October 2, 2025 • 62 mins
We test where AI can actually help veterans build stronger Nexus letters and where it fails hard, from fake citations to the wrong legal phrasing. Bethanie Spangenberg shares practical prompts, research tactics, and quality standards that keep letters credible and readable for VA raters.
• defining what a strong Nexus letter must include
• “at least as likely as not” vs malpractice language
• writing for raters with clear, low-jargon explanations
• how AI helps: summaries, translation, structure, prompts
• where AI fails: hallucinated sources, generic templates
• verifying research, citations, and URLs before use
• handling obesity and other medical risk factors
• statements to fill long gaps in treatment history
• research hierarchy: systematic reviews to cohort studies
• privacy cautions when using public AI tools
• internal workflows, grammar tools, and quality control
• actionable prompts to find relevant medical literature
Tune in live every Thursday at 7 PM EST and join the conversation! Click here to listen and chat with us.
Visit J Basser's Exposed Vet Productions (Formerly Exposed Vet Radioshow) YouTube page by clicking here.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
J Basser (00:07):
Welcome ladies and gentlemen to another episode of
the Exposed Vet Jay BassProductions.
Today is the I guess we sayit's the second day of October
2025.
The year has gone by atlightning speed.
Today our co-host is Mr.
(00:28):
Barry Furley.
He's down in Tennesseesomewhere.
How are you doing, Barry?
I'm hanging in there like aloose tooth today.
Oh, you had to bring that upagain, didn't you?
Anyhow.
Today we've got uh BethanySpangenberg.
She is the owner of the companycalled Fowler for Vet.
(00:48):
Uh Bowler for Vet is a companythat does independent medical
opinions, independent medicalevaluations for VA disability
issues, among other things.
And today we're going todiscuss using artificial
intelligence.
Mainly for Nexus Letters, thenwe'll get into some side
discussion too as to what'sgoing on with this stuff.
(01:09):
And uh Bethany, how are youdoing?
Bethanie Spangenberg (01:12):
I'm doing well.
Staying busy, ready for thetransition into fall, and just
staying keeping like trying torun around, keeping schedules
straight with the kids.
So a lot a lot of moving parts.
J Basser (01:27):
Well, life does have a lot of moving parts.
It does.
But you were talking aboutusing AI for Nexus Letters and
things like that for um, I guessyou do some stuff for uh about
vet, but uh I know it's use it'sgetting more and more popular.
Bethanie Spangenberg (01:48):
It is, and that's actually I had a um
conversation recently withanother business owner, and they
were talking about AI, andthat's kind of what prompted,
you know, the idea of havingthis, especially because the
same same week I saw a questionon a forum about using AI for
(02:08):
Nexus letters, and I'm like, youknow, we really need to have
that discussion.
They've talked about it at Novaa couple sessions, um, brought
a lot of understanding for legaluse.
Uh, and so I I really feel likethere's a gap there that we
need to fill when it comes tothe medical expert side for
Nexus letters and AI.
J Basser (02:30):
I can see that because I still noticed there's a lot
of um, there's one companythat's based their whole
business model is is using AI.
You've probably seen that.
Bethanie Spangenberg (02:40):
Yeah, I have.
Um that kind of worries me alittle bit because of my
experience with AI and thethings that I've seen and the
products that I see it put out.
Um, you know, AI has beenfabulous, and I'm actually
really impressed with where it'sat, but I don't think it's
targeted for VA disabilityclaims just yet.
(03:01):
I think we'll they'll getthere, but you know, there's
some things I feel that AI willnever replace.
Um, but I think there is aplace for it when it comes to
veteran disability and maybeeven in the use of Nexus
letters.
So for this presentation, Iactually wanted to see if you
(03:25):
use AI, the really the skillthat really comes in is learning
how to use that particular typeof AI.
You need to know how to writeyour prompts.
And your prompts will make orbreak the output of what you're
getting.
And so what I tried to do is Iactually had AI create the
(03:49):
content for this presentationtoday.
So when we go through thispresentation, I'm going to talk
about some things where yes, Iagree, no, I don't agree, and
then even point out some thingsthat maybe AI made mistakes on,
and that will be an example forthe limitations with AI.
So just keep that in mind as wego through here.
(04:13):
Uh we are actually seeingveterans give us an example of a
Nexus letter as like toreference or to use as evidence.
So they'll go into, I'm goingto talk mainly about Chat GPT
because it's the most commonright now.
Um, but I see veterans goinginto ChatGPT, they'll create
(04:37):
what they think is the rightNexus letter, they'll drop it
into their Valor for Vet folder,evidence folder for us to look
at as medical experts.
And right off, I know it's AI,I know where they got it from,
and it doesn't quite hit all thelegal points that I would have
hit as a medical expert doingthis for the last 14, almost 15
(04:58):
years.
And so again, that's that'swhat emphasizes the importance
of talking about this today.
So again, these slides arecreated, the content is created
by AI.
And you can plug the contentinto your slides, but even
PowerPoint will create digitalum presentations that are that
(05:23):
are like pleasing.
So for example, this circle onthe edge of my presentation,
this setup, I actually had atype of AI help me that is
plugged into PowerPoint tocreate this slide.
So even your your standardpresentation software is using
(05:45):
AI, your Photoshop uses AI,Grammarly is a type of AI, and
they're expanding.
Our phone lines that we use atValor for Vet are AI, and I'll
talk a little bit about thatlater on in the presentation.
But what is a Nexus letter andwhy AI?
So Nexus Letter is veryimportant for veteran disability
(06:05):
claims.
It takes that current serviceconnection, excuse me, that
current disability and tries torelate it to service.
So it's the part that connectsthese two items.
And majority of veterandisability claims need a Nexus
letter of some sort or a Nexusstatement with a supporting
rationale in order for a veteranto get service connected.
The challenge has been, and ithas always been, getting a Nexus
(06:32):
letter from a qualifiedindividual where the quality
meets the VA standards and it'sat a reasonable price.
Doing this for as long as wehave, we know that majority of
the cost associated with theNexus letter is coming from that
veteran directly.
So if a veteran is working withan attorney, the attorney may
(06:54):
upfront that fee, and then theveteran has to pay that fee on
the back end if they'resuccessful.
So we know at the end of theday, our market needs to target
the veteran's pocket.
And if you look at $1,500 forone Nexus letter, you have to
really weigh, okay, is $1,500worth me getting a Nexus letter
for an ankle condition, orshould I use that for something
(07:17):
like sleep apnea where it mayhave a longer term or a better
benefit for the cost?
And so a lot of the doctorsthat practice medicine, unless
you've been taught by the VA,unless you were taught by
somebody else, the skill ofwriting a Nexus letter is also a
niche, and they're they're hardto find someone who's
(07:39):
experienced enough to put outthe quality that is needed.
So if we use tools likeChatGPT, is it to the rescue, as
this presentation suggests?
So, yes, it can draft Nexusletters quickly, but there are
certain limitations when itcomes to some of the language,
(08:01):
the medical research, the legallanguage, the at least as likely
as not, the more likely thannot kind of phrase.
And then the big question is iscan AI help experts create
strong letters?
The VA will accept.
And so we can, but I don'tthink that that's I don't think
(08:24):
it's a one and done.
I don't think we plug it intoChat GPT and it spits everything
out.
I think we have to one, knowour prompts, and then two, we
have to have the clinicalknowledge to really put it back
like into like a Nexus letterthat flows.
So there's a lot of challengesin that.
J Basser (08:43):
I just hit the nail on the head because you know, if
you're sitting here and you're awriter and all of a sudden you
get a claim, and you get thisstatement, a personal statement
or whatever from this veteranwho's got a tenth grade
education.
Okay.
But he sounds like aPhiladelphia lawyer with uh a
PhD.
Yes.
(09:04):
It makes a big difference.
Bethanie Spangenberg (09:06):
And you know, the other thing is too, is
you know, because you broughtit up, yeah, it's later in my
presentation, but because youbrought it up, I'm gonna go
ahead and talk about it.
There is there are times whenour providers will write a Nexus
letter and they are usingmedical jargon.
And if you think about it, ouraudience is not medical
(09:30):
professionals, our audience isthe raiders or the people
deciding claims at the VA.
And somewhere, I can't rememberthe statistic, it's like more
than 50% of the raiders at theVA have a high school education.
And so, you know, what audienceare you writing this Nexus
letter for?
So if I have a Nexus letterwhere my provider is talking
(09:55):
about a left shift when it comesto white blood cells, they're
not going to immediately knowwhat a left shift is.
And so we use Grammarly to helpus kind of dumb it down in a
layman's language.
Um so then that way there isunderstanding of what what
language or what explanation andrationale that we're trying to
(10:17):
support.
You know, you can get into alot of these numbers and spit
out these numbers for labcounts, kidney count, like your
kidney function, um infectionnumbers, type things like that.
But to a layman, they may notunderstand it.
Well, with Grammarly, we cantake that information and we can
(10:38):
kind of put in that that laymanum explanation so we can really
talk to the reader that isreviewing that Nexus letter.
J Basser (10:47):
Yep, simplicity.
Yeah.
Bethanie Spangenberg (10:51):
So your audience is who you're writing
to.
Yes, you're right.
If a state a veteran who has aneighth grade education is
writing this really detailedstatement, you're like, okay,
this this isn't this is not.
And it's actually very like ifpeople are paying attention,
it's actually very easy to seethat.
So all right, let's move on tothe next slide.
J Basser (11:15):
Okay, go ahead.
Bethanie Spangenberg (11:18):
All right, maybe and go.
Okay, there we go.
So these are the top questionsAI found about Nexus Letters and
AI.
Will the VA accept it?
So content and a doctor'ssignature will matter most.
(11:40):
Absolutely.
So a VA will accept it if thecontent is appropriate to your
claim.
It discusses the veteran'smedical information, the
rationale supports the veteran'sclaim, including the details of
the medical history, and it'swritten or signed by a PA, DO,
(12:05):
nurse practitioner, physician, aPhD, uh, a psychologist with a
PhD.
So, yes, AI can write it andthen the doctor can sign it.
I just yesterday had a VSObring one to me, and they're
like, hey, so and so did thisNexus letter.
(12:28):
Um their VSO helped them writeit.
Will you take a look at it andsign it?
So I I actually know thisindividual pretty well.
So I was like, oh, let me takea look at it.
So I looked at it and I'm like,this is absolutely terrible.
They were trying to connecttheir tinnitus secondary to
migraines, but the medicalhistory discussion was only
(12:55):
about the veterans' tinnitus,and it did not discuss the
veteran's migraines at all.
And you you can't have that.
If you're trying to talk abouthow two conditions are related,
both conditions medically mustbe discussed.
And then you get down into therationale, and the rationale was
like all this great medicalliterature, but it had nothing
(13:17):
in how it pertained to thatveteran.
And so I even question, youknow, did you?
I mean, first of all, your VSO,I should give you a little bit
more, you should actually havemore understanding of what a
Nexus letter needs.
And then two, if you actuallydid use AI, that was terrible
because as far as a qualitystandard for what we put out,
(13:37):
that's not even close.
And so, you know, even VSOsaren't quite getting what is
needed in a Nexus letter, andthat's upsetting to me.
J Basser (13:49):
But did you break up the red pin?
Bethanie Spangenberg (13:52):
I just said um, no, thank you.
It doesn't meet my qualitystandards.
So I didn't want to waste mytime trying to tell them what
was needed or how to make itbetter.
I just pushed it back and said,no, thank you.
I didn't really have the timefor that.
So, yeah, absolutely.
Next question on here Do Istill need a doctor?
(14:14):
Yes.
You need a not necessarily adoctor per se, but you need to
have a qualified medicalprofessional.
