Latest / CLEs You Actually Want to Hear / ChatGPT 5, Claude 4.1, and Gemini: The Next Round with Stephen M. Smith
Transcript
- 0:02Hi there, everyone. And welcome to our program
- 0:08today. Introductions for ChatGPT, Claude 4 .1,
- 0:15and Gemini, the next round. Seems like you can't
- 0:19blink without getting caught behind the eight
- 0:23ball run. What's going on with AI? Anyway, my
- 0:26name is Dean Sirwin. I am the section chair.
- 0:28to the law practice management section here at
- 0:32the Beverly Hills Bar. I'm going to handle this
- 0:35introduction and then hand it off to Stephen
- 0:37Smith. Before I introduce Stephen, let me tell
- 0:39you a little bit about what we do at the BHBA
- 0:42and why it's so wonderful. You're seeing things
- 0:45pop up in the chat here about our CLE library.
- 0:49It is outstanding. I used to get cornered by
- 0:53all these emails. faxes and letters in the middle
- 0:57of days when it was time for me to do my MCLE
- 1:01and I'd realize I hadn't done enough and I'd
- 1:06order these video cassettes, that's how long
- 1:08ago that is, and then DVDs and then links to
- 1:13boring, silly, outdated programs. Instead, what
- 1:18we have with the BHBA is this wonderful on -demand
- 1:22service that's kind of like almost a YouTube
- 1:25for MCLE, and new up -to -date dynamic issues
- 1:32like the one today. If there's a new case that
- 1:35comes down, we've got a program up about it,
- 1:37usually within a week. And so you can go in there,
- 1:42watch these, you get participatory credit for
- 1:45them, you can get the materials, and best of
- 1:48all, once you've watched it, It goes into a locker,
- 1:51so when it comes time for you to do your MCLE
- 1:53compliance, it's a couple of clicks. Last time
- 1:55I did it, it took me two minutes. It's a wonderful
- 1:59system, and this fantastic, I know, dynamic program
- 2:04we're going to have for you today will be up
- 2:06there within 24 hours. So if you liked it and
- 2:09you want to go back and watch it again, you can.
- 2:11You can let your friends know about it. Please
- 2:13do. We have, oh gosh, I don't know, one to three
- 2:18of these a day. So we have hundreds, if not thousands
- 2:22of these programs up there. So anyway, let's
- 2:24get on to telling you about this one. Steven
- 2:28Smith is the CEO of Intelligence by Intent. He's
- 2:31a leading AI trainer and implementation coach
- 2:35specializing in practical applications and use
- 2:38cases for professional service firms. He translates
- 2:42complex AI capabilities into actionable business
- 2:45strategies that deliver measurable results. With
- 2:49over 15 years as a C -level executive and former
- 2:52CFO himself, he brings a unique perspective to
- 2:55AI training, focusing on applications that impact
- 2:59revenue growth, operational efficiency, and strategic
- 3:02decision -making. His approach emphasizes hands
- 3:06-on learning with real -world use cases, not
- 3:09theoretical concepts, which I appreciate greatly
- 3:13because it's wonderful to wax on about what could
- 3:16happen far more interesting and useful to hear
- 3:19what does happen and how to do it for myself.
- 3:23So with that, I'm going to hand it over to you,
- 3:25Stephen, and thank you again for being here today.
- 3:28Thank you, Dean. Let me go ahead and share my
- 3:31screen here. And it is delightful to be back
- 3:36here. I presented, I believe, last March to the
- 3:38Beverly Hills Bar Association. So I'm glad to
- 3:41be back and give you an update about what is
- 3:44happening. So as you can tell from the title,
- 3:47I'm going to focus on some of the newer tools
- 3:49that are out there. You know, Chashi BT5, the
- 3:52new Claude 4 .1, the latest updates on Gemini,
- 3:55and the other advances that are happening out
- 3:57there. So I'm just moving a little chat window
- 4:00out of the way there. OK. And so with that, let
- 4:04me just jump in just to give you a little bit
- 4:06more background and context about who I am. So
- 4:09as Dean said, I've been doing this for a long
- 4:11time. I've got 30 years of experience in technology
- 4:13strategy and operations. I'm a Wharton MBA. I
- 4:17was a CFO and an SVP at a bunch of different
- 4:19tech firms. And the last three and a half years
- 4:22or so, I've really just been working with primarily
- 4:25law firms, but also a couple of valuation firms
- 4:27and accounting firms around how you actually
- 4:30productively use AI on a day -to -day basis.
- 4:33So today is an update to the last presentation
- 4:35I gave. And I would love this to be interactive.
- 4:39So if there are questions throughout this, please
- 4:42add them in the chat. I'll ask Rinda, please
- 4:46let me know. Actually here, I'll move the chat
- 4:48window over there so I can see it. But if I don't
- 4:50see a question, please interrupt. Ask me, I would
- 4:54love to make this more interactive so that it's
- 4:56not me just talking to you for an hour. So I
- 5:00like to start with this. So Scott clopored is
- 5:03a very good friend of mine and this was one of
- 5:05his quotes from about a year ago So Scott was
- 5:07one of the early adopters and using AI and he
- 5:10said to me He said look he's like it is so essential
- 5:13to every activity that I do all day long the
- 5:16only way I stop using it is if you pry it from
- 5:18my cold dead hands and As I've worked with more
- 5:21lawyers with more organizations. That is the
- 5:23sentiment that I get from everybody In fact,
- 5:25I use it day in and day out in everything that
- 5:28I do and I use all the tools I use ChatGP T5,
- 5:32I use Claude, I use Gemini, I will occasionally
- 5:36use GROK and Perplexity. So if you have questions
- 5:39about any of those tools as we go through this
- 5:42today, as I said, please jump in and interrupt.
- 5:46So I'm gonna talk about four things today, really.
- 5:49The first thing is, and this is really important,
- 5:51is privacy updates. You know, one of the things
- 5:54that is incredibly important is that we use these
- 5:56tools to make sure that All the client data that
- 6:00we put in these tools is safe and secure. What
- 6:02you don't want to be doing is using consumer
- 6:05versions of these tools that you have not put
- 6:07the right privacy and security on. You should
- 6:09ideally be using the businesses slash teams or
- 6:13enterprise versions for the highest levels of
- 6:15security. And the reason I say this is so important
- 6:18is a year ago, we talked about Claude. Claude
- 6:22has been one of the best tools out there from
- 6:25a privacy standpoint. In fact, they started focusing
- 6:28on privacy. They said, look, we will not return
- 6:31in your data. All of your things are private.
