:23:57.880
And there has to be an intentional design of how we're going to incorporate AI into assignments, or not, right?
00:23:58.359 --> 00:24:06.440
What is creating a lot of challenges for students is that they'll have one faculty member who says, use AI, I don't care, right?
00:24:06.599 --> 00:24:08.759
Just cite or cite it, whatever.
00:24:08.920 --> 00:24:13.000
And you'll have another faculty member who says, Don't use any AI.
00:24:13.160 --> 00:24:15.559
If I find you using AI, I'm gonna fail you.
00:24:15.799 --> 00:24:34.359
And that's not a problem, like you know, the concept of saying I don't want you to use AI and I want you to critically think, but it's causing confusion because those faculty are not on the same page collectively, and yes, requires a look at the curriculum.
00:24:34.599 --> 00:24:42.200
I'm not saying AI has to be embedded in every course in every assignment, yeah, but it should be somewhere.
00:24:42.440 --> 00:24:50.920
And you know, uh it requires us uh when I was faculty to redesign the assignments, right?
00:24:51.079 --> 00:25:01.559
Like assigning a 15-page paper in today's world is not going to do it because it's too easy to hand it on over to a large language model.
00:25:01.720 --> 00:25:19.240
So maybe, you know, a possible suggestion is to have the large language model refine your uh opening paragraph or give you some suggestions of you know where you should go from here.
00:25:20.039 --> 00:25:30.759
But you have to use it, you know, thoughtfully and not have it take over and sterilize your written word into something that seems generic and fake, you know.
00:25:30.920 --> 00:25:33.319
Yeah, that's not even your original thought.
00:25:33.400 --> 00:25:33.960
Yeah, yeah.
00:25:34.279 --> 00:25:36.680
We like to use the word AI slop.
00:25:37.079 --> 00:25:45.240
Yes, and there's a lot of AI slop out there, and you know, if that's all we're reading, we're gonna start mimicking the large language.
00:25:45.480 --> 00:25:46.200
We're like AI.
00:25:46.279 --> 00:25:53.640
I already hear people talking, like I go to these trainings, and I'm like, he literally sounds like what Claude sounds like because it's like you use it so much.
00:25:53.880 --> 00:25:54.200
Absolutely.
00:25:54.519 --> 00:25:59.720
It's like I don't even know if we realize that we're starting to sound like the AI that we're like talking to, you know.
00:25:59.799 --> 00:26:01.160
It's it's really interesting.
00:26:01.400 --> 00:26:02.120
It is interesting.
00:26:02.200 --> 00:26:07.000
And if I hear the word friction or moat, like moat, that's the new one.
00:26:07.079 --> 00:26:07.400
Mote.
00:26:07.720 --> 00:26:18.039
I don't use those words, like maybe people have used them, but I've never heard them used as much as as they're used today, so it's really interesting.
00:26:18.279 --> 00:26:19.000
That's hilarious.
00:26:19.079 --> 00:26:24.519
It used to be game changer, but literally moat is like all of a sudden showing up at everything I'm doing with Claude these days.
00:26:24.759 --> 00:26:25.880
It's hilarious.
00:26:27.960 --> 00:26:29.160
Oh my gosh, I love it.
00:26:29.240 --> 00:26:30.680
I feel like I could talk about this all day.
00:26:31.799 --> 00:26:39.880
Can we talk a little bit about your programs and what you do and how you help nurse practitioners and other clinicians that are in your ecosystem?
00:26:40.039 --> 00:26:43.799
How, how does your program help with everything that we're talking about?
00:26:44.039 --> 00:26:58.680
So thank you for mentioning AI Nurse Academy, and that is my platform that I built out to provide the literacy for nurses, but not just nurses and all clinicians, right?
00:26:58.759 --> 00:27:01.960
So, right now I have one course that's available.
00:27:02.120 --> 00:27:09.480
It's a one-hour on-demand type format that you know goes over these principles that we talked about.
00:27:09.720 --> 00:27:19.000
In addition to the on-demand format, and there's a post-test at the end, there's the option to join a community.
00:27:19.400 --> 00:27:37.960
And this is the community of people who've taken the course and want to have continuing conversations about AI, and whether that be sharing a really useful prompt that they found works well for them, or sharing some of the limitations of AI.
00:27:38.440 --> 00:27:47.960
It's uh it's an active community, it's a way to, you know, not just make it a static course, but it's to continue the conversation.
00:27:48.279 --> 00:27:55.079
Because again, I think that's the only way we're gonna try and make the best sense of it and the best use of of AI.
00:27:55.400 --> 00:27:59.880
And just having that community dialogue where we can bounce ideas off of each other, that sort of thing.
00:28:00.359 --> 00:28:06.440
Yeah, I will say I did a deep dive before this interview on your website, and I was like, oh, he has a community, and I was like, it's gotta be a paid community.
00:28:06.519 --> 00:28:07.880
I was like, no, it's totally freaky.
00:28:08.039 --> 00:28:12.279
So it comes with the course or the course is the course, the course, right, right.
00:28:12.440 --> 00:28:16.759
But it you know, you get lifetime access to the community again a good way.
00:28:16.839 --> 00:28:26.680
And I'm in there, you know, trying to uh stir conversation and and and trying to, you know, be as thoughtful as I possibly can be.
00:28:27.079 --> 00:28:27.799
I love it.
00:28:27.880 --> 00:28:28.440
It's really good.
00:28:28.519 --> 00:28:29.720
You guys go check out his website.
00:28:29.799 --> 00:28:32.039
Of course, we'll have everything down in the show notes too.
00:28:32.279 --> 00:28:39.559
So we always like to do kind of like a like a hot seat a little bit at the end and just ask what AI tools are you using right now?
00:28:39.720 --> 00:28:52.519
I mean, outside of like the work that we do and all that, like do you have like really like your favorite go-to just for regular productivity or you know, just fun, you know, AI stuff that you're doing right now that you would want to share?
00:28:52.759 --> 00:28:56.039
Anybody that's just like, I just love AI, I want to know what everybody's using it for.
00:28:56.279 --> 00:28:56.759
Sure, sure.
00:28:56.839 --> 00:28:58.440
I'm happy, happy to share.
00:28:58.599 --> 00:29:05.079
And I have no like ownership in any of these companies, so this is just what what I use.
00:29:05.160 --> 00:29:15.079
And and my suggestion really is to people is to use them all because I don't think one works better than another for everything, right?
00:29:15.240 --> 00:29:22.200
So I'm using Claude a lot these days, and I purchased the Claude Coldwork subscription.
00:29:22.279 --> 00:29:33.160
So I'm playing with that because I want to be able to understand you know the idea of an AI agent and what that means, and it's able to do things on my behalf.
00:29:33.319 --> 00:29:35.559
I have very strict guardrails on that.
