4ec792e7-4bcc-492d-a042-b2d873afc809/6-0
00:00:03.328 --> 00:00:06.175
We're here today with Dr.
Paul Testa from NYU Angone,

4ec792e7-4bcc-492d-a042-b2d873afc809/6-1
00:00:06.175 --> 00:00:10.392
the Chief Health Informatics Officer and
an emergency medicine physician. Paul,

4ec792e7-4bcc-492d-a042-b2d873afc809/6-2
00:00:10.392 --> 00:00:12.448
thank you so much for joining us today.

4ec792e7-4bcc-492d-a042-b2d873afc809/5-0
00:00:12.608 --> 00:00:13.568
Thanks so much.

4ec792e7-4bcc-492d-a042-b2d873afc809/8-0
00:00:13.968 --> 00:00:17.209
As background for our listeners,
NYU Langone Health System is

4ec792e7-4bcc-492d-a042-b2d873afc809/8-1
00:00:17.209 --> 00:00:19.510
headquartered in New York,
New York with 2,

4ec792e7-4bcc-492d-a042-b2d873afc809/8-2
00:00:19.510 --> 00:00:22.594
000 beds across 7 hospitals supported by
6,000 physicians.

4ec792e7-4bcc-492d-a042-b2d873afc809/8-3
00:00:22.594 --> 00:00:26.829
We're going to have a lot of fun today,
have a lot of different topics to cover,

4ec792e7-4bcc-492d-a042-b2d873afc809/8-4
00:00:26.829 --> 00:00:30.018
but let's start out with Gen.
AI because why not? It's 2026.

4ec792e7-4bcc-492d-a042-b2d873afc809/8-5
00:00:30.018 --> 00:00:33.312
So you rolled out one of the nation's
first privately managed,

4ec792e7-4bcc-492d-a042-b2d873afc809/8-6
00:00:33.312 --> 00:00:37.181
secure and HIPAA compliant GPT-4
ecosystems in a healthcare organization,

4ec792e7-4bcc-492d-a042-b2d873afc809/8-7
00:00:37.181 --> 00:00:38.488
and you did so enterprise

4ec792e7-4bcc-492d-a042-b2d873afc809/9-0
00:00:38.608 --> 00:00:41.303
wide on Azure.
How do you select that model because

4ec792e7-4bcc-492d-a042-b2d873afc809/9-1
00:00:41.303 --> 00:00:44.206
everyone across the country is trying to
figure it out,

4ec792e7-4bcc-492d-a042-b2d873afc809/9-2
00:00:44.206 --> 00:00:47.887
that hyperscaler in particular,
and what changed between the pilot and

4ec792e7-4bcc-492d-a042-b2d873afc809/9-3
00:00:47.887 --> 00:00:48.768
the full rollout?

4ec792e7-4bcc-492d-a042-b2d873afc809/10-0
00:00:49.568 --> 00:00:52.528
Great question,
particularly the last little bit.

4ec792e7-4bcc-492d-a042-b2d873afc809/12-0
00:00:54.168 --> 00:00:57.709
Folks a lot smarter than me made the
platform decision,

4ec792e7-4bcc-492d-a042-b2d873afc809/12-1
00:00:57.709 --> 00:01:02.642
but Microsoft has been an incredibly
nimble partner with us. Let's be honest,

4ec792e7-4bcc-492d-a042-b2d873afc809/12-2
00:01:02.642 --> 00:01:06.689
we're committed to all the foundation
models and all the hosts.

4ec792e7-4bcc-492d-a042-b2d873afc809/12-3
00:01:06.689 --> 00:01:11.368
So we have presence with all three and
all three. What has been the real,

4ec792e7-4bcc-492d-a042-b2d873afc809/12-4
00:01:11.368 --> 00:01:16.048
the kind of the privilege in my role,
and particularly at NYU Langone, is

4ec792e7-4bcc-492d-a042-b2d873afc809/13-0
00:01:16.288 --> 00:01:20.049
We never had to send out the no don't
e-mail, right? 36 months ago,

4ec792e7-4bcc-492d-a042-b2d873afc809/13-1
00:01:20.049 --> 00:01:22.925
everything gets released.
We see everyone using it.

4ec792e7-4bcc-492d-a042-b2d873afc809/13-2
00:01:22.925 --> 00:01:26.189
We see the traffic.
And the e-mail that gets sent out from

4ec792e7-4bcc-492d-a042-b2d873afc809/13-3
00:01:26.189 --> 00:01:30.448
leadership about generative AI is, yes,
this is an incredibly powerful tool.

4ec792e7-4bcc-492d-a042-b2d873afc809/13-4
00:01:30.448 --> 00:01:34.264
It's going to change a lot of things.
If you want to do cool things,

4ec792e7-4bcc-492d-a042-b2d873afc809/13-5
00:01:34.264 --> 00:01:34.928
let us know.

4ec792e7-4bcc-492d-a042-b2d873afc809/14-0
00:01:36.288 --> 00:01:41.057
And we were able to partner really
quickly with OpenAI and Microsoft in

4ec792e7-4bcc-492d-a042-b2d873afc809/14-1
00:01:41.057 --> 00:01:45.428
getting out a secure platform to say,
and here's your playground.

4ec792e7-4bcc-492d-a042-b2d873afc809/14-2
00:01:45.428 --> 00:01:48.674
Do whatever you want here.
It's HIPAA compliant.

4ec792e7-4bcc-492d-a042-b2d873afc809/14-3
00:01:48.674 --> 00:01:52.317
It will hold our privileged information
and protected.

4ec792e7-4bcc-492d-a042-b2d873afc809/14-4
00:01:52.317 --> 00:01:54.768
And then if you discover cool things,

4ec792e7-4bcc-492d-a042-b2d873afc809/15-0
00:01:55.168 --> 00:01:57.352
Come and talk to us and we'll help you
scale.

4ec792e7-4bcc-492d-a042-b2d873afc809/15-1
00:01:57.352 --> 00:02:00.344
Because innovation without scale is silly.
It's A distraction.

4ec792e7-4bcc-492d-a042-b2d873afc809/15-2
00:02:00.344 --> 00:02:02.528
It actually only matters if you can scale
it.

4ec792e7-4bcc-492d-a042-b2d873afc809/16-0
00:02:03.808 --> 00:02:08.750
So I think what a lot of systems have
found is they roll something out. I mean,

4ec792e7-4bcc-492d-a042-b2d873afc809/16-1
00:02:08.750 --> 00:02:13.075
the most popular and easy one to discuss
is ambient AI, but you know,

4ec792e7-4bcc-492d-a042-b2d873afc809/16-2
00:02:13.075 --> 00:02:17.153
that perhaps is the most covered and
therefore least interesting.

4ec792e7-4bcc-492d-a042-b2d873afc809/16-3
00:02:17.153 --> 00:02:22.034
But as you roll out different kinds of AI
platforms across healthcare delivery

4ec792e7-4bcc-492d-a042-b2d873afc809/16-4
00:02:22.034 --> 00:02:22.528
systems,

4ec792e7-4bcc-492d-a042-b2d873afc809/17-0
00:02:22.928 --> 00:02:27.635
I think a lot of people in your position
find that there is high demand and many

4ec792e7-4bcc-492d-a042-b2d873afc809/17-1
00:02:27.635 --> 00:02:31.180
use cases, many users,
but you have to balance that with the

4ec792e7-4bcc-492d-a042-b2d873afc809/17-2
00:02:31.180 --> 00:02:34.900
high tokenization cost.
So how do you choose which use cases to

4ec792e7-4bcc-492d-a042-b2d873afc809/17-3
00:02:34.900 --> 00:02:37.631
prioritize and which users to grant
access to?

4ec792e7-4bcc-492d-a042-b2d873afc809/17-4
00:02:37.631 --> 00:02:42.048
And how do you kind of limit their access
so that you don't break the bank?