So, how accurate is ChatGPT inany Nexus letter that goes out,
whether it comes from Vali forVet, it comes from one of our
competitors, or if it comes fromChatGPT, you have to verify
(14:34):
everything that it says, and youhave to make sure that it
includes the appropriatelanguage.
When I first started playingaround with ChatGPT, it would
try to give me the medicalmalpractice language as far as
what to use in the Nexus letter.
(14:54):
And medical malpracticelanguage is not the same as what
you use in veteran disability.
So you have to make sure thatthat phrase is in there.
Um the next one is is my dataprivate?
And this is a big thing thatthe NOVA presentation really
leaned on is you know, are youprotecting your veterans'
(15:17):
information if you're putting itinto an open source data
system?
They're using your data to makethings better, and that data is
now, you know, in their systembeing processed.
And so, how safe and secure isthat individual's data if you're
plugging it into this type ofsoftware?
(15:38):
Then the last point that, orlast question it talks about is
could it be biased or misseddetails?
Absolutely.
I've seen this firsthand.
And even, you know, if youthink about what we put into a
Nexus letter, there's we got toreview the medical records, and
then we have to understand themedicine, and then we have to to
(16:02):
find research, and then we haveto take all of that and put it
together.
There's actually an AI datasystem that if I tell it to be
biased in one in one way oranother, it will go and it will
get me research that makes itback research that is biased
(16:24):
towards one direction.
If I click that button for tomake it in support of.
So it's interesting becauseanybody can go in there and use
(16:46):
this AI system and it will pullactual studies from different
countries, from the US, from thegovernment, in order to support
a particular medical opinion.
So, in my in my experience, Idon't like that.
Um, again, that's a qualitystandard thing for us.
(17:06):
It needs to both clinicallymatch and you need to have good
studies to support thatrationale.
And we'll talk towards the endof the presentation.
When it comes to studies andresearch, I have some great
information that you can use andactually how you can use Chat
GPT to find more research forthe claim that you're trying to
pursue, which I think is reallywhere the most value comes from,
(17:30):
you know, AI right now forNexus letters.
Any questions?
Any thoughts before I go intothe next one?
Beri (17:40):
I was gonna ask you, I didn't want to sidetrack too
much, but that one letter thatyou got from the VSO, was it
even a good uh springboard intowhat they needed?
Was it even useful as a tool tostart a good letter, or was it
just completely?
Bethanie Spangenberg (17:56):
I mean, in my opinion, I could see a
veteran writing it the way itwas outlined.
When it comes to knowing a VSOknowing what should be in a
Nexus letter, I think that theyhave a higher standard that they
should know more about reallywhat goes into a nexus letter.
(18:18):
And I was kind of disappointedthat the VSO um didn't have that
quality that I was looking for.
Now, what I ended up telling,because it was one VSO, another
VSO, like one VSO wrote it, theother VSO I know.
He's like, hey, take a look atit.
So, you know, I don't know thisperson directly, but I know
(18:39):
this other person.
And I was just like, you know,this is you know, when I
rejected to sign it, and I said,you know, you can tell the
veteran that you know it doesn'tmeet our quality standards, and
really it's missing the detailsof this individual's migraine
history.
And, you know, it it gave themedical stuff on his tinnitus,
(19:00):
but it gave me nothing fortreatment, how long they've been
going on, um, what remedies hetried.
All of that was omitted in theNexus letter.
So I guess from a standpoint ofa veteran writing it, I'd be
like, okay, I I can understandwhat you're trying to do, but
this isn't quite right.
From a VSO standpoint, I thinktheir standard should have been
(19:21):
a little bit higher.
Beri (19:22):
Right.
So it it might be a useful toolfrom a fetcher that needs to be
filed to claim or something tostart with, but it needs to be
uh worked on.
J Basser (19:32):
I think I've seen that before.
Um I think maybe the veteransalready search connected for
migraines.
So he's trying to get hisanswers based on the context
itself.
Uh he's taking for granted thatthe VA knows he's already
searched connected formigraines, so he's not
elaborating too much on that.
(19:53):
So I think that's that's that'swhere that came from.
That's my personal opinion, butI've seen that several times
before.
But he does need to cover bothbases.
Bethanie Spangenberg (20:03):
Yes.
Any other questions?
No, I will say I also sawanother PDF recently um where it
was like a standard PDFtemplate, and the like the
(20:29):
opinion was like the veteran'sclaim condition is at least as
likely as not related to and hadlike a blank spot.
So it was basically like thedoctor would just like write the
diagnosis and then mark a checkcheckbox, kind of like how we
see the DBQs now.
And then on the back side waslike a rationale, and that
(20:49):
wasn't the PDFs I've seenonline.
I I still don't think thatthey're useful.
I don't I don't think they'recomprehensive enough.
They tried to make it like it'sa form, like you can just, you
know, um, and still kind of hitall those things the VA needs.
And I haven't seen a good PDFout there that that does it, or
(21:12):
a worksheet that does it, thatyou can just take into your
doctor and say, hey, can youfill this out?
Like I haven't found one, and Idon't know that you can quite
create one because there's somany theories of service
connections and so many factorsthat go into like obesity as an
intermediate step.
So you can't take like astandard PDF form or standard
(21:34):
worksheet and be like, here,doctor, fill this out, please.
And that doctor's gonna not bethey're not gonna be able to
look at something like that andreally know what boxes they're
supposed to check.
So there's a lot of limitationsin getting you know a Nexus
letter from your doctor.
It's it's hard.
So this is where AI may be ableto step in at some point, but
(21:56):
anyways, next slide.
So pros of using AI speed.
I would agree it's prettyspeedy, especially we write, I
use AI to write some emails.
When I get annoyed oraggravated and I want to write a
snippy email, I will talk toChat GPT or Grammarly and be
(22:22):
like, make this a kind email.
And it will take all mynegative, annoyed language and
turn it into something that'svery kind and fluffy.
And that has saved me like theheadache of um you know, waiting
the 24 hours or waiting acertain time period to let
(22:43):
yourself like cool down.
And it's like, oh, you see it?
It's like, oh, oh, you did sogood being a nice, a nice little
AI bot.
Like, so it and it does it,it'll rewrite that email just so
quickly and uh you know makesthings a lot easier from the
writing standpoint.
So whoops.
(23:05):
Thoroughness helps integraterecords and literature.
It can be thorough mainly forum when you're trying to discuss
certain research.
What I mean is you can drop theresearch into the chat GPT, and
you can tell them to summarizeit in layman's terms or
(23:28):
summarize the study findings ina seventh grade reading level in
paragraph format or in astructured format, and it'll go
bullet points and make somethingnice and easy to read if you're
trying to understand studies.
So I think that's helpful.
Um clarity, it can dostructured language and required
(23:53):
phrasing.
The other thing I will say, ifwe get so sometimes when we look
for research, we'll find it inanother language.
Most of your abstracts formedical research is written in
English, but the full study andits results may be in another
language.
We can actually drop that intoChat GPT and it will give us the
(24:16):
summary of the data or acertain section of the study
that we're looking for inEnglish.
And so as far as talking aboutlanguage, you know, it can help
us with understanding some ofthat data that comes back on
those studies.
When it comes to cost of usingAI, this is what I want to talk
about with our phone lines.
(24:38):
Um we have set hours, andduring those set hours, if our
phone lines are plugged up, thenthe veteran gets this
continuous voicemail of pleaseleave us a voicemail and we'll
get back to you, you know, yadayada yada.
So we have, and we justrecently put it in, and of
course it's got some learning todo, but we have AI running our
(25:01):
phone lines 95% of the time.
And that is because if one ofour phone lines is tied up, that
bot can still give you basicinformation related to our
services.
Now we're trying to teach it toget more detailed.
So if a veteran is wanting toknow specific details of the
status of their reports or thestatus of the work that we're
(25:24):
doing, we're working on how itcan look up that data and then
provide a summary to thatindividual.
But it's not there yet.
But when we look at the phonecalls that we're getting after
hours, those are customers thatare still engaging with our with
our company because they'regetting answers through that
bot.
(25:45):
So they'll call in, they'lltalk to the bot, then we see
them register online, and thenthey'll go through the process
of using our services.
The other, so that helps us tokeep like a continual phone line
available, you know, forsomebody who's calling in.
And so for us, that's costeffective.
(26:05):
Um, it's not necessarilypreferred.
I know that I don't like totalk to robots, so we have it
set up that if that individualcalls in and they say live
agent, live person, I want totalk to a real person, then it
will automatically transfer toone of our agents.
So for us, that is acost-effective way to keep phone
(26:27):
lines open, to keep an avenueof information for the services
we provide, and we find itcost-effective.
So the last pro of using AI isempowerment, and I would agree.
I think that when veterans lookat um using Chat GPT or similar
(26:54):
for Nexus letters, thequestions you should should
really be asking is if we usethe tinnitus example, how can
tinnitus be related tomigraines?
And so it can tell you in thelayman's terms, and then you can
actually prompt it to help youfind research.
(27:15):
And that is where I think it isthe most powerful for when it
comes to veterans.
And at the end of thepresentation, I actually have a
word-for-word prompt a veterancan use to help them find
appropriate studies that arestrong, that are really
supportive of what they'relooking for as an example.
(27:35):
Any questions here?
Not necessarily for the PTSDbecause PTSD is very specific
(27:58):
for the VA, but what I'm talkingabout is if you have like a
chronic back condition and youheard it in service in 1992, and
your next time of treatment wasin 2022, you've got 30 years
there.
You've got 30 years that youhave to plug in.
(28:19):
And so you may ask Chat GPT andsay, hey, I injured my back in
92, and I'm trying to write astatement for my disability
claim.
I suffered from symptoms andsaw the chiropractor over these
30 years, but I need to helpfill the gap of the lack of
medical record evidence, and itwill walk you through on what
(28:41):
type of information you shouldinclude in that statement as far
as chiropractorover-the-counter medication,
stretching, yoga, uh poolprograms at your YMCA, but it
will prompt you on all of thatsupporting information to help
you fill that gap.
(29:02):
So I do find AI is helpful withwriting statements in support
of your claim.
All right.
Beri (29:13):
I guess the one thing I have noticed, just kind of
trying to plug in stuff becauseuh sometimes I do just a Google
search looking for condition orsomething.
And if you use that, it doeslay out at least all the steps
and the things that you needtogether, which you know we go
over all the time with uhclaimants, but they you know it
(29:38):
it it spells it out step bystep, you know, with the
diagnosis and which form tofile, and it's it's useful for
that.
Yeah.
I'm just not sure you can useit.
A lot of people want it to dothe work of form, and it's it's
I haven't seen it, it's just nottheir Yeah, I would agree.
Bethanie Spangenberg (29:57):
Now uh Google has their Their new like
AI at the top, so where you canactually like, but if you just
use straight up Google for likeresearch, it will give you like
the most popular hits, andthat's where I think using that
AI is more advantageous,advantageous than Google,
(30:18):
because when you use AI, you'resaying find you know this
particular type of study, um, apeer-reviewed article that
discusses tinnitus andmigraines.
And then it will start to scan.
So I think it does a better jobof narrowing down the specific
(30:40):
type of research you're lookingfor, where Google is more of a
broad popularity type hit.
If if you plug that AI portionthat they just added, you know,
a few months ago, several monthsago, I guess, then it actually
has started to populate some ofthat more um useful information,
(31:01):
I think.
So, but I think you're right asfar as walking it step by step.
Um one another advantage for uhusing AI is we use it to check
our references.
And what I mean is in certaincases, you have to use AMA
(31:23):
formatting for writing.
And that's in-text citations,and the end of your report, you
do medical reference citations,and it's got to be, you know, a
certain format.