00:29:35.640 --> 00:29:44.200
I don't let it, you know, write emails for me, I don't let it send emails for me, I don't let it, you know, have access to my entire hard drive.
00:29:44.359 --> 00:29:46.759
You know, it has access to one folder, right?
00:29:46.839 --> 00:29:48.039
And one folder only.
00:29:48.200 --> 00:30:00.839
I'm enjoying using Claude Cowork, but I use Chat GPT, I use Gemini as well for you know all different types of drafting ideas and concept ideas.
00:30:01.079 --> 00:30:20.519
When it comes to LLMs in in the clinical space, there's a host of free tools, and yeah, the ones that I have used are open evidence, there is Vera Health, is another one, Helio H E A L I OK is another good one, and uh Evidence M D.
00:30:21.160 --> 00:30:25.640
These are again shout-outs to those models.
00:30:25.960 --> 00:30:35.559
They all work similarly, in my opinion, but give you different types of output that you may find better for your style.
00:30:35.720 --> 00:30:38.120
Like some are more like bulleted, right?
00:30:38.200 --> 00:30:44.039
So you're gonna get an output that's boom, boom, boom, than reading like paragraphs and paragraphs of content.
00:30:44.279 --> 00:30:46.440
So yeah, I like seeing those.
00:30:46.519 --> 00:30:52.680
And again, those are the models that are trained on the peer-reviewed content.
00:30:52.920 --> 00:31:06.519
It doesn't mean that there's zero possibility of hallucination, still can hallucinate, but it's trained on better data and it's gonna give you, you know, better clinical outputs than the statue.
00:31:07.559 --> 00:31:13.400
Are they node scribes or are they more like just for like I've got a client with major depressive disorder and celiac disease?
00:31:13.480 --> 00:31:17.480
Like, how do I like what's the evidence to support what medication to use?
00:31:17.559 --> 00:31:18.359
Like that kind of stuff.
00:31:18.519 --> 00:31:20.599
Yeah, so okay, interesting.
00:31:20.839 --> 00:31:24.519
Not they all, but some of them also have built-in scribes as well.
00:31:24.839 --> 00:31:25.559
Oh, okay.
00:31:25.720 --> 00:31:26.359
Oh, interesting.
00:31:26.599 --> 00:31:30.120
Yeah, and that's another part of their you know, offerings.
00:31:30.440 --> 00:31:37.160
The part that I focus on is the the large language model, and exactly as you suggest, you know.
00:31:37.240 --> 00:31:46.279
I have a patient with celiac, you know, they're you know, and you give your your history, you know, 21 years old on X, Y, and Z medication.
00:31:46.519 --> 00:31:51.559
I again, these are tools that are free for the most part.
00:31:51.720 --> 00:31:56.120
If you have an NPI number, you could get access to them for the most part.
00:31:56.440 --> 00:31:57.000
Oh, cool.
00:31:57.240 --> 00:32:01.960
So you can compare and contrast, you know, uh outputs from the different models.
00:32:02.200 --> 00:32:07.720
What they aren't yet is they're not embedded in the electronic health record, right?
00:32:07.960 --> 00:32:08.759
Yeah, gotcha.
00:32:09.319 --> 00:32:19.000
Someday they will be, you know, and and we're not that far away from, you know, Epic is using, you know, they have their own suite of tools that they're developing.
00:32:19.160 --> 00:32:25.240
So, but right now they're essentially standalone in, and I have to say it's in my course.
00:32:25.559 --> 00:32:41.079
Do not upload any PHI loud into any large language model unless you know your practice has a BAA, you know, a business associate agreement that you know is is HIPAA compliant.
00:32:41.240 --> 00:32:51.960
But yeah, you know, you can you can anonymize the the data so you're not, you know, yeah, sharing sadding, you know, confidential information.
00:32:52.200 --> 00:32:53.400
Yes, totally love that.
00:32:53.480 --> 00:32:54.839
Yeah, so important to say.
00:32:55.079 --> 00:32:55.640
Thank you.
00:32:55.799 --> 00:32:57.079
What a great conversation.
00:32:57.160 --> 00:32:58.039
I really appreciate it.
00:32:58.120 --> 00:33:04.440
Could you tell everybody your website or your email, whatever, the best way to contact you and follow you and connect with you?
00:33:04.680 --> 00:33:07.000
Yeah, so I would say find me on LinkedIn.
00:33:07.319 --> 00:33:07.640
Okay.
00:33:07.880 --> 00:33:09.400
I'm pretty active there, Stephen A.
00:33:09.559 --> 00:33:10.440
Ferrara.
00:33:10.680 --> 00:33:15.640
There's a few of us out there that have the same name, but I'm the uh I'm the nurse practitioner one.
00:33:16.039 --> 00:33:17.480
Oh, so find me there.
00:33:17.559 --> 00:33:23.640
And then my course, AI Nurse Academy, is at AINurseacademy.ai.
00:33:24.200 --> 00:33:25.240
Easy sign up.
00:33:25.400 --> 00:33:32.359
As I mentioned, it's a$20 course, you know, one-time fee that you could, you know, revisit the course.
00:33:32.440 --> 00:33:37.319
It comes with like a little resource, one pager that you can download.
00:33:37.559 --> 00:33:44.440
And then, as you mentioned, it gives you access to the community where you can actually engage in some some conversations.
00:33:44.599 --> 00:33:48.440
But but yeah, you know, I'll think we all need to have those conversations.
00:33:48.759 --> 00:33:51.880
Yeah, that's amazing to have that community there forever.
00:33:52.120 --> 00:33:53.640
Yeah, thank you so much.
00:33:53.720 --> 00:33:55.400
It's really really been great talking to you.
00:33:55.559 --> 00:33:58.519
Yeah, it's been a pleasure and thank you so much for having me.
00:00:27.559 --> 00:00:28.199
Hi, everybody.
00:00:28.359 --> 00:00:31.640
Welcome back to another episode of Between Sessions with Berries.
00:00:31.799 --> 00:00:33.159
We have the coolest guest today.
00:00:33.239 --> 00:00:34.119
I'm so excited.
00:00:34.359 --> 00:00:34.759
Dr.
00:00:34.839 --> 00:00:41.960
Steven Ferrara is a nurse practitioner, educator, and health policy leader with over 25 years of experience.
00:00:42.200 --> 00:00:47.479
And he's at the forefront of how AI is actually entering clinical practice.
00:00:47.719 --> 00:01:01.879
He's the founder of AI Nurse Academy, the immediate past president of the American Association of Nurse Practitioners, and the founding associate dean for artificial intelligence at Columbia University School of Nursing.
00:01:02.039 --> 00:01:14.920
He's here to talk about something every therapist is wrestling with right now: how to use AI responsibly, how to evaluate the tools flooding the market, and how to cut the documentation burden without cutting corners on care.
00:01:22.920 --> 00:01:24.200
Steven, welcome.
