Latest / CLEs You Actually Want to Hear / AI in Practice: ChatGPT, Claude, and Gemini — 2026 Winter Update
Transcript
- 0:03I'm Mark Hoag, chair of the Beverly Hills Bar
- 0:06Association's recently minted AI and the law
- 0:10section, along with my co -chair, Amanda Harris.
- 0:14I'm the founder of hoaglaw .ai, where I serve
- 0:19as fractional general counsel to AI -first SaaS
- 0:23companies and tech startups. I'm delighted to
- 0:26introduce this morning our speaker, Stephen Smith.
- 0:29Stephen's the CEO of Intelligence by Intent.
- 0:32where he provides AI strategy advisory and training
- 0:35to law firms and professional services organizations.
- 0:39He's delivered AI training and strategic guidance
- 0:42to over 1 ,000 organizations, with a focus on
- 0:45practical adoption, governance, and measurable
- 0:48ROI, rather than just theoretical overviews.
- 0:52Stephen brings over 25 years of executive leadership
- 0:55experience, including at CFO and SVP rev ops
- 0:59roles at multiple organizations. and strategy
- 1:03partner roles at IBM Global Services and TCG
- 1:07Advisors, developing growth strategies for Fortune
- 1:11100 companies in the tech space. Stephen holds
- 1:14an MBA from the Wharton School at the University
- 1:18of Pennsylvania and writes the Intelligence by
- 1:21Intent newsletter at smithsteven .com. And that's
- 1:27Stephen spelled with a P -H, not a V. So thanks,
- 1:31everyone, for being here today. You'll excuse
- 1:33me. I need to bounce off right away, unfortunately.
- 1:36But Stephen, please take it away. Mark, thank
- 1:39you so much. Let me go ahead and jump in. I'm
- 1:41going to share my slides here. So for those of
- 1:45you that have seen me do these presentations
- 1:46before, I don't put it into presentation mode
- 1:50because I am going to bounce around quite a bit
- 1:52and show things on the... on some different tabs
- 1:57and do live examples. I do have the chat Q &A
- 2:00open on the side. If you do have questions as
- 2:02we go through this, please ask questions. I'm
- 2:05happy to answer them. I'll either try and get
- 2:06to them during the presentation or I will get
- 2:09to them at the end of it. So I wanna start with
- 2:12this. As Mark said, there's a newly minted AI
- 2:15group as part of the Beverly Hills Bar Association.
- 2:18And the question I like to ask organizations
- 2:20is, you know, when is the last time that you
- 2:22spent three hours on something that AI can now
- 2:24do in three minutes? You know, the last time
- 2:26I presented in front of this group, I think,
- 2:28was July or so. And these tools have improved
- 2:30dramatically since then. They are absolutely
- 2:33amazing. And I hope that at the end of the session
- 2:35today, you'll see a lot of new use cases and
- 2:37ways that you can be thinking about using AI
- 2:39to benefit your organization. I like to start
- 2:42with this quote. So Scott Kloppert is a partner
- 2:45at Kloppert and Ravdin. He is one of the first
- 2:48uh, law partners that I actually work with extensively
- 2:50and turning to use, starting to use AI a couple
- 2:53of years ago. And I think his quote is just awesome.
- 2:55I mean, he, and he says this and he believes
- 2:57this, you know, he says, Claude is so essential
- 2:58to every activity I do all day long. The only
- 3:01way I would stop using it is if they pride it
- 3:03for my cold dead hands. And most of the people
- 3:06that I work with on a day -to -day basis now
- 3:08have reached that same conclusion. Once you start
- 3:10integrating these tools into your day -to -day
- 3:13activities, it's almost impossible to stop. So
- 3:17One thing I love about this slide is the quote
- 3:20on the right hand side and if you started using
- 3:23chachi BT You know two years ago, right? These
- 3:26tools were trained on everything on the internet
- 3:28and they were basically a really really good
- 3:30answer engine, right? They have improved dramatically
- 3:32since then they have PhD level knowledge in a
- 3:35bunch of areas. They could do incredible reasoning
- 3:37They can win gold medals on you know on math
- 3:40competitions So the key thing I want to mention
- 3:42is like these are the dumbest these tools are
- 3:44gonna be for the rest of your life They literally
- 3:47get better every day. I track adoption in law
- 3:51firms, in AI usage. And you can see some of the
- 3:53latest stats on the right. This is from the middle
- 3:55of 2025. I know these stats have been updated
- 3:59since then. And once they release them, I'll
- 4:00put them in the charts here. But you can see
- 4:03some organizations are as high as basically 50
- 4:05% usage on these tools. And I hope by the end
- 4:08of this, if you're not using them, that you are.
- 4:10I'm going to start using them today. So Mark
- 4:12introduced me. I'm not going to go through this.
- 4:14The only thing I would add to this is I do like
- 4:17the CLE events. I do a lot of keynote events.
- 4:19In fact, I'm speaking at the AAML coming up in
- 4:21March and June. And I just gave a keynote presentation
- 4:25at Humana, a Fortune 50 company back on the East
- 4:28Coast. So what are you going to leave with today?
- 4:32I'm going to talk in the beginning, just about
- 4:34two key concepts. I don't like to go deeply technical
- 4:37around AI and how it works and what tokens are
- 4:40and things like that. But there are two really
- 4:42important concepts we're going to start with
- 4:43that really help you understand these tools and
- 4:46will help you start to maximize value. I'm gonna
- 4:48give updates on some of these tools. There have
- 4:50been huge improvements since we last met in ChatGPT,
- 4:54which is now, you know, the most recent version
- 4:56is 5 .2, Gemini, which is on version three, and
- 5:00Claude, which is on Opus 4 .5. They have added
- 5:03infographic capabilities. In fact, what you're
- 5:05looking at here is an infographic from a very
- 5:07simple description. The ability to generate,
- 5:09you know, mini applications and create timelines
- 5:11and do document workflow, and they can work really
- 5:14well with Excel documents and PowerPoint documents.
- 5:17and Word documents. And again, I'm gonna show
- 5:19you all this over the course of the next hour.
- 5:21But at the end of the day, I hope you walk out
- 5:23with prompts you can use, with ideas about the
- 5:26way you can use these tools and think about the
- 5:28workflow that you do every day. So the two terms
- 5:33that I wanna talk about really quickly to start
- 5:35this off, because I love to spend time in the
- 5:37demos, because the demos is where all the excitement
- 5:39is. The first one is the context window. And
- 5:42I will talk about what that is in a second. And
- 5:44the second thing is hallucinations. And I hope
- 5:46at this point, most of you have heard these terms
- 5:48a little bit, but the context window is effectively
- 5:51just how long of a conversation can you have
- 5:54with AI? And when these tools first started,
- 5:57you know, two years ago, you could have a conversation,
- 6:00you know, that was maybe a couple of thousand
- 6:02words long. Right now you can have conversations
- 6:05that are multiple thousands of pages long. Right.
- 6:07And the best analogy I can give you is these
- 6:10things, as you'll see on the next page, you know,
- 6:12have a certain amount of context that they can
- 6:13have. And the companies talk about tokens, but
- 6:16I'm going to give it to you in sort of words
- 6:17and pages. Cause I think that's easier to reference,
- 6:19but think about it like a desk, right? You can
- 6:21put so many pieces of paper on a desk. You can
- 6:23put so much on. And at some point you're stopped
- 6:26seeing the things that are on there. And that's
- 6:27really what happens to an AI. Once it reaches
- 6:29its limit, it starts forgetting the very first
- 6:31things that you put in. And these numbers are
- 6:34always changing. And so these are the latest
- 6:37numbers from these organizations, at least as
- 6:39of yesterday. So ChatGPT 5 .2, their thinking
- 6:43model, has a context window of 400 ,000 tokens.
- 6:47And if you sort of work your way across here,
- 6:49that's about 300 ,000 words. At 250 words per
- 6:53page, that's about 1 ,200 pages of context length.
- 6:56And then... I actually calculated here something
- 6:58called the conversation length, right? Because
- 7:01these tools don't necessarily always allow you
- 7:03when you're having that in that chat window to
- 7:05access the full context window, right? So in
- 7:08ChatGPT 5 .2 thinking, you can have about roughly
- 7:11700 to 800 pages of conversation before it starts
- 7:14to forget things. With Cloud Opus 4 .5, it's
- 7:17about 600 pages. And with Gemini 3 Pro, it's
- 7:20about 3 ,000 pages. And what I like to do to
- 7:23make it easy is, let me put it in terms that
- 7:25I think everybody understands, right? So think
- 7:27about a Harry Potter book, you know, in terms
- 7:29of benchmarking, right? So the average Harry
- 7:31Potter book is about 487 pages long. So Chat
- 7:34GPT can basically hold the equivalent of about
- 7:36one and a half books, right? Claude, a little
- 7:38over 1 .2 books. And Jim and I can hold almost
- 7:41the entire series. I think there were seven total,
- 7:43right? You can hold six of the seven books in
- 7:46a conversation. So. This is important as you're
- 7:48thinking about tools and as you're loading large
- 7:50documents in, if you've got, you know, pleadings
- 7:52and depositions and video testimony to understand
- 7:55just how much information these tools can hold.
- 7:58And again, this does change all the time. Like
- 8:00I said, Gemini has typically had the biggest
- 8:02about a year and a half ago. They actually had
- 8:04a 2 million token context window and it wouldn't
- 8:07surprise me if we see something that big again
- 8:10later on this year. So the other thing I want
- 8:13to mention that's very, very important is this
- 8:15concept of hallucinations. And in the early days
- 8:19of using these large language models, these tools
- 8:22tended to make up facts. You would ask it for
- 8:24a case, you'd ask it for a citation, you'd ask
- 8:26it to measure something. And they were trained
- 8:28to always try and please you, right? They wanna
- 8:30give you an answer. And when it didn't know the
- 8:32answer, it would confidently tell you, hey, A
- 8:35equals B, even though you knew that A did not
- 8:38equal B, right? So. The organizations have spent
- 8:41a tremendous amount of time trying to minimize
- 8:44hallucinations and with the latest advances in
- 8:47these reasoning models, they've eliminated a
- 8:49lot of it, but not all of it. but it's super
- 8:52important to keep that in mind. And one of the
- 8:54things I will show you is I've actually built
- 8:56a detailed fact checker in ChatGPT, and I'm happy
- 9:00to share with everybody else. I'll show everybody
- 9:02sort of the quick example of how it works as
- 9:04one of the demos. But what it allows you to do
- 9:06is paste in content that you're working on. It
- 9:09will actually go through and try and verify through
- 9:11source documents and Google searches or things
- 9:13you upload to verify that what you've put in
- 9:16there is accurate that you're not, you know,
- 9:17making things up or falsifying things. So again,
- 9:20you still have to be careful around this stuff.
