Latest / CLEs You Actually Want to Hear / Practical & Ethical AI Considerations for Litigators with Joe Stephens
Transcript
- 0:03I just want to welcome everybody to Practical
- 0:05and Ethical AI Considerations for Litigators.
- 0:08Good morning to some and good afternoon to others.
- 0:11My name is Ari Monison. I'm a Senior Executive
- 0:14with Steno Court Reporting, and we are a proud
- 0:17sponsor of today's CLE with the Beverly Hills
- 0:22Bar Association. At Steno, a little bit about
- 0:24us, we're a nationwide court reporting company,
- 0:27and we really try to take pride in our technology,
- 0:31such as our AI tool which is called Transcript
- 0:34Genius allows you really to ask anything you
- 0:36could think of when interrogating your transcripts.
- 0:39Also a nice complementary tool. Beyond technology
- 0:42we take pride in providing dedicated account
- 0:46managers for everybody and our average response
- 0:50time is 20 minutes or less to anybody across
- 0:53the country. We only use certified shorthand
- 0:55co -reporters and even offer complementary rooms
- 0:59for your depositions based upon availability
- 1:01and location. And we really just are grateful
- 1:05to be here today. Without further ado, I would
- 1:08love to introduce our speaker today who is Mr.
- 1:10Joe Stevens. Joe is a consulting attorney here
- 1:14at Steno. He also serves as a clinical lecturer
- 1:17at the and Chief Public Defender at Texas Tech
- 1:20University Law School where he leads the only
- 1:23public defender office housed within a law school.
- 1:28and he's almost 15 years plus of experience in
- 1:31criminal defense and public service. Joe previously
- 1:36founded and directed the Texas largest rural
- 1:39public defending office serving 12 county region.
- 1:43He also serves on the board of the Texas Criminal
- 1:46Defense Lawyers Association and contributes to
- 1:48multiple committees committed to advancing public
- 1:52and rural defense. At the end of today's presentation,
- 1:56Just so you know, there will be a barcode that
- 1:58you can scan if you would like to learn more
- 2:00about Steno and our AI tool and vice versa. And
- 2:04without further ado, let's all please welcome
- 2:06Mr. Joe Stevens today. Ari, thank you for that.
- 2:11And thank you all for letting me be here. I'm
- 2:13coming to you live from West Texas, which is,
- 2:16if there's an opposite of where you all might
- 2:19be practicing, it is the South Plains of the
- 2:22Texas Panhandle. But I teach at the law school
- 2:25here at Texas Tech and use AI all the time in
- 2:30my practice and practice teach with the students
- 2:33too. What we're gonna do here for the next hour,
- 2:38Ari does a wonderful job of setting up exactly
- 2:40what Steno does. And so on top of the work that
- 2:42I do at the law school, I also obviously work
- 2:44with Steno. We are a court reporting company,
- 2:47and I'm really excited for you and us to really
- 2:51spend the next hour. going through a couple of
- 2:55things. As someone who also is used to CLEs and
- 3:02goes to CLEs, I kinda wanna change this up a
- 3:04little bit. We're gonna spend the first half
- 3:05of this really going through the sort of ethical
- 3:07side of this that'll allow you to get your hour
- 3:09of credit when it comes to ethics and especially
- 3:13the intersection of that and AI usage. But then
- 3:16we're gonna do an actual application on the backend.
- 3:20And I'm gonna show you a real specialized AI
- 3:23tool that you can use. Everyone on here can use
- 3:26regardless of practice area, regardless of what
- 3:29side you're on, regardless of exactly, again,
- 3:31the specific type of work that you do that is
- 3:34there to enhance your practice. And it's a magical
- 3:38thing that combines, I think, everything that
- 3:42you're hearing about around AI right now with
- 3:45a practical solution. So whether you're an AI
- 3:49evangelizer or a skeptic or whatever it might
- 3:52be, The back half of this is, I think, where
- 3:54the magic really happens. I love leading webinars.
- 3:58I love CLEs. To the degree to which you want
- 4:02to get involved and active in the chat, please
- 4:05do so. I'll pay attention to it, and we can ask
- 4:08and answer questions throughout. But let's just
- 4:11jump right into it. Here's what I want to do.
- 4:15Everyone obviously knows now, and we hear it
- 4:17all the time. Obviously, AI is everywhere. You
- 4:21can't go to a conference. You can't sort of do
- 4:23anything in any area of your life right now without
- 4:26hearing something about the disruptive force
- 4:29that AI is to it. I'm not going to read all of
- 4:33these slides. You'll have access to the slides
- 4:34at the end. But I really want to spend some time
- 4:36on each slide kind of focusing on one or two
- 4:38bullets. The first one on here really has to
- 4:42do with the idea of how AI actually thinks. And
- 4:45it does a really wonderful job of presenting
- 4:47as though it can think just like a human. It
- 4:52presents itself in a way that demonstrates that
- 4:57it reasons just like a colleague might. And there
- 5:00is some truth to that, right? Like there is some
- 5:02truth to the way it presents its information
- 5:04and the way you can sort of engage with the information
- 5:06it tells you. But the thing I want to focus on
- 5:09slide because this is a question that comes up
- 5:10quite a bit is like how does AI actually even
- 5:13know what your question you're asking and then
- 5:16how does it sort of draw on the information base
- 5:19that it has to tell you its answer and it's an
- 5:22important question because if you don't know
- 5:25how it actually works then you're not going to
- 5:28be using products or you could increase the likelihood
- 5:32of using AI product for its unintended purpose.
- 5:37And I think the most classic example of this
- 5:39is what we see in the headlines all the time
- 5:41now with regularity is the person who used, for
- 5:45example, ChatGPT to do case law research. And
- 5:48then ChatGPT spits out a series of cases that
- 5:51are either incorrectly cited, don't stand for
- 5:53the proposition that the attorney is using them
- 5:57for. And that winds up in front of a judge and
- 6:00sanctions. tend to be the resulting force there.
- 6:03So when we talk about AI and how it actually
- 6:06thinks and processes and how it sort of presents
- 6:08the information, we need to know that AI is really
- 6:11trained on a tremendous amount of information
- 6:14up to a fixed point in time. Based on that information,
- 6:17it really is a strong ability to understand and
- 6:20recognize patterns. Because of that, when you
- 6:23ask it a question, it's gonna draw on that breadth
- 6:26of information. to sort of pick up on the patterns
- 6:29that are associated with the question you ask
- 6:31it. That's what makes AI incredible at some things
- 6:34and less incredible at others. So as long as
- 6:36we remember that as the starting point, we can
- 6:39really start to begin to see how transformative
- 6:42it can be in in multiple other areas of your
- 6:48practice too. Let's talk about actually specific
- 6:51to the legal sector where we're seeing it in
- 6:52litigation. We're seeing it everywhere, obviously.
- 6:55Any product that you're using, any product that
- 6:57you're considering to use, now it seems is gonna
- 7:00have some sort of AI component to it. And having
- 7:04that AI component obviously matters a lot, but
- 7:06because those different products are gonna have
- 7:09totally different functionalities. Knowing what
- 7:14the functionalities are will allow you to do
- 7:16two things at once. One is to decide how it could
- 7:19be beneficial to your practice. The second is
- 7:21to decide whether or not... it could be beneficial
- 7:24to have one or two or three tools sort of working
- 7:26conjunctively so that you can use them together.
- 7:29And I think that that is a real thing that we're
- 7:33gonna focus on here at the beginning and then
- 7:35show you exactly how to vet some of these products
- 7:37at the end. I wanna show you, and for people
- 7:39who are skeptical about AI or people who are
- 7:41sitting there saying, I need to know how to use
- 7:43it more because it's here or the time is now
- 7:46for me to understand it, but I don't know where
- 7:48to begin. The five bullets that you're seeing
- 7:50on this screen right here are five general parts
- 7:54of the legal sort of sector, I suppose, where
- 7:57AI is being a little bit disruptive. So if one
- 7:59of these areas of your practice I love the transcript
- 8:02analysis side of it, for example. Everyone who
- 8:05has had a deposition or a mountain of depositions
- 8:07and is preparing for other depositions or for
- 8:09trial or for any sort of motions practice has
- 8:12had to dig through a tremendous amount of information.
- 8:15Transcript analysis tools using AI are magical
- 8:18at reducing the amount of time that you need
- 8:21to spend digging for information so that you
- 8:23can make all of the key points of that information,
- 8:26the critical moments within those depositions.