- 6:34You can put in anything in here you want, medical
- 6:36information, tax information, things like that.
- 6:38they have just flipped their privacy stance.
- 6:41And so the reason I wanted to put this up front
- 6:43is I know a lot of organizations, a lot of firms
- 6:47use Claude because of the privacy and security.
- 6:49So I encourage you to go back into Claude. You
- 6:53can opt out of it, right? So they've got a little
- 6:55toggle when you go in there now that says, hey,
- 6:57we've changed our privacy and security, right?
- 6:59There's a little toggle on there that says, you
- 7:01know, click this to make sure that your data
- 7:02is not trained on. So it goes into effect on
- 7:05September 28th. So please go in there. or before
- 7:08then if you are a Claude user, I did put a link
- 7:10here to an article I wrote about a week ago on
- 7:13this exact topic. So again, this is brand new,
- 7:17kind of breaking news. Please go in and change
- 7:19that because for Claude, this is a complete change
- 7:21of pace from where they were. Now, I know a lot
- 7:25of people use different tools that are out there.
- 7:27And so I will talk about all of these really,
- 7:30really briefly. So chat GPT and again, so this
- 7:33is the consumer version of the tools. I know
- 7:35a lot of organizations and a lot of small, you
- 7:37know, solo people like to use the consumer tools,
- 7:39which I do too at a certain level and I understand
- 7:42it. So you should just know that, you know, chat
- 7:44GPT by default. Ops you into training on your
- 7:48data, right? So you have to go opt out. So I
- 7:50put the instructions on these pages. So if you're
- 7:52using these tools, you can actually go in and
- 7:54look and make sure you opt out of that data.
- 7:57Anthropic, which is Claude, didn't use to train
- 7:59in your data. They do now. So again, go make
- 8:02that change in there. If you're using Gemini,
- 8:05which is truly an incredible model, and we'll
- 8:08talk about some of the other things that Gemini
- 8:09can do right now. If you're using a true consumer
- 8:13account, it opts you in and you really can't
- 8:16turn out of training. So I tell everyone that's
- 8:18using Gemini, please at least use a business
- 8:21account, even a one -person business account
- 8:23because then you can have the option to not have
- 8:25your data trained on. GROK also opts you into
- 8:29training by default. You can opt out and same
- 8:31thing on perplexity. So again, if you're using
- 8:33any of the consumer grade tools and these are
- 8:35typically the $20 a month versions of these tools,
- 8:38you have to make sure you opt out of training
- 8:40because by default they all started now. So I
- 8:44put in two more pages here. I'm not gonna go
- 8:46through these in detail. I want you to have these
- 8:48as part of the backup content in the presentation.
- 8:51But I actually laid out most of these organizations
- 8:54have teams and enterprise versions. So teams
- 8:56are typically, you know, anywhere from two or
- 8:59five to, you know, a couple of hundred people.
- 9:02And enterprise versions are obviously, you know,
- 9:04for the largest organizations, they're the biggest,
- 9:06most secure, most complex, you know, set of privacy
- 9:09and security tools that you can put on there.
- 9:11So again, really good information on here. If
- 9:14you use any of those services on the business
- 9:18side or on the enterprise side, by default, they
- 9:20don't train in your data. So again, that's why
- 9:23I tell people to go do that. And I'll show you
- 9:25a real -time example right now. So I'm in Gemini.
- 9:29I was actually playing with that image. And so
- 9:31you'll see if I just type something, hello Gemini,
- 9:33something really simple here. Gemini. What you'll
- 9:38see on the bottom, all it says down here is Gemini
- 9:40can make mistakes. I'm actually using here, this
- 9:43is my personal version of this. If I switch over
- 9:46and go to my workspace account, and I go do the
- 9:51same thing here, I'll type hello Gemini. What
- 9:55you'll see on the bottom is a message that your
- 9:58chats are not used to train our data, right?
- 10:00So it's really important. They're very clear.
- 10:02If you're using a consumer grade version, and
- 10:04again, this is Gemini specific, of that, your
- 10:07data can be trained on. So again, it's an amazing
- 10:10tool. It is super powerful. I just want you all
- 10:12to be aware of that as we go forward. So as I
- 10:15said, there's more in here on this, but I really
- 10:16want to get to the heart of the presentation
- 10:18where it's a lot more fun. Let me start by talking
- 10:22about what's new, because we saw a lot of really,
- 10:25really great changes this year, a lot of new
- 10:27capabilities. I'm going to show you over the
- 10:30course of the next hour or so, a lot of great
- 10:34new use cases, a lot of new capabilities. And
- 10:36again, if there are interesting things that you
- 10:37want to see, let me know. I'm starting with a
- 10:40couple of definitions. And this is one of the
- 10:42most important ones. Is there something called
- 10:44a context window? So if you go use ChatGPT or
- 10:48Claude or Gemini, Think of a context window is
- 10:52the short -term memory that you can have a conversation
- 10:54with, right? And the reason I bring this up is
- 10:56I will show you in a minute that when you're
- 10:58using some of the consumer versions of these
- 11:00tools, the chat window you have is actually much
- 11:03smaller than if you go subscribe to professional
- 11:05editions. So, you know, as an attorney, you can
- 11:08put in a lot more exhibits, a lot more content.
- 11:11You can have much more reasons and lengthy discussions
- 11:15with the AI about the content and what you're
- 11:17trying to do. So just think of it as again, the
- 11:21memory that a person has, the more you put in,
- 11:24they're gonna start forgetting things at the
- 11:25beginning if you overwhelm their memory. So it
- 11:28really, as you'll see, some of these have relatively
- 11:30short memory versions, so you really wanna make
- 11:33sure you're using a version that has a lot. And
- 11:36Gemini, by the way, has got the biggest, and
- 11:38we'll talk about the exact numbers in a minute.