00:01:24.280 --> 00:01:25.640
I'm so excited to have you here.
00:01:25.959 --> 00:01:27.319
Thank you so much for having me.
00:01:27.560 --> 00:01:29.480
Yeah, well, let's just dig in.
00:01:29.560 --> 00:01:32.759
I always like to just start with like the real obvious questions.
00:01:33.000 --> 00:01:36.759
What do most clinicians get wrong when they start using AI tools?
00:01:36.920 --> 00:01:38.839
Like what are you seeing with all the clinicians you're working with?
00:01:38.920 --> 00:01:44.599
And when I say clinicians on your end, it's nurse practitioners that you're really talking with the most about these tools, right?
00:01:44.839 --> 00:01:45.079
Sure.
00:01:45.239 --> 00:01:45.479
Sure.
00:01:45.719 --> 00:01:58.039
And other clinicians as well, whether that be physicians and physician associates, therapists, what we have are a flood of AI tools in the market.
00:01:58.359 --> 00:02:02.199
And I do think that it's going to improve practice.
00:02:02.280 --> 00:02:12.199
I mean, that's why I am so passionate about AI, because I think at the end of the day, it can make me a better clinician and provide better care to my patients.
00:02:12.439 --> 00:02:16.120
I will say right from the outset, it's not going to replace us.
00:02:16.360 --> 00:02:19.640
And shame on us if we allow it to, right?
00:02:19.879 --> 00:02:27.319
Because it really is in our control collectively as a society to determine how we're going to use AI.
00:02:27.640 --> 00:02:32.599
And what I want us to do is really have those conversations.
00:02:32.840 --> 00:02:34.120
We may not have the answers.
00:02:34.280 --> 00:02:35.480
We don't have the answers.
00:02:35.960 --> 00:02:38.600
Things are moving incredibly fast.
00:02:39.000 --> 00:02:48.680
It's a full-time job to stay up to date on what is happening in the AI space, especially when it comes to health care and clinical care.
00:02:48.840 --> 00:02:49.080
Yeah.
00:02:49.320 --> 00:02:57.800
But I what I really, you know, want people to look at is, you know, it's not automatic, it's not fail-proof, and to use it critically.
00:02:58.040 --> 00:02:58.600
I love that.
00:02:58.680 --> 00:02:58.920
Yeah.
00:02:59.000 --> 00:03:04.120
So that really leans into that question like, what are you saying that people are doing wrong with AI or clinicians are doing?
00:03:04.280 --> 00:03:06.840
So are you seeing people not using it critically?
00:03:06.920 --> 00:03:10.280
Are you having to correct that behavior and the people that you're working with now?
00:03:10.520 --> 00:03:15.560
Just curious, like how that's coming up in the work you're doing with clinicians, what you're seeing like on the ground level.
00:03:15.800 --> 00:03:16.040
Yeah.
00:03:16.200 --> 00:03:17.960
It's a whole different workflow.
00:03:18.040 --> 00:03:23.719
And AI Scribes is the biggest example that we have in healthcare today.
00:03:23.960 --> 00:03:28.600
We're going from author of the note to reviewer of the note.
00:03:28.680 --> 00:03:29.960
And that's not a bad thing.
00:03:30.120 --> 00:03:30.360
Yes.
00:03:30.600 --> 00:03:31.879
But it's a different workflow.
00:03:32.040 --> 00:03:37.320
I mean, when I went to NP school 26 years ago, like that wasn't a thing.
00:03:37.640 --> 00:03:43.960
So, you know, for the vast majority of people out there, you know, they're using these tools for the first time.
00:03:44.280 --> 00:03:51.640
And we have to look twice when we're reviewing a note that we didn't write ourselves.
00:03:51.879 --> 00:04:03.879
What people shouldn't do is assume that everything in the transcript is correct and look at it, you know, critically and you know, make any necessary corrections and and then sign off on the note.
00:04:04.200 --> 00:04:04.680
Totally.
00:04:04.760 --> 00:04:14.280
I mean, even in my own practice, I catch myself doing that a little bit because you get so comfortable in a way with the AI like taking over that burden that it's almost like, oh, the notes are written.
00:04:14.360 --> 00:04:19.000
I can go like take a walk real quick or go to the bathroom for once, you know, like that kind of thing.
00:04:19.240 --> 00:04:22.680
You know, it's easy to get in that pattern of like, you know, it's good.
00:04:22.920 --> 00:04:26.279
Sign off, go to the next client kind of thing where I feel you on that.
00:04:26.360 --> 00:04:28.040
Like, I hope everybody hears what you're saying.
00:04:28.120 --> 00:04:32.519
It's like AI likes to fill in the blanks, it likes to hallucinate a little bit.
00:04:32.680 --> 00:04:38.040
We have to be super careful and read over it and make sure, like, yes, this is in a court of law.
00:04:38.120 --> 00:04:43.400
If somebody read this note, I want it to be accurately written what I'm saying here, yeah, or what they're saying.
00:04:43.639 --> 00:04:43.960
100%.
00:04:44.199 --> 00:04:48.360
And yeah, and I fear of the the the idea of good enough, right?
00:04:48.519 --> 00:04:57.879
Yeah, this looks good enough, and because we're busy, we're busy clinicians, and you know, you finish with one patient and then you're already thinking about the next one.
00:04:58.040 --> 00:05:07.560
So it really requires us to stop, reflect, make any necessary corrections, and then finalize the note afterwards.
00:05:07.879 --> 00:05:14.600
Because we do a lot of scanning, you know, with our eyes and with our brains, and we need to do better than just scan.
00:05:14.840 --> 00:05:15.400
I love it.
00:05:15.639 --> 00:05:23.480
Any other mistakes or things that you're seeing people getting wrong off the bat right now when they're just hopping into AI without education, like the type of education you offer?
00:05:23.720 --> 00:05:35.079
Yeah, I I think one of the hallmarks is making sure that patients are consented to use, to agree to use an AI scribe in the office.
00:05:35.240 --> 00:05:37.240
I can't assume anything.
00:05:37.400 --> 00:05:46.360
And, you know, I I think it's incumbent upon the clinician that's using, you know, the technology to describe it in plain language to the patient.
00:05:46.519 --> 00:05:53.160
You know, there's a lot of misinformation out there, you know, does it mean it's gonna be discoverable for everyone?
00:05:53.319 --> 00:06:14.920
You know, I I think we have to do a better job collectively on the clinical side to describe what this is, what it's not, and how ultimately it's gonna help, you know, your provider, it's gonna, it's gonna help, you know, the clinician to be able to work more efficiently and more effectively and hopefully provide some some more eye contact during the visit.
00:06:15.240 --> 00:06:16.360
Yes, absolutely.
00:06:16.519 --> 00:06:21.480
You could actually pay attention and not have to like have that, like, how am I gonna write this up like in the back of your brain later?