4ec792e7-4bcc-492d-a042-b2d873afc809/20-0
00:02:42.408 --> 00:02:47.112
Yeah. Also, right question.
And then there's that piece that came out

4ec792e7-4bcc-492d-a042-b2d873afc809/20-1
00:02:47.112 --> 00:02:52.220
from Bain recently on looking at the
tokenization of OpEx and CapEx, right?

4ec792e7-4bcc-492d-a042-b2d873afc809/20-2
00:02:52.220 --> 00:02:56.924
That's going to change our calculus.
We went in, Ambient maybe isn't,

4ec792e7-4bcc-492d-a042-b2d873afc809/20-3
00:02:56.924 --> 00:03:00.688
the conversation's not finished yet
because of nursing.

4ec792e7-4bcc-492d-a042-b2d873afc809/18-0
00:02:59.528 --> 00:03:00.048
Mhm.

4ec792e7-4bcc-492d-a042-b2d873afc809/19-0
00:03:01.408 --> 00:03:01.888
Mhm.

4ec792e7-4bcc-492d-a042-b2d873afc809/23-0
00:03:01.408 --> 00:03:05.600
Right. Yeah. I mean, and for providers,
we get it. It works incredibly well.

4ec792e7-4bcc-492d-a042-b2d873afc809/23-1
00:03:05.600 --> 00:03:09.520
Most patients are very accepting of it.
Providers say, don't, you know,

4ec792e7-4bcc-492d-a042-b2d873afc809/23-2
00:03:09.520 --> 00:03:13.548
it has changed the way we work.
But nursing in the inpatient setting is a

4ec792e7-4bcc-492d-a042-b2d873afc809/23-3
00:03:13.548 --> 00:03:17.686
very different literal conversation.
They're not used to speaking out loud.

4ec792e7-4bcc-492d-a042-b2d873afc809/21-0
00:03:17.128 --> 00:03:17.608
Mhm.

4ec792e7-4bcc-492d-a042-b2d873afc809/23-4
00:03:17.686 --> 00:03:20.408
And there is, I won't say resistance,
but there's

4ec792e7-4bcc-492d-a042-b2d873afc809/24-0
00:03:20.768 --> 00:03:25.683
caution as we get our nurses engaged.
So we haven't had that huge uptick yet

4ec792e7-4bcc-492d-a042-b2d873afc809/22-0
00:03:22.048 --> 00:03:22.288
Yeah.

4ec792e7-4bcc-492d-a042-b2d873afc809/24-1
00:03:25.683 --> 00:03:29.067
and we're learning more what that means
for nursing.

4ec792e7-4bcc-492d-a042-b2d873afc809/24-2
00:03:29.067 --> 00:03:34.046
I actually think it's going to be a
bigger game changer for nurses than it is

4ec792e7-4bcc-492d-a042-b2d873afc809/24-3
00:03:34.046 --> 00:03:37.430
for, you know, LIPs and APPs. That said,
we've been,

4ec792e7-4bcc-492d-a042-b2d873afc809/24-4
00:03:37.430 --> 00:03:39.728
we've had some lessons learned about

4ec792e7-4bcc-492d-a042-b2d873afc809/26-0
00:03:41.128 --> 00:03:45.115
about, you know,
making sure we've got triggers when we

4ec792e7-4bcc-492d-a042-b2d873afc809/26-1
00:03:45.115 --> 00:03:48.533
find things going wild.
We gave Ultraviolet AI,

4ec792e7-4bcc-492d-a042-b2d873afc809/25-0
00:03:46.568 --> 00:03:47.048
Mhm.

4ec792e7-4bcc-492d-a042-b2d873afc809/26-2
00:03:48.533 --> 00:03:53.375
our secure multi-model portal, to 52,
000 members of the workforce.

4ec792e7-4bcc-492d-a042-b2d873afc809/26-3
00:03:53.375 --> 00:03:58.858
We have not woken up with, you know,
essentially everybody has about 5 bucks

4ec792e7-4bcc-492d-a042-b2d873afc809/26-4
00:03:58.858 --> 00:04:01.208
in tokens a month. Most people...

4ec792e7-4bcc-492d-a042-b2d873afc809/27-0
00:04:01.408 --> 00:04:04.475
Many of that goes unused and is
provisioned to others.

4ec792e7-4bcc-492d-a042-b2d873afc809/27-1
00:04:04.475 --> 00:04:09.103
But what we have been doing is grabbing,
we want transparency. So we grab the top,

4ec792e7-4bcc-492d-a042-b2d873afc809/27-2
00:04:09.103 --> 00:04:13.397
you know, 10 users and look at, first off,
we have a limit set, so we won't,

4ec792e7-4bcc-492d-a042-b2d873afc809/27-3
00:04:13.397 --> 00:04:18.026
we'll stop them before they blow through
anything and start generating millions or

4ec792e7-4bcc-492d-a042-b2d873afc809/27-4
00:04:18.026 --> 00:04:20.368
hundreds of thousands of dollars worth of.

4ec792e7-4bcc-492d-a042-b2d873afc809/28-0
00:04:21.248 --> 00:04:24.452
bills and we engage with them.
Like what are you trying to do and why

4ec792e7-4bcc-492d-a042-b2d873afc809/28-1
00:04:24.452 --> 00:04:27.702
and how do we do this differently to make
sure it's scalable? And hey,

4ec792e7-4bcc-492d-a042-b2d873afc809/28-2
00:04:27.702 --> 00:04:29.488
this is a great idea, we'll pay for it.

4ec792e7-4bcc-492d-a042-b2d873afc809/29-0
00:04:31.168 --> 00:04:35.639
So as the AI usage grows,
are you thinking about cost

4ec792e7-4bcc-492d-a042-b2d873afc809/29-1
00:04:35.639 --> 00:04:41.601
predictability at all? I guess,
and the reason I ask is there are ways,

4ec792e7-4bcc-492d-a042-b2d873afc809/29-2
00:04:41.601 --> 00:04:45.576
there are some ways when you're managing
CapEx,

4ec792e7-4bcc-492d-a042-b2d873afc809/29-3
00:04:45.576 --> 00:04:48.888
it could be that you just have unlimited

4ec792e7-4bcc-492d-a042-b2d873afc809/30-0
00:04:49.008 --> 00:04:52.246
tokens or you have a limit and you say
you can hit, you know,

4ec792e7-4bcc-492d-a042-b2d873afc809/30-1
00:04:52.246 --> 00:04:56.528
this many tokens over this amount of time,
but there are other ways to kind of...

4ec792e7-4bcc-492d-a042-b2d873afc809/31-0
00:04:57.328 --> 00:04:59.208
Spread the cost and have like a...

4ec792e7-4bcc-492d-a042-b2d873afc809/34-0
00:04:58.088 --> 00:05:02.017
Well, there's a stage before that, right?
There's a moment before that to say,

4ec792e7-4bcc-492d-a042-b2d873afc809/32-0
00:05:00.368 --> 00:05:00.688
Yeah.

4ec792e7-4bcc-492d-a042-b2d873afc809/34-1
00:05:02.017 --> 00:05:04.404
you're scaling really quickly,
user number 300,

4ec792e7-4bcc-492d-a042-b2d873afc809/33-0
00:05:03.088 --> 00:05:03.568
Mhm.

4ec792e7-4bcc-492d-a042-b2d873afc809/34-2
00:05:04.404 --> 00:05:08.482
and it seems like you're doing incredibly
cool stuff. Can we talk and understand?

4ec792e7-4bcc-492d-a042-b2d873afc809/34-3
00:05:08.482 --> 00:05:12.312
And maybe we've got a cheaper way for you
to do this or do it on a different

4ec792e7-4bcc-492d-a042-b2d873afc809/34-4
00:05:12.312 --> 00:05:13.008
platform. But.

4ec792e7-4bcc-492d-a042-b2d873afc809/35-0
00:05:12.608 --> 00:05:15.808
So do you have any examples of cool stuff
that you've come across?