- 8:28searchable, findable, instantaneously, and incredibly
- 8:31actionable. So pick an area of your practice
- 8:34that is causing you, I don't know, that is maybe
- 8:37most time intensive, and then find the appropriate
- 8:39tool that corresponds to it. And I think that's
- 8:42a good starting point for people who are wondering
- 8:44where to even dip their toe in the water. So
- 8:48interestingly, we hear, I think, a lot of, given
- 8:55how powerfully disruptive the AI is itself. I
- 9:00think there's been a lot of sort of questions
- 9:01about what this means for the future of the industry.
- 9:03We could spend so much time just talking about
- 9:05that. But what I want to sort of emphasize in
- 9:10this is that there's nowhere that I have, I mean,
- 9:14talk about this around the country. There's no
- 9:16sort of air judge, there's no state bar, there's
- 9:19no. local rules that are excited about or anticipating
- 9:24or suggesting that AI is here to take the jobs
- 9:29or really replace the roles of the minds that
- 9:33are actually deploying the AI itself. It's totally
- 9:36here right now as a practice augmenter. And I
- 9:39do think that there is some truth to the idea
- 9:41that if you're not engaging with it in some capacity,
- 9:44you probably are. you know, either losing business
- 9:47to or falling behind the curve when it comes
- 9:50to practitioners who are using it for some in
- 9:53some way, shape or form. So I like to sort of
- 9:56focus on that as well, because It can kind of
- 10:00dispel some of the doomsday narrative around
- 10:02AI. And quite frankly, I think it should get
- 10:05people excited about how transformative it can
- 10:07be today to your practice. The goal of this really
- 10:10is to actually not just dwell in the theory of
- 10:12for the next however long, it's to spend time
- 10:15in that setting it up so that we can deploy the
- 10:18show you an AI product that could you could use
- 10:20today. or many others that you could use today
- 10:23that will bring some great value to your practice.
- 10:26Okay, let's spend the next moment actually getting
- 10:28specific now about the ethical considerations
- 10:30that we're seeing. coming from California when
- 10:35it comes to AI usage. This is a general slide,
- 10:39but three things on this slide I think matter
- 10:41tremendously. And you'll see them reflected in
- 10:43the way that California has thought about AI
- 10:47usage and ethical AI usage for practitioners.
- 10:49The first is in the top left box. It's this idea
- 10:52of verification and due diligence. This might
- 10:55sound obvious to a lot of you, but the idea behind
- 10:57this is obviously AI is a product that you are
- 11:01leaning on to give you information. What information
- 11:04is the AI spitting out and can you verify whether
- 11:07or not the AI is telling you credible, truthful,
- 11:10reliable information or not? You have got to
- 11:13be able to verify the output. of what AI is telling
- 11:16you and courts around the country in their filings
- 11:20where they've called into question the ethical
- 11:23practice or the ethical use of AI by different
- 11:26practitioners have repeatedly said that this
- 11:28is a lawyer problem, not an AI problem. If you
- 11:32go and do traditional case law research and you
- 11:34stumble across a case that you miss site or say
- 11:37stands for a different proposition than it actually
- 11:40does, that's a lawyer problem, not a research
- 11:42problem. Analogously, that's what courts are
- 11:45saying about AI usage. Therefore, when you're
- 11:48using AI, one of the critical components of it
- 11:50is making sure that you can verify how and easily
- 11:54and quickly verify that it's telling you the
- 11:57correct information. Again, that's theory. In
- 12:00about 20 minutes, 30 minutes, I'm gonna show
- 12:02you how to do that. The due diligence side of
- 12:04that also is sort of wrapped up in what I just
- 12:06said. The second part of this slide that matters
- 12:09a lot is the security and client confidentiality.
- 12:12This to me is a really interesting gray area
- 12:15that is being litigated around the country also.
- 12:20But the idea with this is you've got to be able
- 12:22to use a product that generally speaking is not
- 12:28the AI is not getting smarter based off the information
- 12:31that you give it. In other words, if you're putting
- 12:34sensitive client information into AI, you better
- 12:36make sure This is what state bars are telling
- 12:40us. And I think it's backed up by just general
- 12:42ethical obligations that we have to our clients
- 12:45that have long predated the advent of AI. This
- 12:48idea that you got to make sure that the AI is
- 12:50not retaining that information or getting smarter,
- 12:53getting trained off the information that you're
- 12:55giving it. There's ways to vet products to make
- 12:58sure that you can see that this is true. And
- 13:00then there also additionally needs to be some
- 13:04sort of security protocol around that. You see,
- 13:07for example, different levels of compliance,
- 13:09like HIPAA compliance or SOC 2 compliance, different
- 13:11elevated levels of compliance when it comes to
- 13:15engaging with AI itself. So those two things
- 13:18matter a lot. The third thing that you're going
- 13:20to see really reflected, not just in California,
- 13:22but around the country is this idea that competency
- 13:26and growth when it comes to knowledge around
- 13:28AI and AI usage, that's going to matter a lot,
- 13:31right? Like, are you staying up to speed? on
- 13:35what AI is telling, like how it's evolving, its
- 13:37capacities, you know, just the evolution in regulation,
- 13:45the way that state bars are considering it, not
- 13:48just the AI itself, but maybe the way that AI
- 13:50needs to be, AI needs to be disclosed to courts
- 13:52or to clients or things like that. So these are
- 13:56the three things that we're really seeing. predominate
- 14:00when it comes to AI, ethical use of AI, as it's
- 14:04come from the ABA down through state bars around
- 14:07the country. And let's jump into those. Just
- 14:10a quick pause. I love to keep the questions coming.
- 14:12I saw that Jack asked a question about it, First
- 14:14Amendment right to use AI, even if a court or
- 14:17state bar bars its use. I mean, I think that
- 14:20is an amazing question. I'm gonna answer this
- 14:22one sort of live right now because I don't...
- 14:27I mean, talk about a time in our lives where
- 14:31first amendment issues are just increasingly
- 14:33litigated. I could see there being a component
- 14:36to that for sure. And I'd be interested to see
- 14:38how that plays out. I think another area of litigation
- 14:41around AI right now has to do with data itself
- 14:44and the discoverability of your AI usage, depending
- 14:48on the platform that you're using, where the
- 14:52information is being retained. I'm really interested
- 14:56to see how that's that's changing as well as
- 14:59different courts tackle it. I'm going to, David,
- 15:05your question about transcript analysis. I'm
- 15:06gonna answer that when we get to it. Kathleen,
- 15:09you're asking about Lexis protege and the new
- 15:11chat function. We'll talk about that a little
- 15:13later. And then, oh my God, the confidentiality
- 15:16questions. Once we get into that product, I'll
- 15:18actually get into those as well. Certification's
- 15:22been due. They're increasing, they're growing.
- 15:25I think that there is a gap Right now, I'm gonna
- 15:27answer this question live so that people can
- 15:29see it as we consider, because this really belongs,
- 15:31I think, more in the theoretical side of the
- 15:33presentation. Related to competence, are there
- 15:36certifications we'd be seeking out? There are
- 15:38different, I mean, at a law school now, I'm seeing
- 15:41AI and courses around technology really be thrust
- 15:44into the curriculum. That's obviously not necessarily
- 15:46available to us at this stage, but I do think
- 15:49that there are certifications you can get right
- 15:51now through Coursera or other online platforms.
- 15:54The baseline though that I think that state bars,
- 15:56and you'll see this in a couple of slides, that
- 15:58state bars are really calling for is just a level
- 16:00of competence. And we all know really what that
- 16:03means. I mean, competent representation means
- 16:06a lot of different things to a lot of different
- 16:07people. I think if I was to sort of be sort of
- 16:12skeptical about that, sometimes that bar can
- 16:14be pretty low. But when it comes to AI usage,
- 16:17I think that the critical, critical thing that
- 16:20we're seeing that gets people in trouble, which
- 16:22suggests that they're not knowing how to use
- 16:24it, which suggests that increased training would
- 16:25be better, is that they're not verifying what
- 16:28it's telling you. And if there's probably one
- 16:30thing to remember from this, it would be make
- 16:31sure that you verify the output of AI. But Bindu,
- 16:34I love that question about additional certifications.
- 16:38I'm gonna leave the other questions open as we
- 16:39keep rolling along here. So keep them coming.
- 16:43ABA model rule 1 .1 is something we're all familiar
- 16:46with. I mean, this is something that has been
- 16:48mirrored by a lot of state bars and as they've
- 16:51sort of taken this model language and applied
- 16:54it to their own jurisdiction and area of operation.