00:06:21.720 --> 00:06:22.120
Exactly.
00:06:22.280 --> 00:06:22.680
Exactly.
00:06:22.920 --> 00:06:23.240
Yeah.
00:06:23.400 --> 00:06:26.759
Is there like language like in your teaching that you recommend?
00:06:26.840 --> 00:06:28.120
Like, this is what I say to my clients.
00:06:28.199 --> 00:06:30.680
Like, do you have a script or is it different for everybody?
00:06:30.759 --> 00:06:35.160
And just whatever you have, because probably people listening are like, yeah, but I don't know what to say.
00:06:35.319 --> 00:06:36.920
Like half the world hates AI.
00:06:37.079 --> 00:06:38.759
Like, how do I breach this topic?
00:06:38.840 --> 00:06:41.879
You know, do you have like a script or a way to go about it?
00:06:42.120 --> 00:06:42.360
Yeah.
00:06:42.600 --> 00:06:55.160
I try and break it down with with AI literacy, and I know we're gonna get to a little bit of that, but you know, the idea is that it's not an audio recording that is verbatim of the entire conversation, right?
00:06:55.240 --> 00:07:14.759
These are large language models, LLMs, if you will, that have been trained on data to be able to learn, and I use quotes because I am cautious of referring to AI models that when they think or when they reason, right?
00:07:15.240 --> 00:07:17.160
They don't they don't do those things.
00:07:17.400 --> 00:07:19.160
We do those things as human beings.
00:07:19.319 --> 00:07:33.560
What they're doing is a lot of statistical calculations behind the scenes to be able to predict what the next token or word for you and I is going to be in their output.
00:07:33.800 --> 00:07:43.240
So using the the proper language, it's a large language model that's been trained on you know massive amounts of written word and spoken word.
00:07:43.400 --> 00:07:59.000
And what it's going to do is listen to the audio transcript and then give me the output in the format that I've asked it to, you know, whether that be a soap note format or progress note format, this is what it's been trained to do.
00:07:59.480 --> 00:08:12.120
And making sure that there's assurances to the patient that this audio recording is going to be deleted, or doesn't it doesn't exist beyond, you know, this session.
00:08:12.439 --> 00:08:22.199
Again, you know, those clarifying things I think are really important because especially in the mental health space, we're talking about very sensitive and confidential things, right?
00:08:22.360 --> 00:08:34.679
All of our care is confidential, but here particularly, and I think we have to make sure that we're telling our patients exactly to the best of our knowledge what is happening and what's not happening.
00:08:35.000 --> 00:08:35.639
Exactly.
00:08:35.799 --> 00:08:39.240
And that leads into the literacy piece that you were just talking about before.
00:08:39.399 --> 00:08:47.319
Like as a clinician, we're responsible, I think, and you I'm sure you do too, to know what these tools are we're using, what they actually do.
00:08:47.399 --> 00:08:49.799
Like talking about that, yes, it does record the session.
00:08:49.879 --> 00:08:52.279
If you choose that option, a lot of them have lots of ways to do it.
00:08:52.439 --> 00:08:52.679
Correct.
00:08:52.840 --> 00:09:00.200
But if you choose that option and it's generate a recording, like we need to know that it's actually generating a recording and then what happens with the recording, right?
00:09:00.440 --> 00:09:00.919
So absolutely.
00:09:01.240 --> 00:09:03.000
What do you teach around the literacy piece?
00:09:03.080 --> 00:09:10.360
Maybe we could educate the listeners right now, like head AI Literacy 101, just they're like, I haven't even used Describe before, I don't even know anything about it.
00:09:10.519 --> 00:09:12.680
Like, where do we start with the education on that?
00:09:12.919 --> 00:09:24.680
Yeah, and I'll I'll pick up from my last response, you know, large language models or these models, when we talk about AI, for the vast majority, I'm talking about generative AI.
00:09:24.759 --> 00:09:36.279
So this is AI that's generating something, whether that be new text or audio, this is what it is because AI is very, very broad as a term and has existed for a long time.
00:09:36.440 --> 00:09:38.759
So we're talking about generative AI.
00:09:39.159 --> 00:09:41.159
We're talking about tokens, right?
00:09:41.320 --> 00:09:52.680
Tokens are what is being produced behind the scenes in those statistical calculations to make sure that the output that the model is giving you is most correct.
00:09:52.919 --> 00:09:55.639
And you know, those are the main principles.
00:09:55.879 --> 00:09:59.399
There's the idea of weighting and temperature.
00:09:59.560 --> 00:10:08.200
And I go into this very briefly because, again, I think it's important to understand how a token or a word is weighted, right?
00:10:08.519 --> 00:10:16.360
So in the training data, the words that appear more often are going to be weighted more heavily.
00:10:16.600 --> 00:10:23.000
So you're more likely to get a response that has a certain word because of how it appeared in the training data.
00:10:23.159 --> 00:10:27.399
And I think that's important, you know, especially if we're using these tools clinically.
00:10:27.639 --> 00:10:34.919
It's those principles of foundational literacy behind the way that large language models work.
00:10:35.240 --> 00:10:44.680
And I believe that when we understand what they can do and what they can't do, then we're better able to use them in the right application.
00:10:44.840 --> 00:10:58.200
I mean, we're still early at this in generative AI in healthcare, and there is no one AI solution that will solve all the problems out there, right?
00:10:59.000 --> 00:11:08.279
We are now using scribes, we're using large language model, like evidence bases, you know, for clinical lookup.
00:11:08.600 --> 00:11:24.200
There's a number of tools that are out there, but again, they're better than your off-the-shelf like ChatGPTs and Claude's because they have been trained specifically on medical, most times peer-reviewed data.
00:11:24.279 --> 00:11:33.960
So you're getting a better sampling of the training as compared to some of the commercial language models, large language models that have been trained on everything.
00:11:34.200 --> 00:12:00.039
So, you know, these are the nuances that get a little bit boring, gets a little bit, but I think at the same time, will help us better to assess and use these tools and hopefully have us as the clinicians help our patients and be the ones to enable our patients to use AI responsibly and ethically as well.
00:12:00.360 --> 00:12:03.720
Yeah, I mean, it it is really showing up in the therapy room, I tell you.
00:12:03.799 --> 00:12:15.159
I mean, people are coming in with but, well, this is what ChatGPT said, you know, to be able to understand, you know, just technically as a clinician what the LLM is, and that it's not, we don't, we're not gonna humanize it.
00:12:15.240 --> 00:12:17.720
Like I call Claude he and I catch myself all the time.
00:12:17.960 --> 00:12:18.519
I do as well.
00:12:18.919 --> 00:12:20.360
He's not a he, he's a computer.
00:12:20.519 --> 00:12:23.000
It's hard not to because they're trained that way.