4ec792e7-4bcc-492d-a042-b2d873afc809/36-0
00:05:17.008 --> 00:05:17.728
I think.

4ec792e7-4bcc-492d-a042-b2d873afc809/37-0
00:05:18.808 --> 00:05:20.528
Well, I think...

4ec792e7-4bcc-492d-a042-b2d873afc809/38-0
00:05:21.608 --> 00:05:26.709
One thing that's been incredibly powerful
for us is we did enter an agreement with

4ec792e7-4bcc-492d-a042-b2d873afc809/38-1
00:05:26.709 --> 00:05:31.012
NVIDIA in 2017, give or take.
So we have a high performance computing

4ec792e7-4bcc-492d-a042-b2d873afc809/38-2
00:05:31.012 --> 00:05:35.622
cluster that is essentially the largest
high performance computing cluster

4ec792e7-4bcc-492d-a042-b2d873afc809/38-3
00:05:35.622 --> 00:05:40.539
dedicated to life sciences in the world.
And that cluster allows us some of our

4ec792e7-4bcc-492d-a042-b2d873afc809/38-4
00:05:40.539 --> 00:05:42.568
own, we can absorb costs in that.

4ec792e7-4bcc-492d-a042-b2d873afc809/39-0
00:05:42.928 --> 00:05:45.672
So chip access has been incredibly
important.

4ec792e7-4bcc-492d-a042-b2d873afc809/39-1
00:05:45.672 --> 00:05:49.788
We continue to do large investments there.
When you say interesting,

4ec792e7-4bcc-492d-a042-b2d873afc809/39-2
00:05:49.788 --> 00:05:53.248
do you mean like use cases?
Because we got tons of those.

4ec792e7-4bcc-492d-a042-b2d873afc809/41-0
00:05:53.408 --> 00:05:57.014
Yeah, I mean, well,
it's interesting from two ways.

4ec792e7-4bcc-492d-a042-b2d873afc809/40-0
00:05:54.288 --> 00:05:54.768
Yeah.

4ec792e7-4bcc-492d-a042-b2d873afc809/41-1
00:05:57.014 --> 00:06:00.412
Either one,
interesting clinical or business use

4ec792e7-4bcc-492d-a042-b2d873afc809/41-2
00:06:00.412 --> 00:06:04.157
cases, or two,
interesting ways to absorb kind of the

4ec792e7-4bcc-492d-a042-b2d873afc809/41-3
00:06:04.157 --> 00:06:09.913
GPU costs. And I think that's fascinating.
You beat before the stock exploded with

4ec792e7-4bcc-492d-a042-b2d873afc809/41-4
00:06:09.913 --> 00:06:13.034
Gen. AI,
you have access to your own Nvidia,

4ec792e7-4bcc-492d-a042-b2d873afc809/41-5
00:06:13.034 --> 00:06:15.808
I guess, GPU chips. It could be CPU, but

4ec792e7-4bcc-492d-a042-b2d873afc809/44-0
00:06:14.968 --> 00:06:18.744
Yep, yep, yeah, we do,
but you know everybody can't do that,

4ec792e7-4bcc-492d-a042-b2d873afc809/42-0
00:06:16.208 --> 00:06:16.688
Yeah.

4ec792e7-4bcc-492d-a042-b2d873afc809/44-1
00:06:18.744 --> 00:06:23.387
and that that's that ship has sailed,
so where can we use CPU? Absolutely,

4ec792e7-4bcc-492d-a042-b2d873afc809/43-0
00:06:21.168 --> 00:06:21.688
Mhm.

4ec792e7-4bcc-492d-a042-b2d873afc809/44-2
00:06:23.387 --> 00:06:28.463
but another advantage I think we have is
we've been able to develop our own large

4ec792e7-4bcc-492d-a042-b2d873afc809/44-3
00:06:28.463 --> 00:06:33.168
language model with Eric Orman's work
again in Ultraviolet, which is the...

4ec792e7-4bcc-492d-a042-b2d873afc809/46-0
00:06:33.488 --> 00:06:37.238
a massive model.
I think China and the Emirates have a

4ec792e7-4bcc-492d-a042-b2d873afc809/46-1
00:06:37.238 --> 00:06:41.466
larger model based purely on clinical
documentation. So that,

4ec792e7-4bcc-492d-a042-b2d873afc809/46-2
00:06:41.466 --> 00:06:46.376
that's our own model hosted domestically,
makes iteration much cheaper.

4ec792e7-4bcc-492d-a042-b2d873afc809/46-3
00:06:46.376 --> 00:06:51.968
And there are a lot of open source models,
like we all have to source healthcare.

4ec792e7-4bcc-492d-a042-b2d873afc809/45-0
00:06:47.368 --> 00:06:47.848
Mhm.

4ec792e7-4bcc-492d-a042-b2d873afc809/48-0
00:06:52.008 --> 00:06:55.312
We are nonprofit entities,
so we're gonna have to start looking at

4ec792e7-4bcc-492d-a042-b2d873afc809/47-0
00:06:53.968 --> 00:06:54.128
Mm.

4ec792e7-4bcc-492d-a042-b2d873afc809/48-1
00:06:55.312 --> 00:06:57.827
what can we and where can we look to
other models,

4ec792e7-4bcc-492d-a042-b2d873afc809/48-2
00:06:57.827 --> 00:07:01.526
besides the big and frankly somewhat
expensive, you know, frontier models.

4ec792e7-4bcc-492d-a042-b2d873afc809/48-3
00:07:01.526 --> 00:07:03.992
That said,
we're doing a bunch of things that are

4ec792e7-4bcc-492d-a042-b2d873afc809/48-4
00:07:03.992 --> 00:07:07.395
turning out to be a whole lot cheaper
than we thought they would be.

4ec792e7-4bcc-492d-a042-b2d873afc809/48-5
00:07:07.395 --> 00:07:10.848
The thing that excites me,
one of the things that excite me the most.

4ec792e7-4bcc-492d-a042-b2d873afc809/49-0
00:07:11.328 --> 00:07:13.830
is on,
we have a proof of concept right now

4ec792e7-4bcc-492d-a042-b2d873afc809/49-1
00:07:13.830 --> 00:07:18.608
where we are generating a summary of what
happened to you in the hospital that day.

4ec792e7-4bcc-492d-a042-b2d873afc809/51-0
00:07:19.888 --> 00:07:23.153
So many transactions,
the velocity of transactions that occur

4ec792e7-4bcc-492d-a042-b2d873afc809/51-1
00:07:23.153 --> 00:07:25.787
for a particular patient in the ICU,
for example,

4ec792e7-4bcc-492d-a042-b2d873afc809/50-0
00:07:24.048 --> 00:07:24.528
Mhm.

4ec792e7-4bcc-492d-a042-b2d873afc809/51-2
00:07:25.787 --> 00:07:28.368
they cannot keep track of what happened
to them.

4ec792e7-4bcc-492d-a042-b2d873afc809/51-3
00:07:28.368 --> 00:07:31.897
They cannot keep track of who they saw,
what medications they had,

4ec792e7-4bcc-492d-a042-b2d873afc809/51-4
00:07:31.897 --> 00:07:35.795
what's going to happen to them tomorrow,
what happened to them yesterday.

4ec792e7-4bcc-492d-a042-b2d873afc809/51-5
00:07:35.795 --> 00:07:39.008
So we have this extracted model that is
looking at all this.

4ec792e7-4bcc-492d-a042-b2d873afc809/53-0
00:07:39.088 --> 00:07:42.600
the results that come in,
the documentation at 4:00 today and says,

4ec792e7-4bcc-492d-a042-b2d873afc809/53-1
00:07:42.600 --> 00:07:45.182
here's the summary of what happened to
you today.