- 9:23And the warning I give people is look, at the
- 9:26end of the day, AI is an unbelievable tool, right?
- 9:29But the legal responsibility sits with you, right?
- 9:32So the most important thing to remember if you
- 9:35don't walk away with anything else from this
- 9:36presentation. presentation is after you use an
- 9:40AI tool, if it's generating content for you,
- 9:42if it's modifying content, you must do human
- 9:45review after that, right? Because you have a
- 9:47bar number, right? You're gonna have to stand
- 9:49before a judge if there's an error there, it's
- 9:51your reputation, it's your malpractice insurance,
- 9:53it's your career built case by case. So AI doesn't
- 9:57care about all these things, AI can still make
- 9:59mistakes. I will say these tools are truly unbelievable
- 10:03now and what they can do is you'll see in a minute
- 10:05here because we're going to jump into demos.
- 10:07But please, please, please remember this and
- 10:09I'll bring this up at the end again. Always do
- 10:11human review after AI modifies documents. That
- 10:14is very, very important. So let me start showing
- 10:18you some of the things that these tools can do.
- 10:19Right. And so I do a lot of work with family
- 10:22law firms and I've been working with some workers
- 10:24comp firms and some bankruptcy firms. So I've
- 10:27got demos from some of these different kinds
- 10:28of organizations. And so one of the things that
- 10:31these tools have gotten great at since last summer
- 10:33is the ability to generate infographics. So what
- 10:37you're looking at here is a prompt that Nano
- 10:40Banana created for me and Nano Banana is the
- 10:43image generator in Google Gemini. And again,
- 10:45I'm gonna show this to you live in a second,
- 10:47but I gave it a prompt about a divorce settlement.
- 10:49And I said, I need you to create a graphic that
- 10:52simply explains what does this settlement look
- 10:54like, you know, for this $850 million marital
- 10:57estate, right? So you can see it broke up the
- 10:59real estate portfolio. It broke up the business
- 11:01holdings, broke up the financial assets. I am
- 11:04not an artist. I was a strategy consultant for
- 11:06many years, but the ability to generate and do
- 11:09this is incredible, as you'll see in a second.
- 11:12It's also great if you want to workflow out concepts.
- 11:14I was working with something to try and explain
- 11:16simply how did this fraudulent scheme work, right?
- 11:19So how did we unravel the shell company network?
- 11:21And so we listed out all the things again, in
- 11:23a very simple prompt. It was just a text prompt
- 11:25saying, Here's the relationships that we know
- 11:27of. Can you make an image of this? And the nice
- 11:29thing is, as you'll see, you can change the image
- 11:31style. You can make it in any style you want,
- 11:34right? So this was a little bit of a more formal
- 11:36style. This one was obviously a Harry Potter
- 11:38style. This one up here was even a different
- 11:40style yet again. So it's very tunable into whatever
- 11:43style and format you wanna present either to
- 11:46the clients or the court or wherever you're working.
- 11:49And so here's a real world example, right? From
- 11:51a workers comp organization. So there is a, you
- 11:54know, from the defense perspective, there's a
- 11:56timeline in medical history of an alleged injury.
- 11:58And I will show you this in a second, the actual
- 12:01raw starting data for this. But what we cared
- 12:03about is, hey, how do we explain to a judge or
- 12:07a jury? What is the timeline of activities here?
- 12:09What happened, right? So going back to 2014,
- 12:13there was an initial non -industrial back injury,
- 12:15moving furniture. We have more things in August
- 12:17of 2019, right? We go all the way through to
- 12:21September, November, 2023, all the way through
- 12:23to up and through to today, right? Super easy
- 12:27to do this and there are different ways to visualize
- 12:30visualize this and do this but I will show you
- 12:32again Real -world examples of this because it
- 12:34has become an incredibly incredibly powerful
- 12:37tool when you're trying to create these visualizations
- 12:39So and I'll give you a real example right now
- 12:42So that image I showed you at the very beginning
- 12:45here this one, you know When was the last time
- 12:47you spent three hours on something AI could do
- 12:49in three minutes? The prompt for that was literally
- 12:52this prompt here I said, I'm presenting an AI
- 12:55update to the BHBA, and this is gonna be the
- 12:57opening provocative question on a slide. You
- 12:59know, can we create a content? So, and then I
- 13:02did a different one, said, do it in the style
- 13:03of the golden age of air travel. So I'm gonna
- 13:06just grab this prompt. I'm gonna go to Gemini
- 13:08here. And if you have not used Gemini, they've
- 13:12got a tools menu just like ChatGPT does or Claw
- 13:14does. And there's an ability here to create images.
- 13:16I'm gonna say create images. I'm gonna dump in
- 13:19the prompt. I'm going to change it from the style
- 13:20of the golden age of air travel to be, oh, let's
- 13:24say 1950s, you know, home robotics. I don't know
- 13:30what that looks like, but we'll find out. And
- 13:33I'm just going to go ahead and say go. and it
- 13:36will go off, and it will actually now start creating
- 13:37that image for me. And the one thing that's nice
- 13:39about Gemini and NanoBanana is they are quite
- 13:42fast, right? You could paste in a simple prompt,
- 13:45a lengthy prompt, but it will go off and it will,
- 13:47you know, you can see it's loading NanoBanana,
- 13:49it's refining the details, and in a second here,
- 13:51it's gonna go ahead and give me that image. One
- 13:54of the things they've added, I don't wanna interrupt
- 13:55the image creation here, is you'll see on the
- 13:58left -hand side here a thing that says My Stuff.
- 14:00Right. And I can go in there, but they've actually
- 14:02shown you your library now. So here is this image
- 14:05that I created from that single prompt. You know,
- 14:07one is the last time you spent three hours on
- 14:08something AI could do in three minutes, the BHPA
- 14:11and AI update, you know, with 1950s home robotics.
- 14:14So again, a simple and kind of whimsical example,
- 14:17but I just want to show you how easy it is now
- 14:19to start using some of these tools. So the case
- 14:24I was talking about a second ago, this data set,
- 14:26right? So we made up this fictitious data set
- 14:28for someone named Arthur J Pendleton. and that
- 14:31data is here as you can see on this right there
- 14:34is a long long document a history of all the
- 14:37medical appointments and all the things that
- 14:38they did and what we wanted to do was basically
- 14:41create a timeline of that. How do we show a timeline
- 14:44to a jury or to another set of attorneys or a
- 14:48judge about what's really happening here? And
- 14:50so there's two options. One is, and this was
- 14:53the first place we started, we could say, hey,
- 14:54let's go create an infographic. So if I just
- 14:56go grab this text here, and if I go back in here
- 14:59and say I'm going to do a new chat, and I'm going
- 15:02to say, create an infographic that describes
- 15:05the medical history for this person. And I'm
- 15:09gonna paste it in here. And one of the things
- 15:10you'll notice is I often do these demos. I talk
- 15:12to my computer all the time. I use a great product
- 15:14called Whisper Flow. If you haven't used it,
- 15:17it does amazing work at doing dictation and grabbing
- 15:20things and translating what you're doing into
- 15:22basically perfect English. So I'll let that go
- 15:26for a second. And so you can see here. It went
- 15:29and created, you know, a different style, but
- 15:31a timeline again, 2015, 2016, 2019, but it kind
- 15:35of goofed up the dates here. You know, I probably
- 15:36should have told it, hey, make sure it's in sequential
- 15:38order, cross check that information, do things
- 15:40like that. I didn't, but I could go create it
- 15:42again and again and again. But one of the things
- 15:45that's really powerful in these tools is that
- 15:48doesn't give me necessarily the flexibility that
- 15:50I want to see. Right? There's actually a better
- 15:53way to do this. And so with these tools, with
- 15:57ChatGPT or Claude or Gemini, you can actually
- 15:59create interactive tools from this kind of data.
- 16:02So the prompt I used is I basically told Claude
- 16:06and Gemini, go use HTML or react, which is some
- 16:10of their internal tools and create an interactive
- 16:13timeline that allows for scrolling horizontally
- 16:15because the default is usually to scroll vertically,
- 16:17but I want this to scroll horizontally. So make
- 16:20it a timeline that has a center line with activities
- 16:22on the top and bottom, you know, and alternate
- 16:24them for readability. So it actually takes several
- 16:27minutes to go create this. I want to just show
- 16:29you what this output looks like. If I go into
- 16:31Claude here and I go into, I think it was this
- 16:34one, this artifact window here. Let's see if
- 16:36it actually was. If I remember the right one,
- 16:39I did. So if you look at this here, again, the
- 16:44prompt I gave it was literally that prompt is
- 16:47I said, you know, use HTML or React to create
- 16:49this interactive timeline. And I just paste it
- 16:51in all those doctors events, right? It took a
- 16:54little bit of prompting for a second here for
- 16:56me to get exactly what I wanted. But what you
- 16:58can see here is I now have, and here I'll make
- 17:01this a little bit bigger so you can read it here.
- 17:03I now have an interactive timeline that I can
- 17:05kind of walk through year by year and see what's
- 17:08going on here, right? It even color coded different
- 17:10kinds of, you know, primary care visits, specialist
- 17:12visits, physical therapy visits, imaging visits.
- 17:15So as a way to visualize data, I mean, it would
- 17:17have taken, you know, me as an analyst or, you
- 17:20know, a group of paralegals quite a bit of time
- 17:22to do this. This is a much better way to do this.
- 17:25And all of these tools now allow you to do this
- 17:28stuff, to sort of go through and build these
- 17:30really kind of interesting and interactive examples.