Some of it is full first name,full last name of the author,
and you name every singleauthor, and then you name the
title and the journal and thepublication date.
(31:44):
And I mean, there's so muchinformation when it comes to
citation in reports that it'shard to keep up with, and it's
important that we get thatright.
So for us, we typically useChicago style citation where you
put the superscript number intothe text, and then at the
bottom you list it in order.
(32:04):
So when we take an AMA citationand we try to turn it into a
Chicago style citation,sometimes we're not good at
doing that as humans.
And even sometimes ChatGP,ChatGPT is not the best at doing
that, but we will use that tohelp us check our citations to
make sure the formatting isappropriate, and that's
(32:24):
important to you know, givecredit to the authors of those
studies.
All right, let's go on to thenext one here.
Cons of using AI.
It hallucinates.
So I want to tell you the storyabout hallucinations, and I'm
not kidding.
I would love to teach AI how towrite a Nexus letter because it
(32:48):
would save cost, it would savetime.
You know, I think cost is thebiggest factor that we've ran
into as a company because youcan't just buy a franchise of
writing Nexus letters.
This has literally been fromthe ground up finding software
that works for you, findingstuff that works for your
company, and really buildingsomething that that works with
(33:11):
your your flow, with your umyour company's processes.
And something that has actuallyhelped us from the very
beginning is the team structure.
Meaning, if if I take a recordreview and I say, okay, yes, we
can write a Nexus letter, thenit goes off to the next person
and they have to confirm andverify that yes, the information
(33:33):
is here, no, there's not.
That is very time consuming andthat can be difficult because
of cost.
So that helps us if we wouldturn to AI, because we have two
humans verifying anything thatAI puts out.
That that won't change.
But I use AI or I look at toolswhere we can speed things up
(34:00):
and make this process easier forveterans.
And I have played with writingNexus letters, and in Chat GPT,
it created fake references, fakecitations, fake URLs, and it's
like, oh, this is kind oflovely.
And then you go to double checkthe work, and it's like, oh no,
(34:22):
no, no, oh no.
And that's actually very scary.
So even when I talk about theprompts later at the at the end
of the presentation, it isabsolutely necessary that you
check the URLs, that you checkthe titles, because it will
create all of that in order tomake a pretty response.
So, yes, it absolutely willhallucinate and make things up.
(34:44):
So you always have to checkfacts that are in you know,
whatever using for Chat GPT orAI.
Um bias, we talked aboutpardon?
J Basser (34:57):
Disclaimer on this one.
Bethanie Spangenberg (34:59):
Oh, right.
Um bias.
I have not appreciated anysocietal or data set biases.
Um it's maybe that's not reallysomething you know we've really
looked at.
Um so I don't I don't have anexample for that.
The cookie cutter risk draftfeels generic or impersonal.
(35:24):
I tried to teach it how tostructure a nexus letter and I
could not get consistency.
Um I just don't like the waythat it's putting stuff out
right now.
So it just feels, I mean, itlooks good unless you know,
like, you know, like I said, theexperience really gets me here
(35:45):
because it's like I've beendoing this for how long, and
this is like what you're puttingout, like you're not there yet,
bro.
Um black box reasoning may beunclear to defend on an appeal.
Uh I haven't really seen that.
Um, but I can see where thatmay apply if a veteran is trying
to write it, mainly becausethey don't have the medical
(36:08):
understanding.
And so if ChatGPT puts outsomething that sounds good, but
it doesn't clinically makesense, that might be, you know,
and and the veteran won't knowif it doesn't make clinical
sense or not unless they run itby, you know, somebody with an
education on it.
So privacy and ethics, don'tpaste sensitive health data into
(36:29):
public tools.
I I think we've already talkedabout that a little bit.
Any questions before we talkabout this next slide?
J Basser (36:42):
I think we pretty much covered it running.
Bethanie Spangenberg (36:46):
So at um Valley for Vet, our our medical
experts are not allowed to useuh use AI.
They have to explain everythingthat they want to explain and
why this study is important.
They can use AI to findstudies, they are not able or
allowed to use AI to write theirreport.
I'm the one that goes in anddoes grammarly to make things
(37:09):
sound to an appropriate umaudience.
And I'm also, you know, howthey talk about like left brain
and right brain people, medicalpeople aren't necessarily the
best at grammar and where commasand periods go.
And so before we used Grammarlyor had that process in there, I
(37:32):
would have attorneys just riplike comma here, space here.
Uh I mean, just and I'm like,okay, okay, we'll fix it, we'll
fix it.
So by adding that AI, we've hadmuch less complaints for
punctuation and commas andthings like that, which sounds
(37:52):
silly, but you know, some peopleare particular.
Um when uh case study when AIgoes wrong.
Again, AI made thispresentation.
I want to talk about itbriefly.
Um, not gonna spend a lot oftime here.
So there is a case that theytalked about at Nova where
attorneys back in 2023, it was aNew York attorney, they wrote a
(38:16):
brief using AI, and for somereason the attorney did not
check it, which blows my mind.
And they submitted it to, youknow, to court or whatever, and
it made up cases, it made up uma lot of the the findings of the
(38:37):
case just to meet what that umindividual was trying to find.
And so when the judge asked theattorney to talk about these
cases, it was like, oh, can'tfind it, didn't know it.
And so that attorney got likefined like $5,000, and that
attorney was like, or the judgewas like, um, you know, you can
use AI, that that's not theproblem, but you have to check
(38:59):
the work that it does.
You have to be able toindependently verify the
information that is coming outfrom it.
And so I think that was the bigtakeaway is like, you know, you
can use it, it's a tool, butit's not going to replace, you
know, your knowledge, yourexperience, and your education.
So um then there's this VAMunch example.
Sounds like the VA was tryingto use it uh as far as dealing
(39:24):
with contracts, and itincorrectly flagged uh $3,500
number as $35 million.
And so it ended up canceling inreal life some contracts
because it flagged it for theexpense that it was.
And so at that time, they'relike, nope, we're done.
We can't, this is not ready tolaunch.
(39:46):
It has to be um, you know, hasto have the manual oversight
until it's perfected.
So um hypothetical nexusmishap.
Generic letters uh gets alittle weight.
Yeah, yeah.
So right now I don't think it'sputting out enough.
Um takeaway, always doublecheck AI with human judgment,
(40:08):
and I I think that's a goodtakeaway there.
Any questions there?
J Basser (40:17):
Or are you doing a lot?
You do a lot of work doublechecking the uh AI.
Yeah.
Is it not counterproductive todo that, or is it still about
the same time-wise, or is it alittle faster?
Because once you have todouble-check everything, uh it
appears that you're gonna putthe same amount of time into it
(40:40):
that you would without it.
Bethanie Spangenberg (40:43):
Um, so we we're not in a place where we've
completely replaced um, youknow, I what I found when
working with both, and I eventried to pull up some examples
for this, but we won't have timeto go over it.
But what I have found isthere's still a disconnect
between taking the actualveterans case and the facts of
(41:07):
that case and plugging it intothe medical research.
So if I would give AI themedical information, and then I
would give them the studies,they still don't like merge that
understanding.
Um so right now with what we'reusing specifically for grammar,
(41:28):
it's it's not very timeconsuming because it's like keep
the text, make grammaticalcorrections, capitalize the word
veteran when discussing theveteran, period, and that's it.
Um especially with the researchpart too, because you know, the
medical people have to writethat in order to properly
(41:50):
explain their clinical thoughtprocess on that.
And um, you know, I almostthink it's it's it's not very
time consuming for us, but Ican't imagine trying to do like
the whole thing.
Um mainly when we do ourquality reviews, we're making
(42:12):
sure that the the medicalhistory is um heavy enough to
persuade or to feel like thehistory is complete, that
there's any um significant riskfactors are also discussed in
that medical history because theVA is going to try to deny you
(42:33):
and say, well, their obesity istheir primary risk factor for
sleep apnea.
And so if we talk about theirobesity and their medical
history, then we can say, well,yes, they're obese and their
body mass index was 34 at thetime of the study.
However, the prolonged historywith PTSD over 30 years,
including the lack of motivationassociated with PTSD, are
(42:57):
stronger risk factors andcontributing to the veteran's
sleep apnea.
So then, you know, we'reacknowledging those risk
factors.
Um quality review is makingsure that the grammar is
correct.
Um, and then of course, thatthe studies that are listed are
(43:18):
pertinent to the veteran'sclaim, that it fully explains
that relationship, and uh it'slengthy enough.
We don't like one or twoparagraph rationales.
We want something that hasthree or four medical studies,
even more to support a case.
If I have a Nexus letter thathas two medical literature
(43:39):
citations, it's getting kickedback.
I want more than that.
Does that answer the question,or is that like that was way too
much?
J Basser (43:47):
Your goal is you mean your goal is to at least kick in
the benefit of the doubt.
Guest (43:52):
Yeah.
J Basser (43:53):
At a minimum.
Yeah.
But you have to do what youhave to do.
On problem is, I mean, you knowthe VM is going to use negative
evidence, but you don't haveaccess to that evidence unless
you are, you know, accredited orwhatever, you're representing a
veteran.
So it's uh kind of catch-22 insome ways.
But as long as you, you know,you present it and you know you
(44:14):
stack the deck in your in thebest way you can, so that's what
it takes.
Bethanie Spangenberg (44:19):
Yeah.
We actually um we do get a lotof claims files.
So we're um so I'm happy, I'mexcited about that.
They're big files, but that'sthat's what we need.
The only limitation that Idon't like is that sometimes we
can still get the C file, but wecan't get um VB, uh excuse me,
(44:41):
can't get CPRS, the medicalrecords from the VA.
And so those aren't directlyplugged into VBMS.
So, you know, there's stillinformation there that we're not
fully getting.
So, but if we don't have enoughinformation, we don't have
enough information and we rejectto write the nexus.
And we'll actually tell theveteran, hey, we want more
(45:03):
records related to yourtreatment for migraines.
If you provide that to us,we'll take another look at it.
No cost to you.
We need more of those treatmentrecords.
Did you see a neurologist?
Did you have a head CT?
You know, what have you donefor treatment?
You we need more from you.
Right now we can't write it.
We may be able to.
If you get more records, getthem back to us and we'll take a
look at them.
J Basser (45:25):
Well, that's also an issue too when you deal with
older vets, because you know, alot of vets are, you know,
especially vets in their 50s and60s, you know, most of their
records are not computerized.
They're mostly paper.
2013, I think the VAs geteverything computerized.
So it's hard, you know, becausegoing through like the old days
(45:47):
that used to send the claimsfolder over to a C and P
examiner, here comes this500-pound block of paper.
And you've probably seen thatbefore.
And uh, you know, that'd gothrough each one of them.
I I've been sitting with a Cand P exam helping the examiner
go through the record.
So if everything'scomputerized, you could actually
(46:08):
you actually you could actuallyAI the entire the entire
veteran's medical history, youknow, from A to Z.
And it would probably cook up awhole bunch of stuff.
Bethanie Spangenberg (46:21):
So we have tested a few of those systems.
Um the biggest limitation ishandwritten records.
You're never gonna get rid ofhand record handwritten records
because of like in service orlike if you're you know in the
field or if you're at a basethat doesn't have a data system
(46:44):
or the computers are down.
I think we're always gonna seehandwritten records within the
military file.
Um and so that was the biggestlimitation with one of the
companies that I saw is theythey could their systems
couldn't read handwrittenwriting.
The other thing is check boxes.
So with the DBQs, the uminformation, if it's positive,
(47:10):
then they mark a box.
So we had some cases where theAI told us that they had hip
surgery, and here it was justreading the DBQ and the history
section where it was blank forit to state the history, and
here it read it as if thatindividual had hip surgery.