4ec792e7-4bcc-492d-a042-b2d873afc809/53-2
00:07:45.182 --> 00:07:48.798
Right now we have a clinician reading
each one and signing off on it,

4ec792e7-4bcc-492d-a042-b2d873afc809/52-0
00:07:48.168 --> 00:07:48.768
Mm-hmm.

4ec792e7-4bcc-492d-a042-b2d873afc809/53-3
00:07:48.798 --> 00:07:51.535
and then releasing it into our NYU Lingo
Health app,

4ec792e7-4bcc-492d-a042-b2d873afc809/53-4
00:07:51.535 --> 00:07:55.409
which goes to the patient's proxies.
So they've got a summary, hey, today,

4ec792e7-4bcc-492d-a042-b2d873afc809/53-5
00:07:55.409 --> 00:07:59.128
this is what happened to you.
You got questions, let's talk about that.

4ec792e7-4bcc-492d-a042-b2d873afc809/56-0
00:07:59.728 --> 00:08:03.379
Dialing into that for a second,
and we have a lot of topics to cover,

4ec792e7-4bcc-492d-a042-b2d873afc809/54-0
00:08:01.408 --> 00:08:01.648
Yeah.

4ec792e7-4bcc-492d-a042-b2d873afc809/56-1
00:08:03.379 --> 00:08:07.342
but this is just really interesting.
You said the clinician reason releases

4ec792e7-4bcc-492d-a042-b2d873afc809/56-2
00:08:07.342 --> 00:08:11.201
each one. This is a really cool example.
I asked you for a cool use case.

4ec792e7-4bcc-492d-a042-b2d873afc809/56-3
00:08:11.201 --> 00:08:13.966
You gave me one.
But AI kind of got popular with the

4ec792e7-4bcc-492d-a042-b2d873afc809/56-4
00:08:13.966 --> 00:08:17.564
promise of saving time and effort and
reducing burden on clinicians.

4ec792e7-4bcc-492d-a042-b2d873afc809/56-5
00:08:17.564 --> 00:08:21.528
Now I've spoken to other executives and
they say, oh, ambient listening has

4ec792e7-4bcc-492d-a042-b2d873afc809/55-0
00:08:18.528 --> 00:08:19.048
Absolutely.

4ec792e7-4bcc-492d-a042-b2d873afc809/57-0
00:08:21.648 --> 00:08:25.543
actually increased provider time in the
EHR, not reduced pajama time.

4ec792e7-4bcc-492d-a042-b2d873afc809/57-1
00:08:25.543 --> 00:08:29.105
I see you shaking your head.
You can speak to that in a moment,

4ec792e7-4bcc-492d-a042-b2d873afc809/57-2
00:08:29.105 --> 00:08:33.557
but more specifically in this use case,
it sounds like you're adding additional

4ec792e7-4bcc-492d-a042-b2d873afc809/57-3
00:08:33.557 --> 00:08:36.006
tasks to the already busy provider's
plate.

4ec792e7-4bcc-492d-a042-b2d873afc809/57-4
00:08:36.006 --> 00:08:39.568
How are they responding and how is that
affecting patient care?

4ec792e7-4bcc-492d-a042-b2d873afc809/59-0
00:08:39.648 --> 00:08:42.045
Absolutely we are,
but that is because we are in a

4ec792e7-4bcc-492d-a042-b2d873afc809/59-1
00:08:42.045 --> 00:08:44.208
validation phase.
This is a proof of concept.

4ec792e7-4bcc-492d-a042-b2d873afc809/58-0
00:08:42.928 --> 00:08:43.328
OK.

4ec792e7-4bcc-492d-a042-b2d873afc809/61-0
00:08:45.248 --> 00:08:48.498
And as I said before,
innovation is not interesting if I can't

4ec792e7-4bcc-492d-a042-b2d873afc809/61-1
00:08:48.498 --> 00:08:51.078
scale it.
So what we're doing is we're looking at

4ec792e7-4bcc-492d-a042-b2d873afc809/61-2
00:08:51.078 --> 00:08:54.587
the delta, the distance and edit,
and we're building three kinds of

4ec792e7-4bcc-492d-a042-b2d873afc809/60-0
00:08:54.048 --> 00:08:54.528
Mhm.

4ec792e7-4bcc-492d-a042-b2d873afc809/61-3
00:08:54.587 --> 00:08:58.869
judgment models that will look at those,
and then we will get the human out of the

4ec792e7-4bcc-492d-a042-b2d873afc809/61-4
00:08:58.869 --> 00:09:02.326
loop. This is summarization.
This is actually one of the more easy

4ec792e7-4bcc-492d-a042-b2d873afc809/61-5
00:09:02.326 --> 00:09:04.648
tasks.
This is extraction and summarization.

4ec792e7-4bcc-492d-a042-b2d873afc809/63-0
00:09:04.928 --> 00:09:08.474
So in the validation phase,
you find good clinical partners who are

4ec792e7-4bcc-492d-a042-b2d873afc809/63-1
00:09:08.474 --> 00:09:11.239
willing to work with you and yet put in
some effort.

4ec792e7-4bcc-492d-a042-b2d873afc809/63-2
00:09:11.239 --> 00:09:15.307
We're going to ask something of them
because they appreciate in a longer run,

4ec792e7-4bcc-492d-a042-b2d873afc809/62-0
00:09:13.808 --> 00:09:14.288
Mhm.

4ec792e7-4bcc-492d-a042-b2d873afc809/63-3
00:09:15.307 --> 00:09:19.323
the patient and clinician experience will
be better when we move to a judge,

4ec792e7-4bcc-492d-a042-b2d873afc809/63-4
00:09:19.323 --> 00:09:22.818
an AI judge-based model that extracts the
clinician from it. I am,

4ec792e7-4bcc-492d-a042-b2d873afc809/63-5
00:09:22.818 --> 00:09:26.208
we rely too much on saying, oh,
but there's a human in the loop.

4ec792e7-4bcc-492d-a042-b2d873afc809/64-0
00:09:26.928 --> 00:09:29.944
We find that that adds bias.
It engenders inequity.

4ec792e7-4bcc-492d-a042-b2d873afc809/64-1
00:09:29.944 --> 00:09:33.540
It will make something less empathic.
It will make something,

4ec792e7-4bcc-492d-a042-b2d873afc809/64-2
00:09:33.540 --> 00:09:38.180
it will often add into the reading level
when we're exactly trying to lower the

4ec792e7-4bcc-492d-a042-b2d873afc809/64-3
00:09:38.180 --> 00:09:42.646
reading level. So I will tell you,
I firmly believe always having a human in

4ec792e7-4bcc-492d-a042-b2d873afc809/64-4
00:09:42.646 --> 00:09:45.488
the loop is not as comforting as we think
it is.

4ec792e7-4bcc-492d-a042-b2d873afc809/65-0
00:09:45.888 --> 00:09:48.905
Yeah,
so you talk about clinical workflows and

4ec792e7-4bcc-492d-a042-b2d873afc809/65-1
00:09:48.905 --> 00:09:53.977
about how we'll be eventually removing
humans from the loop in order to enable

4ec792e7-4bcc-492d-a042-b2d873afc809/65-2
00:09:53.977 --> 00:09:55.968
AI-driven workflows. Fantastic.

4ec792e7-4bcc-492d-a042-b2d873afc809/66-0
00:09:55.248 --> 00:09:58.464
But,
but eventual is like 6 weeks from now,

4ec792e7-4bcc-492d-a042-b2d873afc809/66-1
00:09:58.464 --> 00:09:59.488
not six years.

4ec792e7-4bcc-492d-a042-b2d873afc809/67-0
00:10:13.808 --> 00:10:16.808
Oh, like the world, yeah, yeah, yeah,
yeah, the world's fairs, yeah.