- 17:32They're really great. So another one I want to
- 17:35show you that was another really good interactive
- 17:37example was a family law calendar. So I was actually
- 17:41presenting to a firm a couple of weeks ago. And
- 17:43the question was, Hey, we have a hard time sometimes
- 17:47creating these real time instant. interactive
- 17:50family a lot of calendars just to show people
- 17:51hey how much time is mom gonna get how much time
- 17:53is dad gonna get if we separate the organization
- 17:55this way so you know our first attempt was hey
- 17:58let's go dump it into gemini and let's actually
- 18:01have it create an image in fact i think if i
- 18:03go look at my stuff here it might even just pop
- 18:06up look if i pop over here just looking to see
- 18:09if i saw it um here we go so we dumped in that
- 18:14exact same prompt Well it doesn't want to load
- 18:16figures the one time I want to have it go do
- 18:18something historical it doesn't do it but it
- 18:20basically it didn't show a perfect example of
- 18:23what it was trying to do it really basically
- 18:24said hey I'm gonna make an attempt but. creating
- 18:27infographic from that wasn't the right thing
- 18:28to do, but saying, Hey, will you go build code
- 18:31and allow me to do this? And I'm going to click
- 18:33here. It actually went ahead and created for
- 18:35me an interactive family law calendar. And the
- 18:38nice thing is, is I can actually now share this
- 18:40with others. I can share with other people in
- 18:42my firm, but you'll see here, it basically said,
- 18:44Hey, here's the June distribution. That's 53
- 18:46% mom, 47 % dad, right? If I want to set something
- 18:49else as dad, I can click set dad and let's give
- 18:51dad that day. Hey, now we're at 50, 50, right?
- 18:54So again, Literally from this one single simple
- 18:58prompt, these tools can now build interactive
- 19:01applications that you can go back and access
- 19:03and then leverage internally with your clients,
- 19:07things like that. So again, just another really
- 19:10good example of the kind of things that you can
- 19:12do here. So I've got tons of these examples.
- 19:14And again, I'm watching the webinar chat. So
- 19:17if people have questions, comments, if they want
- 19:19to see live examples, please let me know. But
- 19:22I want to go into a bunch more, you know, real
- 19:24world practical examples today of the kind of
- 19:26things that you're going to see. And so one of
- 19:29the things again, in family law is sometimes
- 19:31we have to go find, you know, we have to get
- 19:33a financial declaration inconsistency detector,
- 19:35right? So the challenge is opposing parties routinely
- 19:38underestimate income and overstate expenses.
- 19:41Surprise, surprise. Catching inconsistencies
- 19:43requires cross -referencing tax returns, bank
- 19:46statements, and lifestyle evidence, which, as
- 19:48most of you know, can take a tremendous amount
- 19:50of time, no matter what kind of case you're working
- 19:52on. So by the way, I'm not gonna read all the
- 19:55text and all the slides on these pages. I'm just
- 19:56gonna do this one a little bit more in detail.
- 19:59And as I go through these other examples, I will
- 20:01just sort of show you what they're doing and
- 20:03how they're working. So... The prompt, and this
- 20:07is where I say I've given you guys a bunch of
- 20:09prompts in here. So there are family law prompts,
- 20:11there are workers comp prompts, there are general
- 20:12prompts, there are all sorts of them as you'll
- 20:16see here, right? But the prompt is, hey, don't
- 20:18analyze the opposing party's financial documents
- 20:20for inconsistencies and hidden income, right?
- 20:22I'm gonna give you a bunch of documents that
- 20:23are provided. I'm gonna give you the known lifestyle,
- 20:26you know, the value, the rent, the vehicles,
- 20:28children's schools, things like that. And this
- 20:30is what I want you to do. And so again, even
- 20:32a year ago, it would have been very hard for
- 20:34these tools to do this amount of analysis at
- 20:36one time, right? To go through and find income
- 20:39discrepancies, right? Compare the W -2s, the
- 20:411099s declared income, identify unreported schedule
- 20:44C or K -1 income, calculate average monthly.
- 20:47you know, income. Look for expense discrepancies,
- 20:50right? Compare declared expenses against actual
- 20:52bank credit card spending, identify understated
- 20:53categories. Look for the lifestyle gap, right?
- 20:56Does the declared income support the apparent
- 20:58lifestyle? You know, is there a gap there? Can
- 21:00you help me calculate that, right? Find any red
- 21:02flags. Were there any transfers to unknown accounts,
- 21:05right? Are there round number deposits not matching
- 21:07paychecks? Are there missing accounts referenced
- 21:08but not disclosed? And then discovery recommendations,
- 21:12you know, specific follow -up requests and subpoenas
- 21:14based on the findings. Apple, this is a chart
- 21:17showing declared versus actual for use in court
- 21:21or negotiations. So I'm gonna grab this. I'm
- 21:23gonna actually go run this live in two different
- 21:27sessions right now. I'm gonna go run this in
- 21:29ChatGPT and I'm gonna go grab here some sample
- 21:33files that I've got here to make this work. So
- 21:36this is number one, financial declarations. I'm
- 21:39gonna go grab, so there's pay stubs, there's
- 21:41known lifestyle evidence, there's income and
- 21:42expenses, bank statements, things like that.
- 21:44I'm gonna just hit open here. I'm gonna go ahead
- 21:47and paste this prompt in. And I'm gonna tell
- 21:50it to go ahead and go. And then I'm gonna go
- 21:52launch the same thing over here in Gemini. I'm
- 21:55gonna upload some files here. I'm gonna grab
- 21:57all these. I'm gonna do a slightly faster model
- 22:01than their most advanced one, just for purposes
- 22:04of speed. I'm gonna say go ahead and go. While
- 22:07those are running though, I wanna show you something
- 22:09in Claude. So Claude has done an incredible job
- 22:14now at incorporating the ability to work with
- 22:16Word documents and PowerPoint documents and Excel
- 22:18documents. And so it takes longer when you want
- 22:22to have the output really highly formatted in
- 22:25a tool like Claude, but the output you get is
- 22:27great. So I've actually pre done some of these
- 22:29just for purposes of speed. So this is the financial
- 22:31discrepancy analysis for litigation. So I'm just
- 22:33going to go ahead and open this up here. So I
- 22:35ran these prompts last night. I'm going to click
- 22:37on this here. And what you'll see is it went
- 22:40through here. and created a nicely formatted
- 22:43financial discrepancy analysis, right? Here's
- 22:45the opposing party, here's the case number, here's
- 22:47the superior court where it is, here's an executive
- 22:50summary, right? Total monthly income was understated
- 22:52by this amount, actual monthly gross income versus
- 22:56declared, monthly expenses understatement, hidden
- 22:58income source, undeclosed 1099 consulting, right?
- 23:01Again, you would have a team of folks going through
- 23:04and looking for this, maybe first years or second
- 23:06years or a paralegal. With these tools now, you
- 23:09can easily start to go in and find this information.
- 23:12So then it went through and it broke down, hey,
- 23:14income discrepancies, right? W -2 income versus
- 23:16declared income, undisclosed 1099. And it's actually
- 23:19giving me all the details here. So one of the
- 23:22things, like I said, that Claude has done is
- 23:24they've introduced this thing called skills and
- 23:26they work really, really well now with, as I
- 23:28said, you know, Word doc formats. And you can
- 23:30save it straight out. If I download it, I can
- 23:32save it right as a Word doc. or as a PDF, I can
- 23:35do things like that. I'll scroll to the bottom
- 23:37here. You know, red flag is a suspicious activity,
- 23:39you know, undisclosed income streams, right?
- 23:42Discovery recommendations, document subpoenas,
- 23:44third party subpoenas, interrogatories, deposition
- 23:46questions. And then here's a, you know, a nice
- 23:48summary chart of what it found here. So again,
- 23:51I just wanted to highlight. the kind of format
- 23:53and the kind of analysis you can get from these
- 23:55tools now. So ChatGPD is going through, it's
- 23:58analyzing these incomes. And again, these things
- 23:59take a while to go do, right? So you can see
- 24:01here though, it's finding the declared income
- 24:04discrepancies, right? So it's gonna go through
- 24:06and document all this stuff. Gemini is going
- 24:08through and doing the same thing, your declared
- 24:10versus actual salary. So all these tools can
- 24:13do this and just sort of the output formats are
- 24:16a little bit different in terms of what it's
- 24:17creating here and doing. But again, I wanted
- 24:19to sort of see you or have you see a live real
- 24:22example like that. It's really interesting. So
- 24:25another good example, and this is a bankruptcy
- 24:27example is right, you know, determining the best
- 24:30chapter requires, you know, analyzing income
- 24:32assets, debts and goals, explaining trade off.
- 24:35So clients can make informed decisions, right?
- 24:37If a client is trying to decide between chapter
- 24:39seven or chapter 13, right? They're going to
- 24:41look at things like, you know, what assets do
- 24:43you have? What debts do you have, right? You
- 24:45know, which things are secured? What's the priority?
- 24:47What are your unsecured assets? And then go through
- 24:49and analyze, you know, create a, you know, a
- 24:51means test, you know, does client pass chapter
- 24:53seven? If not, you know, projected disposable
- 24:55income question mark, right? So using Arizona
- 24:58exemptions, any non -exempt equity at risk for
- 25:00asset protection, right? What's the chapter seven
- 25:02outcome? What's the chapter 13 outcome? So again,
- 25:05I'm leaving these prompts with you guys to give
- 25:07you a sense of what these tools can do. I'm gonna
- 25:10go ahead and run this again. Let's close this
- 25:13one. Let's go back into ChatGPT and I'm gonna
- 25:16fire it off again here. And while these are going,
- 25:19so I wanna again show you the differences in
- 25:21some of these tools. Let me grab these. Paste
- 25:25this in and I'm gonna say go and Make sure it
- 25:30doesn't ask me any questions and I'm gonna go
- 25:31ahead and I'm gonna run this again in this one
- 25:35Let me go grab some files here We got these Let's
- 25:41do thinking. Let's paste this in Let's say go
- 25:44and again while those are running and I'll show
- 25:46you the output I actually want to go show you
- 25:48now again the output that I ran in Claude because
- 25:50again the you know the formatting is just it's
- 25:53just different right and it looks quite nice
- 25:55in this case I didn't ask it for a word doc right
- 25:57so this should look pretty similar to what we
- 26:00saw in what we'll see in Chachi BT and in Gemini
- 26:03right but here's the Martinez bankruptcy analysis
- 26:06right the means test analysis so it's going through
- 26:08and looking at all the different things so this
- 26:11was a case in Arizona What do they have? How
- 26:14much is protected? How much is exposed? Things
- 26:15like that all the way down. So the chapter seven
- 26:18outcome projection, what happens, right? The
- 26:21chapter 13 outcome projection for comparison.
- 26:23And at the end, you know, it says Martina's family
- 26:26cannot afford a chapter 13 plan. They would need
- 26:28to find over a thousand month of additional income.