And so for DBQs, it hasn't beengreat either.
(47:34):
So right now we work with acompany that does both AI and
manual to index the records.
Um you know how they do it,their AI, you know, I'm not
familiar with, but um, you know,if we have issues, we we kick
it back to them to help them umlike re-educate or reteach their
(47:57):
system.
So ethical guidelines for usingAI, it recommends that you
verify the accuracy, you protectconfidentiality, the doctors
sign what matches their clinicaljudgment, that there's
(48:18):
transparency, that AI was usedas a drafting aid if asked.
Um to some extent, I thinkthat's appropriate.
I think the limitation is nowis that there is some type of AI
in every component.
So even in Word doc, there'swhat they call copilot, and I've
(48:39):
never used it, but that's aform of AI, and so I I can't use
a Word doc without copilotpopping in my face every time I
open it.
So I think if your AI is usedto draft more than 50%, then
maybe yeah, you you providethat.
Um avoid over reliance, so youreally have to keep humans in
(49:03):
the decision loop.
I that's what I think won'tever change, and then bias
checks.
Um I've never appreciated thatfrom the Nexus Letter side of
things, but certainly uh if it'sin there, it should be
considered.
J Basser (49:24):
So this verify Yes.
Go ahead.
Bethanie Spangenberg (49:29):
So bias accuracy and source risks.
Um let's see here.
There are some AI data systemsthat you can build or teach, and
the limitation is with that isthey have to have the most
up-to-date information.
So, for example, if somebody'strying to use AI in order to
(49:50):
keep up with the the CFR andratings, then every time it
changes, they're gonna have toupdate that into the system.
And so we call it Yeah.
They call it learning.
Yeah.
So I'm gonna push through theselast few slides here.
Um, I would agree that tone,um, there's a difference in the
tone between human and AI.
(50:12):
The there are certain wordsthat AI likes to fixate on.
And so when I read an articleonline, I'm like, oh, AI wrote
that.
Because I've seen it so much.
Um you want to make sure it hasconsistency, the rationale,
credibility.
We've talked about all of that.
So, as you can actually see,like in this AI-created
(50:34):
presentation, a lot of it isredundant.
So we're beating a dead horsewith some of this.
Predictive analytics, I don'tthink that this applies for
veterans.
So I taught or I told the AI tothat my audience was veterans
for disability claims.
Now, this predictive analytics,I actually know some large
(50:54):
attorney groups use this.
So they're using AI for all ofthese claims, and they're trying
to use some predictions on howthe VA is going to come back or
what things work or what thingsdon't work, which I think is
valuable for those largeattorney firms, but not
necessarily to you knowveterans.
(51:15):
Again, protect privacy, verifythe information.
I think that's pretty redundantthere.
So I would agree on thisconclusion that the AI is an
enabler, it's not a replacement.
Um, you have to balance thepros and cons, keep human in the
(51:37):
loop.
I do think that the VA is goingto use more AI.
I think they have to.
Um, in order to keep up withthe claims and keep to keep up
with government shutdowns, tokeep up with budgets, all of
that.
And then these couple slideshere.
Yeah.
Say that again.
J Basser (51:59):
I'll give you some insight in a minute.
Bethanie Spangenberg (52:01):
Oh, okay.
All right.
I'm going to push push throughthese last few things.
What I really want to talkabout are the types of studies.
When we look at medicalliterature, these are the all
the different kinds of studies,okay?
Your three at the top are goingto be your strongest evidence
to use.
The four in the middle aregoing to be mid, and then at the
bottom, the last four arethings you kind of want to steer
(52:22):
away from.
Um, ecological studies can berelevant when it comes to like
Agent Orange or toxicology typethings, but the case studies and
case reports, it's usuallysomething that's on one
individual person and they'renot very strong.
So when we take the fact oflike the strength of these, if
(52:43):
we plug into AI to find a study.
So help me find a study forhearing loss and dementia.
That was my example.
I prefer studies that are Did Ispell that wrong?
Oh boy.
Systematic.
I put systemic.
Systematic reviews andmeta-analysis, randomized
(53:06):
controlled trials, or pros yeah,or prospective or retrospective
cohort studies.
Studies from the Department ofVeterans Affairs should be
prioritized.
Additionally, PubMed orNational Institutes of Health
are important.
Provide a list of five studiesif found for each result.
Give me the title of the study,a brief summary in layman's
terms at a seventh grade readinglevel, and the URL to link to
(53:28):
the study.
So you put that, it should besystematic, not systemic.
But you plug that into Chat GPTfor what you're trying, the two
conditions that you're tryingto discuss, and you can actually
find quality studies that youcould submit to the VA for your
claim.
It'll explain it in layman'sterms.
(53:49):
And then let's say you get thePDF of that study and you're
like, you know what, I reallywant to verify this again.
So you open up a new chat andyou attach that study and you
put this prompt in here.
It says, analyze this medicalresearch article for me in plain
language.
Please explain only these fourthings in a way that a
non-medical person canunderstand.
(54:09):
The study type, what kind ofstudy it is, and where that
ranks in strength of evidence.
Relevance, how closely itrelates to the medical
connection I'm trying to prove,weight for legal use, whether
it's strong, moderate, or weakevidence for a Nexus letter, and
the recommendation (54:24):
should I use it as primary evidence,
supportive evidence, or not atall?
And keep it simple and clear,no heavy medical or research
jargon, do not use acronyms orabbreviations.
So using that prompt with thatPDF of the study, it can give
you useful information that youcan apply to the veterans
disability claim.
(54:45):
And that's all the slides.
Those last few slides were myslides.
Those weren't AI.
But like I said, you could seethat what it created is
redundant.
So if I would have gone inthere and tweaked it a little
bit, it could have a smootherflow.
We could talk about, you know,group some of those
conversations differently.
(55:06):
But I wanted you to see that,yeah, it's great, but you still
need the human touch in there.
J Basser (55:13):
Note to agents if you follow this criteria.
You can also work using Chat Bthis AI.
You can also type in verysimilar criteria, and you can
actually dig out uh all the BBAdecisions you need on certain
issues, and all the veteranscourt issues and the entire
history of the veterans court ispublished that you can also
(55:35):
find for whatever you're workingon.
So that's one good thing tohave, and that's that's a good
avenue.
So but uh given the fact thatartificial intelligence is
artificial intelligence and it'sartificially based, it's based
off of information availableacross the you know the airways
(55:56):
and the internet.
Every published article,whatever everything's on there
is available.
Um certain levels of thegovernment do not have access,
or not the access, but they donot have the power because once
you start going into learningmodes of AI, um the power
(56:17):
consumption just in order to douh the updates and things like
that, or to write the programs,we don't have enough power,
especially you know, in withinthe government scope itself.
It's gonna be 10 or 15, 20years before very much speed.
So I've had that up with onlygood authority.
(56:40):
So uh as far as what you guysare doing and what everybody you
know what needs to be done on ageneral basis as long as you're
using already established uhthat's all said and said and
done, you know.
You look at the big picture andthen AI's uh it's gonna be
probably the biggest thing thatthe human race has ever come
(57:01):
across once it gets going.
Bethanie Spangenberg (57:05):
You know, it's kind of scary because I
really do like I see what itdoes now.
I use it for mainly my kids.
I'll I'll create like uhbirthday cards tailored to
things that they like.
Like I did a birthday invitewith a Cappy Bear wearing a
party hat for my youngest.
That's what she wanted.
And so, you know, I was able tohave Chat GPT, you know, create
(57:27):
that cute little image, and Iprinted it on a card at Walmart
and you know had it tailored toher and said something that she
loved.
But um, you know, what doesthat do for our you know our our
artists?
What does that do for our youknow digital artists?
I mean, it's gonna replace somany things.
J Basser (57:47):
It's gonna replace a lot of things.
I mean, you know, you sit hereand it it's all it's all it's
all over Facebook, it's overeverywhere.
Guest (57:53):
Yeah.
J Basser (57:54):
You know, there's so much fraud and scams going on
using AI, AI generated actorsand things like that.
You gotta be careful becausethey're doing all it's uh but I
think our enforcementcommunities need to get their
hands on this to get a handle onit because it's gonna get ugly.
Okay for every good person thathas it.
(58:14):
Oh no, they're underneath theeight ball.
Not behind it, they're underit.
I think it's a magic eightball.
Shake it up.
Bethanie Spangenberg (58:30):
Denied.
J Basser (58:34):
But again, more good, really good content.
We'll probably get a lot offeedback on this one.
Um if you're using AI, guys, goahead and keep on using it.
I mean it works, but if you'rea veteran and you're using AI,
make sure that you kind of dumbit down.
You don't want to be a highschool educated veteran and
(58:54):
sound like a PhD when you submitsomething to the VA, well,
they're gonna know something'sup.
Guest (59:00):
Yeah.
Bethanie Spangenberg (59:01):
And I do want to mention that even like
even if you put in there, like,okay, so let's say you graduated
from high school, if you put a12th grade leaving reading level
into Chat GPT, it it's I don'tthink it's accurate.
I think you have to target likeninth, eighth, ninth grade
(59:21):
reading level to really hit UShigh school level.
Like I when I try to do it indifferent things, it get the
words get bigger and bigger andbigger, and I'm like, uh-uh.
So even if you're a veteranwriting a statement and you're
trying to get it down to youreducation, you you need to play
with it a little bit.
You need to understand yourprompts.
J Basser (59:41):
Look at the veteran's age.
Go back in history and forexample, do a um um do a little
math and say, okay, it's got tograduate high school, maybe you
run this time.
Put the time in there and do itthat way.
Because education has actuallychanged that much.
Yeah.
Bethanie Spangenberg (01:00:00):
And you're not quitting in the eighth
grade to go tend to the farmsanymore, so they do.
J Basser (01:00:06):
I mean used to.
They probably didn't know howto though, didn't they?
Bethanie Spangenberg (01:00:11):
They sure did.
J Basser (01:00:12):
Yeah, here too.
Well it used to be sixteen theyquit, but I guess they changed
that here to eighteen.
I know a lot of folks do that.
I know a lot of them stillthere too and done nothing in
their life, you know what Imean?
Bethanie Spangenberg (01:00:32):
Yeah.
J Basser (01:00:34):
Okay.
Well, guys, it's uh eighto'clock.
We want to thank Beth forcoming on.
Barry.
I'm sorry you didn't get a lotof words at me, but I do
appreciate you.
Beri (01:00:46):
Oh, I I appreciate this because I mean nothing and I
think those last two slides, yougotta really uh discriminate
what you ask that, and it'suseful total, but not ready for
prime time.
J Basser (01:01:02):
No, it's c it's gonna be complicated.
It's gonna be complicated.
Beri (01:01:06):
So it's gonna change a lot of, you know, it's gonna be
this AI is gonna be like thetelephone, but the TV is gonna
be a major thing.
It changes society.
J Basser (01:01:17):
Well, I mean, they just now got the degree.
The degree programs for AI havejust been had just come out.
I mean, this is something thatuh, you know, a lot of college
universities like UK or OhioState are just starting.
You know, some people alreadyhave the degrees in it, you
know, but uh, you know, they hadto go over and above and do
extra stuff just to get thedegree and certificates in AI.
Beri (01:01:38):
Well, and it changes so much.
J Basser (01:01:41):
And these kids are some of the smartest, these are
some of the smartest people thatI know in the world, but I'm
telling you what, I mean they'regeniuses doing this stuff.
But with that, uh guys, thankyou for coming on.
And we'll have another partynext week, Thursday, about eight
uh about seven o'clock eastEastern Standard Time.
And uh we'll have some otherparty and uh we'll uh we'll
(01:02:03):
drink some water and haveanother good discussion on
something VA related anyways.
Hopefully it'll be back towork.