- 16:58So this is what I just talked about. This is
- 17:00the obligation for lawyers to provide competent
- 17:03representation to clients. The language in the
- 17:05bullet you see is something that we've been familiar
- 17:07with before AI as well, right? Like this is not
- 17:10new. The concept of this is not new. The interesting
- 17:13thing about it though is that last year we saw
- 17:16that the umbrella of this rule got a little bit
- 17:19further widened and extended to AI usage. So
- 17:23AI usage now is, if you're going to be incorporating
- 17:26AI usage into your practice, One, it needs to
- 17:30be done competently. And secondly, if we're looking
- 17:33at the bullets down sort of underneath that,
- 17:35the idea of what it means to use AI in a competent
- 17:39way is kind of broad and nebulous, which is rather
- 17:43unsurprising. But what we're seeing... effectively
- 17:46is that you've got to stay up to speed on how
- 17:48things are changing, which means you've got to
- 17:50understand how the products are changing and
- 17:53the capacities of AI are changing so that you're
- 17:56using them in the right way. And again, that
- 17:59sounds theoretical, but once we get into a specific
- 18:01product, you'll be able to see how we can actually
- 18:03vet this specifically. And here is really the
- 18:08main sort of slide when it comes to the California
- 18:12bar and the guidance that they're giving around
- 18:14ethical use of generative AI. 1 .1 on the left
- 18:19is, as you can see, rather identical to the slide
- 18:23before when it comes to competent representation
- 18:25and AI usage. You gotta know what the tech is.
- 18:28You gotta know how it functions. It doesn't necessarily
- 18:30mean 1 .3 in the middle. talks more specifically
- 18:34about this, but it doesn't necessarily mean that
- 18:36you have to be able to sit inside a conference
- 18:39room with a product team and a group of engineers
- 18:43and explain the ins and outs of how AI actually
- 18:46works. But you do have to know if the thing that
- 18:49you're using is being deployed for the purpose
- 18:52that you want. If you're using co -counsel in
- 18:57Westlaw to do case law research, that makes a
- 19:01great deal of sense. There's some really specific
- 19:03other prompts within co -counsel, right? Like
- 19:06to do some sort of email drafting or motion drafting
- 19:10or document analysis. That makes it really clear
- 19:14that you're using it for its intended purpose.
- 19:16But again, the classic example of using some
- 19:19sort of AI product outside of Lexis or Westlaw
- 19:22to do case law research, That's problematic.
- 19:26And then if you jump over to the right sort of
- 19:28column here, 1 .6 and 1 .82, this is really interesting.
- 19:32This kind of contemplates internal firm policies.
- 19:36In other words, if you're part of a firm, does
- 19:40your firm have an internal policy about how it
- 19:42vets AI products? I've seen this across the board.
- 19:46Some firms sort of say, let's use it as long
- 19:50as it. checks a bunch of compliances. Some say
- 19:53we're not using any AI until our internal team
- 19:56vets the security of it and we have a real understanding
- 20:00and an explicit guideline for our employees to
- 20:04follow when it comes to using the technology
- 20:06itself. So, you know, in other state bars, New
- 20:10York is a little bit more comprehensive when
- 20:11it comes to this. There really is a strong encouragement
- 20:14towards having some sort of policy around AI
- 20:18usage. There's also additionally in New York
- 20:21that doesn't exist in California to my knowledge,
- 20:24a real encouragement to disclose to clients usually
- 20:27in the initial agreement that AI could be used
- 20:30as a part of the representation. So New York
- 20:33considers things, other state bars consider things
- 20:35that you're not seeing here in California just
- 20:37yet. And let's see, then the last part of that
- 20:44that I was just trying to sort of. Touch on here
- 20:47is the idea of security and confidentiality.
- 20:49We talked about that before. We're going to talk
- 20:51about that again. It's easy to vet this. Um,
- 20:53and, uh, I think that, uh, as long as you're
- 20:57paying attention to the security protocol that
- 20:59comes from these AI products, that should be
- 21:02easy to access, should be incredibly transparent,
- 21:04um, and basically reduced to a number of basic
- 21:07questions that you can ask when you're vetting
- 21:09the product. You should be in good, in a good
- 21:12spot. We're going to shift to data security and
- 21:14AI here. Um, I saw. question just pop up from
- 21:17Chris Rose here. He says, is lawyer or law firm?
- 21:22So solos are now burdened with this. Yeah, I
- 21:26mean, I don't know, is the short answer. I've
- 21:30sort of wondered about the, I mean, enforceability
- 21:33is too strong of a word here. Enforceability
- 21:37to me, the way that this is happening is coming
- 21:39from AI usage that winds up leading to some sort
- 21:44of questionable practice that like generally
- 21:47speaking means that AI has given misinformation
- 21:49that winds up in front of a court. I haven't
- 21:52seen a lot of like more local, let's say a client
- 21:56goes after a lawyer for work that they did because
- 22:00they weren't happy with the results. And it turns
- 22:04out that one of the sticking points was that
- 22:05this person was not happy that their lawyer used
- 22:07AI. So for solos, there's a couple of things
- 22:12that I think, like to me, all of that is still
- 22:14very nebulous and we're in the throes of that
- 22:15being sort of addressed. The answer to me on
- 22:19that front would be one of two things. One is
- 22:22I don't think it really takes much if in case
- 22:25you wanted to just sort of paper it up to suggest,
- 22:28to document the fact that internally you're going
- 22:31to be using AI in accordance with all of the
- 22:34most recent state bar rules. I mean, I don't
- 22:35know where that needs to live, but I think if
- 22:37that's an internal firm policy, Which then so
- 22:41be it then that's obviously covered. I think
- 22:42another answer approach would be I have seen
- 22:45people disclose this more in their initial agreements
- 22:48this idea that Hey, just like before when I was
- 22:51using digital case law like Lexus or Westlaw
- 22:54digitized case law databases to do things more
- 22:57efficiently than going to books I might also
- 22:58be using AI in again in accordance with the ethical
- 23:02guidelines that that are most recently promulgated.
- 23:04So yeah, I think Chris that that the encouragement
- 23:10here is to add another layer of, burden is one
- 23:13word, but another layer of explicitness, if that's
- 23:16even a word, around how and when you're using
- 23:19AI. But that is as specific as it gets. Hopefully
- 23:24that relatively unsatisfying broad answer is
- 23:27somewhat satisfying, but anyway. And Sedar, Sedar
- 23:34has just asked a question about guidance for
- 23:37experts on the use of AI in their reports. Amazing,
- 23:39we're gonna talk about that on the backend of
- 23:40this as well, because I wanna get into expert
- 23:42uses of AI. All right, so this AI security checklist,
- 23:48we've got this idea that not all AI tools are
- 23:50created equal. I love this idea tremendously
- 23:52because I actually think that, I mean, I use
- 23:55a lot of sort of wider... broader tool clawed
- 23:59in my personal life, like outside of work, sometimes
- 24:02in work. Specialized tools to me are really where
- 24:05we're seeing a lot of incredible value at. It's
- 24:07this idea that we've taken the power of AI and
- 24:11really tailored it to a specific use, single
- 24:14or double, like brilliant specific use case.
- 24:17And I think that there's real value there. Sometimes
- 24:20the generalist tools do some, a number of things
- 24:24really well, and sometimes actually very, very
- 24:26well. But I think that specialized tools, especially
- 24:29for people who are just trying to sort of weave
- 24:31AI in into specific parts of their practice,
- 24:34specialized tools are where I think a lot of
- 24:36gold actually lies. Again, you see the other
- 24:39bullets down here about security verification
- 24:41and having a smart approach to AI security. And
- 24:44we are going to talk more about that. And some
- 24:47of the questions that you're asking actually
- 24:48pertain to that as well. One of the easiest ways
- 24:52while I'm on this to have a smart approach to
- 24:54AI security is if, for example, you're using
- 24:57chat GPT. You know, different levels, your account
- 25:01is going to have different levels of security
- 25:03protocol, depending on, you know, I mean, enterprise
- 25:05level accounts have different levels than sort
- 25:07of your just free account. And so vetting that
- 25:10as a minimum is something that would be critical
- 25:13to ensuring that you have a smart approach to
- 25:15AI security. Um, but again, we'll talk more about
- 25:18that when we get into the specific product itself.
- 25:20Here are two great questions to ask. We talked
- 25:22about transcript genius on the Ari mentioned
- 25:24it from the jump. That's what we're going to
- 25:25show you at the end here, but here are two great
- 25:28general. questions to ask that suggest that you're
- 25:30on the right track when it comes to vetting products,
- 25:32AI tools for security. One is what we talked
- 25:35about before. What is the AI doing with the data
- 25:38that you're giving it? And is there any retention
- 25:40whatsoever? The answer really should be no. I
- 25:42mean, you don't want your... I mean, for a whole
- 25:45host of reasons, right? I mean, this just goes
- 25:46back to... ethical practice of law, confidentiality,
- 25:50making sure that the data that you're using in
- 25:52any product, even before AI, when it was your
- 25:55case law research, like making sure that that
- 25:57data is not being retained and that the product
- 26:01is getting smarter on the backs of what you are
- 26:03telling it and the information you're providing
- 26:04it. Then there's a flip side to this or another
- 26:07side to this, which is just general security,
- 26:09like how secure is the product itself? And again,
- 26:13right now, I personally think from a lawyer,
- 26:16obviously the legal side, of things. You should
- 26:18be using AI products that are HIPAA compliant.