- 26:30If it were feasible, it could look like this.
- 26:32So again, these tools, the new reasoning capability
- 26:35in these tools really drives through a lot of
- 26:38differences and lets you get a lot more insight
- 26:41into this stuff. So it's going through, analyzing
- 26:44the same thing. So it's getting the data for
- 26:47the Williams family and just doing all these
- 26:49different things in here. So again, I'm not going
- 26:51to let all these build. One of the reasons I
- 26:53wanted to pre -build some of these, particularly
- 26:55in Claude, was to be able to show you the output
- 26:57and show you what some of these things can look
- 26:59like. So let me do another live example. And
- 27:04this is a great example. And let me just take
- 27:06a quick pause for a sip of water here before
- 27:08I do this. And I'll let you look at the screen.
- 27:14So one of the things that is really unique and
- 27:18really helpful for attorneys is to be able to
- 27:20do, you know, video testimony analysis and a
- 27:23lot of times, you know, audio analysis. And these
- 27:27tools, and this is really a focus on Gemini in
- 27:30particular and Google's model has gotten so good.
- 27:33So Google's model Gemini is natively multimodal.
- 27:36And what that means is it natively understands
- 27:38video and audio and images and text. and is absolutely
- 27:43incredible at being able to understand what's
- 27:46happening. So you can actually upload a video.
- 27:48So I'm going to show you this. I'm not going
- 27:50to upload the video because it would take a long
- 27:51time to process, but I have it sort of pre done
- 27:53here where I can say, look, I'm a trial lawyer.
- 27:56I'm analyzing the 60 minute video deposition
- 27:58of a plaintiff in a workspace discrimination
- 28:00case. You know, I've uploaded a video which shows
- 28:03the different, you know, alleging harassment
- 28:04by a supervisor, but hesitating when asking for
- 28:06dates. So I've uploaded all this stuff and I
- 28:08had to analyze the video and you can see sort
- 28:10of the summary here. But what I care about and
- 28:13one of the things you can do is you can actually
- 28:15ask the tool something like hey Find the times
- 28:18in this video when this witness became evasive
- 28:20or when they became angry or when they changed
- 28:22their message So it's not just looking at text.
- 28:25It's actually looking at the video itself So
- 28:27if I go load up this video, so you'll see here
- 28:30that this was literally an hour and one minute
- 28:32long Right, and it took 370 ,000 tokens. So if
- 28:37I went back for a second and I looked at this
- 28:39token list, this chart that we had, right? You
- 28:43could theoretically load it into ChatGPD 5 .2,
- 28:46but it doesn't analyze video nearly as well as
- 28:50Gemini 3 does. You could not load it into Claude.
- 28:52And by the way, Claude really doesn't analyze
- 28:53video like the others do. But Gemini Pro, Gemini
- 28:563, actually has the ability to go do that. And
- 28:59so let me just jump back to kind of where I was
- 29:01for a second. But I want to show you so I can
- 29:03say I loaded this video, identified it, and I
- 29:06just said, hey, identify every time in this deposition
- 29:09where the person being interviewed was evasive
- 29:11in their testimony. So it went through. And again,
- 29:14I'll make this a little bit bigger here. It said
- 29:16the following timestamps identify moments in
- 29:18the deposition. where the witness, Mr. Patrick
- 29:20Dennis Paul, appears evasive, vague, or, you
- 29:23know, um, you know, feigns lack of recollection
- 29:25regarding specific events. And so he gave me
- 29:26the exact timestamps and he gave me a description
- 29:29about, you know, what he said and what he's doing.
- 29:31And again, so this is actually based on it's
- 29:34looking at the image. It's looking at his tone
- 29:36of voice. It's looking at the stress in his face
- 29:38and answering questions that way, not just from
- 29:40the text. The other thing you can easily do is
- 29:44you could actually go through and create a, um,
- 29:47a text summary. Of this, I don't have it in this
- 29:50one, but I did it in a different one where I
- 29:53took an audio file. This was actually from a
- 29:55YouTube video I was watching yesterday called
- 29:56the AI superiority and software architecture.
- 29:59And I just said, give me a word for word transcript
- 30:01of this file, right? So again, these tools have
- 30:04gotten so good. I mean, they're not perfect,
- 30:06but the error rate is generally less than 2 %
- 30:09now on the most advanced models. And they are
- 30:11very, very fast when you ask it to go do this.
- 30:14I mean, this whole thing took. Let me see if
- 30:16it tells me how long it took to go do this. I
- 30:18don't remember if it shows me the timestamp on
- 30:20here. Yep. So it took 50, you know, 59 seconds
- 30:25to basically go through and do a 24 minute audio
- 30:29file to give me a word for word transcript. And
- 30:32so if I go through and look at it, right, it
- 30:34literally is all this, you know, all this here.
- 30:36And one of the things that not everybody knows
- 30:38is let's say for example, that you wanted to
- 30:40now, Have this in Korean or if you wanted to
- 30:43have this in Japanese you can actually ask the
- 30:45model just quite simply Translate this word -for
- 30:48-word Transcript to I'm gonna say Korean if I
- 30:52can type right here. See our IPT to Korean. Oh
- 30:57Let's get transcript right here. Even though
- 30:59the reality is you generally don't have to fix
- 31:01errors It knows what you wanna do, but I'm gonna
- 31:04go ahead and fix it here. I'm gonna say run.
- 31:05We'll let it go for a second and then we'll come
- 31:07back to it. But it's gonna go through now and
- 31:09actually translate this. So one of the things
- 31:11is if you're working in cases and you have documents
- 31:14from multiple different languages. One of the
- 31:16things you can do is you can actually go through
- 31:19and have these things read images and text and
- 31:22translate multiple languages. So again, you can
- 31:24see just how fast this is at going on and running
- 31:27this translation. I mean, it is extraordinarily
- 31:30fast and we could even then take it and translate
- 31:32it to another language from this and we can go
- 31:34back and forth and things like that. So. I was
- 31:39just showing you can create audio transcript
- 31:40from a video file. I was doing it from that audio
- 31:44one, but I could have gone through and said,
- 31:46hey, give me the word for word transcript from
- 31:47this video file. I won't do it for purposes of
- 31:50time, because I want to spend some more time
- 31:52on some of these other more interesting examples.
- 31:54And I was just looking at the questions really
- 31:56quick. I don't see any yet. But again, if you
- 31:59guys have any, or if you have ideas or things
- 32:01you'd like to know about, please let me know.
- 32:04So here's another really good general use case
- 32:06for these AI tools. Right and a lot of times
- 32:09and I do a lot of these with my organization.
- 32:11I want to create custom partnership agreements,
- 32:13right? So I'm a corporate lawyer I'm drafting
- 32:15a partnership agreement between two individuals,
- 32:17you know starting a small tech firm in California
- 32:19I want them to go ahead and create, you know
- 32:22a starting draft document for me right and the
- 32:24reality is This isn't gonna be the final version
- 32:27you get right? You're not gonna get something
- 32:28perfect from day You know from the very first
- 32:30prompt that you put in there, but you will get
- 32:32something very very good And the thing that I
- 32:35want to highlight here, because I'm not really
- 32:36going to focus on prompting techniques today,
- 32:40is really focus on giving the AI tools as much
- 32:44detail as you can, right? So I'm saying, create
- 32:47a comprehensive partnership agreement, include
- 32:49the names of the partners, purposes of the partnership,
- 32:53how much money they're putting in, dispute resolution,
- 32:55non -compete clauses, ensure that the language
- 32:58is precise, legally enforceable under California
- 33:00law, tailor to the tech industry, include boilerplate
- 33:03clauses, and you may have more that you like.
- 33:05I'm gonna go ahead and grab this here again.
- 33:07Cause again, you know, showing these live. Oh,
- 33:09if I go back to this other bankruptcy one, you
- 33:11can see here that again, Chachi BT, you know,
- 33:13did a really good job of going through, um, you
- 33:15know, how to go through the, uh, the different
- 33:17means tests and things like that. And so, and
- 33:19again, this one had the same conclusion, you
- 33:21know, feasibility problem that you're going to
- 33:22run a deficit on chapter 13 outcomes. So if I
- 33:26go paste this in here, and by the way, one thing
- 33:29I want to mention to you, if you're using Chachi
- 33:32BT, um, and I'm not really going to talk too
- 33:35much about privacy today, but please make sure
- 33:37that privacy is turned off on these tools. In
- 33:39the model selector up here, you have a set of
- 33:41choices, right? You can use auto, which allows
- 33:44ChatGPT to think, you know, to decide how long
- 33:46it wants to think. You can use instant. You can
- 33:50use thinking, which is the one that I recommend.
- 33:52So thinking means always use your reasoning model,
- 33:54right? If you use auto, it may choose to use
- 33:58a thinking model. It may try to use a less smart
- 34:00model and give you just a really quick answer.
- 34:03My view is you want these tools to think for
- 34:06you as much as possible, right? Don't let them
- 34:08sort of shortcut the process and go short and
- 34:11not do things, right? You want them to help you,
- 34:13you want them to think. So I tell all my clients.
- 34:16Always, always, always do thinking. Now I happen
- 34:19to have a subscription. They have something called
- 34:21a pro subscription for on the consumer side for
- 34:23$200 a month that also has something called pro
- 34:25here, which is their absolute most advanced model.
- 34:28But again, use thinking. And then if you look
- 34:31at here, you actually have a dropdown. You can
- 34:33actually tell it how much thinking to do my default.
- 34:36So all of, you know, for purposes of demo today,
- 34:38I'm doing standard thinking, which is a little
- 34:39bit faster. I always either use extended thinking
- 34:43or heavy thinking when I'm doing analysis. In
- 34:45fact, If you use heavy thinking and you give
- 34:47it detailed writing instructions, you can get
- 34:50absolutely phenomenal writing output that sounds
- 34:52very, very like a human being, right? So my default
- 34:55is I must use heavy, but again, for purposes
- 34:58of speed and demo today, I'm using standard here.
- 35:01So I'm going to go ahead and say, go, um, I'm
- 35:03going to just run this in all of them. I will
- 35:04just sort of, you know, plop this in. So you
- 35:06can see these things here and let's go into Gemini.
- 35:09Let's do the same thing here. And again, I'll
- 35:12use Pro on this one, which is their most advanced
- 35:14model. So I'm kind of going to compare now the
- 35:16output that you're going to get from ChatGPT
- 35:19and from Claude and from... from Gemini to go
- 35:23ahead and take a look at the sample agreement.