00:12:23.240 --> 00:12:26.519
You think they're your best friend, you know, and it's a problem with therapy too.
00:12:26.600 --> 00:12:32.440
I mean, I know we could go down that rabbit hole too, just like, you know, clients turning to that for support and that sort of thing.
00:12:32.519 --> 00:12:45.960
But but yes, back to like AI literacy as the clinician and tool literacy, and then being able to kind of break it down to layman's terms to explain to our clients, this is what's happening when it's involved in the session and how you're protected and all of that.
00:12:46.200 --> 00:12:46.440
Exactly.
00:12:46.600 --> 00:12:47.240
Yeah, exactly.
00:12:47.320 --> 00:12:50.279
And we're using AI all generative AI all over, right?
00:12:50.440 --> 00:12:56.519
So, you know, people are using it at home and they're redesigning their backyards with generative AI.
00:12:56.600 --> 00:12:58.039
And I think that's amazing, right?
00:12:58.120 --> 00:13:06.039
Like, you know, it's not so amazing for the landscape architects that may not be getting, you know, the business that they want.
00:13:06.200 --> 00:13:23.080
But I think when we use it thoughtfully and intentionally, right, it's not to replace those skilled individuals, it's to use it to be more informed as a coexistence and being able to have better ideas and a better idea of what it is you want.
00:13:23.399 --> 00:13:24.200
That's amazing.
00:13:24.360 --> 00:13:24.759
Yeah.
00:13:25.000 --> 00:13:29.080
And something you talk about is full practice authority.
00:13:29.159 --> 00:13:39.240
And we're kind of talking about that a little bit, but can you expand upon that concept and how it applies to us as clinicians using AI tools, whether it's Claude or, you know, a note scribe that you're using?
00:13:39.399 --> 00:14:00.519
What does that mean exactly when you're talking about So full practice authority is the idea of clinicians being able to practice at the top of their licenses to be able to use the education and preparation that you have garnered in school to be a clinician that could be effective in your patient care.
00:14:00.759 --> 00:14:04.120
So it's mostly used for nurse practitioners.
00:14:04.680 --> 00:14:04.919
Okay.
00:14:05.159 --> 00:14:15.320
And you know, the idea is that uh we have 50 states and when you count DC, we'll count them as as an extra as a 51.
00:14:15.639 --> 00:14:28.120
And for NP practice, there's a lot of variability across the states, and that's because the states are the ones who regulate and legislate clinical practice.
00:14:28.279 --> 00:14:34.519
There is no one like law that any healthcare provider follows, right?
00:14:34.600 --> 00:14:38.440
You have to follow a law in the state that you are licensed.
00:14:38.600 --> 00:14:42.039
And if you're not licensed in that state, then you can't practice in that state.
00:14:42.279 --> 00:15:04.120
So the idea is that you know, the board of nursing in the state that regulates the practice of nursing and nurse practitioners has the sole authority to set professional standards and discipline and everything that goes along without the requirement of another discipline being involved in the provision of patient care.
00:15:04.279 --> 00:15:06.440
So that's the idea of full practice authority.
00:15:06.519 --> 00:15:24.039
Right now in the US, there are 27 states that are considered full practice authority, and all of those other states that are not, it is legislation that must be introduced and passed in order for those other states to be considered full practice authority.
00:15:24.440 --> 00:15:25.320
So interesting.
00:15:25.480 --> 00:15:27.080
Yeah, I mean, I'm on the other side of it.
00:15:27.159 --> 00:15:39.080
I I have a whole nother business called the traveling therapist, and it is such I live in Airbnb's, I'm fully nomadic, and it is such an issue, like that we're not just under one regulatory board across the entire United States.
00:15:39.159 --> 00:15:39.799
It's so complicated.
00:15:40.200 --> 00:15:41.799
It's very, very complicated.
00:15:42.360 --> 00:15:50.120
You have different states and different sets of rules, and it's up to you as the licensed individual to make sure that you're following the rules of that state.
00:15:50.200 --> 00:15:54.519
And what makes things so much harder is telehealth, right?
00:15:54.680 --> 00:16:09.639
So yeah, you can be literally seeing a patient in any one of the 50 states, and you have to follow the law, and you are in your state, but you have to follow the law and be licensed in that state that you're providing that care to where that patient is.
00:16:09.720 --> 00:16:12.680
And it's very, very difficult and very confusing.
00:16:12.840 --> 00:16:17.720
And again, it's incumbent upon yourself to know what those laws are and follow them.
00:16:17.960 --> 00:16:26.759
Otherwise, you don't want to get a knock at the door from the board of whatever profession you're practicing and saying, you know, you're not practicing, you know, correctly.
00:16:27.080 --> 00:16:27.480
Yeah.
00:16:27.720 --> 00:16:28.840
Yeah, it's so interesting.
00:16:28.919 --> 00:16:32.440
And I'm seeing so much AI legislation come down in different states.
00:16:32.600 --> 00:16:34.039
I'm just so curious.
00:16:34.120 --> 00:16:42.440
I mean, this is probably a future conversation to see like how is all this going to play out for different states and professions and boards, and how are they gonna interpret these rules?
00:16:42.600 --> 00:16:48.039
And it's just kind of like spins your mind up a little bit, just trying to keep track of all of it.
00:16:48.120 --> 00:16:49.639
It's really an interesting time right now.
00:16:49.879 --> 00:16:52.200
It's nearly impossible to keep track of that as well.
00:16:52.360 --> 00:16:53.480
The legislative side.
00:16:53.560 --> 00:16:58.840
I mean, there are states that are saying no use of chatbots for therapy, right?
00:16:59.159 --> 00:17:03.480
I I mean, even states have passed legislation.
00:17:03.720 --> 00:17:14.359
Washington state is one of the first states in Oregon passed a law that said you can't call an AI bot, chatbot a nurse, right?
00:17:14.440 --> 00:17:24.119
You yes, the term nurse is a protected term, and you need a license and you need to be a human being that can be called a nurse.
00:17:24.279 --> 00:17:34.759
So there are states that have passed these laws that you know make it illegal for an AI company to use the term nurse so-and-so as a chat box.
00:17:35.000 --> 00:17:39.000
But my bigger question is where does this end, right?
00:17:39.079 --> 00:17:44.440
Do we need a law for every profession in every state to do this?
00:17:44.599 --> 00:17:51.240
Now, I believe that you know, if you're a therapist, if you if you're a nurse practitioner, that should be a protected title.
00:17:51.319 --> 00:17:54.440
And and not not anyone should be able to call themselves that.
00:17:54.680 --> 00:18:02.759
But I believe we need a comprehensive strategy that isn't individual, state by state, for every profession because it's it's messy.
00:18:02.920 --> 00:18:05.720
It makes it that much harder to track.
00:18:05.960 --> 00:18:12.839
Yeah, and I mean, don't get me wrong, I'm grateful for the legislation because I don't want AI therapy bots taking over the world, you know.