- 26:20There's additional layers of security. You see
- 26:22SOC 2 security more commonly in products now,
- 26:25too. But these are the things that AI products
- 26:29should be very transparent and loudly transparent
- 26:32about and very proud of how secure their platforms
- 26:34actually are. And those are two just generic
- 26:39questions to ask. OK, one little sort of I think
- 26:44sticking point and it's not really an ethical
- 26:47thing. This is prompt writing. But let's now
- 26:51let's kind of like understand the flow a little
- 26:53bit. We understand what the state bar has told
- 26:55us about ethical use of AI. Now we're sort of
- 26:58sitting here being like, okay, let's vet the
- 27:00product secure. So we're in accordance with,
- 27:02you know, making sure that it is secure and it's
- 27:05something that we can use. Okay, now we've picked
- 27:07our product and now it's like, it's go time,
- 27:10right? The prompt here. is obviously like the
- 27:15point of engagement with the AI itself. Like
- 27:17this is the thing that is going to, how you engage
- 27:21with it, the quality of the question that you
- 27:23ask it, the way that you interact with it is
- 27:27going to be the thing that governs what kind
- 27:30of output you get. I think six months ago, this
- 27:32was really kind of... presented as a mystifying
- 27:37art. Like you've got to ask the right question.
- 27:40You've got to really drill in perfectly. You've
- 27:42got to know where to start. You've got to be
- 27:43as specific as possible. Some of that lingers,
- 27:46but as AI continues to evolve, it's also getting
- 27:49better at giving you answers. to questions that
- 27:52you might also be curious about based on the
- 27:54question that you asked it. Some products explicitly
- 27:57do this. Like if you're using Adobe and you're
- 28:00using the AI functionality inside Adobe, you're
- 28:02going to see that it'll suggest other questions
- 28:05for you based off the question that you asked.
- 28:07So this happens in a lot of places. The evolution
- 28:09of prompt writing means that you don't have to
- 28:11be so concerned about getting the question right
- 28:13from the jump. thing is to just get started and
- 28:17then make sure that you refine and refine and
- 28:19refine. Ideally, you're using a product that
- 28:21allows you to do that. AI, I think it's at its
- 28:26most powerful when you have complete free reign
- 28:29over how you engage with it and how you can follow
- 28:31up with it, just the way you would have witnessed.
- 28:34If you're going to ask a witness a question,
- 28:36you're going to ask a specific question. You're
- 28:37probably going to ask a follow -up question.
- 28:39You're not going to ask three questions in one.
- 28:42You're probably going to make sure it's not vague.
- 28:43These are the things, if you're going to give
- 28:45an assignment to a colleague, or you're going
- 28:47to ask a witness a question, following those
- 28:48same set of rules about prompt engagement I think
- 28:51is a really good idea. Daniel, you're asking
- 28:56what is Transcript Genius? That's going to be
- 28:58the practical application, the product that I'm
- 29:00going to show you that sort of reduces all of
- 29:03this theory here to a real life demonstration.
- 29:08So we'll get to that here in a second. Key elements
- 29:11of a prompt, just as I said, I mean, clarity,
- 29:13context, specificity, and output. So how clear
- 29:15are you with your question? What kind of additional
- 29:17context are you giving that question? Are you
- 29:19specifying the format in which you want the answer?
- 29:22To me, this is no different than the way that
- 29:25you would sort of engage with a witness and the
- 29:28way you would teach someone to examine a witness,
- 29:30right? That clarity does matter because the flip
- 29:33side is going to make things a little bit more
- 29:35murky or perhaps difficult to obtain the answer
- 29:38right away. It's not going to stop you from getting
- 29:40there, but it's not the ideal starting point.
- 29:42Sometimes being too vague, right? Like not giving
- 29:45the context that you need to give, you're assuming
- 29:49that it knows information that it doesn't, or
- 29:51you're just over broad. So you ask five questions
- 29:53in one, it might answer each of those five questions,
- 29:56but because you asked five in one, the length,
- 29:59the quality, the value of each answer is a little
- 30:02bit more diluted. All right, so here we have
- 30:07a sort of a step -by -step guide that makes sense
- 30:09when it comes to how you actually go from the
- 30:13thought of a prompt to crafting one that is of
- 30:17a higher quality. And here's a prompt itself.
- 30:20This is a much more detailed prompt than I use
- 30:22most of the time, but this gives you a sense
- 30:24of how specific you can be. You know, when people
- 30:29very often, when I talk to people about AI and
- 30:32they'll sit there and they'll be like, can it
- 30:34do this? And I'm like, well, you can just ask
- 30:35it to do that. What if I ask it to focus on this?
- 30:38Well, you should ask it to focus on that. I mean,
- 30:40the idea is that it will respond to the information
- 30:44and the context and the parameters that you give
- 30:45it. So feel free to run wild with that. I was
- 30:48talking to another bar association maybe a few
- 30:50weeks ago and... the guy who was sort of leading
- 30:54that meeting was talking about how specific he
- 30:57was from the jump. And my approach is almost
- 31:00the complete opposite. I'll start specifically
- 31:02but broad and then drill in just the way I would
- 31:05sort of flow a line of questioning with a witness.
- 31:08So the idea is like, you don't need to hesitate
- 31:11to use it. You should really just jump in and
- 31:15just see what it tells you and let it guide you
- 31:17a little bit, have a little bit of faith in that
- 31:19process. Daniel is asking a question about using
- 31:24an AI for editing a draft brief for Clarity and
- 31:27products that are out there. I'm going to save
- 31:29some of the product specific questions for when
- 31:30I jump into the demo here in about, I don't know,
- 31:33two or three minutes. So here's a sum of the
- 31:36best practices for AI and legal work. The verification
- 31:38matters tremendously. It is there to augment
- 31:41your practice, continue learning. These CLEs
- 31:44are only going to be more prevalent. They're
- 31:45going to be more practice area specific. They're
- 31:47going to be product specific. It's amazing to
- 31:51see the rapid pace of change. It's also kind
- 31:52of interesting to see how legislatures that traditionally
- 31:56move a little bit slower. I mean, in Texas, the
- 31:59legislature here meets every second year for
- 32:01140 days. So for them to grapple with AI regulation,
- 32:05I mean, You can't keep a pace with the speed
- 32:09of innovation that the technology is actually
- 32:11presenting. So it's incumbent upon the practitioner
- 32:13to stay aware of that. And I think every single
- 32:15conference is going to have an ethical AI component
- 32:17to it from here on out. And let's end on this
- 32:22before I sort of stop sharing this screen and
- 32:24jump into another. As I said earlier, These generic
- 32:28products are really valuable, and they can do
- 32:30some things really wonderfully. I love, from
- 32:33a practitioner's point of view, the specialized
- 32:34products. So Daniel's question here about using
- 32:37AI for editing a draft brief for clarity, what
- 32:41products are out there? That's actually an area
- 32:42where we might see a nice sort of middle ground,
- 32:44because I think you can use generic products
- 32:46well, as long as the security component to them
- 32:50has been vetted. But specialized products on
- 32:53that front are going to be fantastic as well.
- 32:56That brings us to this practical application.
- 32:58And what I'm going to do here is I'm going to
- 33:00stop sharing this screen for a second. And then
- 33:03I'm going to open up a different window here.
- 33:07And I want to get into an actual demo so that
- 33:12you can actually see what I have been talking
- 33:17about as theory for the past little bit. Let
- 33:21me pull this up. All right. I'm going to start
- 33:26sharing this. And again, this to me is where
- 33:29the magic really happens, because theory is one
- 33:31thing, but practical application is another.
- 33:35So we've got 25 minutes, maybe a little bit more,
- 33:38depending on how long we go on this. Keep your
- 33:40questions coming. I'm going to try and answer
- 33:41them as I feel like they really slot into this
- 33:44side of the presentation. OK. All the theory
- 33:49basically suggested upfront that ethical use
- 33:52of AI first and foremost requires that you can
- 33:55ensure that the product you're using is secure.
- 33:58I'm showing you now, Daniel, this is your question
- 34:01earlier. You asked what is Transcript Genius?