- 11:41The biggest change this year was the movement
- 11:44from reasoning to not, sorry, from non -reasoning
- 11:47models to reasoning models. So if you were a
- 11:50big user of Chat GPT over the last year, and
- 11:53you were using 4 .0, which the vast majority
- 11:55of people were, right, that is a non -reasoning
- 11:58model. It was basically saying, I'm gonna go
- 12:00look at whatever data that I've been trained
- 12:02on, and try and give you the best answer from
- 12:05that data. A reasoning model, on the other hand,
- 12:08takes your input, it takes what it's been trained
- 12:11on, it will do real -time web searches, it will
- 12:13actually go... enable and call different tools
- 12:15to process data and try and come up with the
- 12:18best answer. In fact, it will often go through
- 12:21five, 10, 20, 50 different answers to give you
- 12:25the best possible answer to your question. And
- 12:27what you'll see when we go through these demos
- 12:28is it does take a lot longer to get the answers
- 12:31that once you start using reasoning models, I
- 12:34think you will never go back to using the non
- 12:36-reasoning models because the quality answer
- 12:38is so much better. It is so far superior. In
- 12:41fact, I didn't put the chart in here this time,
- 12:44but the IQ, because they can actually measure
- 12:47these models on the Mensa test and things like
- 12:48that, the IQ of the ChatGPT -5 is, I think it's
- 12:52like 135 IQ, right? So these models have gotten
- 12:55really, really good, whereas ChatGPT -4 .0 was
- 12:58around 100. So again, you'll see... Gemini is
- 13:04a, 2 .5 Pro is a reasoning model, Claude 4 .1
- 13:07is a reasoning model. They've all moved in that
- 13:09direction and you'll see that they take a little
- 13:11bit longer to answer, but you get much better
- 13:13answers. And the last thing I wanna mention here
- 13:16are these things called agents. If you read anything
- 13:19about AI, you know that you can't go a day without
- 13:23seeing people talk about, hey, we've got a new
- 13:25agent that does this, we have an agent that does
- 13:27that, or we have agenda capabilities. Think of
- 13:31an agent as basically a tool built around an
- 13:34AI that … works autonomously for you to solve
- 13:37a problem, right? And so one of the examples
- 13:39we'll do live is a deep research issue. Deep
- 13:42research is an agent. You give it instructions,
- 13:44it will go off, and it will go run that task
- 13:47for you. I will show you a live example of another
- 13:49one. Agents now can control your browser. They
- 13:52can go do research. They can fill out web forms.
- 13:55They can do all sorts of things. They can take
- 13:56input from an email and a spreadsheet and combine
- 13:59things, right? So you can set these things up.
- 14:01They'll go off and run and do things for you.
- 14:03So if you have tasks, that you repeat on a frequent
- 14:06basis, you can set up agents now to go do these
- 14:08things. We're not gonna go deep in this session
- 14:10today around that, but I just wanted you to be
- 14:12aware. So these are the only terms I wanted to
- 14:14define upfront, but I want you to see these as
- 14:17we went through. So let's talk about OpenAI,
- 14:21right? ChatGPT 5 was the big launch of the summer.
- 14:24It launched just about a month ago. They definitely
- 14:27had some hiccups in the beginning. They have
- 14:29straightened those out. It is an absolutely unbelievable
- 14:33tool. It is really incredible, especially when
- 14:36you use the thinking model and the pro model.
- 14:38So if you go into ChatGPT now, right, this is
- 14:42their default. They call it the one system model.
- 14:44So I will just switch the ChatGPT screen for
- 14:47a second. If you do a new chat, which you'll
- 14:49see up here, is you just see ChatGPT 5, right?
- 14:52This is basically their auto model. So they were
- 14:55trying to simplify it, because in the past, if
- 14:57you remember, they had, you know, 4 .0, 0 .4
- 15:01mini low, 0 .4 mini high, they had 0 .3, they
- 15:04had 0 .3 pro, they had 0 .1 pro, right? I mean,
- 15:07I can see some of the legacy models here. They
- 15:09had a lot of choices. They tried to make it really,
- 15:12really simple for most users, just by saying,
- 15:14look, if you leave it on auto, it will figure
- 15:16out whether it can do a quick web. search to
- 15:18answer your question or if it uses reasoning.
- 15:20I will tell you as someone who lives and breathes
- 15:24in these tools all day long I just start with
- 15:27the thinking model right and I would encourage
- 15:29you if you use Chat CPT to start using the thinking
- 15:32model to try it out and I'll show you the differences
- 15:34as we go through here. I also use the Pro model
- 15:37a lot for incredibly advanced research. And I've
- 15:41worked with a bunch of attorneys on very, very
- 15:43advanced cases. And again, using the Pro model
- 15:46gives it its absolute best thinking. You can
- 15:47put in a tremendous amount of context and get
- 15:50absolutely brilliant answers. And I'll show you
- 15:52some examples of that. But they made this to
- 15:55say, hey, we're trying to be simpler. They have
- 15:58continued to evolve their deep research tool,
- 16:00right? As I said, that's an agentic web searcher.
- 16:03So you can create an incredibly detailed query
- 16:06for it to go off and do deep research on. It
- 16:07will go off and it will spend anywhere from,
- 16:10you know, five to 30 minutes to go search the
- 16:13web, clarify the question, do the research, you
- 16:15know, cite the sources, verify all the facts
- 16:18that are in there and get you the most robust,
- 16:20best answer that it can as it does that. They've
- 16:24also incorporated web search, right? So if you
- 16:26don't wanna go to Google, you can go into ChatGPT
- 16:30and do really simple questions and it will actually
- 16:32go search the web real time for you as well.
- 16:35They have continued to evolve projects in Canvas
- 16:38and memory and we'll talk about those things.
- 16:40One of the beautiful things now about projects
- 16:42is you can go create a project. In fact, I have
- 16:46one here that I was doing for the presentation
- 16:48today where I ran a bunch of the prompts through
- 16:51here as an example. So let's say that you create
- 16:53a project on a particular client and you wanna
- 16:56have a string of chats about that client. You
- 16:58can upload a ton of documents about that client.