- 35:25So it's going through, it's drafting it carefully.
- 35:27And so the thing is here is what you'll see now
- 35:29is these models are showing you how it's thinking,
- 35:31right? It's outlining partnership agreement details.
- 35:33You can actually hit the dropdown and see on
- 35:35the right hand side here. You can actually see
- 35:37all the thinking it's doing. You can see that
- 35:38it's searching, by the way, on the web to make
- 35:41sure it has very current and good answers. And
- 35:45this is what I talk about in Claude using documents.
- 35:49So it's actually reading something called the
- 35:51doc X skill, the word doc skill, and it's gonna
- 35:54create a properly formatted word document. So
- 35:56a lot of times I will tell it, please don't use
- 35:58that skill, cause I want it to go faster. But
- 36:00what you'll see here is it will actually create
- 36:01a beautifully formatted nice document, which
- 36:04again, you can download as a word doc, which
- 36:07is quite nice. So, and here Gemini is already
- 36:10done. Here's a journal partnership agreement,
- 36:12right? Article one formation and name. purpose,
- 36:14capital contributions, right? Again, it's a great
- 36:18starting point when you wanna create these things
- 36:19in the models. So let's see how ChachiPT is.
- 36:23All right, so ChachiPT has created a comprehensive
- 36:25partnership agreement as well. Right? We've got
- 36:27all this information here and all these things.
- 36:30And one of the things that I didn't do, in fact,
- 36:33I'm going to show it. I'll add it to this because
- 36:35it's a really good example, is ChatGPT has something
- 36:38called a canvas window. Cloud has something called
- 36:40an artifact window. And Google also has a canvas
- 36:43window. Normally when you're having a conversation
- 36:46like this in chat GPT, right? It's a threaded
- 36:48conversation. You're going back and forth and
- 36:50it goes and updates the next thing. But a lot
- 36:52of times you might want to only just work on
- 36:54one section of the document, right? You might
- 36:56say, Hey, you know what? I really want to go
- 36:58work just in that IP section and in section seven.
- 37:01So one of the things that, um, Chat GPT has here
- 37:05is called the canvas window. So when this is
- 37:08done, I'm gonna tell it to just open up this
- 37:10document in the canvas window and hopefully it
- 37:12will do that. Sometimes you have to create it
- 37:13initially in the canvas window. But what the
- 37:15canvas window does is it allows you to go and
- 37:18actually edit just portions of the document without
- 37:21editing all the rest of it. So you can say, I'm
- 37:23just gonna go work on chapter seven. I wanna
- 37:24just change this one term or do this. And I can
- 37:27rewrite it in a particular style or I can make
- 37:28it longer or shorter. So I'll let this finish
- 37:31here. I'll see if I can get Gemini to do it.
- 37:33I've never actually tried to do this with Gemini,
- 37:35but I'm gonna just go ahead and say, hey, open
- 37:38this up in the canvas window so I can edit it.
- 37:40We'll just see what it does and we'll see if
- 37:43it happens real time. We'll see if Claude is
- 37:46still working on the beautiful version here.
- 37:48And okay, this is almost done here. A very long
- 37:53document. And again, this is why I use thinking
- 37:55because you get, you know, super, super good
- 37:58detail on all these things. All right, ChatGPT,
- 38:01finish up here. All right, well, here's an example.
- 38:03So you'll see here in Gemini, you'll see here
- 38:05I was now opened up a document on the right -hand
- 38:08side, which is great. And you actually have formatting
- 38:11and you can actually, you know, you can share
- 38:13and export this. You can actually, from this,
- 38:15you can create infographics and web pages and
- 38:17other things. But you can actually highlight
- 38:19sections of the document right here, like article
- 38:22one formation name, and just say, ask Gemini.
- 38:25And I can just say, add three more terms, three
- 38:27more key terms here. So all of the canvas documents
- 38:32work the same. What it's gonna do is it's gonna
- 38:34just work on that one section of the document.
- 38:37So I know I've showed this to some organizations
- 38:38in the past, but this is a super powerful way
- 38:41when you're editing a long document to just focus
- 38:43on a piece of it. That also saves tokens because
- 38:46it's actually not rewriting everything. So you'll
- 38:48see here, everything else stayed the same, but
- 38:50it now added these three new terms here that
- 38:52I've got in that chapter. So we'll go through
- 38:56here. I won't do it since I showed it to you
- 38:57there. And I'll just show this pop up here. We'll
- 39:01see what this looks like for Claude. So again,
- 39:04it actually made it a nice Word doc. If I were
- 39:06to hit Save here, it actually saved it as a doc
- 39:08x already, right? Did a great job of formatting
- 39:11and indentation. So again, Claude has come a
- 39:14long ways in terms of its ability to go do that
- 39:16stuff. So one of the things that we can do, and
- 39:20I want to be cognizant of time here, is I'm going
- 39:22to go through a couple more quick examples and
- 39:23wrap up on some privacy things and some next
- 39:25steps. One of the great things these tools are
- 39:28good at is extracting information from financial
- 39:30statements. So sometimes you have to open up
- 39:32credit card statements or 10K filings or tax
- 39:36returns. These tools are amazing at pulling things
- 39:39out. So you can see here, I'm gonna actually
- 39:41have it extract some information out of a 10K.
- 39:43So I've got a copy of Apple's 10K. I'm gonna
- 39:46grab this prompt here. I'm gonna go ahead and
- 39:48say go. And in this case, I'm actually gonna
- 39:51do. I'm gonna actually let it run as fast as
- 39:54it can, because this is a relatively simple example
- 39:56for one of these AI tools to do. I'm gonna grab
- 39:59Apple's 10K. I'm gonna go ahead and just say,
- 40:02hey, go extract all this information out of here.
- 40:05And I will go fire this off also in Gemini. And
- 40:09again, I'm gonna use, I'll use thinking, which
- 40:11is sort of their middle speed model here. Let
- 40:13me go upload the 10K again. I'm gonna go ahead
- 40:16and say, go. And, you know, one of the things
- 40:18you might be noticing is some of these tools
- 40:20respond more quickly than others, right? They're
- 40:22always trying to optimize for speed and doing
- 40:23things. And so, you know, your mileage may vary.
- 40:26I have personally found that Gemini, even the
- 40:29most advanced models, as you can see, are very,
- 40:31very fast these days. Google has done a really
- 40:34good job of optimizing these tools for speed.
- 40:37And obviously Claude is, because it's actually
- 40:40creating these very detailed Word documents,
- 40:42is doing some other things there. So we'll let
- 40:44this run. So it's going to do a clean extraction
- 40:46on this. But you can see that it's actually going
- 40:48through. It's creating a really nicely formatted
- 40:50table. And again, my point is I'm trying to give
- 40:52you real old examples of ways to use these tools
- 40:55and show you really good prompts that let you
- 40:58maximize the value of what these things can do.
- 41:01So here's another family law example I'm going
- 41:04to jump into. Let's say you got to do your guideline
- 41:07support for child support. We want to do some
- 41:11deviation argument building. And the reality
- 41:13is deviations are you know, are available. Building
- 41:17a persuasive argument requires evidence, statutory
- 41:19factors, and clear math. So I'm going to again
- 41:22say, hey, look, here's the build arguments for
- 41:24child support deviation in this California case.
- 41:26Here's the guideline calculations, you know,
- 41:28payer's net disposable income, payee's net disposable
- 41:30income, right, the deviation support, analyze
- 41:33the legal frameworks, the factor by factor analysis,
- 41:36the math for it, evidence checklists. And so
- 41:39I'm going to grab this. I'm going to go pop this
- 41:41into Gemini. And while it's running, I'm actually
- 41:44going to pull up, because I've already run this
- 41:45in Claude, again, to show you what this looks
- 41:49like here. So if I go grab PHP demo files, this
- 41:54is the child support one here. I'm going to grab
- 41:57these files as a demo. Because there's a couple
- 42:00other cool demos I want to get to in a few minutes
- 42:02here that I think you're going to find really
- 42:03interesting. I'm going to tell it to go do that.
- 42:06But let me just go in here and show you what
- 42:08the output looks like. in this one. So I'm gonna
- 42:14load this up again. And again, this is the advantage
- 42:15of being able to sort of pre -do some of these
- 42:17demos while we do them live. So you can really
- 42:19see what's happening here. So again, it went
- 42:24through, it created a nicely formatted document,
- 42:26right? It went through, here's the executive
- 42:27summary, right? We want to court order that,
- 42:30you know, child support in the amount of 8 ,500
- 42:32per month representing a 75 % upward deviation
- 42:34from the guideline amount. It's warranty because
- 42:37and then it goes through and it gives you the
- 42:38guidelines and it gives you the rules. So if
- 42:40you tell it what county you're in, what you know,
- 42:41what state you're in, the guidelines you have
- 42:43to focus on, it'll go through and give you a
- 42:46starting point. And again, the the key thing
- 42:48I would like to mention is like. These documents
- 42:50will not be perfect, right? You need to collaborate
- 42:52with these tools. In fact, think of these as
- 42:54an incredibly, incredibly smart intern that's
- 42:57been trained on tons of stuff. It will probably
- 42:59get you 70 to 80 % of the way there on the first
- 43:02go around, but then it's up to you to edit it,
- 43:04to go through and to make the changes and to
- 43:06have this thing keep going. Right. So it's going
- 43:08through creating all this great analysis. And
- 43:10at the end, you give you that starting point
- 43:12to go justify what you're trying to do there.
- 43:14And right. So you can see here, um, Gemini is
- 43:17going through and creating the same thing. You
- 43:18know, it's obvious. see not formatted as a Word
- 43:20document. I could have opened this up in the
- 43:22Canvas window and said, hey, make this a nicely
- 43:24formatted document for me versus the bullet point
- 43:26structure. Because this is my normal output structure
- 43:29in this tool is I tell it to do bullet points.