00:18:13.079 --> 00:18:21.400
We said earlier you're not gonna be replaced, but like trying or attempting to replace our professions, any of our professions, you know, it's it's a super interesting time right now.
00:18:21.720 --> 00:18:25.880
What's great about a chatbot, right, is that it's available 24-7.
00:18:26.440 --> 00:18:26.759
Right.
00:18:26.920 --> 00:18:30.200
And for the most part, a lot of them are free, right?
00:18:30.440 --> 00:18:32.599
Yes, and we know nothing is free.
00:18:32.920 --> 00:18:33.079
Right.
00:18:33.400 --> 00:18:48.839
So when you're using an AI chatbot, and if you don't turn off the setting that says don't train on my data, your data and your conversations are being used to train future models.
00:18:49.000 --> 00:18:51.559
And again, is that a bad thing?
00:18:51.880 --> 00:18:58.599
Maybe, maybe not, but you know, maybe if it's getting better, it'll be more useful.
00:18:58.759 --> 00:18:59.640
But at what cost?
00:18:59.799 --> 00:19:09.000
Again, and I think that's where I take the position that we as a society are the ones that need to decide how how to use this or not.
00:19:09.400 --> 00:19:09.960
Yeah.
00:19:10.279 --> 00:19:11.640
And you brought that up at the beginning.
00:19:11.720 --> 00:19:13.079
I think it's such an important topic.
00:19:13.240 --> 00:19:20.839
Do you have sort of your own guardrails now or guardrails that you teach around how to make sure or ensure that we're not crossing that line?
00:19:21.000 --> 00:19:34.599
And I mean, you're talking about like turn off the settings so it's private, they're not using your data in the large LLMs and that sort of thing, but I'm just wondering, are there other things that maybe somebody hasn't thought about that we can keep our autonomy and not like, you know, totally feed the machine kind of thing?
00:19:34.839 --> 00:19:35.160
Yeah.
00:19:35.400 --> 00:19:38.359
And I think, I mean, the biggest is the privacy settings.
00:19:38.599 --> 00:19:39.240
Yeah, okay.
00:19:39.400 --> 00:19:39.640
Yeah.
00:19:39.880 --> 00:19:41.240
That's first and foremost.
00:19:41.559 --> 00:19:46.359
And then it's, you know, it's making sure that you're verifying outputs.
00:19:46.599 --> 00:19:55.240
It's making sure that when you see a reference link that you actually click on the reference link and see what it goes to.
00:19:55.559 --> 00:20:02.120
Because we know that LLMs are very good at we call them hallucinations.
00:20:02.279 --> 00:20:04.839
I also like to call them confabulations.
00:20:05.240 --> 00:20:05.880
Oh, nice.
00:20:06.120 --> 00:20:10.680
Because these have been programmed to give you an answer, a response.
00:20:10.759 --> 00:20:11.000
Yeah.
00:20:11.240 --> 00:20:18.519
Because what would happen if you put in your prompt in your favorite LLM and it said, I don't know.
00:20:18.759 --> 00:20:19.000
Right?
00:20:19.160 --> 00:20:19.319
Yeah.
00:20:19.480 --> 00:20:20.519
I don't know the answer to that.
00:20:20.759 --> 00:20:21.799
You wouldn't use it.
00:20:21.880 --> 00:20:24.200
You'd stop reading and say that that didn't help me.
00:20:24.279 --> 00:20:25.240
So unhelpful.
00:20:25.400 --> 00:20:25.480
Yeah.
00:20:25.720 --> 00:20:26.039
Exactly.
00:20:26.359 --> 00:20:26.839
Unhelpful.
00:20:27.400 --> 00:20:34.039
So they are trained to give you a response, and sometimes we don't see what's happening behind the scenes.
00:20:34.200 --> 00:20:39.160
Sometimes it's not very confident in that response and it's going to just spit out something.
00:20:39.319 --> 00:20:49.559
I've clicked on links that have that DOI number that you know looks like a legit, you know, like it goes to PubMed, and then you click it and it doesn't exist.
00:20:50.039 --> 00:20:53.319
Or for some different topic, the paper.
00:20:53.480 --> 00:21:03.000
So you have to always, always verify the outputs and not assume that, you know, because it looks good, that it actually is good.
00:21:03.400 --> 00:21:05.079
Yes, I I love that.
00:21:05.240 --> 00:21:06.120
Such good advice.
00:21:06.359 --> 00:21:20.279
And I also think, I mean, what you're doing, like with your education, that sort of thing, like people like us showing up and trying to educate, you know, the people in our field around all of these topics can just help everyone have a seat at the table when the stuff's being developed.
00:21:20.359 --> 00:21:23.480
Like Barries, being the clinical consultant as a therapist, you know.
00:21:23.559 --> 00:21:28.519
It's I just think that stuff's so important for us to show up to the rooms too, where the stuff's being developed.
00:21:28.599 --> 00:21:49.400
If we can't do we have to partner with the smart engineers and the software developers, they're good at what they do, we're good at what we do, and you know, we need to work together on the solutions to make sure that, you know, again, we're using it ethically, responsibly, in the best way possible, based on, you know, the current limitations.
00:21:49.480 --> 00:22:04.839
And, you know, a limitation today, I'm borrowing this or I'm stealing this from very other prolific AI people who've said, you know, this is the worst that AI is gonna be today is the worst it's ever going to be, right?
00:22:04.920 --> 00:22:06.920
It's only gonna get better from here.
00:22:07.160 --> 00:22:09.240
And you know, again, what what does that mean?
00:22:09.319 --> 00:22:10.519
Uh, how do we use it?
00:22:10.680 --> 00:22:14.599
And just because we know knew something six months ago, it could have changed.
00:22:14.759 --> 00:22:16.359
You know, could have changed uh already.
00:22:16.440 --> 00:22:18.519
So we have to be up uh up to date on that.
00:22:19.079 --> 00:22:19.480
Wow.
00:22:19.640 --> 00:22:21.720
Yeah, there's so much to think about and all that.
00:22:22.039 --> 00:22:27.000
We kind of jumped around a little bit, but I would love to go back to the Columbia piece and how you introduced.
00:22:27.480 --> 00:22:30.279
The founding AI program with the nursing program there.
00:22:30.359 --> 00:22:33.559
I know that you're not doing that anymore, but you started it, right?
00:22:33.720 --> 00:22:34.039
Right.
00:22:34.200 --> 00:22:38.440
I'm just curious, like, like, where do you see AI in the role of education?
00:22:38.599 --> 00:22:42.920
It's I mean, it's so cool that you're starting that conversation in Columbia, no less.
00:22:43.079 --> 00:22:43.960
I mean, that is amazing.
00:22:44.120 --> 00:22:47.480
And I'm just curious, like, where do you see that going?