4ec792e7-4bcc-492d-a042-b2d873afc809/69-0
00:10:16.888 --> 00:10:21.451
But there's the other side of the coin.
As AI becomes part of these clinical

4ec792e7-4bcc-492d-a042-b2d873afc809/69-1
00:10:21.451 --> 00:10:24.236
workflows,
an intrinsic inherent part of these

4ec792e7-4bcc-492d-a042-b2d873afc809/69-2
00:10:24.236 --> 00:10:28.680
workflows, what does downtime mean?
And how has that changed how you think

4ec792e7-4bcc-492d-a042-b2d873afc809/69-3
00:10:28.680 --> 00:10:29.688
about resilience?

4ec792e7-4bcc-492d-a042-b2d873afc809/68-0
00:10:30.888 --> 00:10:31.448
Um...

4ec792e7-4bcc-492d-a042-b2d873afc809/72-0
00:10:32.648 --> 00:10:37.908
It is part and parcel of our job to
ensure enterprise resilience, right?

4ec792e7-4bcc-492d-a042-b2d873afc809/72-1
00:10:37.908 --> 00:10:43.673
That's an easy thing to say out loud.
But there are enough proactive monitoring

4ec792e7-4bcc-492d-a042-b2d873afc809/72-2
00:10:43.673 --> 00:10:46.915
tools.
And we also have to acknowledge where

4ec792e7-4bcc-492d-a042-b2d873afc809/72-3
00:10:46.915 --> 00:10:50.879
these tools sit.
So can a daily summary go down and we

4ec792e7-4bcc-492d-a042-b2d873afc809/71-0
00:10:47.528 --> 00:10:48.008
Mhm.

4ec792e7-4bcc-492d-a042-b2d873afc809/72-4
00:10:50.879 --> 00:10:52.248
keep the lights on?

4ec792e7-4bcc-492d-a042-b2d873afc809/74-0
00:10:52.288 --> 00:10:55.640
provide the number one quality care in
the country. Yes,

4ec792e7-4bcc-492d-a042-b2d873afc809/74-1
00:10:55.640 --> 00:11:00.344
these are important systems right now,
but we can all envision a time when they

4ec792e7-4bcc-492d-a042-b2d873afc809/73-0
00:10:59.248 --> 00:10:59.448
Mm.

4ec792e7-4bcc-492d-a042-b2d873afc809/74-2
00:11:00.344 --> 00:11:05.048
make or break the quality care we provide.
And that's why we got to build these

4ec792e7-4bcc-492d-a042-b2d873afc809/74-3
00:11:05.048 --> 00:11:08.988
systems and Hardin them now.
Absolutely agree with you. That said,

4ec792e7-4bcc-492d-a042-b2d873afc809/74-4
00:11:08.988 --> 00:11:12.928
a vast majority of these. So for example,
when we have them in the

4ec792e7-4bcc-492d-a042-b2d873afc809/75-0
00:11:13.928 --> 00:11:17.825
education environment or the research
environment,

4ec792e7-4bcc-492d-a042-b2d873afc809/75-1
00:11:17.825 --> 00:11:24.016
the expectation of resilience and uptime
is frankly lower and our doors can stay

4ec792e7-4bcc-492d-a042-b2d873afc809/75-2
00:11:24.016 --> 00:11:29.060
opened if we have a downtime in a place
that we, in, for example,

4ec792e7-4bcc-492d-a042-b2d873afc809/75-3
00:11:29.060 --> 00:11:32.728
in a researcher education,
we cannot have that.

4ec792e7-4bcc-492d-a042-b2d873afc809/76-0
00:11:32.888 --> 00:11:38.690
In clinical care, and we measure those,
we measure those times in seconds and

4ec792e7-4bcc-492d-a042-b2d873afc809/76-1
00:11:38.690 --> 00:11:41.368
minutes, not hours, for a year over.

4ec792e7-4bcc-492d-a042-b2d873afc809/77-0
00:11:42.168 --> 00:11:47.608
So do you have kind of paper workarounds
for if you have been subjected to a cyber

4ec792e7-4bcc-492d-a042-b2d873afc809/77-1
00:11:47.608 --> 00:11:52.918
attack or is it more like an independent
recovery environment? Or is it, I mean,

4ec792e7-4bcc-492d-a042-b2d873afc809/77-2
00:11:52.918 --> 00:11:55.933
I know the answer is always all of the
above,

4ec792e7-4bcc-492d-a042-b2d873afc809/77-3
00:11:55.933 --> 00:12:01.046
but can you elaborate on what hardening
your system to ensure resilience in a

4ec792e7-4bcc-492d-a042-b2d873afc809/77-4
00:12:01.046 --> 00:12:02.488
clinical care delivery

4ec792e7-4bcc-492d-a042-b2d873afc809/78-0
00:12:02.728 --> 00:12:04.008
Setting looks like.

4ec792e7-4bcc-492d-a042-b2d873afc809/80-0
00:12:02.848 --> 00:12:06.249
Yeah.
So an IRE is absolutely core to our

4ec792e7-4bcc-492d-a042-b2d873afc809/80-1
00:12:06.249 --> 00:12:10.299
strategies.
But let's also acknowledge that we've

4ec792e7-4bcc-492d-a042-b2d873afc809/80-2
00:12:10.299 --> 00:12:15.806
been digital for 15 years,
and we know how to thoughtfully take the

4ec792e7-4bcc-492d-a042-b2d873afc809/79-0
00:12:15.048 --> 00:12:15.528
Mhm.

4ec792e7-4bcc-492d-a042-b2d873afc809/80-3
00:12:15.806 --> 00:12:20.584
system down and bring it back up in a
matter of, you know,

4ec792e7-4bcc-492d-a042-b2d873afc809/80-4
00:12:20.584 --> 00:12:22.528
it used to take 6 hours.

4ec792e7-4bcc-492d-a042-b2d873afc809/81-0
00:12:22.808 --> 00:12:26.648
and now takes 35 minutes.
But those 35 minutes,

4ec792e7-4bcc-492d-a042-b2d873afc809/81-1
00:12:26.648 --> 00:12:32.488
there are people who are much more
informed than I that can speak to the

4ec792e7-4bcc-492d-a042-b2d873afc809/81-2
00:12:32.488 --> 00:12:36.728
technical aspects.
I will tell you from the bedside.

4ec792e7-4bcc-492d-a042-b2d873afc809/82-0
00:12:37.608 --> 00:12:41.843
I need a clinical population.
I need peers of nurses and physicians

4ec792e7-4bcc-492d-a042-b2d873afc809/82-1
00:12:41.843 --> 00:12:44.708
that say to me,
I know why you're doing this.

4ec792e7-4bcc-492d-a042-b2d873afc809/82-2
00:12:44.708 --> 00:12:49.752
And it's not just to make our night on a
Saturday night in June worse. That, oh,

4ec792e7-4bcc-492d-a042-b2d873afc809/82-3
00:12:49.752 --> 00:12:55.046
good things are coming out of this change.
And that's actually an important cultural

4ec792e7-4bcc-492d-a042-b2d873afc809/82-4
00:12:55.046 --> 00:12:56.728
thing that they don't view.

4ec792e7-4bcc-492d-a042-b2d873afc809/83-0
00:12:56.768 --> 00:13:01.148
change as a bad thing,
but that we knew about it, it comes down,

4ec792e7-4bcc-492d-a042-b2d873afc809/83-1
00:13:01.148 --> 00:13:05.664
and when we didn't know about it,
and it's an unexpected downtime,

4ec792e7-4bcc-492d-a042-b2d873afc809/83-2
00:13:05.664 --> 00:13:10.786
which fortunately are quite rare,
we rally the same way we would rally with

4ec792e7-4bcc-492d-a042-b2d873afc809/83-3
00:13:10.786 --> 00:13:14.088
the same playbook at a threat to the
enterprise.