- 34:03This is what I'm about to show you, but let's
- 34:05use Transcript Genius as a lens through which
- 34:07we can sort of explore the first 30 minutes of
- 34:10what I just talked about. We're looking at the
- 34:13security side of Transcript Genius and you're
- 34:16seeing upfront that there is some real proud
- 34:19broadcasting about the security of the product
- 34:22itself. Is there data retention? No, there isn't.
- 34:26What are the levels of security? This is what
- 34:29the levels of security are. We meet heightened
- 34:31levels of SOC 2 compliance, of HIPAA standards
- 34:34as well. And you can click further into the privacy
- 34:37policy. You can see the different compliances
- 34:39over here. In this case, you're seeing... up
- 34:42to the minute different levels of controls and
- 34:44security controls that are monitored by a third
- 34:47party. So you can really dig into and request
- 34:50actually access to a deeper understanding of
- 34:53the controls so that you or if you're part of
- 34:55a larger firm can vet the security of the product.
- 34:58Now, This, to me, checks the box as a practitioner
- 35:02of making sure that you're using a secure product.
- 35:05Like, this is the presentation of how they are
- 35:08sort of saying the security of their product
- 35:10is, and this is, I think, incredibly, incredibly,
- 35:14incredibly thorough when it comes to the different
- 35:16reporting that you have access to, the different
- 35:18controls that you can sort of request access
- 35:20to, different organizational infrastructure security.
- 35:23It's amazing. In my opinion, Box has been checked.
- 35:27Now it comes time to actually engage with the
- 35:30product itself. So I'm clicking into Transcript
- 35:32Genius. As Ari said from the jump, Transcript
- 35:34Genius is one of the tech pieces that Steno offers
- 35:37as a part of their court reporting services.
- 35:40It is a court reporting company at its core that
- 35:43does a number of things tech -wise to reduce
- 35:45the friction of the deposition booking and taking
- 35:48and then subsequently on the back end analyzing
- 35:51process, right? So a component of this is the
- 35:54AI itself. That's what Transcript Genius is.
- 35:58And that's what we're going to show you. Transcript
- 36:00Genius, when you click into it, you're seeing
- 36:01immediately a list of the sort of unfiltered.
- 36:05transcripts that have arrived automatically at
- 36:08this dashboard because you have booked the job
- 36:09through Steno. So what that means, and I say
- 36:11that, that means that it's coming to this dashboard
- 36:14in the optimal form for AI to engage with, right?
- 36:18What I'm going to do is I'm going to filter out
- 36:20by case because I want to know which cases have
- 36:24been actually like are relevant to this one case
- 36:28itself. And I'm going to click into one of the
- 36:31transcripts associated with this case. When I
- 36:34do that, you see that there is a transcript on
- 36:40the right side and then some suggested questions
- 36:42on the left. This to me is actually an optimal
- 36:46way to use AI because what you're having is the
- 36:49information that you're analyzing on the same
- 36:51screen as the questions and answers you're gonna
- 36:53get. The questions here are suggestive. Right?
- 36:56You're not locked into these questions. But if
- 36:58I click into one, you're going to see that in
- 37:00real time, it's going to consider a couple hundred
- 37:02pages of transcript and do what I asked it to
- 37:06do, which is generate a full length transcript
- 37:08summary with page and line numbers. AI is masterful
- 37:11at churning out sort of narrative summaries and
- 37:15giving you the information. But if you popped
- 37:17this question with this document into ChatGPT,
- 37:21First of all, there'd be some security concerns
- 37:23around that depending on your level of access,
- 37:27your account level and the security compliance
- 37:29and standards that go along with that account
- 37:30level. Secondly, You're going to have a different
- 37:33process of verification of the answers. So we've
- 37:36vetted the product for security. Now what we're
- 37:38doing is complying with the other critical component
- 37:40of, and I'm not sure how to say your name, Uduak.
- 37:43I'm not sure if that's correct. But you're saying
- 37:47California does not yet have a binding professional
- 37:49conduct rule specifically on AI usage. Yeah,
- 37:53absolutely. So this is in accordance with what
- 37:55they're suggesting and recommending for how AI
- 37:58should be used. verification matters. So next
- 38:02to every single statement of fact associated
- 38:04with this summary, you're gonna see that there's
- 38:07a hyperlinked reference to the source material.
- 38:10So you can click in and on the same page, you're
- 38:12looking at what it told you that the witness
- 38:14was married to this person and this was their
- 38:16wedding date. You can click in, you can see that
- 38:19on page 10, line 14 through 17, which is exactly
- 38:22where we are, the information has now been highlighted
- 38:24so you can verify that what it's telling you
- 38:26is true and correct. This to me is This obviously
- 38:32checks the verification box. From a practitioner's
- 38:35point of view when it comes to AI usage, this
- 38:37is optimal because what I have been given now
- 38:40is something dynamic and not static. This has
- 38:43not been given to me and now I just have to trust
- 38:46it and then verify on my own. It's given to me
- 38:49in a way that I can dynamically verify it. It's
- 38:51thorough, it's still writing its answer. And
- 38:54let's say the thing that I was actually most
- 38:56curious about wasn't, let's go to the top of
- 38:59the answer again, right? I don't, background
- 39:01information is great. I need to know that maybe
- 39:03what I actually really want out of this transcript
- 39:06is a real hyper -focus on the incident that was
- 39:11at issue in this case itself. So now. because
- 39:16it's not a static presentation, because I can
- 39:19actually subsequently engage with it, I can drill
- 39:22into it more and more and more. So if I just
- 39:24ask it to tell me more about the incident itself,
- 39:29and then also give me a timeline of events in
- 39:36table format, then this is what I love. I want
- 39:43you to think about this as slowly ramping up
- 39:46the functionality of being able to use an AI
- 39:50product thoughtfully, right? Summaries are wonderful
- 39:54because obviously it is exactly what we said.
- 39:56It's condensing this massive document into a
- 39:59much shorter version with all the key details
- 40:01there. Data extraction is a really powerful use
- 40:03of AI as well. And in this case, I've asked it
- 40:06now to... just focus on one area of the transcript,
- 40:09and then subsequently give me some more data
- 40:13in a more palatable format. Again, in this case,
- 40:16it was just a timeline of events, a sequence.
- 40:18What did this witness testify to across the course
- 40:20of this deposition when it came to how things
- 40:22played out? And then make sure that you cross
- 40:25reference exactly what parts of the transcript
- 40:27it's getting its information from. The thing
- 40:31I love about this is it's not going to tell you
- 40:33what information that doesn't exist. So here
- 40:37we've got that on this day, 40 to 60 minutes
- 40:41later, this is the thing that happened and this
- 40:43is the result of the thing that happened and
- 40:45we can verify that that's exactly what the transcript
- 40:48is telling us. So again, it's dynamic. It allows
- 40:51you to follow up. It really allows you to sort
- 40:54of drill into it and make sure that you can get
- 40:58the information that you want. You're not being
- 41:00delivered something that might be useful, but
- 41:02then you can't follow up on. You have complete
- 41:04control over how you prep for the next deposition,
- 41:06the information you get out of depositions, how
- 41:08you prep for your trial, how you prep for your
- 41:10next hearing, et cetera, et cetera. Christopher
- 41:12asked another question. I'm going to answer this
- 41:14live. So is it included in the base cost or extra?
- 41:17Steno has built, you pay for the transcript,
- 41:20you pay for the court reporting. This, everything
- 41:23you're seeing now, everything you're about to
- 41:25see, does not come at any additional cost. So
- 41:27when it comes to being able to engage with this
- 41:30technology and the AI itself, you can, it's not
- 41:34only a secure product, but you can engage with
- 41:35it time and time and time again, with no limit.
- 41:38At no additional cost. This is part of what makes
- 41:41I think this product really compelling is that
- 41:44you're never at a in the middle of prepping for
- 41:46something and asking follow -up questions and
- 41:48then Realizing that you have to change a subscription
- 41:51or anything like that doesn't happen use this
- 41:53because this We believe that obviously the AI
- 41:56is incredibly valuable I think I answered that
- 42:00live So I'm gonna be done answering that live
- 42:02now So that is single transcript interrogation,
- 42:10right? I mean, think of the use cases here. Did
- 42:13this witness contradict themselves? Did this
- 42:17witness make any key admissions? Did this witness
- 42:18deflect responsibility in any way, shape or form?
- 42:21If so, who did they deflect responsibility to?