So but good news is guys, sinceyesterday the VA has not denied
one single claim.
All right.
Well, guys, I want to shut herdown.
You guys have a good evening.
Bethanie Spangenberg (01:02:26):
I am too.
J Basser (01:02:27):
All right.
Bethanie Spangenberg (01:02:28):
Bye.
Welcome ladies and gentlemen to another episode of
the Exposed Vet Jay BassProductions.
Today is the I guess we sayit's the second day of October
2025.
The year has gone by atlightning speed.
Today our co-host is Mr.
(00:28):
Barry Furley.
He's down in Tennesseesomewhere.
How are you doing, Barry?
I'm hanging in there like aloose tooth today.
Oh, you had to bring that upagain, didn't you?
Anyhow.
Today we've got uh BethanySpangenberg.
She is the owner of the companycalled Fowler for Vet.
(00:48):
Uh Bowler for Vet is a companythat does independent medical
opinions, independent medicalevaluations for VA disability
issues, among other things.
And today we're going todiscuss using artificial
intelligence.
Mainly for Nexus Letters, thenwe'll get into some side
discussion too as to what'sgoing on with this stuff.
(01:09):
And uh Bethany, how are youdoing?
Bethanie Spangenberg (01:12):
I'm doing well.
Staying busy, ready for thetransition into fall, and just
staying keeping like trying torun around, keeping schedules
straight with the kids.
So a lot a lot of moving parts.
J Basser (01:27):
Well, life does have a lot of moving parts.
It does.
But you were talking aboutusing AI for Nexus Letters and
things like that for um, I guessyou do some stuff for uh about
vet, but uh I know it's use it'sgetting more and more popular.
Bethanie Spangenberg (01:48):
It is, and that's actually I had a um
conversation recently withanother business owner, and they
were talking about AI, andthat's kind of what prompted,
you know, the idea of havingthis, especially because the
same same week I saw a questionon a forum about using AI for
(02:08):
Nexus letters, and I'm like, youknow, we really need to have
that discussion.
They've talked about it at Novaa couple sessions, um, brought
a lot of understanding for legaluse.
Uh, and so I I really feel likethere's a gap there that we
need to fill when it comes tothe medical expert side for
Nexus letters and AI.
J Basser (02:30):
I can see that because I still noticed there's a lot
of um, there's one companythat's based their whole
business model is is using AI.
You've probably seen that.
Bethanie Spangenberg (02:40):
Yeah, I have.
Um that kind of worries me alittle bit because of my
experience with AI and thethings that I've seen and the
products that I see it put out.
Um, you know, AI has beenfabulous, and I'm actually
really impressed with where it'sat, but I don't think it's
targeted for VA disabilityclaims just yet.
(03:01):
I think we'll they'll getthere, but you know, there's
some things I feel that AI willnever replace.
Um, but I think there is aplace for it when it comes to
veteran disability and maybeeven in the use of Nexus
letters.
So for this presentation, Iactually wanted to see if you
(03:25):
use AI, the really the skillthat really comes in is learning
how to use that particular typeof AI.
You need to know how to writeyour prompts.
And your prompts will make orbreak the output of what you're
getting.
And so what I tried to do is Iactually had AI create the
(03:49):
content for this presentationtoday.
So when we go through thispresentation, I'm going to talk
about some things where yes, Iagree, no, I don't agree, and
then even point out some thingsthat maybe AI made mistakes on,
and that will be an example forthe limitations with AI.
So just keep that in mind as wego through here.
(04:13):
Uh we are actually seeingveterans give us an example of a
Nexus letter as like toreference or to use as evidence.
So they'll go into, I'm goingto talk mainly about Chat GPT
because it's the most commonright now.
Um, but I see veterans goinginto ChatGPT, they'll create
(04:37):
what they think is the rightNexus letter, they'll drop it
into their Valor for Vet folder,evidence folder for us to look
at as medical experts.
And right off, I know it's AI,I know where they got it from,
and it doesn't quite hit all thelegal points that I would have
hit as a medical expert doingthis for the last 14, almost 15
(04:58):
years.
And so again, that's that'swhat emphasizes the importance
of talking about this today.
So again, these slides arecreated, the content is created
by AI.
And you can plug the contentinto your slides, but even
PowerPoint will create digitalum presentations that are that
(05:23):
are like pleasing.
So for example, this circle onthe edge of my presentation,
this setup, I actually had atype of AI help me that is
plugged into PowerPoint tocreate this slide.
So even your your standardpresentation software is using
(05:45):
AI, your Photoshop uses AI,Grammarly is a type of AI, and
they're expanding.
Our phone lines that we use atValor for Vet are AI, and I'll
talk a little bit about thatlater on in the presentation.
But what is a Nexus letter andwhy AI?
So Nexus Letter is veryimportant for veteran disability
(06:05):
claims.
It takes that current serviceconnection, excuse me, that
current disability and tries torelate it to service.
So it's the part that connectsthese two items.
And majority of veterandisability claims need a Nexus
letter of some sort or a Nexusstatement with a supporting
rationale in order for a veteranto get service connected.
The challenge has been, and ithas always been, getting a Nexus
(06:32):
letter from a qualifiedindividual where the quality
meets the VA standards and it'sat a reasonable price.
Doing this for as long as wehave, we know that majority of
the cost associated with theNexus letter is coming from that
veteran directly.
So if a veteran is working withan attorney, the attorney may
(06:54):
upfront that fee, and then theveteran has to pay that fee on
the back end if they'resuccessful.
So we know at the end of theday, our market needs to target
the veteran's pocket.
And if you look at $1,500 forone Nexus letter, you have to
really weigh, okay, is $1,500worth me getting a Nexus letter
for an ankle condition, orshould I use that for something
(07:17):
like sleep apnea where it mayhave a longer term or a better
benefit for the cost?
And so a lot of the doctorsthat practice medicine, unless
you've been taught by the VA,unless you were taught by
somebody else, the skill ofwriting a Nexus letter is also a
niche, and they're they're hardto find someone who's
(07:39):
experienced enough to put outthe quality that is needed.
So if we use tools likeChatGPT, is it to the rescue, as
this presentation suggests?
So, yes, it can draft Nexusletters quickly, but there are
certain limitations when itcomes to some of the language,
(08:01):
the medical research, the legallanguage, the at least as likely
as not, the more likely thannot kind of phrase.
And then the big question is iscan AI help experts create
strong letters?
The VA will accept.
And so we can, but I don'tthink that that's I don't think
(08:24):
it's a one and done.
I don't think we plug it intoChat GPT and it spits everything
out.
I think we have to one, knowour prompts, and then two, we
have to have the clinicalknowledge to really put it back
like into like a Nexus letterthat flows.
So there's a lot of challengesin that.
J Basser (08:43):
I just hit the nail on the head because you know, if
you're sitting here and you're awriter and all of a sudden you
get a claim, and you get thisstatement, a personal statement
or whatever from this veteranwho's got a tenth grade
education.
Okay.
But he sounds like aPhiladelphia lawyer with uh a
PhD.
Yes.
(09:04):
It makes a big difference.
Bethanie Spangenberg (09:06):
And you know, the other thing is too, is
you know, because you broughtit up, yeah, it's later in my
presentation, but because youbrought it up, I'm gonna go
ahead and talk about it.
There is there are times whenour providers will write a Nexus
letter and they are usingmedical jargon.
And if you think about it, ouraudience is not medical
(09:30):
professionals, our audience isthe raiders or the people
deciding claims at the VA.
And somewhere, I can't rememberthe statistic, it's like more
than 50% of the raiders at theVA have a high school education.
And so, you know, what audienceare you writing this Nexus
letter for?
So if I have a Nexus letterwhere my provider is talking
(09:55):
about a left shift when it comesto white blood cells, they're
not going to immediately knowwhat a left shift is.
And so we use Grammarly to helpus kind of dumb it down in a
layman's language.
Um so then that way there isunderstanding of what what
language or what explanation andrationale that we're trying to
(10:17):
support.
You know, you can get into alot of these numbers and spit
out these numbers for labcounts, kidney count, like your
kidney function, um infectionnumbers, type things like that.
But to a layman, they may notunderstand it.
Well, with Grammarly, we cantake that information and we can
(10:38):
kind of put in that that laymanum explanation so we can really
talk to the reader that isreviewing that Nexus letter.
J Basser (10:47):
Yep, simplicity.
Yeah.
Bethanie Spangenberg (10:51):
So your audience is who you're writing
to.
Yes, you're right.
If a state a veteran who has aneighth grade education is
writing this really detailedstatement, you're like, okay,
this this isn't this is not.
And it's actually very like ifpeople are paying attention,
it's actually very easy to seethat.
So all right, let's move on tothe next slide.
J Basser (11:15):
Okay, go ahead.
Bethanie Spangenberg (11:18):
All right, maybe and go.
Okay, there we go.
So these are the top questionsAI found about Nexus Letters and
AI.
Will the VA accept it?
So content and a doctor'ssignature will matter most.
(11:40):
Absolutely.
So a VA will accept it if thecontent is appropriate to your
claim.
It discusses the veteran'smedical information, the
rationale supports the veteran'sclaim, including the details of
the medical history, and it'swritten or signed by a PA, DO,
(12:05):
nurse practitioner, physician, aPhD, uh, a psychologist with a
PhD.
So, yes, AI can write it andthen the doctor can sign it.
I just yesterday had a VSObring one to me, and they're
like, hey, so and so did thisNexus letter.
(12:28):
Um their VSO helped them writeit.
Will you take a look at it andsign it?
So I I actually know thisindividual pretty well.
So I was like, oh, let me takea look at it.
So I looked at it and I'm like,this is absolutely terrible.
They were trying to connecttheir tinnitus secondary to
migraines, but the medicalhistory discussion was only
(12:55):
about the veterans' tinnitus,and it did not discuss the
veteran's migraines at all.
And you you can't have that.
If you're trying to talk abouthow two conditions are related,
both conditions medically mustbe discussed.
And then you get down into therationale, and the rationale was
like all this great medicalliterature, but it had nothing
(13:17):
in how it pertained to thatveteran.
And so I even question, youknow, did you?
I mean, first of all, your VSO,I should give you a little bit
more, you should actually havemore understanding of what a
Nexus letter needs.
And then two, if you actuallydid use AI, that was terrible
because as far as a qualitystandard for what we put out,
(13:37):
that's not even close.
And so, you know, even VSOsaren't quite getting what is
needed in a Nexus letter, andthat's upsetting to me.
J Basser (13:49):
But did you break up the red pin?
Bethanie Spangenberg (13:52):
I just said um, no, thank you.
It doesn't meet my qualitystandards.
So I didn't want to waste mytime trying to tell them what
was needed or how to make itbetter.
I just pushed it back and said,no, thank you.
I didn't really have the timefor that.
So, yeah, absolutely.
Next question on here Do Istill need a doctor?
(14:14):
Yes.
You need a not necessarily adoctor per se, but you need to
have a qualified medicalprofessional.
So, how accurate is ChatGPT inany Nexus letter that goes out,
whether it comes from Vali forVet, it comes from one of our
competitors, or if it comes fromChatGPT, you have to verify
(14:34):
everything that it says, and youhave to make sure that it
includes the appropriatelanguage.
When I first started playingaround with ChatGPT, it would
try to give me the medicalmalpractice language as far as
what to use in the Nexus letter.
(14:54):
And medical malpracticelanguage is not the same as what
you use in veteran disability.
So you have to make sure thatthat phrase is in there.
Um the next one is is my dataprivate?
And this is a big thing thatthe NOVA presentation really
leaned on is you know, are youprotecting your veterans'
(15:17):
information if you're putting itinto an open source data
system?