- 17:00You can have all of your conversations. And as
- 17:03they've added memory now, within this particular
- 17:06project, the chats will have an understanding
- 17:08of the other chats that you've had, right? So
- 17:10this is something that's relatively new for ChatGDT.
- 17:12They're adding memory, and you can have simply
- 17:15within project memories. So if I have a chat
- 17:18about one thing, you can actually reference things
- 17:20that I've done in earlier chats in that project.
- 17:22So for client work, that is a super powerful
- 17:25use case. I'll show you a cannabis again in a
- 17:28minute. Again, a super powerful tool. I'll give
- 17:31you an example of agent mode. So this is one
- 17:33where you can basically have Chat Sheep ET go
- 17:35off and act as an independent agent. You can
- 17:38give it instructions to go search the web, to
- 17:40go find content, to fill out forms, to go do
- 17:43research, to bring it all together. It will act
- 17:45autonomously for you. So we're not going to run
- 17:48a full session today because I find that they
- 17:50often take a tremendous amount of time to go
- 17:52run. But just know that this is the mode that
- 17:54they're evolving to where they can automate all
- 17:57these different things. Then the last thing around
- 18:00OpenAI that they launched this summer was they've
- 18:03introduced a bunch of business connections, right?
- 18:05So they can connect to, if you are a Microsoft
- 18:08shop and you use SharePoint or OneDrive, right,
- 18:11they can connect to that. They can now connect
- 18:12to your Google Drive if you are a Google shop.
- 18:15They can connect to Box and a bunch of other
- 18:17things. So they actually will allow you to use
- 18:19deeper research and other questions against all
- 18:22those very secure business tools now. So you
- 18:25can actually bring, you know, whatever knowledge
- 18:28of the documents you have into these conversations
- 18:30and actually use it in a very safe and secure
- 18:31way. So they are really making ChatGPT a very
- 18:35business -friendly tool. So as I said here, GPT
- 18:39-5 is like having two brains in one, but this
- 18:42is where I wanted to talk about the token context
- 18:44because this is really important. If for some
- 18:47reason you are using the free version of ChatGPT,
- 18:49and again, I strongly encourage you not to do
- 18:51that because you really don't have any privacy,
- 18:53you're basically limited to an 8 ,000 token conversation,
- 18:56which is about 12 pages of text. If you're just
- 19:00using the $20 a month version, it's 32 ,000 tokens
- 19:04per conversation, which is about 48 pages. So
- 19:06if you load in a couple of reasonably long documents,
- 19:09you're gonna run out of context. And so what's
- 19:11gonna happen is it's gonna forget the beginning
- 19:13of your conversations or the documents you've
- 19:15uploaded. So I encourage you to go get a business
- 19:18subscription, a pro subscription, because obviously
- 19:20you can have much longer conversations now, right?
- 19:2396 ,000 words, 192 pages, or if you use the API,
- 19:27you can get even longer access after that. But
- 19:30again, this is something that most people don't
- 19:32realize, and they don't realize that when you're
- 19:35drafting content and doing things, If you have
- 19:37a relatively small context window, it's going
- 19:40to get cut off when you do things in the beginning.
- 19:43So I'll show you a couple of things in the beginning
- 19:44here just in terms of the ability to do like
- 19:47real time web search and things like that is
- 19:50I'm going to grab some text here and I'm just
- 19:52going to turn off the air conditioning in my
- 19:54house because it's a little bit cold in here.
- 19:57But you can actually say, look, I want to go
- 19:59do a quick web search. I'm going to just grab
- 20:01this content here. and I'm gonna go over to ChatGPT,
- 20:06I'm gonna leave it on auto, and let's say new,
- 20:09and I'm also gonna tell it to actually just do
- 20:11more here and do web search, right? So you can
- 20:14basically tell it, you know, go look for 2020,
- 20:162024, 2025, your prenuptial agreement updates
- 20:19on this content, and it's gonna basically say,
- 20:22as you can see here, it's gonna go search the
- 20:23web right now, and as you'll see, you get a very,
- 20:27very fast answer, right? And I've got citations
- 20:29in here as well, but it is very, very fast. because
- 20:32it was not using a thinking model. If you then
- 20:36wanna basically say, all right, I want you to
- 20:38think a little bit more about this, right? I
- 20:40want you to summarize the signal from those articles
- 20:42into 150 word thesis. So I'm gonna just grab
- 20:45this. I'm gonna come up here and I'm gonna switch
- 20:47this now to the thinking model. And I'm gonna
- 20:50go ahead and run this query. I'm gonna turn off
- 20:52web search because I don't need to do that again.
- 20:54And I'm gonna say go. And what you'll see here
- 20:56is now it says thinking. And so this is an area
- 20:59where you will see that this may take 10 seconds,
- 21:03it may take a minute to actually go through and
- 21:06figure out based on the research that it did,
- 21:08how to answer my question. So you can see it's
- 21:11gonna go through some different steps. I'm gonna
- 21:13step away and come back to it for a second, just
- 21:15so I can keep talking about what's new in here
- 21:17and what we've got, because there's some really
- 21:19interesting things. And then I won't do the second
- 21:22one here, but I could say, you know, you know,
- 21:25I can tell, think hard about this. And if you
- 21:27use that phrase, it will actually do its deepest
- 21:30level of thinking that actually has multiple
- 21:31levels of thinking within chat GPT. So I could
- 21:34say, Hey, using the articles above, you know,
- 21:36write a one paragraph thesis on, you know, relocation
- 21:38case postures in LA, give me five risks, rank,
- 21:41you know, fact, three gaps that we have to close
- 21:43things like that. So they've added, like I said,
- 21:46some really, really powerful new tools. As I
- 21:49said, as you go through reasoning, yeah, it's
- 21:51still working on it, you're gonna see it really
- 21:53give you very, very detailed answers. So I would
- 21:56encourage you to primarily use that unless you
- 21:58know you're simply asking a question that it's
- 22:00likely been trained on, that it can just give
- 22:02you a quick answer to. So as I said earlier,
- 22:08Deep Research launched early this year and I
- 22:10know a lot of people haven't tried it. It is
- 22:12one of the most powerful tools out there, right?