- 42:25Who else should? I consider deposing based off
- 42:28this, et cetera, et cetera, et cetera. That is
- 42:31a single transcript interrogation. I'm gonna
- 42:33go back to this case though, because what I wanna
- 42:36do is I wanna show you that if I pick multiple
- 42:40transcripts associated with this case, this is
- 42:43where the value really is added, right? Like
- 42:46now. As we ramp up, it's not just data extraction
- 42:49on a single transcript. It's not a summary that's
- 42:51being given to you on a transcript that you can't
- 42:54follow up on. This is now where you can sit there
- 42:56and you can say, did these witnesses ever contradict
- 43:01each other? And now I'm going to violate all
- 43:05of the rules that I just told you. If so, where,
- 43:08let's say, when, I don't know, and how. And then
- 43:11I'm going to just kind of compound this question
- 43:13and say, also, give me a list of questions to
- 43:22challenge their credibility at trial. Okay, I
- 43:30would never compound this question if I was just
- 43:32doing this by myself, but I want you to sort
- 43:34of see in the time that we have together that
- 43:36it can really tackle these more nuanced things
- 43:39as well. So we're not just assessing it for data
- 43:42extraction now. We're really assessing it for
- 43:44witness contradiction. And it's going to now
- 43:46triple the amount of transcript it's interrogating.
- 43:49It's looking for nuances in where people, memory
- 43:53gaps and things like that, obviously. I'm also
- 43:56asking it basically now to assess all of these
- 43:58witnesses and give me a line of questioning I
- 44:01can ask them that will really reflect challenging
- 44:03their credibility on these issues. As it writes
- 44:07this answer, a couple of things. This is sort
- 44:13of party agnostic plaintiff defendant criminal
- 44:17civil like it doesn't matter You can deploy this
- 44:20however you like You might want to know this
- 44:23on the defense side because you want to see if
- 44:25there might be weaknesses in your OK, I mean,
- 44:27same on the plaintiffs. It doesn't matter. You
- 44:29can sort of assign it a role. How would opposing
- 44:32counsel consider the strengths and weaknesses
- 44:33of my case? What line of questioning should I
- 44:36anticipate opposing counsel would ask these three
- 44:37witnesses based on the weakness in their testimony?
- 44:40And you're seeing, again, that it's giving you
- 44:42this information. It's not static. It's dynamic.
- 44:45You could sit there and be like, oh, Now what
- 44:47I want to do is just focus on this one witness
- 44:49and give me 50 questions that I can ask that
- 44:53really flow a logical progression that will allow
- 44:55me to get to this point where this witness acknowledges
- 44:58this one weakness, whatever it might be. So as
- 45:01long as your first prompt is not your last, which
- 45:02transcript genius allows you to do really well,
- 45:05you're going to be golden. This is where, again,
- 45:08multiple transcript interrogation. I want to
- 45:11show one more thing and then I want to answer
- 45:13people's questions. Okay, let me go back to this
- 45:18environment. And I'm going to pick a different
- 45:20case, because someone asked a question about
- 45:23exhibits. And I'm going to kind of filter out
- 45:25here. And let's pick a case that I know has.
- 45:32Let's do this. A witness. Let's see I want to
- 45:43take the let's take another witness here and
- 45:51There should be maybe it's still in here, maybe
- 45:54it's not but you'll see that I Think I'm just
- 46:02gonna use these two there's a complaint that's
- 46:04not associated with this case, but if I uploaded
- 46:08the complaint in here as well What you'd see
- 46:12is I could run a prompt that is fantastic like
- 46:15I could run this prompt I'm gonna copy and paste
- 46:18it so everyone can see I just wrote this one
- 46:19up in advance because I was using it this morning
- 46:21on a different case if I say With these two depositions
- 46:26and the complaint create a chart showing the
- 46:28complaint allegations supporting testimony to
- 46:30contradicting testimony and gaps in evidence
- 46:33It's going to churn out an answer that's going
- 46:35to map every single allegation of the complaint
- 46:37against all of the supporting testimony, any
- 46:40contradicting testimony, and some of the gaps
- 46:42that live exist in the evidence. I don't have
- 46:45the actual complaint for this case uploaded in
- 46:48a trans. I have it on another case, but I can't
- 46:51publicly demo that case. But the idea is to the
- 46:54question, Jack, that you asked, can you load
- 46:57in exhibits too? Transcript genius is really
- 47:00good at understanding because you can upload
- 47:02PDFs, and I'm going to actually show you this
- 47:04right now. You can upload the transcripts here
- 47:06in text or PDF format, and it's going to understand
- 47:09what those are. So it'll understand and it'll
- 47:13obviously recognize because of the format that
- 47:15an expert's report is not a transcript, but an
- 47:18expert report in conjunction with the expert's
- 47:21testimony. They'll be able to map all of the
- 47:23sort of the weak spots, the similarities, the
- 47:27points of challenge. obviously make it understand
- 47:30that. And transcript genius is good at knowing
- 47:33the differences there. So yeah, it recognizes
- 47:35different documents, motions, motions for summary
- 47:38judgment, hearings on motions for summary judgment.
- 47:40It doesn't just have to be a transcript. Quite
- 47:42frankly, it was built for that, right? Like the
- 47:44critical point of engagement, because we are
- 47:46a court reporting company, was taking the transcript
- 47:49and making it actionable. but it's getting AI
- 47:53as it continues to evolve understands the different
- 47:55document types and understands that a trial has
- 47:57a different flow than a hearing. A trial has
- 48:01a different flow than a deposition or maybe an
- 48:03art emotions argument before the court. So if
- 48:06it's a, you OCR that PDF and you want to put
- 48:08it in here, I've used many different documents
- 48:10in here. And as transcript genius continues to
- 48:13evolve, you know, it's, it's as AI continues
- 48:17to evolve, the capacity gets more and more robust.
- 48:19When you upload through transcript genius. I'm
- 48:21going to go back to this. If you want to upload,
- 48:23you can upload it right here. And again, no additional
- 48:26costs. You have other transcripts. You have other
- 48:28documents. You put it in here, but you want to
- 48:30make sure this allows you to assign it to a case,
- 48:32associate it with a case, and then you can run
- 48:34an interrogation that isn't just a deposition
- 48:36or multiple depositions. It's the deposition
- 48:38plus other pieces of discovery that might be
- 48:42relevant to the question that you're trying to
- 48:43ask it. OK. It's 1250. And so I'm going to keep
- 48:53this screen open. I want to show a couple more
- 48:56things. You can take the output here, and you
- 48:59can copy and paste it, put it in whatever document
- 49:02that you are building to create. Because once
- 49:03we close out of here, again, security and making
- 49:06sure that there's no retention of the data, especially
- 49:09the AIs not retaining the data, it's gone. So.
- 49:12build your trial prep, your hearing prep notebook.
- 49:15I use this over the weekend to build about a
- 49:17hundred page anticipated cross of a witness.
- 49:20Like it's incredible. You can use this, do this
- 49:22in the course of an afternoon, take all the information
- 49:25with you very easily so that you hang onto it.
- 49:29Anyway, that's one thing I really wanted to show.
- 49:30Now I really wanna kind of pivot into some of
- 49:32these questions. And I'm gonna start from the
- 49:37top. David. The tech questions, what is the context
- 49:40window size for transcript analysis? David is
- 49:43basically asking where, yeah, how much can this
- 49:50do at once? There's AI limitations. So any product
- 49:54that's using AI, that's built on the back of
- 49:56another AI vendor, in this case we use Claude,
- 49:59Anthropic, we use Anthropic's most recent AI
- 50:02product. There are limitations. Like AI can't
- 50:06take 100 % of all of the things associated with
- 50:08the case and then give you an answer. It's going
- 50:09to be too much. The context is too great. So
- 50:13Transcript Genius is built with the idea that
- 50:15we prioritize accuracy. So we take the context
- 50:18window of Claude and basically we say, this can
- 50:21handle about 500 pages at once. And that's a
- 50:24volume thing. Maybe if there's 10 documents,
- 50:2710, which I've done before 20 that are much smaller
- 50:30in size. It can sort of understand that, but
- 50:32the volume, total volume of a single interrogation
- 50:35is about 500 pages. Um, there's so many ways
- 50:39to work around that too, but as a practitioner,
- 50:43The greatest value AI can bring me is reliability.
- 50:47And so if I'm doing, I sometimes will have multiple
- 50:49interrogations running at once based on the answers
- 50:51that it gives. And I can show you how to do that,
- 50:53David. At the end of this, I'm going to pop my
- 50:56information in the chat so that you can reach
- 50:58out to me directly. But that's the limitation.