They're using your data to makethings better, and that data is
now, you know, in their systembeing processed.
And so, how safe and secure isthat individual's data if you're
plugging it into this type ofsoftware?
(15:38):
Then the last point that, orlast question it talks about is
could it be biased or misseddetails?
Absolutely.
I've seen this firsthand.
And even, you know, if youthink about what we put into a
Nexus letter, there's we got toreview the medical records, and
then we have to understand themedicine, and then we have to to
(16:02):
find research, and then we haveto take all of that and put it
together.
There's actually an AI datasystem that if I tell it to be
biased in one in one way oranother, it will go and it will
get me research that makes itback research that is biased
(16:24):
towards one direction.
If I click that button for tomake it in support of.
So it's interesting becauseanybody can go in there and use
(16:46):
this AI system and it will pullactual studies from different
countries, from the US, from thegovernment, in order to support
a particular medical opinion.
So, in my in my experience, Idon't like that.
Um, again, that's a qualitystandard thing for us.
(17:06):
It needs to both clinicallymatch and you need to have good
studies to support thatrationale.
And we'll talk towards the endof the presentation.
When it comes to studies andresearch, I have some great
information that you can use andactually how you can use Chat
GPT to find more research forthe claim that you're trying to
pursue, which I think is reallywhere the most value comes from,
(17:30):
you know, AI right now forNexus letters.
Any questions?
Any thoughts before I go intothe next one?
Beri (17:40):
I was gonna ask you, I didn't want to sidetrack too
much, but that one letter thatyou got from the VSO, was it
even a good uh springboard intowhat they needed?
Was it even useful as a tool tostart a good letter, or was it
just completely?
Bethanie Spangenberg (17:56):
I mean, in my opinion, I could see a
veteran writing it the way itwas outlined.
When it comes to knowing a VSOknowing what should be in a
Nexus letter, I think that theyhave a higher standard that they
should know more about reallywhat goes into a nexus letter.
(18:18):
And I was kind of disappointedthat the VSO um didn't have that
quality that I was looking for.
Now, what I ended up telling,because it was one VSO, another
VSO, like one VSO wrote it, theother VSO I know.
He's like, hey, take a look atit.
So, you know, I don't know thisperson directly, but I know
(18:39):
this other person.
And I was just like, you know,this is you know, when I
rejected to sign it, and I said,you know, you can tell the
veteran that you know it doesn'tmeet our quality standards, and
really it's missing the detailsof this individual's migraine
history.
And, you know, it it gave themedical stuff on his tinnitus,
(19:00):
but it gave me nothing fortreatment, how long they've been
going on, um, what remedies hetried.
All of that was omitted in theNexus letter.
So I guess from a standpoint ofa veteran writing it, I'd be
like, okay, I I can understandwhat you're trying to do, but
this isn't quite right.
From a VSO standpoint, I thinktheir standard should have been
(19:21):
a little bit higher.
Beri (19:22):
Right.
So it it might be a useful toolfrom a fetcher that needs to be
filed to claim or something tostart with, but it needs to be
uh worked on.
J Basser (19:32):
I think I've seen that before.
Um I think maybe the veteransalready search connected for
migraines.
So he's trying to get hisanswers based on the context
itself.
Uh he's taking for granted thatthe VA knows he's already
searched connected formigraines, so he's not
elaborating too much on that.
(19:53):
So I think that's that's that'swhere that came from.
That's my personal opinion, butI've seen that several times
before.
But he does need to cover bothbases.
Bethanie Spangenberg (20:03):
Yes.
Any other questions?
No, I will say I also sawanother PDF recently um where it
was like a standard PDFtemplate, and the like the
(20:29):
opinion was like the veteran'sclaim condition is at least as
likely as not related to and hadlike a blank spot.
So it was basically like thedoctor would just like write the
diagnosis and then mark a checkcheckbox, kind of like how we
see the DBQs now.
And then on the back side waslike a rationale, and that
(20:49):
wasn't the PDFs I've seenonline.
I I still don't think thatthey're useful.
I don't I don't think they'recomprehensive enough.
They tried to make it like it'sa form, like you can just, you
know, um, and still kind of hitall those things the VA needs.
And I haven't seen a good PDFout there that that does it, or
(21:12):
a worksheet that does it, thatyou can just take into your
doctor and say, hey, can youfill this out?
Like I haven't found one, and Idon't know that you can quite
create one because there's somany theories of service
connections and so many factorsthat go into like obesity as an
intermediate step.
So you can't take like astandard PDF form or standard
(21:34):
worksheet and be like, here,doctor, fill this out, please.
And that doctor's gonna not bethey're not gonna be able to
look at something like that andreally know what boxes they're
supposed to check.
So there's a lot of limitationsin getting you know a Nexus
letter from your doctor.
It's it's hard.
So this is where AI may be ableto step in at some point, but
(21:56):
anyways, next slide.
So pros of using AI speed.
I would agree it's prettyspeedy, especially we write, I
use AI to write some emails.
When I get annoyed oraggravated and I want to write a
snippy email, I will talk toChat GPT or Grammarly and be
(22:22):
like, make this a kind email.
And it will take all mynegative, annoyed language and
turn it into something that'svery kind and fluffy.
And that has saved me like theheadache of um you know, waiting
the 24 hours or waiting acertain time period to let
(22:43):
yourself like cool down.
And it's like, oh, you see it?
It's like, oh, oh, you did sogood being a nice, a nice little
AI bot.
Like, so it and it does it,it'll rewrite that email just so
quickly and uh you know makesthings a lot easier from the
writing standpoint.
So whoops.
(23:05):
Thoroughness helps integraterecords and literature.
It can be thorough mainly forum when you're trying to discuss
certain research.
What I mean is you can drop theresearch into the chat GPT, and
you can tell them to summarizeit in layman's terms or
(23:28):
summarize the study findings ina seventh grade reading level in
paragraph format or in astructured format, and it'll go
bullet points and make somethingnice and easy to read if you're
trying to understand studies.
So I think that's helpful.
Um clarity, it can dostructured language and required
(23:53):
phrasing.
The other thing I will say, ifwe get so sometimes when we look
for research, we'll find it inanother language.
Most of your abstracts formedical research is written in
English, but the full study andits results may be in another
language.
We can actually drop that intoChat GPT and it will give us the
(24:16):
summary of the data or acertain section of the study
that we're looking for inEnglish.
And so as far as talking aboutlanguage, you know, it can help
us with understanding some ofthat data that comes back on
those studies.
When it comes to cost of usingAI, this is what I want to talk
about with our phone lines.
(24:38):
Um we have set hours, andduring those set hours, if our
phone lines are plugged up, thenthe veteran gets this
continuous voicemail of pleaseleave us a voicemail and we'll
get back to you, you know, yadayada yada.
So we have, and we justrecently put it in, and of
course it's got some learning todo, but we have AI running our
(25:01):
phone lines 95% of the time.
And that is because if one ofour phone lines is tied up, that
bot can still give you basicinformation related to our
services.
Now we're trying to teach it toget more detailed.
So if a veteran is wanting toknow specific details of the
status of their reports or thestatus of the work that we're
(25:24):
doing, we're working on how itcan look up that data and then
provide a summary to thatindividual.
But it's not there yet.
But when we look at the phonecalls that we're getting after
hours, those are customers thatare still engaging with our with
our company because they'regetting answers through that
bot.
(25:45):
So they'll call in, they'lltalk to the bot, then we see
them register online, and thenthey'll go through the process
of using our services.
The other, so that helps us tokeep like a continual phone line
available, you know, forsomebody who's calling in.
And so for us, that's costeffective.
(26:05):
Um, it's not necessarilypreferred.
I know that I don't like totalk to robots, so we have it
set up that if that individualcalls in and they say live
agent, live person, I want totalk to a real person, then it
will automatically transfer toone of our agents.
So for us, that is acost-effective way to keep phone
(26:27):
lines open, to keep an avenueof information for the services
we provide, and we find itcost-effective.
So the last pro of using AI isempowerment, and I would agree.
I think that when veterans lookat um using Chat GPT or similar
(26:54):
for Nexus letters, thequestions you should should
really be asking is if we usethe tinnitus example, how can
tinnitus be related tomigraines?
And so it can tell you in thelayman's terms, and then you can
actually prompt it to help youfind research.
(27:15):
And that is where I think it isthe most powerful for when it
comes to veterans.
And at the end of thepresentation, I actually have a
word-for-word prompt a veterancan use to help them find
appropriate studies that arestrong, that are really
supportive of what they'relooking for as an example.
(27:35):
Any questions here?
Not necessarily for the PTSDbecause PTSD is very specific
(27:58):
for the VA, but what I'm talkingabout is if you have like a
chronic back condition and youheard it in service in 1992, and
your next time of treatment wasin 2022, you've got 30 years
there.
You've got 30 years that youhave to plug in.
(28:19):
And so you may ask Chat GPT andsay, hey, I injured my back in
92, and I'm trying to write astatement for my disability
claim.
I suffered from symptoms andsaw the chiropractor over these
30 years, but I need to helpfill the gap of the lack of
medical record evidence, and itwill walk you through on what
(28:41):
type of information you shouldinclude in that statement as far
as chiropractorover-the-counter medication,
stretching, yoga, uh poolprograms at your YMCA, but it
will prompt you on all of thatsupporting information to help
you fill that gap.
(29:02):
So I do find AI is helpful withwriting statements in support
of your claim.
All right.
Beri (29:13):
I guess the one thing I have noticed, just kind of
trying to plug in stuff becauseuh sometimes I do just a Google
search looking for condition orsomething.
And if you use that, it doeslay out at least all the steps
and the things that you needtogether, which you know we go
over all the time with uhclaimants, but they you know it
(29:38):
it it spells it out step bystep, you know, with the
diagnosis and which form tofile, and it's it's useful for
that.
Yeah.
I'm just not sure you can useit.
A lot of people want it to dothe work of form, and it's it's
I haven't seen it, it's just nottheir Yeah, I would agree.
Bethanie Spangenberg (29:57):
Now uh Google has their Their new like
AI at the top, so where you canactually like, but if you just
use straight up Google for likeresearch, it will give you like
the most popular hits, andthat's where I think using that
AI is more advantageous,advantageous than Google,
(30:18):
because when you use AI, you'resaying find you know this
particular type of study, um, apeer-reviewed article that
discusses tinnitus andmigraines.
And then it will start to scan.
So I think it does a better jobof narrowing down the specific
(30:40):
type of research you're lookingfor, where Google is more of a
broad popularity type hit.
If if you plug that AI portionthat they just added, you know,
a few months ago, several monthsago, I guess, then it actually
has started to populate some ofthat more um useful information,
(31:01):
I think.
So, but I think you're right asfar as walking it step by step.
Um one another advantage for uhusing AI is we use it to check
our references.
And what I mean is in certaincases, you have to use AMA
(31:23):
formatting for writing.
And that's in-text citations,and the end of your report, you
do medical reference citations,and it's got to be, you know, a
certain format.
Some of it is full first name,full last name of the author,
and you name every singleauthor, and then you name the
title and the journal and thepublication date.
(31:44):
And I mean, there's so muchinformation when it comes to
citation in reports that it'shard to keep up with, and it's
important that we get thatright.
So for us, we typically useChicago style citation where you
put the superscript number intothe text, and then at the
bottom you list it in order.
(32:04):
So when we take an AMA citationand we try to turn it into a
Chicago style citation,sometimes we're not good at
doing that as humans.
And even sometimes ChatGP,ChatGPT is not the best at doing
that, but we will use that tohelp us check our citations to
make sure the formatting isappropriate, and that's
(32:24):
important to you know, givecredit to the authors of those
studies.
All right, let's go on to thenext one here.
Cons of using AI.