4ec792e7-4bcc-492d-a042-b2d873afc809/85-0
00:13:14.488 --> 00:13:17.645
Just to be clear,
and I think everyone listening knows the

4ec792e7-4bcc-492d-a042-b2d873afc809/85-1
00:13:17.645 --> 00:13:20.749
answer to this,
but do you have the equivalent of a third

4ec792e7-4bcc-492d-a042-b2d873afc809/85-2
00:13:20.749 --> 00:13:23.157
grade elementary school fire drill?
Meaning,

4ec792e7-4bcc-492d-a042-b2d873afc809/85-3
00:13:23.157 --> 00:13:25.566
do you guys do have these planned
downtimes?

4ec792e7-4bcc-492d-a042-b2d873afc809/85-4
00:13:25.566 --> 00:13:29.847
And you're saying that these nurses and
physicians now understand why they have

4ec792e7-4bcc-492d-a042-b2d873afc809/85-5
00:13:29.847 --> 00:13:33.968
the inconvenience of a planned downtime.
That's what you're saying, correct?

4ec792e7-4bcc-492d-a042-b2d873afc809/84-0
00:13:34.008 --> 00:13:35.928
Yes, thank you for taking that, yes.

4ec792e7-4bcc-492d-a042-b2d873afc809/86-0
00:13:35.048 --> 00:13:35.288
No.

4ec792e7-4bcc-492d-a042-b2d873afc809/87-0
00:13:36.968 --> 00:13:37.768
OK, so...

4ec792e7-4bcc-492d-a042-b2d873afc809/88-0
00:13:38.968 --> 00:13:41.825
All right,
so AI is helping facilitate care.

4ec792e7-4bcc-492d-a042-b2d873afc809/88-1
00:13:41.825 --> 00:13:46.142
We have plans for when there is downtime,
hardening our resiliency.

4ec792e7-4bcc-492d-a042-b2d873afc809/88-2
00:13:46.142 --> 00:13:51.284
But how has AI changed your threat model?
Do you ever feel like you benefit from

4ec792e7-4bcc-492d-a042-b2d873afc809/88-3
00:13:51.284 --> 00:13:55.728
outside expertise? Obviously,
AI is generating more frequent attacks,

4ec792e7-4bcc-492d-a042-b2d873afc809/88-4
00:13:55.728 --> 00:14:00.109
sometimes novel kinds of attacks,
including impersonation of CEOs of

4ec792e7-4bcc-492d-a042-b2d873afc809/88-5
00:14:00.109 --> 00:14:03.728
vendors who you think you're talking to,
but you're not.

4ec792e7-4bcc-492d-a042-b2d873afc809/89-0
00:14:04.808 --> 00:14:06.808
You know,
how has AI changed your threat model?

4ec792e7-4bcc-492d-a042-b2d873afc809/91-0
00:14:08.968 --> 00:14:11.890
Lots of secret sauce,
but we're all facing it,

4ec792e7-4bcc-492d-a042-b2d873afc809/91-1
00:14:11.890 --> 00:14:16.428
so it's important to be transparent with
each other in the right venues.

4ec792e7-4bcc-492d-a042-b2d873afc809/91-2
00:14:16.428 --> 00:14:21.028
I am learning more from our CISO,
our Chief Information Security Officer,

4ec792e7-4bcc-492d-a042-b2d873afc809/91-3
00:14:21.028 --> 00:14:25.007
in the last year than I have,
probably in 15 years in this job.

4ec792e7-4bcc-492d-a042-b2d873afc809/90-0
00:14:23.368 --> 00:14:23.528
Mm.

4ec792e7-4bcc-492d-a042-b2d873afc809/91-4
00:14:25.007 --> 00:14:28.488
And he's an important partner to me.
I would say we are

4ec792e7-4bcc-492d-a042-b2d873afc809/92-0
00:14:28.728 --> 00:14:30.728
You know, we are finding...

4ec792e7-4bcc-492d-a042-b2d873afc809/93-0
00:14:31.888 --> 00:14:36.519
vectors of attack that have made us close
doors regularly. And it's,

4ec792e7-4bcc-492d-a042-b2d873afc809/93-1
00:14:36.519 --> 00:14:41.888
they're getting really creative and we
have to ensure that we're using the same

4ec792e7-4bcc-492d-a042-b2d873afc809/93-2
00:14:41.888 --> 00:14:43.768
models and the same weapons.

4ec792e7-4bcc-492d-a042-b2d873afc809/94-0
00:14:44.728 --> 00:14:46.088
to prevent these attacks.

4ec792e7-4bcc-492d-a042-b2d873afc809/95-0
00:14:46.368 --> 00:14:46.888
Mmh.

4ec792e7-4bcc-492d-a042-b2d873afc809/97-0
00:14:47.688 --> 00:14:52.955
So it's kind of like in the Cold War.
I have 15 nukes, I have 20 nukes,

4ec792e7-4bcc-492d-a042-b2d873afc809/97-1
00:14:52.955 --> 00:14:56.248
I have 25 nukes.
It's an arms race in a way.

4ec792e7-4bcc-492d-a042-b2d873afc809/96-0
00:14:56.648 --> 00:14:57.368
In a way it is.

4ec792e7-4bcc-492d-a042-b2d873afc809/98-0
00:14:58.328 --> 00:15:01.386
As AI usage grows,
how do you decide what runs in the cloud

4ec792e7-4bcc-492d-a042-b2d873afc809/98-1
00:15:01.386 --> 00:15:04.648
versus on-prem and how are you keeping
those costs predictable?

4ec792e7-4bcc-492d-a042-b2d873afc809/99-0
00:15:05.848 --> 00:15:06.728
Ohh.

4ec792e7-4bcc-492d-a042-b2d873afc809/100-0
00:15:09.688 --> 00:15:13.641
Great, great question.
My pause is I don't know if I have the

4ec792e7-4bcc-492d-a042-b2d873afc809/100-1
00:15:13.641 --> 00:15:18.488
firm answer yet and we are learning.
I think FinOps is an entire field unto

4ec792e7-4bcc-492d-a042-b2d873afc809/100-2
00:15:18.488 --> 00:15:22.888
itself of experts who we have employed in
that situation. That said.

4ec792e7-4bcc-492d-a042-b2d873afc809/101-0
00:15:25.528 --> 00:15:30.599
The ability to remain flexible and
thoughtful in how, for example,

4ec792e7-4bcc-492d-a042-b2d873afc809/101-1
00:15:30.599 --> 00:15:34.686
if we have 40 million images,
clinical images stored,

4ec792e7-4bcc-492d-a042-b2d873afc809/101-2
00:15:34.686 --> 00:15:40.816
and how do we keep them in deep storage,
but still have some latency acceptable?

4ec792e7-4bcc-492d-a042-b2d873afc809/101-3
00:15:40.816 --> 00:15:43.768
Or as we move to all digital pathology,

4ec792e7-4bcc-492d-a042-b2d873afc809/103-0
00:15:44.088 --> 00:15:48.150
We have millions of pathology images.
The reality is we don't need

4ec792e7-4bcc-492d-a042-b2d873afc809/103-1
00:15:48.150 --> 00:15:52.515
instantaneous access to those.
So we have to engage with the vendors to

4ec792e7-4bcc-492d-a042-b2d873afc809/102-0
00:15:48.688 --> 00:15:49.168
Mhm.

4ec792e7-4bcc-492d-a042-b2d873afc809/103-2
00:15:52.515 --> 00:15:57.244
talk about what does it mean for deep
storage and what do we get to put there

4ec792e7-4bcc-492d-a042-b2d873afc809/103-3
00:15:57.244 --> 00:16:00.943
and not always be extracting and then
store locally as well.

4ec792e7-4bcc-492d-a042-b2d873afc809/103-4
00:16:00.943 --> 00:16:03.368
It's an absolute hybrid model. And we've

4ec792e7-4bcc-492d-a042-b2d873afc809/104-0
00:16:03.448 --> 00:16:08.412
Been, I think,
fortunate to not commit too early to a

4ec792e7-4bcc-492d-a042-b2d873afc809/104-1
00:16:08.412 --> 00:16:14.111
hosted environment,
but really maintain a hybridized model of

4ec792e7-4bcc-492d-a042-b2d873afc809/104-2
00:16:14.111 --> 00:16:18.248
on-prem and all three large hosting
vendors.