- 51:01And that's partially an AI limitation. And that's
- 51:03our internal engineering team making sure that
- 51:05the answer it gives you is quality. So you'll
- 51:09know when it's given you, when you overload it
- 51:13and it can't give you, when you've exceeded the
- 51:16capacity, right? Kathleen, you asked about Lexus
- 51:19protege and the chat function. I don't, I use
- 51:21Westlaw mostly and co -counsel so I don't really
- 51:23know about the chat function in there. Uh, I'm
- 51:26going to kick that around. Actually, through
- 51:29the school, my university, I'm going to be working
- 51:31with a Lexus rep next week. So if you want to
- 51:32reach out with me, to me directly, I can follow
- 51:35up with you on that next week. Uh, Joseph confidentialities.
- 51:38Is it sufficient? We do not input client names.
- 51:41I think that that checks a huge box. Um, but
- 51:44again, it depends on what you're using the AI
- 51:46for, uh, anonymizing things and redacting information
- 51:51is really important and can be really important.
- 51:54I prefer not to take that step. I prefer to have
- 51:56this tool and product secure enough for me to
- 51:59engage with it the way that I would a colleague.
- 52:03But that definitely goes a long way in obviously
- 52:06respecting client confidentiality. Let's see.
- 52:11I'm going to bounce around here. Some of the
- 52:13easier questions. Bindu, if we toggle off on
- 52:15allowing the AI to train based off our inputs,
- 52:17does that satisfy? I would guess so. But again,
- 52:20I'm trying to think of a product that you're
- 52:22talking about here. I prefer to go the extra
- 52:25mile and use a product that ensures that there's
- 52:29no risk of that from the outset. And so I've
- 52:32actually never seen the toggle off. When it comes
- 52:34to Transcript Genius, that doesn't exist. Other
- 52:36products certainly will provide that capacity.
- 52:39But I would definitely make sure that if you're
- 52:41using a product and it has the option for you
- 52:43not to, for the AI to get smarter based off the
- 52:47client information and what you're putting into
- 52:49it, then I would definitely encourage that. Daniel,
- 52:53you asked this question about drafting a brief,
- 52:56editing a draft brief for Clarity. You can use
- 53:00a general tool for that. Now, if you have, you
- 53:02want to make sure that things are redacted, if
- 53:03there's no security that or not an up to snuff
- 53:06security, but you can use even transcript genius
- 53:08for that. Like I would put a copy of, I would
- 53:12ask it, you know, for the information that you
- 53:15want to include in that brief in a format that's
- 53:17copy and pasteable into your brief. And then
- 53:20I would. Then once you've got it into a spot
- 53:22that you like, I would put it back in here and
- 53:24say, what do you think about this? What changes
- 53:27would you make? Where can I tighten this up?
- 53:29Where to make sure it's not repetitive? Make
- 53:31it really thorough. What else am I missing? If
- 53:34you were opposing counsel, how would you poke
- 53:36holes in? So this is where, to me, a good interactive
- 53:40AI product allows you to Use it to get the information,
- 53:44make the information actionable, then float it
- 53:46right back to the AI itself and ask for feedback.
- 53:50So I love that use case. And I think you can
- 53:52use a number of AI products to get there, including
- 53:54the one that I just showed you. This is a wonderful
- 54:00question and thought. This is, I think, a really
- 54:04important piece of information I just want to
- 54:05highlight. And I love this idea of treating AI
- 54:07like social media, knowing the settings. I mean,
- 54:10yeah, if you're using tech, you got to understand
- 54:11this in your practice. You got to understand
- 54:13the security side of it. And even though there
- 54:15aren't binding, as you said, professional conduct
- 54:17rules through the California State Bar, there
- 54:19are notes of encouragement. And you're seeing
- 54:22if you really want to any sort of motions for
- 54:25sanctions will reflect exactly where the lawyer
- 54:28fell short and what they're going to have to
- 54:30sort of demonstrate and what the court is concerned
- 54:32about their AI usage if it gets that far. So
- 54:35we get a sense of how these regulatory bodies
- 54:40are considering it and also courts themselves.
- 54:43Jack, can you load in all exhibits too? I think
- 54:47I showed you that a second ago. Yes and no. Yes.
- 54:52I mean, as long as you have an OCR PDF, put it
- 54:54in there and see what. But it's not going to
- 54:56be able to take your entire case file and interrogate
- 54:58it once. There has to be some thought into how
- 54:59it's deployed. But this is built for transcript
- 55:02analysis. And then because AI is what AI is,
- 55:05it has a really deep recognition that other documents
- 55:09are not transcripts and how they all relate to
- 55:11each other. The citation format will look different
- 55:13because obviously not everything has page line
- 55:15citations the way a transcript does. But yeah,
- 55:18I mean, I people upload other documents into
- 55:20it all the time. And I certainly do. And over
- 55:23time, it's only going to get stronger and stronger.
- 55:25You just have to realize this was built for transcripts.
- 55:30Here's another question. In a version of Transcript
- 55:32Genius included in real time during depots? Yeah.
- 55:35Transcript Genius is not a real time product.
- 55:40That's not the strength of this. The strength
- 55:42of this. is in the prep around your hearing,
- 55:45after your hearing, the same day. Take your rough
- 55:48draft. If you can get it that day, put it in
- 55:50here, ask questions. But real -time transcription
- 55:53and digital transcription during a hearing, that's
- 55:56not something that Transcript Genius does. This
- 56:00allows you to be much more prepared for the hearing,
- 56:03take the information from the hearing afterwards
- 56:04or during or at a lunch break or at the end of
- 56:06the day and put it in here. But this is designed
- 56:09for you to actually have a copy of the transcript.
- 56:12on hand. Heather, can you put all your complaint
- 56:17documents, the answer, and motions into steno
- 56:19analysis before or after deposition transcripts?
- 56:22Yes, do it all. That's perfect. The more context
- 56:25it has, the better. Ask away. That is a great
- 56:28use case. And again, the upload function and
- 56:31the prompts themselves aren't going to cost you
- 56:32anything extra. So do that for sure. Can chat
- 56:37GPT be used ethically? Oh my goodness, this is
- 56:40an amazing question. I feel like we could get
- 56:41really philosophical here. Yes. I mean, I guess
- 56:46a bigger question is like, can AI in general
- 56:48be used ethically? I suppose yes. The evolving
- 56:51standards of the practice around what ethical
- 56:53is with AI. I think your real question, Bertrand,
- 56:57that I want to answer and I think you're getting
- 57:01at is. Is it ethical in terms of the compliance
- 57:05and the security that a lawyer should be engaging
- 57:09with tech? And the answer is yes, you just have
- 57:12to have a heightened account level. And sometimes
- 57:14that can be, I don't actually know off the top
- 57:16of my head how much those accounts cost and what
- 57:19those subscriptions are. I don't use ChatGPT
- 57:22very commonly. Claude, I use, it's a personal
- 57:25preference. The most recent models especially
- 57:27really are, I found to be really good when it
- 57:30comes to legal nuance. So I do know that their
- 57:34heightened security is you pay for different
- 57:36levels of access and different subscription levels.
- 57:38It comes with different security levels as well.
- 57:41So I can't really speak to ChatGPT. um luke this
- 57:46is just what i'm talking about yeah there's a
- 57:48ton of uh of their sanctions left and right and
- 57:51courts around the country are sanctioning lawyers
- 57:53who are not using ai correctly verify the security
- 57:56verify the output and make sure that before you
- 57:59put it in a court pleading you can verify that
- 58:01everything inside it is what you know it's accurate
- 58:03right And then Jack, you're asking if Transcript
- 58:08Genius wipes out all these extensive lit management
- 58:10systems. It's not designed to be your case management.
- 58:12It's not designed to replace case law research.
- 58:16This is a specialized product, right? This is
- 58:17designed to be something that if you use one
- 58:20of the Steno, like if you book a job through
- 58:22Steno, you have access to all of this that allows
- 58:24you to interact with your deposition and associated
- 58:27documents that relate to your case really well.
- 58:30So, and at no additional cost. So to me, A lot
- 58:33of these products can actually exist conjunctively.
- 58:37Ask it to give you legal research terms that
- 58:41you can use in Westlaw or Lexis in different
- 58:43formats. And then those two things kind of are
- 58:45married nicely. Like the different products have
- 58:47different strengths. The industry is probably
- 58:49going to consolidate, I mean, to some degree.
- 58:51But for right now, these specialized products
- 58:53really can occupy a nice role. So if you're booking
- 58:56a deposition and you decide to use Steno, this
- 58:58is just part of the suite of technology that
- 59:00you get. Joseph is asking a question about cost.
- 59:05Joseph, I really want you to reach out so we
- 59:07can kind of talk about this offline. What you
- 59:10saw today isn't costing you anything additional.