- 22:15It will basically go off and you can give it
- 22:17a very, very detailed prompt with lots of different
- 22:20examples, a lot of different things, and say,
- 22:22go research this topic for me. What we'll come
- 22:24back with is an unbelievably detailed citation
- 22:28rich document to answer your question. And so
- 22:33if I show you this example, right? So I want
- 22:36to understand the treatment of stock -based compensation
- 22:39of RSUs and PSUs and spousal and child support.
- 22:42So I can tell it, hey, this is my scope and jurisdiction.
- 22:44hear the core questions I care about, right?
- 22:47When do courts treat, you know, vested versus
- 22:49unvested RSUs and bonuses as income for support?
- 22:52What are the standards for imputing income, right?
- 22:55Use the structure of the percentage, you know,
- 22:56of this order, you know, what are the evidentiary
- 22:59showings? Look at primarily most recent case
- 23:02studies, right? And then what I want is a deliverable
- 23:04150 word thesis, the five strongest counter arguments,
- 23:08all these different things. So this is a single
- 23:10query that I'm gonna go launch. And you'll see
- 23:14here, by the way, so this was the thinking. So
- 23:16it thought for two minutes and nine seconds from
- 23:18that prior one to go give me this last topic
- 23:22here. And if I go search here on stock -based
- 23:25compensation, that deep research query that I
- 23:29was running earlier, and I'm going to actually
- 23:31just go back to the very top of this, because
- 23:34what you'll see here is I put in here that I
- 23:38want to go do deep research on this topic. And
- 23:40one of the things that's great about ChatGPT's
- 23:43deep research, and Jim and I does this too, is
- 23:46it will say, hey, this is a very well detailed
- 23:48in scope request. Before I begin the research,
- 23:51can you confirm these things that I want to know
- 23:53just to make sure I'm focused on the right things,
- 23:55right? So it's saying, am I looking for differences
- 23:57in how course treat RSUs versus PSUs versus options?
- 24:01Should I include analysis of temporary versus
- 24:02permanent support? So it's trying to clarify,
- 24:05what do I really want to get the best answer?
- 24:08And so I answered those questions for it. And
- 24:10it went off, it says, I'm gonna go think about
- 24:12all these things. So look, it searched for eight
- 24:14minutes. It went to 18 sources, it did 114 searches.
- 24:18And I'm just gonna flip here. I pulled up the
- 24:21actual document it created. So it created, this
- 24:23was the thesis that I asked for, right? Sort
- 24:25of the 150 word thesis up front here. These are
- 24:28the best counter arguments to that thesis. Then
- 24:30it went through and identified all the cases
- 24:32for me. So again, this is a super, super powerful
- 24:37tool to use on this stuff. It is incredible.
- 24:40I've worked with lots of organizations. PE firms
- 24:44use it. Valuation firms use it. When you want
- 24:46to research an industry, a set of laws, a set
- 24:48of cases, if you're working in a new specialty
- 24:51area, it is a super, super powerful tool to get
- 24:54you up to speed as fast as it can. part is at
- 24:57the end, it will actually give you all of its
- 24:58sources and all of its linked. So again, super,
- 25:01super powerful tool built into ChatGBT. I already
- 25:07talked really briefly about Projects. So again,
- 25:10Projects is a custom workspace, and you can create
- 25:12as many as you want, where you can have chats,
- 25:14you can have files, you can have custom instructions.
- 25:17And again, because it now has memory, it has
- 25:21the ability to basically have knowledge of the
- 25:24other chats that you have in there, right? So
- 25:26I actually have a number of projects. I have
- 25:29a detailed fact checker project where I've got
- 25:31incredibly detailed instructions and you can
- 25:33see I've got all these different things that
- 25:34are in here. And so you can actually create,
- 25:37again, very, very detailed instructions about
- 25:39what you wanted to do. You can upload files,
- 25:42but it becomes contained as a project. So again,
- 25:45if you haven't tried projects and you're working
- 25:47on a single case, it is a great way to put that
- 25:50information in there. Canvas is another powerful
- 25:54tool. So if you've historically used ChatGPT
- 25:57or Claude or Gemini, you know that the conversation
- 26:00is just like a long threaded conversation. Well,
- 26:03ChatGPT has a Canvas tool. I'm going to leave
- 26:05this in auto for a second that I'm going to put
- 26:07on Canvas here. And I'll do something silly and
- 26:11simple, but I'm just going to say, write me a
- 26:13three paragraph story about St. Bernard's and
- 26:18use the Canvas window. And so when you use Canvas
- 26:22here, what it will do is it will create a very
- 26:24quick story and you'll see here it's going to
- 26:26open it up into a different interface. And so
- 26:29if you're working on a document, if you're working
- 26:31on a deposition, you know, or a draft memo for
- 26:33something, you can do it in a canvas window.
- 26:35And the beauty of using canvas is you can now
- 26:38go in and simply edit certain portions of it,
- 26:41right? Versus having to edit the whole thing.
- 26:43So I'm now in the canvas window here and you'll
- 26:46see here you've got some options here. You can
- 26:47suggest edits, right? I can adjust the length.
- 26:49I can do some different things. But what I can
- 26:52really do is I might want to say, you know what?
- 26:54I just want to change this one paragraph. Just
- 26:57rewrite this one sentence, rewrite this one sentence.
- 27:01So if you've got a 20 page document, historically,
- 27:04if you told ChatGPT to simply rewrite a portion
- 27:07of it, you're gonna see it go do everything again
- 27:10and again. What you saw here is it left everything
- 27:12else exactly the same and just edited in that
- 27:15canvas window. And then once you're in the canvas
- 27:18window, you can then share, you can create links
- 27:20and you can do different things to it. You can
- 27:22create, by the way, interactive infographics.
- 27:25You can create charts and dashboards in the canvas
- 27:27window. It is super super powerful. I'll show
- 27:30you some examples as we go through here. So agent
- 27:34mode. Again, I will fire one off so you can see
- 27:37what it does because it really is cool. But again,
- 27:40agent mode is something that's new. It's exploratory
- 27:42from chat GPT where you can actually have it
- 27:46go off and do research, find reports, you know,
- 27:49go to the SEC website, download things, extract
- 27:52information, do anything you want. I'm going
- 27:55to fire off an agent that basically, you know,
- 27:56it's related to that RSU stuff that I was doing
- 27:58earlier where I want to tell it to go do these
- 28:01things. I want it to go find, you know, 2024,
- 28:0325, you know, appellate decisions on imputation
- 28:06of income and RSU -based compensation, create
- 28:09a case matrix, issues holding citations, draft
- 28:12a five -slide outline for hearing prep, but don't
- 28:14download any files. And so if I go fire this
- 28:17off, I'm going to go into agent mode here, and
- 28:21I'm just going to go ahead and say, do this.