4ec792e7-4bcc-492d-a042-b2d873afc809/105-0
00:16:17.768 --> 00:16:20.931
Got it.
You mentioned FinOps is a field of

4ec792e7-4bcc-492d-a042-b2d873afc809/105-1
00:16:20.931 --> 00:16:25.198
experts, and of course,
everything is a field of experts.

4ec792e7-4bcc-492d-a042-b2d873afc809/105-2
00:16:25.198 --> 00:16:30.200
What is hardest to hire for right now?
There's a lot of AI experts,

4ec792e7-4bcc-492d-a042-b2d873afc809/105-3
00:16:30.200 --> 00:16:34.393
a lot of demand,
probably more demand than supply of the

4ec792e7-4bcc-492d-a042-b2d873afc809/105-4
00:16:34.393 --> 00:16:36.968
best kind of experts that you need.

4ec792e7-4bcc-492d-a042-b2d873afc809/106-0
00:16:37.288 --> 00:16:41.975
for anything from cybersecurity to
developing AI models to doing your FinOps,

4ec792e7-4bcc-492d-a042-b2d873afc809/106-1
00:16:41.975 --> 00:16:45.160
et cetera.
How do you decide what to build in-house,

4ec792e7-4bcc-492d-a042-b2d873afc809/106-2
00:16:45.160 --> 00:16:49.848
what expertise to build institutional
knowledge in versus what to partner on?

4ec792e7-4bcc-492d-a042-b2d873afc809/107-0
00:16:50.208 --> 00:16:52.774
Yeah,
there's actually probably a couple of

4ec792e7-4bcc-492d-a042-b2d873afc809/107-1
00:16:52.774 --> 00:16:54.408
questions in there, I think.

4ec792e7-4bcc-492d-a042-b2d873afc809/108-0
00:16:57.288 --> 00:17:00.377
I,
we're always doing the calculus of what

4ec792e7-4bcc-492d-a042-b2d873afc809/108-1
00:17:00.377 --> 00:17:05.406
to build, what to buy, what to partner on.
There are, I think we are,

4ec792e7-4bcc-492d-a042-b2d873afc809/108-2
00:17:05.406 --> 00:17:11.441
there are certain things that we've come
to realize are simply commodities in this,

4ec792e7-4bcc-492d-a042-b2d873afc809/108-3
00:17:11.441 --> 00:17:15.608
and we have enough teams that we can
build our own tools.

4ec792e7-4bcc-492d-a042-b2d873afc809/109-0
00:17:16.088 --> 00:17:18.408
that there's just a level of...

4ec792e7-4bcc-492d-a042-b2d873afc809/110-0
00:17:19.688 --> 00:17:23.377
inappropriate pricing.
We're not a large healthcare,

4ec792e7-4bcc-492d-a042-b2d873afc809/110-1
00:17:23.377 --> 00:17:28.319
we're not a large financial institution.
We're a nonprofit healthcare.

4ec792e7-4bcc-492d-a042-b2d873afc809/110-2
00:17:28.319 --> 00:17:32.426
And we have to be incredibly responsible
with that margin.

4ec792e7-4bcc-492d-a042-b2d873afc809/110-3
00:17:32.426 --> 00:17:37.647
To do that means building things locally,
but also an incredibly important

4ec792e7-4bcc-492d-a042-b2d873afc809/110-4
00:17:37.647 --> 00:17:41.128
strategic relationship with our EHR
vendor, Epic.

4ec792e7-4bcc-492d-a042-b2d873afc809/112-0
00:17:41.608 --> 00:17:45.290
and they are responsive to things we need.
And they,

4ec792e7-4bcc-492d-a042-b2d873afc809/112-1
00:17:45.290 --> 00:17:49.945
but why are they responsive to things we
need? Why do they listen?

4ec792e7-4bcc-492d-a042-b2d873afc809/112-2
00:17:49.945 --> 00:17:54.738
Because we will deploy at scale.
I can't have five different dietary

4ec792e7-4bcc-492d-a042-b2d873afc809/112-3
00:17:54.738 --> 00:17:59.324
management tools. I need one.
I can't have 5 instances of my EHR.

4ec792e7-4bcc-492d-a042-b2d873afc809/111-0
00:17:57.128 --> 00:17:57.608
Mhm.

4ec792e7-4bcc-492d-a042-b2d873afc809/112-4
00:17:59.324 --> 00:18:00.088
I need one.

4ec792e7-4bcc-492d-a042-b2d873afc809/113-0
00:18:00.408 --> 00:18:03.791
If the same goes across all from our
tripartite mission,

4ec792e7-4bcc-492d-a042-b2d873afc809/113-1
00:18:03.791 --> 00:18:07.708
it goes across all three environments of
the education, research,

4ec792e7-4bcc-492d-a042-b2d873afc809/113-2
00:18:07.708 --> 00:18:12.338
and clinical that we should be based,
we should use to the fullest that which

4ec792e7-4bcc-492d-a042-b2d873afc809/113-3
00:18:12.338 --> 00:18:15.781
we are already paying for.
And when we are at the margin,

4ec792e7-4bcc-492d-a042-b2d873afc809/113-4
00:18:15.781 --> 00:18:18.808
think about whether we need to build it
ourselves.

4ec792e7-4bcc-492d-a042-b2d873afc809/114-0
00:18:19.008 --> 00:18:24.129
which is increasingly becoming viable,
or partner with either upstarts,

4ec792e7-4bcc-492d-a042-b2d873afc809/114-1
00:18:24.129 --> 00:18:28.752
we find a lot of those.
Those are the back hauls at HIMS and the

4ec792e7-4bcc-492d-a042-b2d873afc809/114-2
00:18:28.752 --> 00:18:34.442
smaller booths. We find partners early.
And if they're willing to try to scale,

4ec792e7-4bcc-492d-a042-b2d873afc809/114-3
00:18:34.442 --> 00:18:39.208
meaning deploy everywhere for us,
then it's a good partner for us.

4ec792e7-4bcc-492d-a042-b2d873afc809/115-0
00:18:40.568 --> 00:18:45.317
So I just want to ask a fun question
because we're approaching the end of this

4ec792e7-4bcc-492d-a042-b2d873afc809/115-1
00:18:45.317 --> 00:18:47.962
podcast episode.
You mentioned Ultraviolet,

4ec792e7-4bcc-492d-a042-b2d873afc809/115-2
00:18:47.962 --> 00:18:51.208
which I believe is the name of NYU
Langone's own LLM.

4ec792e7-4bcc-492d-a042-b2d873afc809/115-3
00:18:51.208 --> 00:18:55.957
And then you mentioned something that is
quite obvious to every listener here,

4ec792e7-4bcc-492d-a042-b2d873afc809/115-4
00:18:55.957 --> 00:18:59.384
which is your nonprofit health system
with slim margins.

4ec792e7-4bcc-492d-a042-b2d873afc809/115-5
00:18:59.384 --> 00:19:01.368
Everybody can identify with that.

4ec792e7-4bcc-492d-a042-b2d873afc809/116-0
00:19:01.808 --> 00:19:06.407
So something I think that's interesting
is that different health systems across

4ec792e7-4bcc-492d-a042-b2d873afc809/116-1
00:19:06.407 --> 00:19:10.087
the country are saying,
maybe we're not only in the business of

4ec792e7-4bcc-492d-a042-b2d873afc809/116-2
00:19:10.087 --> 00:19:13.881
providing clinical care.
Given your background in business and JD

4ec792e7-4bcc-492d-a042-b2d873afc809/116-3
00:19:13.881 --> 00:19:18.423
and your MPH, I said, maybe, you know,
you're bringing something to the table,

4ec792e7-4bcc-492d-a042-b2d873afc809/116-4
00:19:18.423 --> 00:19:21.068
which is maybe there are new revenue
streams.