- 59:12It's basically book a job through Steno in a
- 59:15competitive price to how you would be getting
- 59:18a deposition transcribed, and then you get this
- 59:20on the back end. But let's talk about that offline,
- 59:23Joseph, and I'll make sure that you follow up
- 59:24on that. And I just see, how do you reach out?
- 59:31Let me, I don't know if we can put a contact
- 59:33in the chat. Maybe I can do this. I'm just gonna
- 59:37put my email in the chat. And I don't know if
- 59:40this can be shared more broadly, but this is
- 59:42my email. Maybe someone else can share this,
- 59:44but we'll put that in the chat, Joseph, so you
- 59:45can get an answer to that question. It's the
- 59:48top of the hour, and I understand that this was
- 59:49built for an hour. I think that there's credit
- 59:51adjustable. So I'm going to hang on until we
- 59:54can ramp it up a little bit. I'm going to answer
- 59:55all these questions because they're good ones.
- 59:57So feel free to keep the questions coming. And
- 1:00:00I'm going to go back to Sedar. What is the best
- 1:00:02practices for providing guidance to experts on
- 1:00:04the use of AI in their reports? I just think
- 1:00:07disclosure. I mean, look, at the end of the day,
- 1:00:10If you're letting people know that you're using
- 1:00:11AI, the expectation is that it's not, it's augmenting
- 1:00:15something, right? Like people, I used AI to get
- 1:00:18something transcribed. And I just wanted to make
- 1:00:21sure that we were very forthright, that AI was
- 1:00:25used in this transcription. And then there was
- 1:00:27a spot for opposing counsel at this hearing to
- 1:00:29question about AI usage. And so disclosure is
- 1:00:33the key. Where people get into trouble, I think,
- 1:00:36is where they pretend that they're not using
- 1:00:37it. And then it's clear that they have been.
- 1:00:39And now sort of the credibility of the analysis
- 1:00:41falls apart. So disclosure, I think, is the only
- 1:00:45way forward. And then the question becomes, how
- 1:00:47is that sort of assessed if that's ever challenged?
- 1:00:50But that's what people are challenging now more
- 1:00:53than ever. And as long as you're using a credible
- 1:00:54product that has security, make sure that the
- 1:00:57person who's testifying has verified the output
- 1:00:59of that. And in their experience and training,
- 1:01:02it's accurate. I don't think that there's a better,
- 1:01:05I don't think there's a playbook outside of that,
- 1:01:07but I think it's a really good question. How
- 1:01:12about evidence that are AI generated, like voicemail
- 1:01:14and videos and photos? I mean, there's so many,
- 1:01:16that's a great question, Nicole. I kind of think
- 1:01:17I was touching on that a second ago. You know,
- 1:01:20I think that there's authentication issues potentially.
- 1:01:22I mean, there's got to be ways. I mean, in Texas,
- 1:01:25you know, once you get up into, you know, rules
- 1:01:27of 900 and above, we're talking about authentication.
- 1:01:30It hasn't been litigated in a way that's that
- 1:01:33I know of just yet in a way that's been binding.
- 1:01:36But I would be challenging if I'm trying to keep
- 1:01:38it out, I suppose the authentication of them.
- 1:01:41How do you authenticate this? That would be an
- 1:01:44interesting process. I think if you are in support
- 1:01:47of authenticating them and you are the sort of
- 1:01:49introducing party who is trying to introduce
- 1:01:52evidence that has been AI generated, again, it
- 1:01:54boils down to disclosure. You know, AI existed
- 1:01:58before like... like we've been discussing this
- 1:02:02in different ways for a while. And I do think
- 1:02:06that authentication is possible. So is challenging
- 1:02:08the authentication, but this is really ripe for
- 1:02:11litigation right now. And I haven't seen anything.
- 1:02:14Granted, my eyes are really focused in Texas
- 1:02:16because that's where I practice. I haven't seen
- 1:02:18anything move the needle on that outside of just
- 1:02:21the authentication side of it. Well, let's see.
- 1:02:26Can Steno be integrated with Zoom AI companion?
- 1:02:29Yeah, well, so Steno has a Zoom app in the app
- 1:02:33store that is the place that you would take your
- 1:02:35deposition virtually. And this allows you to
- 1:02:38annotate all your exhibits, load them up in advance
- 1:02:41so that you have them right there for you to
- 1:02:42introduce. And it's built, again, as a digital
- 1:02:45banker's box that is designed to take virtual
- 1:02:48depositions swimmingly. So that product is part
- 1:02:56of the suite of products that Steno, obviously
- 1:02:57not showing that because there's no real AI component
- 1:03:00to that. But yeah, the Zoom app is fantastic.
- 1:03:07Yes, Joseph, my email is joe .stevens at steno
- 1:03:11.com. J -O -E dot S -T -E -P -H -E -N -S at steno
- 1:03:15.com. If someone who, a host, could put that
- 1:03:18in the chat, I don't know, then go for it, but
- 1:03:20otherwise. All right. Chat is closed. Thanking
- 1:03:27you here. Great. Thank you, Chris. I appreciate
- 1:03:29it. Wow, Luke, this is a huge question. What
- 1:03:34do you make of The Reason Book? If anyone builds
- 1:03:35it, everyone dies. I didn't know that was a book.
- 1:03:41If that is a book, that's amazing. I'm scared.
- 1:03:44I'm an AI user and evangelizer in my practice,
- 1:03:47but obviously the larger repercussions are...
- 1:03:50I don't even know, but it does. There's a scary
- 1:03:53side of it. So if you have my email and you want
- 1:03:55to chat about that, I'd be down to have that
- 1:03:57conversation. Last question that I've got in
- 1:04:00here. I know that sort of we're coming up on
- 1:04:01the end. Confidentiality, my subscription of
- 1:04:05any of these claims to be private edits, letter
- 1:04:08writing and briefing. Well, but how safe if I'm
- 1:04:10putting in clients names and scarier still facts.
- 1:04:13Yeah, you know. The only thing you have to go
- 1:04:16off is the way that these providers are claiming
- 1:04:19that their security holds up. You know, I mean,
- 1:04:23and think of it outside the legal context. Like
- 1:04:25when you're logging on and you're on a bank app
- 1:04:28in your phone and it's supposedly this like very
- 1:04:31isolated encrypted engagement. And once you,
- 1:04:35there's no sort of lingering record of to some
- 1:04:38degree, there's some trust that goes into that.
- 1:04:41I don't know. I mean, beyond what they're presenting,
- 1:04:46it's hard to truly vet. And so I think you're
- 1:04:51kind of drilling in Christine to something that
- 1:04:53also occupies space in my head, which is how
- 1:04:56are these things gonna be litigated? You know,
- 1:04:58when Snapchat says that things disappear and
- 1:05:01then subpoenas go out to Snapchat and suddenly
- 1:05:03they're not gone, what does that mean? I mean,
- 1:05:07so... I don't know the answer to that question.
- 1:05:09I think it's a brilliant question. And it's one
- 1:05:11that gives me pause and maybe Luke pause as well
- 1:05:13from the substance of that book that we was just
- 1:05:16talking about. But that's all I really know how
- 1:05:18to answer on that front is there's there is a
- 1:05:21little bit of faith, which is a weird thing to
- 1:05:23say, especially in a world where we need to be
- 1:05:25a little bit more skeptical that the way that
- 1:05:27they're presenting the security actually is certifiable
- 1:05:30and true and honest and correct. But I wonder
- 1:05:34how that is going to be litigated. And I think
- 1:05:36that the next six months a year are going to
- 1:05:38be really enlightening on that front. All right,
- 1:05:43guys, I am that comes to the end here. I'm going
- 1:05:46to go back to one thing. I'm going to share one
- 1:05:49more QR code for everyone, because I want people
- 1:05:53to be able to scan it. This is going to give
- 1:05:56you information how to get in touch with me.
- 1:06:01And if you have any other questions. So let's
- 1:06:04go to this right now. If you have any other questions,
- 1:06:08feel free to reach out. And thank you all for
- 1:06:15joining today. This has been a delight to actually
- 1:06:17share some space and time with you. And if I
- 1:06:20can ever be of help, scan this barcode, learn
- 1:06:23more about getting in touch with us, ask me,
- 1:06:25ask away, Ari's email is on this screen too.
- 1:06:29So you'll be able to reach out that way as well.
- 1:06:32I think that covers it. Let me make sure. Christine,
- 1:06:36thank you. I appreciate your feedback. This concludes
- 1:06:40today's presentation. Thank you so much, Joe,
- 1:06:43for your time and thank you to our audience.
- 1:06:45Have a great day, everybody.