- 28:23And what you'll see here is you'll see it's gonna
- 28:26say, okay, I'm gonna now set up a virtual desktop.
- 28:28So it is literally creating a virtual set of
- 28:31computers that are running in the cloud that
- 28:33are gonna go off now and go do this research
- 28:35for me. So I'm not doing any of this. And by
- 28:38the way, if there was a site you wanted to point
- 28:40into that you had, that say you had to log into,
- 28:43you could actually have it log in for you. It
- 28:45could go do those things. So again, super powerful
- 28:48tool. It will go off. It will do all this stuff.
- 28:49You can create presentations. And again, they're
- 28:52pretty... rudimentary, I would say, this day.
- 28:55But again, a very powerful tool and use case
- 28:57for doing that. So again, this will go on for
- 29:0015, 20, 30 minutes. If it gets done by the end,
- 29:03I'll show the results, but I want to keep going
- 29:04to make sure that we can get through the content.
- 29:07I mentioned earlier these business connectors,
- 29:10I won't belabor this, but you can actually, like
- 29:12I said, if you are a Microsoft shop, you can
- 29:14connect it to your SharePoint. You can use deep
- 29:16research and the other tools to access content
- 29:18in there, to pull content out of there, to do
- 29:20things like that. So it is really great. You
- 29:23can use SharePoint box. It can read your email
- 29:25and Outlook. It can connect to Teams. It can
- 29:27connect to HubSpot. It has a lot of great different
- 29:30things. So that's a quick summary of what's in
- 29:33GPT -5. And again, it is super, super powerful.
- 29:37So one thing I will show you is, let me just
- 29:40duplicate this tab here. And if you are using
- 29:44ChatGPT 5 Pro mode, I've been running a big thing
- 29:48to overhaul my website. And what you'll see here
- 29:50is. I put in a very long prompt here and a couple
- 29:53of PDFs. And I said, go off and do a bunch of
- 29:56research. So when you do Pro Mode, it will go
- 29:59think for a long time. So in this case, it thought
- 30:01for eight minutes. It came back with a very specific
- 30:04set of recommendations and a bunch of issues.
- 30:06And then you can keep going. And this is their
- 30:08deepest type of thinking. You can have it create
- 30:10content and do things for me. this is an incredible
- 30:14amount of output from a single prompt, right?
- 30:16And then I actually took additional content from
- 30:18other, from Claude and from Gemini and added
- 30:21all the stuff in here and said, now consolidate
- 30:23their best thinking and your best thinking, right?
- 30:26So, you know, thought again, hey, I've incorporated
- 30:27this stuff into here. So again, super powerful
- 30:31use case. I use Pro several times a day. It is
- 30:35their single most powerful model. And if you're
- 30:37integrating a lot of different data, it is unbelievably
- 30:40powerful. So not too many updates on the Claude
- 30:44side. They did launch in August, 2025, a big
- 30:47update to their Claude Opus 4 .1 model. So the
- 30:50one thing I will say about Claude, it is absolutely
- 30:53fantastic at creating beautiful visualizations
- 30:56and infographics. Or if you've got a lot of financial
- 30:59data on a case and you wanna have that consolidated
- 31:01and like create insights and create charts. All
- 31:04the tools can do it, but you'll get the most
- 31:06beautiful output from Claude if you're doing
- 31:09that, right? They have really stepped up their
- 31:11game. I'll just see if I have any here that I
- 31:16can look at really quickly. I don't know that
- 31:17I had any visualizations set up. That was a video
- 31:21one. I don't see any. I was just looking to see
- 31:23if I had a quick, simple one. Oh, I was doing
- 31:25some. I was doing an interactive chart around
- 31:29temperature change. So I was saying, hey, go
- 31:31model out global temperature changes. Show me
- 31:33what this looks like. Can you automate it? So
- 31:36again, through simple prompting, you can create
- 31:39really, really good interactive charts and graphs
- 31:42in Claude. You can do similar things in ChatGPT
- 31:45and Gemini. It's just not as good. And then you
- 31:47can actually publish these to the web. You can
- 31:49download them as images, things like that. But
- 31:51they have continued to update their reasoning.
- 31:53They have historically been one of the absolute
- 31:55best writers out there. Most of the attorneys
- 31:59that I know actually work with Claude because
- 32:00Claude's writing capabilities are absolutely
- 32:04fabulous. They also have a Claude research capability
- 32:07as well, just like ChachiPT and Gemini does.
- 32:09But that's been their big update for the year.
- 32:13So Google introduced a couple of things that
- 32:15I want to highlight here, some things that are
- 32:17interesting. So back in April, they launched
- 32:19their Gemini 2 .5 Pro model. That is their flagship
- 32:23multimodal model. The thing that is different
- 32:26about... Gemini versus chat GPT or Claude is
- 32:30really two things. One is this multimodal. And
- 32:32what that means is it natively understands audio
- 32:35and video from the ground up. So if you have
- 32:38a video of a deposition, if you have an audio
- 32:41file of something and you want to transcript,
- 32:43or if you want to understand the sentiment of
- 32:45what people are saying, Gemini can go get that
- 32:48for you. In fact, I'll show you because I'll
- 32:51bounce back to it later. But I pulled in a video
- 32:54earlier. I grabbed this 20 minute 21 minute video
- 32:59summary, you know off of YouTube It was a lawyer
- 33:01being deposed and I said hey just give me a detailed
- 33:04analysis of this video, right? So it went through
- 33:07and it created here I'll make the text a little
- 33:09bit bigger so you guys can see it right a summary
- 33:12of what's in this video So the nice thing is
- 33:14like I said Gemini, it doesn't just read the
- 33:17text of the video It natively understands what's
- 33:20going on there. And the best way I can showcase
- 33:23this is Gemini has another tool called AI Studio,
- 33:27and it is without a doubt one of the most powerful
- 33:30tools out there if you know how to use it. And
- 33:33you have to set it up correctly so that it doesn't
- 33:35train in your data. But it lets you access the
- 33:37most powerful models and do a lot of tweaking
- 33:39on the models. So I uploaded that same video,
- 33:42which, by the way, was, as you can see here,
- 33:44386 ,000 tokens. So, you know, ChatGPT's limit
- 33:48is 128 ,000. Claude's is 200 ,000. So. even if
- 33:52it took native video, it couldn't actually take
- 33:54this video. But what you'll see here is I asked
- 33:57it to explain the range of emotions shown in
- 33:59this video. Does this supposed attorney ever
- 34:02get mad, flustered or angry? If so, tell me when
- 34:05and give me a timestamp. So it went through and
- 34:07thought about it. And it says, yep. In fact,
- 34:10here's when the attorney got flustered and agitated.