- 43:31But again, another good example of what it can
- 43:33do there. So here's another one, business valuation
- 43:37discovery. So this is an example that's not just
- 43:40in family law. But valuing a business requires
- 43:42extensive discovery that some business owners
- 43:44resist, knowing what to request and why requires
- 43:47forensic accounting expertise that many family
- 43:49lawyers lack, right? I've worked with a number
- 43:51of valuation organizations as well, and I've
- 43:54seen the kind of, you know, forensic expertise
- 43:56that's needed. And these tools, because I've
- 43:58actually built some custom tools for different
- 44:00valuation companies, are absolutely amazing at
- 44:03actually doing this kind of analysis. So I'm
- 44:05gonna jump in here again, and I'm gonna go ahead
- 44:07and let this run, and I will show you what an
- 44:10output looks like. So let me go grab all these
- 44:13business valuation items here. We will use, paste
- 44:19this in. I want to make sure I got the right
- 44:22one. Yep, business owner stuff. I believe this
- 44:25is the right one. Yeah, it is. Okay. I'm going
- 44:28to go ahead and just say go. And while I do that,
- 44:31I will go pull this up again to show this output
- 44:34to you. And this is the business valuation one.
- 44:38And let's open this up. So, you know, and this,
- 44:44by the way, probably took Claude, you know, five
- 44:46minutes or so to do, because it does, it does
- 44:48all the analysis and it, you know, carefully
- 44:50constructs a Word document kind of line by line,
- 44:53but trying to create this, right? So it's gone
- 44:55through for the Pacific Coast HVAC services organization,
- 44:59you know, 2020 -23 revenue, revenue claimed,
- 45:02right? 30 % decline, right? It goes through and
- 45:05helps you can do the forensic accounting on all
- 45:08these different documents and processes these.
- 45:09And these tools can analyze PDF tax returns they
- 45:13can analyze bank statements credit card statements,
- 45:15you know payroll records You can see the list
- 45:17of the kind of things that can do here Any document
- 45:19you can kind of throw in here. These things can
- 45:21go through and process these days So this one's
- 45:23still going through and thinking about doing
- 45:25that because this is a more complex case But
- 45:27again, I want to show you the kind of output
- 45:29that you get from these tools when you use them
- 45:32because they are fantastic I'm not going to do
- 45:36this case. There's another case in here. I want
- 45:38to get to another example that I think is really
- 45:40interesting and show you something else new that's
- 45:42in Excel. And again, let me just pop over. I
- 45:45don't see any questions. So if there are questions,
- 45:50at least I don't think there are any. Let me
- 45:52look at the chat. Yep. No, I don't see anything.
- 45:56OK. Here's another one that's an objection response
- 46:00generator, right? I will run, that's when I think
- 46:02this was an interesting, actually I'll just show
- 46:03you the output from it. But you know, the trustees
- 46:06and creditors file objections to exemptions,
- 46:08confirmation and discharge, you know, responding
- 46:10under a deadline pressure requires quick legal
- 46:12analysis. Again, here's the prompt, help me respond
- 46:15to this bankruptcy objection, right? Here's the
- 46:17case type, you know, the objection is filed by
- 46:19this person. You generate the legal standards,
- 46:21the response outline, the evidence needed, the
- 46:23draft response. So again, I'm going to jump into
- 46:26this just to show you what the output on something
- 46:29like this can look like. So bankruptcy objection
- 46:33response here. And again, this one, I didn't
- 46:35have to create a word doc. But it went through
- 46:38and you'll see it created a very, very detailed
- 46:41response. I picked the case because I had three
- 46:44sample cases, you know, response to, you know,
- 46:46desert community banks, objection to confirmation.
- 46:49Here's the legal standards, right? Here's the,
- 46:51you know, the particular guidelines that it's
- 46:53focused on, you know, point by point response
- 46:55outline, what to do here, the value, the source,
- 46:57you know, the standards, all the things that
- 46:59you would expect, you know, in a really good
- 47:01output, whether you created or, you know, or
- 47:03some of your attorneys on staff created it. So.
- 47:07I want to switch gears here. I'm going to actually
- 47:09show you something before I show you this. So
- 47:12one of the things that's really unique now is
- 47:15the ability for these tools to work on files
- 47:19on your computer, right? And so there's tools
- 47:21like cloud code. There's chat, GPT codecs or
- 47:24Gemini CLI. They are advanced. They really require
- 47:28you to really start to understand how this AI
- 47:30technology works, but it lets you do really interesting,
- 47:33innovative things. And I'll show you that in
- 47:35a second before I do that. There is something
- 47:37really unique from Claude and I'm going to stop
- 47:41sharing this. I'm going to share something else
- 47:42here for a second. I'm going to share this Excel
- 47:44file. So I'm going to go grab this. So one of
- 47:48the things that I know is a challenge for a lot
- 47:50of people is the ability to actually, you know,
- 47:52create business models, to understand settlements,
- 47:55to do financial analysis. And one of the things
- 47:57that Claude from Anthropic has built in is they
- 48:01have built in a new tool here called Claude for
- 48:03Excel. And you can actually just talk now to
- 48:06Claude in plain English. And it will actually
- 48:09create spreadsheets for you. It will do analysis
- 48:11for you. It will create charts for you. It will
- 48:13identify key things. So I use this as a really
- 48:16simple example here. I basically said, hey, go
- 48:18create a fictitious divorce settlement for a
- 48:21high net worth couple. Make up all the details
- 48:23and make it look visually beautiful. And you
- 48:24can see it did a pretty amazing job here of actually
- 48:28doing this. I'm not gonna go through in detail
- 48:30about all the different things that these tools
- 48:32can do, right? But I could actually, if you had
- 48:35uploaded a, you know, 10 ,000 lines of credit
- 48:38card statements, you could literally just ask
- 48:40Claude in here, hey, go analyze these credit
- 48:42card statements, find any unusual expenditures
- 48:45when they were, create a new sheet with a table
- 48:47of those for me, or create a chart of expenditures
- 48:50by month. In fact, if I actually tell this, hey,
- 48:52create a chart of the categories. and total values
- 48:58on a new tab. And I will let that run in the
- 49:02background for a second. Sometimes this can take
- 49:04a few seconds to do. So it's creating a new worksheet
- 49:07for a chart. You can see it already popped up
- 49:09down here, the asset chart. Hopefully, it's going
- 49:11to go through and quickly generate this. So it's
- 49:14adding chart data table, it's writing the categories.
- 49:16I mean, you can see it's actually building this
- 49:17real time. So again, I know not everybody on
- 49:20this call has deep Excel skills. I have spent
- 49:22a lot of time in Excel, but you can see here
- 49:25that from that one command, it went through and
- 49:27actually created for me a starting point chart
- 49:29of the marital estates by category. So I wanted
- 49:33to show this to you because this has become a
- 49:34really, really powerful tool. and a great way
- 49:37to do things. So if I go back to the presentation
- 49:40for one second here, because then I'm gonna bounce
- 49:42to one more thing here, is I wanna talk about
- 49:45this thing, like I said, this thing called Gemini
- 49:47CLI or Cloud Code or ChatGBD Codec. So normally
- 49:51these tools only operate on the files that you
- 49:54upload into their chat window, right? Or if you've
- 49:57created a project, you can do things like that
- 49:59too. but it really is powerful. So let's say,
- 50:02for example, I'm gonna describe two use cases
- 50:04and I'll show at least one of them to you really
- 50:07quickly. So one of the use cases might be, let's
- 50:10say that you've got hundreds of financial documents
- 50:12that you've gotten from... Either from depositions
- 50:16or from discovery and you want to rename them
- 50:19based on the name of you know, the case you're
- 50:21working on, right? So let's say that I want to
- 50:24have them in a certain format where it's like
- 50:25like I have the sterling case number and a certain
- 50:27date and I and I like your your month month date
- 50:30date format, you know If you've got hundreds
- 50:32of documents and let's say that they you've got
- 50:34bank statements and tax statements and credit
- 50:36card statements Right and you want to categorize
- 50:37in a certain way You know for someone in your
- 50:40team to do that that could take several hours
- 50:42of work What tools like Gemini CLI and Cloud
- 50:45Code can do is because they can operate on files
- 50:48on your computer, they can do this automatically.
- 50:50So I'm going to go grab a prompt. And the prompt
- 50:53I'm going to give this, and I'm going to let
- 50:53you watch this run for a second, is I'm going
- 50:55to say there are 100 financial documents in this
- 50:58folder. Rename them using this format, right?
- 51:01Five character case codes, date and year, year,
- 51:03year, month, month, day, day, document type,
- 51:05vendor. And then I give it an example and the
- 51:07case code is Sterling. So then I tell it, hey,
- 51:10create a subdirectory that's categorized, analyze
- 51:12every file. So open them all up, identify what
- 51:15they are. And then basically, you know, copy
- 51:18a version of that new file with the renamed information
- 51:20over there. So I'm going to go ahead and I'm
- 51:23going to switch over and I'm going to just show
- 51:25you the Gemini CLI window. And again, this is
- 51:27a very advanced case. I'm not expecting people
- 51:29here to know this or to do this, but I want to
- 51:32show you what's possible and where these are
- 51:34going. So this is the tool. It's a command line
- 51:36tool. I'm going to just paste this prompt in
- 51:38here and I'm going to go ahead and just say go.
- 51:41And so it's gonna go look and it's gonna say,
- 51:43okay, I'm gonna go look in the directories that
- 51:44you've got here. I see that there's one called
- 51:47financial statements. And it's gonna go through
- 51:50and it's gonna read those financial statements.
- 51:51And it's gonna actually then go through and follow
- 51:54my instructions and actually start renaming those
- 51:58documents, right? So it's reading the folder
- 51:59financial statements, it listed 101 items. And
- 52:02you can see here, it's now going through and
- 52:03reading one file at a time. It's gonna then understand
- 52:07what that file is. It's going to figure out what
- 52:09the year is, what the date of the file is, is
- 52:11it a credit card statement, is it a bank statement,
- 52:13whatever it happens to be. And then it's going
- 52:15to go ahead and rename it and copy that renamed
- 52:17file into that categorized directory. So again,
- 52:20what took a firm or what would take a firm three
- 52:22or four hours of a paralegal's time or your time,
- 52:26you can now do in literally under 90 seconds.
- 52:28It will go through and do all those files. So
- 52:31I'm going to let it run. I'm going to go back
- 52:32to the presentation because I want to wrap up
- 52:34here in a minute. The other thing that I want
- 52:37to mention is the other example, let me share
- 52:41this document again, that I've often done is
- 52:44I know that there's a lot of need often to redact
- 52:47PDFs. You can use these tools very carefully
- 52:51to do that, right? Now you have to set up privacy
- 52:54statements correctly. But one of the things I
- 52:56showed in a live example two weeks ago was I
- 52:58took a, uh, a divorce document and I said, go
- 53:01redact all the names, flatten the file, the images,
- 53:04right? Remove all the hidden data, then re OCR
- 53:06to make it searchable and a clean PDF again.