00:22:47.720 --> 00:22:49.960
Are all the programs just going to be AI forward now?
00:22:50.120 --> 00:22:54.759
We're just all going to like enter into the workforce with AI forward thinking, or what do you think about that?
00:22:55.000 --> 00:23:01.640
Well, I think we need to have the conversation to determine how we use AI in the classroom.
00:23:02.200 --> 00:23:11.640
And the challenge that exists, like you have your universities, and then you know, in the universities, you have your schools of you know the different health professions.
00:23:11.960 --> 00:23:18.440
There's very little discussion that's happening other than like, you know, a policy statement.
00:23:18.599 --> 00:23:20.440
Now, a policy statement is important.
00:23:20.599 --> 00:23:23.400
And at heart, I am a policy wonk, right?
00:23:23.559 --> 00:23:37.160
I love I love health policy, but these this technology requires more than what has been written on a piece of paper in trying to anticipate something that we really don't even have a good grip on, right?
00:23:37.319 --> 00:23:48.279
So what I believe needs to happen across every health profession in every school that teaches the health professions is that there needs to be a deep dive by the faculty.
00:23:48.680 --> 00:23:57.880
And there has to be an intentional design of how we're going to incorporate AI into assignments, or not, right?
00:23:58.359 --> 00:24:06.440
What is creating a lot of challenges for students is that they'll have one faculty member who says, use AI, I don't care, right?
00:24:06.599 --> 00:24:08.759
Just cite or cite it, whatever.
00:24:08.920 --> 00:24:13.000
And you'll have another faculty member who says, Don't use any AI.
00:24:13.160 --> 00:24:15.559
If I find you using AI, I'm gonna fail you.
00:24:15.799 --> 00:24:34.359
And that's not a problem, like you know, the concept of saying I don't want you to use AI and I want you to critically think, but it's causing confusion because those faculty are not on the same page collectively, and yes, requires a look at the curriculum.
00:24:34.599 --> 00:24:42.200
I'm not saying AI has to be embedded in every course in every assignment, yeah, but it should be somewhere.
00:24:42.440 --> 00:24:50.920
And you know, uh it requires us uh when I was faculty to redesign the assignments, right?
00:24:51.079 --> 00:25:01.559
Like assigning a 15-page paper in today's world is not going to do it because it's too easy to hand it on over to a large language model.
00:25:01.720 --> 00:25:19.240
So maybe, you know, a possible suggestion is to have the large language model refine your uh opening paragraph or give you some suggestions of you know where you should go from here.
00:25:20.039 --> 00:25:30.759
But you have to use it, you know, thoughtfully and not have it take over and sterilize your written word into something that seems generic and fake, you know.
00:25:30.920 --> 00:25:33.319
Yeah, that's not even your original thought.
00:25:33.400 --> 00:25:33.960
Yeah, yeah.
00:25:34.279 --> 00:25:36.680
We like to use the word AI slop.
00:25:37.079 --> 00:25:45.240
Yes, and there's a lot of AI slop out there, and you know, if that's all we're reading, we're gonna start mimicking the large language.
00:25:45.480 --> 00:25:46.200
We're like AI.
00:25:46.279 --> 00:25:53.640
I already hear people talking, like I go to these trainings, and I'm like, he literally sounds like what Claude sounds like because it's like you use it so much.
00:25:53.880 --> 00:25:54.200
Absolutely.
00:25:54.519 --> 00:25:59.720
It's like I don't even know if we realize that we're starting to sound like the AI that we're like talking to, you know.
00:25:59.799 --> 00:26:01.160
It's it's really interesting.
00:26:01.400 --> 00:26:02.120
It is interesting.
00:26:02.200 --> 00:26:07.000
And if I hear the word friction or moat, like moat, that's the new one.
00:26:07.079 --> 00:26:07.400
Mote.
00:26:07.720 --> 00:26:18.039
I don't use those words, like maybe people have used them, but I've never heard them used as much as as they're used today, so it's really interesting.
00:26:18.279 --> 00:26:19.000
That's hilarious.
00:26:19.079 --> 00:26:24.519
It used to be game changer, but literally moat is like all of a sudden showing up at everything I'm doing with Claude these days.
00:26:24.759 --> 00:26:25.880
It's hilarious.
00:26:27.960 --> 00:26:29.160
Oh my gosh, I love it.
00:26:29.240 --> 00:26:30.680
I feel like I could talk about this all day.
00:26:31.799 --> 00:26:39.880
Can we talk a little bit about your programs and what you do and how you help nurse practitioners and other clinicians that are in your ecosystem?
00:26:40.039 --> 00:26:43.799
How, how does your program help with everything that we're talking about?
00:26:44.039 --> 00:26:58.680
So thank you for mentioning AI Nurse Academy, and that is my platform that I built out to provide the literacy for nurses, but not just nurses and all clinicians, right?
00:26:58.759 --> 00:27:01.960
So, right now I have one course that's available.
00:27:02.120 --> 00:27:09.480
It's a one-hour on-demand type format that you know goes over these principles that we talked about.
00:27:09.720 --> 00:27:19.000
In addition to the on-demand format, and there's a post-test at the end, there's the option to join a community.
00:27:19.400 --> 00:27:37.960
And this is the community of people who've taken the course and want to have continuing conversations about AI, and whether that be sharing a really useful prompt that they found works well for them, or sharing some of the limitations of AI.
00:27:38.440 --> 00:27:47.960
It's uh it's an active community, it's a way to, you know, not just make it a static course, but it's to continue the conversation.
00:27:48.279 --> 00:27:55.079
Because again, I think that's the only way we're gonna try and make the best sense of it and the best use of of AI.
00:27:55.400 --> 00:27:59.880
And just having that community dialogue where we can bounce ideas off of each other, that sort of thing.
00:28:00.359 --> 00:28:06.440
Yeah, I will say I did a deep dive before this interview on your website, and I was like, oh, he has a community, and I was like, it's gotta be a paid community.
00:28:06.519 --> 00:28:07.880
I was like, no, it's totally freaky.
00:28:08.039 --> 00:28:12.279
So it comes with the course or the course is the course, the course, right, right.
00:28:12.440 --> 00:28:16.759
But it you know, you get lifetime access to the community again a good way.
00:28:16.839 --> 00:28:26.680
And I'm in there, you know, trying to uh stir conversation and and and trying to, you know, be as thoughtful as I possibly can be.
00:28:27.079 --> 00:28:27.799
I love it.
00:28:27.880 --> 00:28:28.440
It's really good.
00:28:28.519 --> 00:28:29.720
You guys go check out his website.
00:28:29.799 --> 00:28:32.039
Of course, we'll have everything down in the show notes too.
00:28:32.279 --> 00:28:39.559
So we always like to do kind of like a like a hot seat a little bit at the end and just ask what AI tools are you using right now?