- 34:13And it gave me a time stamp. It says, here's
- 34:14what he was saying. Here's what was going on.
- 34:16And here's when there's clear anger. So I've
- 34:19worked with a number of attorneys on when they
- 34:21have videos of kids or other things they're trying
- 34:24to figure out. Were they sad? Or when did they
- 34:26get happy? Or when did they appear to be deceitful?
- 34:29Gemini is the only model, like I said, that understands
- 34:31sentiments, emotion, voice tone. It is incredibly
- 34:35powerful. And I would encourage you to use it.
- 34:38They also introduced something called DeepThink
- 34:40this year, which is basically similar to ChatGPT's
- 34:435 Pro, a super advanced deep reasoning model.
- 34:46It's for your hardest problems. NotebookLM, I
- 34:50don't think this was out the last time I presented.
- 34:52It is an incredibly powerful tool. If you have
- 34:55not used NotebookLM, what NotebookLM is, I'm
- 34:59gonna jump to this page on it for a second. It
- 35:01lets you create these workbooks where you can
- 35:04dump in up to 300 different sources. And each
- 35:07one of these sources can be quite large. So you
- 35:08can put in a tremendous amount of content about
- 35:10a case. What happens in these workbooks is their
- 35:14AI model only goes when you ask it questions
- 35:17and analyzes the content that's in that workbook.
- 35:20It does not bring in any outside knowledge. It
- 35:22does not do any web searching. It just searches
- 35:25what's in there to give you the best answers.
- 35:27What made it really unique is its ability to
- 35:30do audio podcasts and now video podcasts and
- 35:34now the ability to create reports. So I'm going
- 35:36to go into this one and I could ask it a simple
- 35:39question. I could say when did this show me when
- 35:44Let's talk to it. Show me when this highlighted
- 35:47a 45 % boost in the text. So what it's going
- 35:51to do is I uploaded this single PDF file. So
- 35:54it's gonna go through and says, hey, here's when
- 35:56it talked about that. And if you hover over it,
- 35:58it actually shows you the text. So if I had had
- 36:0020 or 30 or 40 pages of content in here, it would
- 36:03actually show me all the places related to the
- 36:06question that I'm asking. And if I click on the
- 36:09button, it will actually then open up the exact
- 36:11document. So if you've got 100 different sources
- 36:13in here, it is a great way to answer questions,
- 36:16to find things, to say, hey, did we talk about
- 36:18this? Did we cover X or Y? It is super powerful.
- 36:22But the other really great use case is the ability
- 36:24to create audio and video podcasts about the
- 36:27content. So I'm going to play about 20 seconds
- 36:29of this audio podcast from this called Gemini
- 36:33Unleashed. And I'm just going to hit play here.
- 36:36Welcome back to The Deep Dive. Today, we're tackling
- 36:38a topic that's, well, it's everywhere, isn't
- 36:40it? AI, specifically. generative AI and the really
- 36:45profound impact it's starting to have on businesses.
- 36:48Absolutely. It's moving so fast. It really is.
- 36:50And for anyone, maybe not fully immersed, gen
- 36:53AI. So I'm going to pause there. So what this
- 36:56tool does and what makes it so unique is it takes
- 36:59all the content that you have uploaded into here
- 37:01again, whether you've got one document or a hundred
- 37:03documents and creates a very, very detailed and
- 37:06insightful podcast about that content to help
- 37:08you learn things. They also recently added a
- 37:11video version of this, and I'll just hit play
- 37:13on this for a second, where what it does is it
- 37:16actually will take whatever content you have,
- 37:18create slides about that content, and also create
- 37:21an audio podcast. All right, let's talk about
- 37:24AI, but let's cut through all the hype, okay?