4ec792e7-4bcc-492d-a042-b2d873afc809/116-5
00:19:21.068 --> 00:19:23.368
Maybe we can also be a software company.

4ec792e7-4bcc-492d-a042-b2d873afc809/117-0
00:19:23.568 --> 00:19:29.705
Maybe we can license out Ultraviolet to
other institutions that don't own their

4ec792e7-4bcc-492d-a042-b2d873afc809/117-1
00:19:29.705 --> 00:19:34.078
own GPU stack in their data center from
pre-Gen AI days.

4ec792e7-4bcc-492d-a042-b2d873afc809/117-2
00:19:34.078 --> 00:19:40.368
And maybe we can take these resources and
we can be the licenser and allow others

4ec792e7-4bcc-492d-a042-b2d873afc809/117-3
00:19:40.368 --> 00:19:45.048
to buy from us. And now we can have 30,
40% software margins

4ec792e7-4bcc-492d-a042-b2d873afc809/118-0
00:19:45.328 --> 00:19:49.577
that helps enable us to do our care
delivery mission. What are your thoughts?

4ec792e7-4bcc-492d-a042-b2d873afc809/118-1
00:19:49.577 --> 00:19:52.138
And it's just kind of a fun,
oddball question.

4ec792e7-4bcc-492d-a042-b2d873afc809/118-2
00:19:52.138 --> 00:19:55.842
What are your thoughts on NYU being more
than just a care provider,

4ec792e7-4bcc-492d-a042-b2d873afc809/118-3
00:19:55.842 --> 00:19:57.368
but also a software company?

4ec792e7-4bcc-492d-a042-b2d873afc809/120-0
00:19:59.128 --> 00:20:03.447
I have very strong thoughts about that.
And this comes up not infrequently

4ec792e7-4bcc-492d-a042-b2d873afc809/120-1
00:20:03.447 --> 00:20:07.996
because I think we're good at what we do.
But what we do is provide number one

4ec792e7-4bcc-492d-a042-b2d873afc809/120-2
00:20:07.996 --> 00:20:11.624
quality care and the safest care of the
health in the country.

4ec792e7-4bcc-492d-a042-b2d873afc809/120-3
00:20:11.624 --> 00:20:14.561
That's Visient ranking and that's
important to us.

4ec792e7-4bcc-492d-a042-b2d873afc809/119-0
00:20:13.208 --> 00:20:13.368
Mm.

4ec792e7-4bcc-492d-a042-b2d873afc809/120-4
00:20:14.561 --> 00:20:19.168
I think there are health systems that are
able to or wish to live in that space

4ec792e7-4bcc-492d-a042-b2d873afc809/121-0
00:20:19.848 --> 00:20:24.400
of being a software shop and maybe a SAS
shop as well.

4ec792e7-4bcc-492d-a042-b2d873afc809/121-1
00:20:24.400 --> 00:20:28.704
I know my mission is to provide high
quality, safe,

4ec792e7-4bcc-492d-a042-b2d873afc809/121-2
00:20:28.704 --> 00:20:34.332
high patient experience care.
So I want to be candid that I feel at

4ec792e7-4bcc-492d-a042-b2d873afc809/121-3
00:20:34.332 --> 00:20:40.208
times that can be a little bit of a
distraction from the core mission.

4ec792e7-4bcc-492d-a042-b2d873afc809/123-0
00:20:40.688 --> 00:20:45.557
We have to generate the next generation
of clinicians, of nurses and doctors.

4ec792e7-4bcc-492d-a042-b2d873afc809/123-1
00:20:45.557 --> 00:20:50.551
We got to do cutting edge research that
will change the way we provide the care

4ec792e7-4bcc-492d-a042-b2d873afc809/123-2
00:20:50.551 --> 00:20:55.607
that we provide and do that in a learning
health system loop where we learn from

4ec792e7-4bcc-492d-a042-b2d873afc809/123-3
00:20:55.607 --> 00:20:59.103
every patient.
So I think we're happy to engage when we

4ec792e7-4bcc-492d-a042-b2d873afc809/123-4
00:20:59.103 --> 00:21:01.288
can in that sort of co-development.

4ec792e7-4bcc-492d-a042-b2d873afc809/124-0
00:21:01.808 --> 00:21:03.977
But that's not where our secret sauce
lies.

4ec792e7-4bcc-492d-a042-b2d873afc809/124-1
00:21:03.977 --> 00:21:06.048
High quality care with a great experience.

4ec792e7-4bcc-492d-a042-b2d873afc809/125-0
00:21:07.168 --> 00:21:10.358
So Paul,
I'd like to turn it over to you with the

4ec792e7-4bcc-492d-a042-b2d873afc809/125-1
00:21:10.358 --> 00:21:15.206
final question of this interview.
What advice would you give to yourself or

4ec792e7-4bcc-492d-a042-b2d873afc809/125-2
00:21:15.206 --> 00:21:18.843
anybody listening,
but yourself maybe a year or two ago,

4ec792e7-4bcc-492d-a042-b2d873afc809/125-3
00:21:18.843 --> 00:21:23.628
or anybody else listening across the
country who's a little bit earlier in

4ec792e7-4bcc-492d-a042-b2d873afc809/125-4
00:21:23.628 --> 00:21:27.328
their AI data center journey than clearly
NYU Langone is?

4ec792e7-4bcc-492d-a042-b2d873afc809/126-0
00:21:32.928 --> 00:21:34.048
Take the bet.

4ec792e7-4bcc-492d-a042-b2d873afc809/127-0
00:21:35.328 --> 00:21:41.185
that this is not about making yourself
ready for some pivot down the road,

4ec792e7-4bcc-492d-a042-b2d873afc809/127-1
00:21:41.185 --> 00:21:47.433
but commit early and be able to scale.
And I really want to emphasize that last

4ec792e7-4bcc-492d-a042-b2d873afc809/127-2
00:21:47.433 --> 00:21:53.290
part. There are times we, not just me,
but more senior than me folks here,

4ec792e7-4bcc-492d-a042-b2d873afc809/127-3
00:21:53.290 --> 00:21:58.288
we've had to make that decision of saying
no to certain things.

4ec792e7-4bcc-492d-a042-b2d873afc809/128-0
00:21:59.048 --> 00:22:03.467
where, because it won't scale.
And I want to reiterate what we opened

4ec792e7-4bcc-492d-a042-b2d873afc809/128-1
00:22:03.467 --> 00:22:08.329
with. Innovation that doesn't scale,
using things we have already paying for

4ec792e7-4bcc-492d-a042-b2d873afc809/128-2
00:22:08.329 --> 00:22:13.317
at scale is the fastest way for us to
protect our margin and to be responsible

4ec792e7-4bcc-492d-a042-b2d873afc809/128-3
00:22:13.317 --> 00:22:16.979
with our resources.
So use with what we're paying for and

4ec792e7-4bcc-492d-a042-b2d873afc809/128-4
00:22:16.979 --> 00:22:18.368
stop paying for things

4ec792e7-4bcc-492d-a042-b2d873afc809/129-0
00:22:18.608 --> 00:22:19.408
Three times over.

4ec792e7-4bcc-492d-a042-b2d873afc809/131-0
00:22:20.688 --> 00:22:25.104
I appreciate all of your insights.
We've covered a lot of ground today.

4ec792e7-4bcc-492d-a042-b2d873afc809/131-1
00:22:25.104 --> 00:22:29.642
For our listeners, this has been Dr.
Paul Testa of the NYU Langone Health

4ec792e7-4bcc-492d-a042-b2d873afc809/131-2
00:22:29.642 --> 00:22:33.568
System, the CHIO. Paul,
thank you so much for joining us today.

4ec792e7-4bcc-492d-a042-b2d873afc809/130-0
00:22:33.688 --> 00:22:34.528
Thank you so much.