00:28:39.720 --> 00:28:52.519
I mean, outside of like the work that we do and all that, like do you have like really like your favorite go-to just for regular productivity or you know, just fun, you know, AI stuff that you're doing right now that you would want to share?
00:28:52.759 --> 00:28:56.039
Anybody that's just like, I just love AI, I want to know what everybody's using it for.
00:28:56.279 --> 00:28:56.759
Sure, sure.
00:28:56.839 --> 00:28:58.440
I'm happy, happy to share.
00:28:58.599 --> 00:29:05.079
And I have no like ownership in any of these companies, so this is just what what I use.
00:29:05.160 --> 00:29:15.079
And and my suggestion really is to people is to use them all because I don't think one works better than another for everything, right?
00:29:15.240 --> 00:29:22.200
So I'm using Claude a lot these days, and I purchased the Claude Coldwork subscription.
00:29:22.279 --> 00:29:33.160
So I'm playing with that because I want to be able to understand you know the idea of an AI agent and what that means, and it's able to do things on my behalf.
00:29:33.319 --> 00:29:35.559
I have very strict guardrails on that.
00:29:35.640 --> 00:29:44.200
I don't let it, you know, write emails for me, I don't let it send emails for me, I don't let it, you know, have access to my entire hard drive.
00:29:44.359 --> 00:29:46.759
You know, it has access to one folder, right?
00:29:46.839 --> 00:29:48.039
And one folder only.
00:29:48.200 --> 00:30:00.839
I'm enjoying using Claude Cowork, but I use Chat GPT, I use Gemini as well for you know all different types of drafting ideas and concept ideas.
00:30:01.079 --> 00:30:20.519
When it comes to LLMs in in the clinical space, there's a host of free tools, and yeah, the ones that I have used are open evidence, there is Vera Health, is another one, Helio H E A L I OK is another good one, and uh Evidence M D.
00:30:21.160 --> 00:30:25.640
These are again shout-outs to those models.
00:30:25.960 --> 00:30:35.559
They all work similarly, in my opinion, but give you different types of output that you may find better for your style.
00:30:35.720 --> 00:30:38.120
Like some are more like bulleted, right?
00:30:38.200 --> 00:30:44.039
So you're gonna get an output that's boom, boom, boom, than reading like paragraphs and paragraphs of content.
00:30:44.279 --> 00:30:46.440
So yeah, I like seeing those.
00:30:46.519 --> 00:30:52.680
And again, those are the models that are trained on the peer-reviewed content.
00:30:52.920 --> 00:31:06.519
It doesn't mean that there's zero possibility of hallucination, still can hallucinate, but it's trained on better data and it's gonna give you, you know, better clinical outputs than the statue.
00:31:07.559 --> 00:31:13.400
Are they node scribes or are they more like just for like I've got a client with major depressive disorder and celiac disease?
00:31:13.480 --> 00:31:17.480
Like, how do I like what's the evidence to support what medication to use?
00:31:17.559 --> 00:31:18.359
Like that kind of stuff.
00:31:18.519 --> 00:31:20.599
Yeah, so okay, interesting.
00:31:20.839 --> 00:31:24.519
Not they all, but some of them also have built-in scribes as well.
00:31:24.839 --> 00:31:25.559
Oh, okay.
00:31:25.720 --> 00:31:26.359
Oh, interesting.
00:31:26.599 --> 00:31:30.120
Yeah, and that's another part of their you know, offerings.
00:31:30.440 --> 00:31:37.160
The part that I focus on is the the large language model, and exactly as you suggest, you know.
00:31:37.240 --> 00:31:46.279
I have a patient with celiac, you know, they're you know, and you give your your history, you know, 21 years old on X, Y, and Z medication.
00:31:46.519 --> 00:31:51.559
I again, these are tools that are free for the most part.
00:31:51.720 --> 00:31:56.120
If you have an NPI number, you could get access to them for the most part.
00:31:56.440 --> 00:31:57.000
Oh, cool.
00:31:57.240 --> 00:32:01.960
So you can compare and contrast, you know, uh outputs from the different models.
00:32:02.200 --> 00:32:07.720
What they aren't yet is they're not embedded in the electronic health record, right?
00:32:07.960 --> 00:32:08.759
Yeah, gotcha.
00:32:09.319 --> 00:32:19.000
Someday they will be, you know, and and we're not that far away from, you know, Epic is using, you know, they have their own suite of tools that they're developing.
00:32:19.160 --> 00:32:25.240
So, but right now they're essentially standalone in, and I have to say it's in my course.
00:32:25.559 --> 00:32:41.079
Do not upload any PHI loud into any large language model unless you know your practice has a BAA, you know, a business associate agreement that you know is is HIPAA compliant.
00:32:41.240 --> 00:32:51.960
But yeah, you know, you can you can anonymize the the data so you're not, you know, yeah, sharing sadding, you know, confidential information.
00:32:52.200 --> 00:32:53.400
Yes, totally love that.
00:32:53.480 --> 00:32:54.839
Yeah, so important to say.
00:32:55.079 --> 00:32:55.640
Thank you.
00:32:55.799 --> 00:32:57.079
What a great conversation.
00:32:57.160 --> 00:32:58.039
I really appreciate it.
00:32:58.120 --> 00:33:04.440
Could you tell everybody your website or your email, whatever, the best way to contact you and follow you and connect with you?
00:33:04.680 --> 00:33:07.000
Yeah, so I would say find me on LinkedIn.
00:33:07.319 --> 00:33:07.640
Okay.
00:33:07.880 --> 00:33:09.400
I'm pretty active there, Stephen A.
00:33:09.559 --> 00:33:10.440
Ferrara.
00:33:10.680 --> 00:33:15.640
There's a few of us out there that have the same name, but I'm the uh I'm the nurse practitioner one.
00:33:16.039 --> 00:33:17.480
Oh, so find me there.
00:33:17.559 --> 00:33:23.640
And then my course, AI Nurse Academy, is at AINurseacademy.ai.
00:33:24.200 --> 00:33:25.240
Easy sign up.
00:33:25.400 --> 00:33:32.359
As I mentioned, it's a$20 course, you know, one-time fee that you could, you know, revisit the course.
00:33:32.440 --> 00:33:37.319
It comes with like a little resource, one pager that you can download.
00:33:37.559 --> 00:33:44.440
And then, as you mentioned, it gives you access to the community where you can actually engage in some some conversations.
00:33:44.599 --> 00:33:48.440
But but yeah, you know, I'll think we all need to have those conversations.
00:33:48.759 --> 00:33:51.880
Yeah, that's amazing to have that community there forever.
00:33:52.120 --> 00:33:53.640
Yeah, thank you so much.
00:33:53.720 --> 00:33:55.400
It's really really been great talking to you.
00:33:55.559 --> 00:33:58.519
Yeah, it's been a pleasure and thank you so much for having me.