- 53:08So a tool like Gemini CLI or cloud code can do
- 53:11that. But, but, but you have to set up the privacy
- 53:14stuff correctly. Please don't try that without
- 53:17going ahead and doing that. So I see there's
- 53:20a couple of questions are in the Q &A. We have
- 53:2312 questions. I'm going to leave some time for
- 53:25that in a second. I'm going to just hit these
- 53:27last couple of slides and then I will wrap up.
- 53:30If you haven't seen projects yet, please go take
- 53:34a look at projects. Take a look at the content
- 53:36that I'm sending out, and I'll let you take a
- 53:39look at those there. But projects let you basically
- 53:41have information in a particular area. And the
- 53:44reason I wanted to mention projects is I have
- 53:46this one called Detailed Fact Checker. And what
- 53:48I mentioned earlier was That you can actually
- 53:52use this to help you dramatically improve the
- 53:55facts in the case, right? And I will if anyone
- 53:57wants this information my emails at the end of
- 53:59the presentation just go ahead and email me but
- 54:01there is a Mechanism where you can dump this
- 54:04information in and it will literally go through
- 54:07Analyze what you put in and look at every single
- 54:09fact and assertion in the in what you've pasted
- 54:11in and give you is this accurate? Is it not accurate?
- 54:14Like you know, it'll give you a score 95 % 0
- 54:16% right this fact is plate blatantly untrue.
- 54:19So again, this is a project. Go ahead and go
- 54:21look at that. Again, privacy is super important,
- 54:26right? I talk a lot about privacy and setting
- 54:28things up correctly. My recommendation is, first
- 54:32of all, don't use the free tools at all. The
- 54:34tools can train on your data and they will train
- 54:36on your data. If you are going to use a consumer
- 54:38grade tool, please enable the privacy settings,
- 54:41which is on the next slide. I'm not going to
- 54:42go through this in detail. I strongly recommend
- 54:44that everybody use the business or the enterprise
- 54:46tools, right? You get data encryption, you get
- 54:49stock to compliance, data is encrypted. They
- 54:51guarantee that will never train on your data.
- 54:53So again, I put these two slides in here. So
- 54:55if you're using consumer grade tools, this tells
- 54:58you how to opt out of it. Here's some more information
- 55:00about the teams, the enterprise one. So by default.
- 55:03If you use ChatGBT, a business or an enterprise
- 55:05one, they don't train in your data. You get SOC
- 55:09compliance, things like that. Anthropic is the
- 55:10same way, as is Google Gemini on the workspace.
- 55:15Grok doesn't have an enterprise front end. You
- 55:17can go through the API. And Perplexity also does
- 55:20that if you're using Perplexity. So I'm going
- 55:23to end again with, hey, these are the dumbest
- 55:25tools that they will be for the rest of your
- 55:27life. I want to remind you one more time. You
- 55:30have ultimate responsibility. Like I said earlier,
- 55:32if you take just one thing away, please know
- 55:35you have to do human review after you go through
- 55:37this stuff. And if you're looking for more help
- 55:41on this, just 10 seconds on me. Look, I do a
- 55:44lot of executive workshops where I take teams
- 55:46through. how you can use AI in your organization,
- 55:49right? Understanding your processes, your tools,
- 55:52cut through the noise. I'm obviously, as you
- 55:53can tell, I'm very agnostic about these tools.
- 55:55I do firm -wide training. For organizations,
- 55:57I do litigation support. One of the things I
- 55:59was mentioning this morning was that I actually,
- 56:02before the call started, is I'm working with
- 56:04an organization. We've actually uploaded 10 ,000
- 56:06of their documents into Gemini Enterprise and
- 56:10create an automatic AI search on it. Using tools
- 56:13like that, because you can only upload, say,
- 56:1610 documents at a time. We can actually have
- 56:18these tools analyzed across all thousands of
- 56:20documents at once in a completely secure environment.
- 56:24I've got smithsteven .com for subscribers, and
- 56:27my contact info is here. So I'm going to pause
- 56:30on this. I'm looking for the Q &A. The one I'm
- 56:34seeing is I'm not seeing a whole Q &A. I'm only
- 56:39seeing the... Let me find... Oh, Q &A, here we
- 56:44go. I found it now, sorry. Let me go back and
- 56:48look at these and try and get to these in the
- 56:49next two minutes. So John asks, is there anything
- 56:54like a GitHub for AI prompts? I use Cloud and
- 56:56the only thing I've seen is very limited artifact
- 56:58library, odd given how easy it is to build and
- 57:01share prompts. No, no one that I've seen, John,
- 57:03has really solved that. Organizations, what I've
- 57:05seen is they create shared documents internally,
- 57:07either on Slack or in like a shared Google Doc.
- 57:10And they go ahead and do it that way. But I've
- 57:12not seen any of these companies do that. Although,
- 57:15if you're using some of these enterprise and
- 57:17teams versions, you actually can share prompts
- 57:20and projects and gems across the organization.
- 57:25I tried creating a Gemini infographic. It had
- 57:27many basic typos, undiagnosed liabilities instead
- 57:30of undiagnosed liabilities. Look, if you're using.
- 57:35the paid versions and you're using it with the
- 57:38pro model, right? Make sure that you have pro
- 57:40turned on when you're doing it, like, cause you
- 57:42can have fast thinking or pro. Make sure it's
- 57:45on pro, cause then it takes the longest amount
- 57:46of time to actually think about the model and...
- 57:49If you're not using a paid version, you're getting
- 57:51sort of like the old version of NanoBanana is
- 57:54not NanoBanana Pro. So you want to be using the
- 57:56paid ones, and that should eliminate those problems.
- 57:58I don't get those anymore. The prior version
- 58:01that was out last summer, you definitely got
- 58:03typos like that, but that's the best way to address
- 58:05that. I use ChatGBT already. They allow us to
- 58:08use projects. Can you explain the limits of that
- 58:11information that will track within a project?
- 58:13Will all chats in the project be considered when
- 58:15you ask questions? Awesome question. I didn't
- 58:17have time to go through this. You can set up
- 58:19a project so it actually has memories. When you
- 58:21set up a project, there is a memory option on
- 58:23there. So go ahead and say, hey, have memory
- 58:27of all the other chats in this project and not
- 58:29the rest of all your chats. Then you can actually,
- 58:31when you're having a chat in that project, it
- 58:33will have memory of the other conversations you
- 58:35have. I use it all the time. It's absolutely
- 58:37fabulous. All right. What about privilege issues
- 58:40and privacy when using these AI programs? Look,
- 58:43that's a longer conversation. The reason I tell
- 58:45people to be using the secure versions of the
- 58:47tools is you want to actually have, you know,
- 58:49the best security and privacy settings that you've
- 58:52got there. So look, I know we're at the top of
- 58:54the hour. Let me just ask, do you guys want me
- 58:58to keep going? I can spend a few more minutes.
- 59:00I don't know if the host wants me to go on for
- 59:02a few more minutes to answer these questions.
- 59:06Then I'll go through the rest of these. And I
- 59:07just didn't know if we had a truly hard stop.
- 59:10So I had ChatGPD extract transactions from bank
- 59:13statements into Excel, maybe 100 pages, asked
- 59:15to categorize expenses. The first set was great.
- 59:17Then said, do the same thing for the following
- 59:19years. It would not create the first really good
- 59:20output in subsequent years. I asked why. It said
- 59:23it was learning, and that learning would change
- 59:24the output. So this is an example where you start
- 59:28running into context window limitations. And
- 59:30these tools, again, they try to be inherently
- 59:33lazy. The best way to solve that problem, Linda,
- 59:37would be to use a tool like Gemini CLI or Claude
- 59:39code, or in ChachiBD they have a tool called
- 59:41Codex. What it will do then is if you would go
- 59:44into Codex and you would give it those exact
- 59:46same instructions and point it at that Excel
- 59:48file, it will effectively write a Python program
- 59:50to go extract that, and then it won't get lazy.
- 59:52It will do it again and again, and it will do
- 59:54all the formatting. It'll do any calculations.
- 59:56You can make it do formulas. The tools in the
- 59:58context windows, you're basically just sort of
- 1:00:00running out of memory is what's happening. Like
- 1:00:01I said, and they tend to get lazy. They try to
- 1:00:03take shortcuts. So the best way, and again, I
- 1:00:06run into this all the time. I would just go open
- 1:00:08up one of those terminal windows, or if you also
- 1:00:10happen to have Claude, Claude just introduced
- 1:00:13something called Claude Cowork, which is their
- 1:00:15version of Claude code, but it's in the web.
- 1:00:17So you can actually have it operate on files,
- 1:00:19your computer through a really nice web interface.
- 1:00:22Your financial discrepancy analysis, how does
- 1:00:24your admonition to check AI work jive with the
- 1:00:26voluminous data and results generated by AI?
- 1:00:28It's always seemed to me that human verification
- 1:00:30would be very time consuming. How is AI? Well,
- 1:00:34look, so one of the reasons I wrote that fact
- 1:00:36check tool was to actually give a set of text
- 1:00:40messages so you can actually go. look at that
- 1:00:43content, have it analyze this content. What I
- 1:00:45will also do, Lin, is I will often take the content
- 1:00:47from one tool and dump it into another tool and
- 1:00:50have it cross -check and cross -reference that.
- 1:00:52So when you turn on the most advanced thinking
- 1:00:54and you tell it, hey, go check every calculation,
- 1:00:57check every analysis. I mean, the best tool at
- 1:00:59doing that, honestly, is ChachiBT, their pro
- 1:01:02-level model, will truly go. calculation by calculation
- 1:01:05and verify everything that's in there. But those
- 1:01:07are the two ways I would do it. I would either,
- 1:01:09you know, have ChatGPD Pro do it or run it through
- 1:01:11a different tool or actually use something like
- 1:01:14that fact checker, which again, if people want
- 1:01:16it, please just email me. How do you predict
- 1:01:19the names of clients and others in using various
- 1:01:21AI products? So again, if you're using the security
- 1:01:25and privacy ones where it's not training your
- 1:01:27data, that is not a bigger concern. All the firms
- 1:01:31I've worked with have reached the conclusion
- 1:01:33that they're using enterprise -grade security
- 1:01:35and tools and that they're putting the data in
- 1:01:37there. One of the examples I gave was on that
- 1:01:39redaction. Most people still want to do redaction
- 1:01:42internally, which I completely understand. I
- 1:01:44have found a way to allow these tools to do it,
- 1:01:47but again, it requires a lot of advanced... Privacy
- 1:01:50setup and a lot of nuance. So don't try that
- 1:01:52out of the box. But that's one way to do that
- 1:01:54is to actually have these tools do the reaction.