It hallucinates.
So I want to tell you the storyabout hallucinations, and I'm
not kidding.
I would love to teach AI how towrite a Nexus letter because it
(32:48):
would save cost, it would savetime.
You know, I think cost is thebiggest factor that we've ran
into as a company because youcan't just buy a franchise of
writing Nexus letters.
This has literally been fromthe ground up finding software
that works for you, findingstuff that works for your
company, and really buildingsomething that that works with
(33:11):
your your flow, with your umyour company's processes.
And something that has actuallyhelped us from the very
beginning is the team structure.
Meaning, if if I take a recordreview and I say, okay, yes, we
can write a Nexus letter, thenit goes off to the next person
and they have to confirm andverify that yes, the information
(33:33):
is here, no, there's not.
That is very time consuming andthat can be difficult because
of cost.
So that helps us if we wouldturn to AI, because we have two
humans verifying anything thatAI puts out.
That that won't change.
But I use AI or I look at toolswhere we can speed things up
(34:00):
and make this process easier forveterans.
And I have played with writingNexus letters, and in Chat GPT,
it created fake references, fakecitations, fake URLs, and it's
like, oh, this is kind oflovely.
And then you go to double checkthe work, and it's like, oh no,
(34:22):
no, no, oh no.
And that's actually very scary.
So even when I talk about theprompts later at the at the end
of the presentation, it isabsolutely necessary that you
check the URLs, that you checkthe titles, because it will
create all of that in order tomake a pretty response.
So, yes, it absolutely willhallucinate and make things up.
(34:44):
So you always have to checkfacts that are in you know,
whatever using for Chat GPT orAI.
Um bias, we talked aboutpardon?
J Basser (34:57):
Disclaimer on this one.
Bethanie Spangenberg (34:59):
Oh, right.
Um bias.
I have not appreciated anysocietal or data set biases.
Um it's maybe that's not reallysomething you know we've really
looked at.
Um so I don't I don't have anexample for that.
The cookie cutter risk draftfeels generic or impersonal.
(35:24):
I tried to teach it how tostructure a nexus letter and I
could not get consistency.
Um I just don't like the waythat it's putting stuff out
right now.
So it just feels, I mean, itlooks good unless you know,
like, you know, like I said, theexperience really gets me here
(35:45):
because it's like I've beendoing this for how long, and
this is like what you're puttingout, like you're not there yet,
bro.
Um black box reasoning may beunclear to defend on an appeal.
Uh I haven't really seen that.
Um, but I can see where thatmay apply if a veteran is trying
to write it, mainly becausethey don't have the medical
(36:08):
understanding.
And so if ChatGPT puts outsomething that sounds good, but
it doesn't clinically makesense, that might be, you know,
and and the veteran won't knowif it doesn't make clinical
sense or not unless they run itby, you know, somebody with an
education on it.
So privacy and ethics, don'tpaste sensitive health data into
(36:29):
public tools.
I I think we've already talkedabout that a little bit.
Any questions before we talkabout this next slide?
J Basser (36:42):
I think we pretty much covered it running.
Bethanie Spangenberg (36:46):
So at um Valley for Vet, our our medical
experts are not allowed to useuh use AI.
They have to explain everythingthat they want to explain and
why this study is important.
They can use AI to findstudies, they are not able or
allowed to use AI to write theirreport.
I'm the one that goes in anddoes grammarly to make things
(37:09):
sound to an appropriate umaudience.
And I'm also, you know, howthey talk about like left brain
and right brain people, medicalpeople aren't necessarily the
best at grammar and where commasand periods go.
And so before we used Grammarlyor had that process in there, I
(37:32):
would have attorneys just riplike comma here, space here.
Uh I mean, just and I'm like,okay, okay, we'll fix it, we'll
fix it.
So by adding that AI, we've hadmuch less complaints for
punctuation and commas andthings like that, which sounds
(37:52):
silly, but you know, some peopleare particular.
Um when uh case study when AIgoes wrong.
Again, AI made thispresentation.
I want to talk about itbriefly.
Um, not gonna spend a lot oftime here.
So there is a case that theytalked about at Nova where
attorneys back in 2023, it was aNew York attorney, they wrote a
(38:16):
brief using AI, and for somereason the attorney did not
check it, which blows my mind.
And they submitted it to, youknow, to court or whatever, and
it made up cases, it made up uma lot of the the findings of the
(38:37):
case just to meet what that umindividual was trying to find.
And so when the judge asked theattorney to talk about these
cases, it was like, oh, can'tfind it, didn't know it.
And so that attorney got likefined like $5,000, and that
attorney was like, or the judgewas like, um, you know, you can
use AI, that that's not theproblem, but you have to check
(38:59):
the work that it does.
You have to be able toindependently verify the
information that is coming outfrom it.
And so I think that was the bigtakeaway is like, you know, you
can use it, it's a tool, butit's not going to replace, you
know, your knowledge, yourexperience, and your education.
So um then there's this VAMunch example.
Sounds like the VA was tryingto use it uh as far as dealing
(39:24):
with contracts, and itincorrectly flagged uh $3,500
number as $35 million.
And so it ended up canceling inreal life some contracts
because it flagged it for theexpense that it was.
And so at that time, they'relike, nope, we're done.
We can't, this is not ready tolaunch.
(39:46):
It has to be um, you know, hasto have the manual oversight
until it's perfected.
So um hypothetical nexusmishap.
Generic letters uh gets alittle weight.
Yeah, yeah.
So right now I don't think it'sputting out enough.
Um takeaway, always doublecheck AI with human judgment,
(40:08):
and I I think that's a goodtakeaway there.
Any questions there?
J Basser (40:17):
Or are you doing a lot?
You do a lot of work doublechecking the uh AI.
Yeah.
Is it not counterproductive todo that, or is it still about
the same time-wise, or is it alittle faster?
Because once you have todouble-check everything, uh it
appears that you're gonna putthe same amount of time into it
(40:40):
that you would without it.
Bethanie Spangenberg (40:43):
Um, so we we're not in a place where we've
completely replaced um, youknow, I what I found when
working with both, and I eventried to pull up some examples
for this, but we won't have timeto go over it.
But what I have found isthere's still a disconnect
between taking the actualveterans case and the facts of
(41:07):
that case and plugging it intothe medical research.
So if I would give AI themedical information, and then I
would give them the studies,they still don't like merge that
understanding.
Um so right now with what we'reusing specifically for grammar,
(41:28):
it's it's not very timeconsuming because it's like keep
the text, make grammaticalcorrections, capitalize the word
veteran when discussing theveteran, period, and that's it.
Um especially with the researchpart too, because you know, the
medical people have to writethat in order to properly
(41:50):
explain their clinical thoughtprocess on that.
And um, you know, I almostthink it's it's it's not very
time consuming for us, but Ican't imagine trying to do like
the whole thing.
Um mainly when we do ourquality reviews, we're making
(42:12):
sure that the the medicalhistory is um heavy enough to
persuade or to feel like thehistory is complete, that
there's any um significant riskfactors are also discussed in
that medical history because theVA is going to try to deny you
(42:33):
and say, well, their obesity istheir primary risk factor for
sleep apnea.
And so if we talk about theirobesity and their medical
history, then we can say, well,yes, they're obese and their
body mass index was 34 at thetime of the study.
However, the prolonged historywith PTSD over 30 years,
including the lack of motivationassociated with PTSD, are
(42:57):
stronger risk factors andcontributing to the veteran's
sleep apnea.
So then, you know, we'reacknowledging those risk
factors.
Um quality review is makingsure that the grammar is
correct.
Um, and then of course, thatthe studies that are listed are
(43:18):
pertinent to the veteran'sclaim, that it fully explains
that relationship, and uh it'slengthy enough.
We don't like one or twoparagraph rationales.
We want something that hasthree or four medical studies,
even more to support a case.
If I have a Nexus letter thathas two medical literature
(43:39):
citations, it's getting kickedback.
I want more than that.
Does that answer the question,or is that like that was way too
much?
J Basser (43:47):
Your goal is you mean your goal is to at least kick in
the benefit of the doubt.
Guest (43:52):
Yeah.
J Basser (43:53):
At a minimum.
Yeah.
But you have to do what youhave to do.
On problem is, I mean, you knowthe VM is going to use negative
evidence, but you don't haveaccess to that evidence unless
you are, you know, accredited orwhatever, you're representing a
veteran.
So it's uh kind of catch-22 insome ways.
But as long as you, you know,you present it and you know you
(44:14):
stack the deck in your in thebest way you can, so that's what
it takes.
Bethanie Spangenberg (44:19):
Yeah.
We actually um we do get a lotof claims files.
So we're um so I'm happy, I'mexcited about that.
They're big files, but that'sthat's what we need.
The only limitation that Idon't like is that sometimes we
can still get the C file, but wecan't get um VB, uh excuse me,
(44:41):
can't get CPRS, the medicalrecords from the VA.
And so those aren't directlyplugged into VBMS.
So, you know, there's stillinformation there that we're not
fully getting.
So, but if we don't have enoughinformation, we don't have
enough information and we rejectto write the nexus.
And we'll actually tell theveteran, hey, we want more
(45:03):
records related to yourtreatment for migraines.
If you provide that to us,we'll take another look at it.
No cost to you.
We need more of those treatmentrecords.
Did you see a neurologist?
Did you have a head CT?
You know, what have you donefor treatment?
You we need more from you.
Right now we can't write it.
We may be able to.
If you get more records, getthem back to us and we'll take a
look at them.
J Basser (45:25):
Well, that's also an issue too when you deal with
older vets, because you know, alot of vets are, you know,
especially vets in their 50s and60s, you know, most of their
records are not computerized.
They're mostly paper.
2013, I think the VAs geteverything computerized.
So it's hard, you know, becausegoing through like the old days
(45:47):
that used to send the claimsfolder over to a C and P
examiner, here comes this500-pound block of paper.
And you've probably seen thatbefore.
And uh, you know, that'd gothrough each one of them.
I I've been sitting with a Cand P exam helping the examiner
go through the record.
So if everything'scomputerized, you could actually
(46:08):
you actually you could actuallyAI the entire the entire
veteran's medical history, youknow, from A to Z.
And it would probably cook up awhole bunch of stuff.
Bethanie Spangenberg (46:21):
So we have tested a few of those systems.
Um the biggest limitation ishandwritten records.
You're never gonna get rid ofhand record handwritten records
because of like in service orlike if you're you know in the
field or if you're at a basethat doesn't have a data system
(46:44):
or the computers are down.
I think we're always gonna seehandwritten records within the
military file.
Um and so that was the biggestlimitation with one of the
companies that I saw is theythey could their systems
couldn't read handwrittenwriting.
The other thing is check boxes.
So with the DBQs, the uminformation, if it's positive,
(47:10):
then they mark a box.
So we had some cases where theAI told us that they had hip
surgery, and here it was justreading the DBQ and the history
section where it was blank forit to state the history, and
here it read it as if thatindividual had hip surgery.
And so for DBQs, it hasn't beengreat either.
(47:34):
So right now we work with acompany that does both AI and
manual to index the records.
Um you know how they do it,their AI, you know, I'm not
familiar with, but um, you know,if we have issues, we we kick
it back to them to help them umlike re-educate or reteach their
(47:57):
system.
So ethical guidelines for usingAI, it recommends that you
verify the accuracy, you protectconfidentiality, the doctors
sign what matches their clinicaljudgment, that there's
(48:18):
transparency, that AI was usedas a drafting aid if asked.
Um to some extent, I thinkthat's appropriate.
I think the limitation is nowis that there is some type of AI
in every component.
So even in Word doc, there'swhat they call copilot, and I've
(48:39):
never used it, but that's aform of AI, and so I I can't use
a Word doc without copilotpopping in my face every time I
open it.