- 1:01:56Again, you either do it internally if you want
- 1:01:58to. But again, if you're using enterprise -grade
- 1:02:01security tools from ChatGPT, Claude, or Gemini,
- 1:02:05they don't access your data. They don't see your
- 1:02:07data. They can't train in your data. It's locked
- 1:02:08down to you and your organization, assuming you've
- 1:02:10set up your security properly. Which, in your
- 1:02:14opinion, is the best AI model at this time? Look,
- 1:02:17I don't answer that question generally because
- 1:02:19they change every week to be perfectly honest.
- 1:02:22The way I will answer it though is this way,
- 1:02:24is because they leapfrog each other all the time.
- 1:02:27And so last November, December, all three of
- 1:02:29these companies came out with their latest and
- 1:02:31greatest models, right? So six months ago, if
- 1:02:33you had asked me, I was basically using ChatGPT
- 1:02:36for probably 70 % of what I did. And then the
- 1:02:39rest of the time was split between Claude and
- 1:02:41Gemini. Today, I would say, you know, 60 to 70
- 1:02:45% of my time, probably 60 % of my time is in
- 1:02:47Claude because of these new document skills and
- 1:02:48things that it could do there. And some of the
- 1:02:50formatting it can do and probably 30 % in Gemini
- 1:02:52because I do a lot of super long context things.
- 1:02:55And I use the Gemini enterprise where we can
- 1:02:56analyze, you know, 10 ,000 documents at a time.
- 1:02:59Um, but I use chat GPT for the pro model and
- 1:03:01some super heavy thinking. So when I want deep
- 1:03:03analysis, I go there. So the thing is I use these
- 1:03:06tools for different things, depending on what
- 1:03:07the use case is. So it's really hard to answer
- 1:03:09that. But hopefully you saw what some of the
- 1:03:11outputs look like. And look, I'm happy to consult
- 1:03:13with organizations and to figure out what are
- 1:03:15your workflows and figure out what the best ones
- 1:03:19are based on what you're trying to do. Because
- 1:03:20depending on what you're trying to do, there's
- 1:03:22never a single answer. I use other AI platforms
- 1:03:27that have access to Gemini through Google Workspace.
- 1:03:29I'm evaluating from enterprise AI governance
- 1:03:30and data privacy. Could you speak to the key
- 1:03:33safeguards Gemini provides, including training
- 1:03:35use of customer? Yeah. So look, if you're using
- 1:03:37Gemini in the workspace, and actually I didn't
- 1:03:40show this, but I can actually show this really
- 1:03:42quickly, really easily. If I share my screen
- 1:03:47again, so I can show you perfectly an example
- 1:03:50of how this works. So this Gemini over here,
- 1:03:53this is my personal version on my personal account.
- 1:03:56And you can tell because I got my St. Bernard
- 1:03:57logo up there. It says ultra. And I'm just going
- 1:03:59to type hi. I'm going to do a new chat here.
- 1:04:04Let's refresh this. It's being slow today. New
- 1:04:06chat, here we go. I'm just going to say hi. What
- 1:04:10you'll see here is it says Gemini can make mistakes,
- 1:04:12so double check it. This version of Gemini can
- 1:04:15train on my data. This version of my Gemini here,
- 1:04:17this is my workspace account. This is my Intelligence
- 1:04:20by Intent account. And you can see it explicitly
- 1:04:22says, your Intelligence by Intent accounts chats
- 1:04:25are not used to improve our models. It can still
- 1:04:28make mistakes, double check it. They explicitly
- 1:04:30tell you in Gemini, if you're using Workspace,
- 1:04:33that it does not train in your data. And you
- 1:04:35can actually go and even set higher restrictions,
- 1:04:37how long it retains chats, that by default, anything
- 1:04:40you do in the Workspace one does not train on
- 1:04:43your data. So that's a great way to know that.
- 1:04:45And again, even if you use their most advanced
- 1:04:47enterprise ones with Vertex Studio, it really
- 1:04:49doesn't do it there. A couple more, wow, there's
- 1:04:54a lot more questions. All right, I'll spend like
- 1:04:55five more minutes. If I don't get to the rest
- 1:04:57of these, I'm happy to answer them offline if
- 1:04:59we can capture some of them. The power of these
- 1:05:01programs is mind blowing. I am new to this and
- 1:05:03I briefly use ChatGVT. Should I be using a minimum
- 1:05:06of 5 .2? Yeah, as I said earlier, I mean, 5 .2
- 1:05:10is our newest model. And again, please use the
- 1:05:13thinking version of it, right? If you use auto
- 1:05:15or instant, you're just not gonna get the best
- 1:05:17answers. Always use thinking, because that's
- 1:05:20when it applies its best amount of reasoning.
- 1:05:23In the bankruptcy example, there were references
- 1:05:24to case law, given the recent exposure of hallucinations.
- 1:05:27How does one know if the cases are real, other
- 1:05:29than having to jeopardize each case? So this
- 1:05:32is, again, rice, two things. One, human review,
- 1:05:35any time AI creates things, especially around
- 1:05:37cases of citations. I can't emphasize enough.
- 1:05:40That is super, super important. You or someone
- 1:05:42on your team has to verify that, right? The other
- 1:05:44thing that I do is I will go run it through,
- 1:05:47like I said, that fact checker prompt is a first
- 1:05:49starting point and say, hey, I need you to go
- 1:05:51verify that all these are real. Now there is
- 1:05:53a new technology that's emerging where these
- 1:05:56tools are starting to be able to open browsers
- 1:05:57for you. So if you have a subscription to LexisNexis
- 1:06:00or KSlaw or something else, It can actually open
- 1:06:03up a browser and actually kind of go do manual
- 1:06:05searches for you. So you could literally fire
- 1:06:06up an agent. It will go do that. Again, this
- 1:06:09is a new experimental area. It will definitely
- 1:06:10mature over the course of this year. So that's
- 1:06:12going to get a lot better. And that will start
- 1:06:14help to eliminate that. But at the moment, those
- 1:06:16are the approaches that you're doing. For legal
- 1:06:19research, pleadings, motions, is there an AI
- 1:06:21program that is better than the others? Chat,
- 1:06:23Claude, Gemini? Honestly, I think that ChatGPT,
- 1:06:28so they have something called Deep Research,
- 1:06:30which we didn't cover, and Gemini has Deep Research.
- 1:06:32Those are absolutely amazing. They are generally
- 1:06:35better at searching, doing pure raw search than
- 1:06:38Claude is. Claude is actually phenomenal at reasoning
- 1:06:42and writing. So again, they all have different
- 1:06:44strengths. But for legal research, obviously,
- 1:06:46look, there are dedicated tools out there. Please
- 1:06:48use those. These tools only have access to what's
- 1:06:51available on the public internet today, right?
- 1:06:53Although, as I said, they are going to be able
- 1:06:54to start connecting to some of these services
- 1:06:55and access this stuff. Aren't we violating protective
- 1:07:00orders by allowing Gemini or Claude access to
- 1:07:02attorney's eyes only documents? So I've worked
- 1:07:05with a number of firms. Look, I will give you
- 1:07:07my opinion on that one. It's like using any other
- 1:07:10enterprise tool. It's like if you're using Dropbox
- 1:07:15or if you're using Microsoft Office. storing
- 1:07:20things in the cloud, right? As long as you're
- 1:07:21using the enterprise security and you're doing
- 1:07:23your best that you can do there, right? You're
- 1:07:25providing the best protections in place, right?
- 1:07:28you know, the rules around privilege are not
- 1:07:30completely settled at all. But as long as you're
- 1:07:33applying, you know, your best practices using
- 1:07:35enterprise tools, using security, you're doing
- 1:07:37everything you can to put all the right kind
- 1:07:39of privacy protections in place. And it basically
- 1:07:40is giving you, you know, the arguments I've seen
- 1:07:43is the same level of protection as using, you
- 1:07:44know, email and sending that stuff back and forth
- 1:07:46because you're using cloud services. So again,
- 1:07:48don't use consumer grade ones and don't use anything
- 1:07:50that can train on your data. For local products,
- 1:07:54Gemini, Cloud, GBT, is there an integration with
- 1:07:57Dropbox to pull files? Yes, there is. In fact,
- 1:08:00Cloud and ChatGBT have it built in. Gemini has
- 1:08:03a special version called Gemini for Business,
- 1:08:06which I didn't show. I was showing the general
- 1:08:08web app one. It also has box integrations, Dropbox
- 1:08:11integrations, Microsoft Office integrations,
- 1:08:13OneDrive. All those things are built into all
- 1:08:15those now. Sensitive attorney client privilege
- 1:08:18documents. That one's gone. No, Gemini does.
- 1:08:22Cannot opt out of unpaid subscription, correct?
- 1:08:25Yeah, if you're on unpaid ones, you can not opt
- 1:08:27out of privacy training. Don't use an unpaid
- 1:08:29one. I just, I can't recommend that enough. Don't
- 1:08:31do it. I would never tell anybody to use an unpaid
- 1:08:33one. And which should I try first in my law firm,
- 1:08:36Claude, Chachapiti, or Gemini? You know, pick
- 1:08:39one of them, honestly. They're all amazing. You
- 1:08:42know, pick a set of work cases and try it out.
- 1:08:44So I can't, you know, I mean, you saw the different
- 1:08:47demos and if you've already got access to one
- 1:08:48of them at home, you know, go try it, you know,
- 1:08:50try it. If you have a consumer version, try it
- 1:08:52with non -sensitive data, but they're all amazing.
- 1:08:55They all have incredible strengths. I know we're
- 1:08:57at the end. Look, thank you all for the time
- 1:08:59today. Again, you know, subscribe to the newsletter
- 1:09:02if you're interested. I write several times a
- 1:09:04week. If you're looking for help, please reach
- 1:09:05out. And again, thank you all. I hope you all
- 1:09:08enjoy this. And if you have following questions,
- 1:09:10feel free to email me. Okay. Thank you so much,
- 1:09:12everybody. Bye bye.