So I think if your AI is usedto draft more than 50%, then
maybe yeah, you you providethat.
Um avoid over reliance, so youreally have to keep humans in
(49:03):
the decision loop.
I that's what I think won'tever change, and then bias
checks.
Um I've never appreciated thatfrom the Nexus Letter side of
things, but certainly uh if it'sin there, it should be
considered.
J Basser (49:24):
So this verify Yes.
Go ahead.
Bethanie Spangenberg (49:29):
So bias accuracy and source risks.
Um let's see here.
There are some AI data systemsthat you can build or teach, and
the limitation is with that isthey have to have the most
up-to-date information.
So, for example, if somebody'strying to use AI in order to
(49:50):
keep up with the the CFR andratings, then every time it
changes, they're gonna have toupdate that into the system.
And so we call it Yeah.
They call it learning.
Yeah.
So I'm gonna push through theselast few slides here.
Um, I would agree that tone,um, there's a difference in the
tone between human and AI.
(50:12):
The there are certain wordsthat AI likes to fixate on.
And so when I read an articleonline, I'm like, oh, AI wrote
that.
Because I've seen it so much.
Um you want to make sure it hasconsistency, the rationale,
credibility.
We've talked about all of that.
So, as you can actually see,like in this AI-created
(50:34):
presentation, a lot of it isredundant.
So we're beating a dead horsewith some of this.
Predictive analytics, I don'tthink that this applies for
veterans.
So I taught or I told the AI tothat my audience was veterans
for disability claims.
Now, this predictive analytics,I actually know some large
(50:54):
attorney groups use this.
So they're using AI for all ofthese claims, and they're trying
to use some predictions on howthe VA is going to come back or
what things work or what thingsdon't work, which I think is
valuable for those largeattorney firms, but not
necessarily to you knowveterans.
(51:15):
Again, protect privacy, verifythe information.
I think that's pretty redundantthere.
So I would agree on thisconclusion that the AI is an
enabler, it's not a replacement.
Um, you have to balance thepros and cons, keep human in the
(51:37):
loop.
I do think that the VA is goingto use more AI.
I think they have to.
Um, in order to keep up withthe claims and keep to keep up
with government shutdowns, tokeep up with budgets, all of
that.
And then these couple slideshere.
Yeah.
Say that again.
J Basser (51:59):
I'll give you some insight in a minute.
Bethanie Spangenberg (52:01):
Oh, okay.
All right.
I'm going to push push throughthese last few things.
What I really want to talkabout are the types of studies.
When we look at medicalliterature, these are the all
the different kinds of studies,okay?
Your three at the top are goingto be your strongest evidence
to use.
The four in the middle aregoing to be mid, and then at the
bottom, the last four arethings you kind of want to steer
(52:22):
away from.
Um, ecological studies can berelevant when it comes to like
Agent Orange or toxicology typethings, but the case studies and
case reports, it's usuallysomething that's on one
individual person and they'renot very strong.
So when we take the fact oflike the strength of these, if
(52:43):
we plug into AI to find a study.
So help me find a study forhearing loss and dementia.
That was my example.
I prefer studies that are Did Ispell that wrong?
Oh boy.
Systematic.
I put systemic.
Systematic reviews andmeta-analysis, randomized
(53:06):
controlled trials, or pros yeah,or prospective or retrospective
cohort studies.
Studies from the Department ofVeterans Affairs should be
prioritized.
Additionally, PubMed orNational Institutes of Health
are important.
Provide a list of five studiesif found for each result.
Give me the title of the study,a brief summary in layman's
terms at a seventh grade readinglevel, and the URL to link to
(53:28):
the study.
So you put that, it should besystematic, not systemic.
But you plug that into Chat GPTfor what you're trying, the two
conditions that you're tryingto discuss, and you can actually
find quality studies that youcould submit to the VA for your
claim.
It'll explain it in layman'sterms.
(53:49):
And then let's say you get thePDF of that study and you're
like, you know what, I reallywant to verify this again.
So you open up a new chat andyou attach that study and you
put this prompt in here.
It says, analyze this medicalresearch article for me in plain
language.
Please explain only these fourthings in a way that a
non-medical person canunderstand.
(54:09):
The study type, what kind ofstudy it is, and where that
ranks in strength of evidence.
Relevance, how closely itrelates to the medical
connection I'm trying to prove,weight for legal use, whether
it's strong, moderate, or weakevidence for a Nexus letter, and
the recommendation (54:24):
should I use it as primary evidence,
supportive evidence, or not atall?
And keep it simple and clear,no heavy medical or research
jargon, do not use acronyms orabbreviations.
So using that prompt with thatPDF of the study, it can give
you useful information that youcan apply to the veterans
disability claim.
(54:45):
And that's all the slides.
Those last few slides were myslides.
Those weren't AI.
But like I said, you could seethat what it created is
redundant.
So if I would have gone inthere and tweaked it a little
bit, it could have a smootherflow.
We could talk about, you know,group some of those
conversations differently.
(55:06):
But I wanted you to see that,yeah, it's great, but you still
need the human touch in there.
J Basser (55:13):
Note to agents if you follow this criteria.
You can also work using Chat Bthis AI.
You can also type in verysimilar criteria, and you can
actually dig out uh all the BBAdecisions you need on certain
issues, and all the veteranscourt issues and the entire
history of the veterans court ispublished that you can also
(55:35):
find for whatever you're workingon.
So that's one good thing tohave, and that's that's a good
avenue.
So but uh given the fact thatartificial intelligence is
artificial intelligence and it'sartificially based, it's based
off of information availableacross the you know the airways
(55:56):
and the internet.
Every published article,whatever everything's on there
is available.
Um certain levels of thegovernment do not have access,
or not the access, but they donot have the power because once
you start going into learningmodes of AI, um the power
(56:17):
consumption just in order to douh the updates and things like
that, or to write the programs,we don't have enough power,
especially you know, in withinthe government scope itself.
It's gonna be 10 or 15, 20years before very much speed.
So I've had that up with onlygood authority.
(56:40):
So uh as far as what you guysare doing and what everybody you
know what needs to be done on ageneral basis as long as you're
using already established uhthat's all said and said and
done, you know.
You look at the big picture andthen AI's uh it's gonna be
probably the biggest thing thatthe human race has ever come
(57:01):
across once it gets going.
Bethanie Spangenberg (57:05):
You know, it's kind of scary because I
really do like I see what itdoes now.
I use it for mainly my kids.
I'll I'll create like uhbirthday cards tailored to
things that they like.
Like I did a birthday invitewith a Cappy Bear wearing a
party hat for my youngest.
That's what she wanted.
And so, you know, I was able tohave Chat GPT, you know, create
(57:27):
that cute little image, and Iprinted it on a card at Walmart
and you know had it tailored toher and said something that she
loved.
But um, you know, what doesthat do for our you know our our
artists?
What does that do for our youknow digital artists?
I mean, it's gonna replace somany things.
J Basser (57:47):
It's gonna replace a lot of things.
I mean, you know, you sit hereand it it's all it's all it's
all over Facebook, it's overeverywhere.
Guest (57:53):
Yeah.
J Basser (57:54):
You know, there's so much fraud and scams going on
using AI, AI generated actorsand things like that.
You gotta be careful becausethey're doing all it's uh but I
think our enforcementcommunities need to get their
hands on this to get a handle onit because it's gonna get ugly.
Okay for every good person thathas it.
(58:14):
Oh no, they're underneath theeight ball.
Not behind it, they're underit.
I think it's a magic eightball.
Shake it up.
Bethanie Spangenberg (58:30):
Denied.
J Basser (58:34):
But again, more good, really good content.
We'll probably get a lot offeedback on this one.
Um if you're using AI, guys, goahead and keep on using it.
I mean it works, but if you'rea veteran and you're using AI,
make sure that you kind of dumbit down.
You don't want to be a highschool educated veteran and
(58:54):
sound like a PhD when you submitsomething to the VA, well,
they're gonna know something'sup.
Guest (59:00):
Yeah.
Bethanie Spangenberg (59:01):
And I do want to mention that even like
even if you put in there, like,okay, so let's say you graduated
from high school, if you put a12th grade leaving reading level
into Chat GPT, it it's I don'tthink it's accurate.
I think you have to target likeninth, eighth, ninth grade
(59:21):
reading level to really hit UShigh school level.
Like I when I try to do it indifferent things, it get the
words get bigger and bigger andbigger, and I'm like, uh-uh.
So even if you're a veteranwriting a statement and you're
trying to get it down to youreducation, you you need to play
with it a little bit.
You need to understand yourprompts.
J Basser (59:41):
Look at the veteran's age.
Go back in history and forexample, do a um um do a little
math and say, okay, it's got tograduate high school, maybe you
run this time.
Put the time in there and do itthat way.
Because education has actuallychanged that much.
Yeah.
Bethanie Spangenberg (01:00:00):
And you're not quitting in the eighth
grade to go tend to the farmsanymore, so they do.
J Basser (01:00:06):
I mean used to.
They probably didn't know howto though, didn't they?
Bethanie Spangenberg (01:00:11):
They sure did.
J Basser (01:00:12):
Yeah, here too.
Well it used to be sixteen theyquit, but I guess they changed
that here to eighteen.
I know a lot of folks do that.
I know a lot of them stillthere too and done nothing in
their life, you know what Imean?
Bethanie Spangenberg (01:00:32):
Yeah.
J Basser (01:00:34):
Okay.
Well, guys, it's uh eighto'clock.
We want to thank Beth forcoming on.
Barry.
I'm sorry you didn't get a lotof words at me, but I do
appreciate you.
Beri (01:00:46):
Oh, I I appreciate this because I mean nothing and I
think those last two slides, yougotta really uh discriminate
what you ask that, and it'suseful total, but not ready for
prime time.
J Basser (01:01:02):
No, it's c it's gonna be complicated.
It's gonna be complicated.
Beri (01:01:06):
So it's gonna change a lot of, you know, it's gonna be
this AI is gonna be like thetelephone, but the TV is gonna
be a major thing.
It changes society.
J Basser (01:01:17):
Well, I mean, they just now got the degree.
The degree programs for AI havejust been had just come out.
I mean, this is something thatuh, you know, a lot of college
universities like UK or OhioState are just starting.
You know, some people alreadyhave the degrees in it, you
know, but uh, you know, they hadto go over and above and do
extra stuff just to get thedegree and certificates in AI.
Beri (01:01:38):
Well, and it changes so much.
J Basser (01:01:41):
And these kids are some of the smartest, these are
some of the smartest people thatI know in the world, but I'm
telling you what, I mean they'regeniuses doing this stuff.
But with that, uh guys, thankyou for coming on.
And we'll have another partynext week, Thursday, about eight
uh about seven o'clock eastEastern Standard Time.
And uh we'll have some otherparty and uh we'll uh we'll
(01:02:03):
drink some water and haveanother good discussion on
something VA related anyways.
Hopefully it'll be back towork.
So but good news is guys, sinceyesterday the VA has not denied
one single claim.
All right.
Well, guys, I want to shut herdown.
You guys have a good evening.
Bethanie Spangenberg (01:02:26):
I am too.
J Basser (01:02:27):
All right.
Bethanie Spangenberg (01:02:28):
Bye.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Las Culturistas with Matt Rogers and Bowen Yang
Ding dong! Join your culture consultants, Matt Rogers and Bowen Yang, on an unforgettable journey into the beating heart of CULTURE. Alongside sizzling special guests, they GET INTO the hottest pop-culture moments of the day and the formative cultural experiences that turned them into Culturistas. Produced by the Big Money Players Network and iHeartRadio.
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by audiochuck Media Company.