- 37:28We're not going to talk about some, for you,
- 37:31for your team, brand new ways to work. And they
- 37:35have added a lot of capabilities. So they're
- 37:37allowing you now to create mind maps, to create
- 37:38flashcards. And right now, you can create different
- 37:41reports in Notebook and LAM briefing docs. They
- 37:44just announced today that they are announcing
- 37:45a ton of new formats. financial reports, stock
- 37:49reports, strategy briefings. So that's gonna
- 37:51be rolling out really soon. But again, it is
- 37:54an incredibly powerful tool. And the one that's
- 37:56nice thing about this is from the first day they
- 37:59launched this tool, they said that everything
- 38:01you put in here is gonna be 100 % private. So
- 38:03that's fantastic. They have a great deep research
- 38:08tool just like ChatGPT does. I use it all the
- 38:10time. They also launched a VO3 video tool. I'll
- 38:14show you that for a second. They've also launched
- 38:15something that was making the rounds if you've
- 38:17been following AI called NanoBanana, which is
- 38:20a flash image, an image generator, but it lets
- 38:23you create very tailored edits. I'll show you
- 38:25a very simple version of what you can do. But
- 38:29they continue to innovate. So one of the things
- 38:31I did is I created a very, very simple video
- 38:34on here. I was saying, hey, let's do a history
- 38:37of the Beverly Hills Bar Association. So let
- 38:39me just go grab this. I won't play the whole
- 38:41thing. But I gave it a prompt. And you can actually
- 38:44have it create incredibly long, detailed videos
- 38:47now. Different scenes, different cuts. You can
- 38:52have character consistency. All the kind of things
- 38:55that you want to do there. And if you want to
- 38:57do nano banana, like I said, video editing, this
- 39:02is. as I said, taking the internet by storm right
- 39:05now, because what you can do is you can take
- 39:07an original image and say, go change one thing
- 39:10about this, right? You know, go change a profile
- 39:12picture, like make this a professional headshot,
- 39:14or I said, hey, take this crime scene and add
- 39:17notes to the crime scene, right, based on where
- 39:19the items are. You can change colors, you can
- 39:21change images. So if you're looking at, you know,
- 39:24different images from different things, you can
- 39:25actually have it tailor things, change things
- 39:27to see what might look like from a different
- 39:29direction. Here's the fun version of it. So I
- 39:32took my giant Saint Bernard who's sleeping to
- 39:34Oregon last week and had a picture. I said, you
- 39:36know, make an image of us driving a bike, riding
- 39:38a bike around Crater Lake, right? Put us in a
- 39:40spaceship. Put us in the opera house in Paris
- 39:42or make him a making a chef. It is super, super
- 39:45powerful. In fact, the image that I had at the
- 39:48very beginning today. Was this one that I did
- 39:51of Scott or I started with a shot Scott and I
- 39:54are friends. I took a friend shot I said, you
- 39:57know make it a professional headshot. I said,
- 39:58that's not very friendly looking make him smile
- 40:00Then I said, you know, give him a purple tie
- 40:03and then you know, give him a charcoal suit So
- 40:05my point is you can actually do really interesting
- 40:08things with these tools to drive them forward
- 40:11But what I want to do now is get into some of
- 40:13the more interesting use cases because this is
- 40:15where the tools get really, really powerful.
- 40:17And I'm going to bounce back and forth across
- 40:19the different tools because I want to show you
- 40:20some of these tools live and how they interact
- 40:23and how they do things differently. So I've got
- 40:25a number of examples in here to highlight different
- 40:28capabilities. And actually, just before I do
- 40:30this, let me just go back to this agent. So all
- 40:33right. that agent we fired off to go do that
- 40:36research, right? It worked for eight minutes.
- 40:38It did it all. We could then download the report.
- 40:41We could get all the content and we could get
- 40:42all the stuff that it did for us and all the
- 40:44files that it created. So it went off and again,
- 40:47did that autonomously. For those of you that
- 40:50saw my first presentation, one of the things
- 40:53that these tools are amazing at is language translation.
- 40:56They have gotten absolutely incredible at language
- 41:00translation. And so when I ran this back in March,
- 41:04I said, hey, here's a press release. Convert
- 41:07this into French, Spanish, German, and simplified
- 41:10Chinese. And sure enough, boom, 30 seconds later,
- 41:13it's converted the entire press release to it.
- 41:16One of these cases I was working with someone
- 41:18recently is they said, hey, I got a case where
- 41:21I got a note in a divorce case from a prescription
- 41:26in a language I don't know. I can't read it.
- 41:29Can you help me translate it? But all I have
- 41:30is a scan of the image. So you can see the image
- 41:33down here. And I'm going to go grab this. I'm
- 41:36just going to go grab this text here. And we'll
- 41:39go off into Chat GPT here. And I'll leave it
- 41:42on auto for a second. I'm going to just go grab
- 41:44this image. Right this doctor's note and I'm
- 41:47basically just gonna go tell it hey attaches
- 41:50the doctor's note That's relevant from divorce
- 41:51case. I'm working on you know, it's from the
- 41:53Philippines It's a combination of English and
- 41:55Filipino give me detailed translation and description
- 41:58from this image, right? So the thing is it's
- 42:00gonna do two things one. It's gonna process the
- 42:02image And the second thing is, it's actually
- 42:05translating the languages. So it can read handwriting.
- 42:08It can read even really chicken scratch handwriting.
- 42:10So again, you can see, I mean, it was very, very
- 42:13fast and very good at reading that note and diagnosing
- 42:16that. So if you're working on international cases,
- 42:19if you're doing those kinds of things, these
- 42:21tools are incredibly powerful at doing that.
- 42:24And again, I could actually have it now translated
- 42:26back. I could literally say, you know, Can you
- 42:30translate this into German for me, please? And
- 42:34so it'll go off now, and it'll just start translating
- 42:37that entire thing into German. So again, if you're
- 42:41working in international cases, the ability of
- 42:43all these tools, not just ChatGPT, Gemini is
- 42:46amazing at it. Cloud is amazing at it. The capabilities
- 42:49are fantastic. Another really good use case is
- 42:54often you might be working with financial statements.
- 42:56In fact, I worked with a large accounting firm
- 42:58here recently in LA and I was showing how if
- 43:01you're pulling 10 Ks or 10 Qs or other financial
- 43:04statements. and you want to get information out
- 43:06of it, you can actually say, hey, go read everything
- 43:10in this document. And you can see the prompt
- 43:12here. You're an AI assistant. Go grab this 10k.
- 43:15I want you to go get EBITDA, revenue growth rates,
- 43:17net income, debt to equity ratios, things like
- 43:20that. And if the metric isn't in the document,
- 43:23I want you to calculate it for me. So this is
- 43:25a case where I'm going to run this quickly and
- 43:28I'm going to run it in a thinking model at the
- 43:30same time because I want you to see the difference.
- 43:33So there is going to be a version that's going
- 43:34to run here. I'll go run. Actually, I'll make
- 43:37this the thinking one. And shoot, I got to add
- 43:43it again. Hold on. Let me add that file again
- 43:44here. Hopefully, we'll add it quickly. We can
- 43:48add that prompt in. And while that's loading,
- 43:50I'm going to go start a new chat over here and
- 43:53do the same thing and say, let's go grab this
- 43:57presentation. All right. Hopefully. All right,
- 44:01so this one's loaded. This is the thinking one.
- 44:04And now we'll go run it off on this one. So while
- 44:07this is running, we'll let it go in a second.
- 44:09So it's going to read that document. And now
- 44:11it's going to go extract out all the relevant
- 44:13financial data out of that table, out of this
- 44:17PDF, no matter where it is. And so the output,
- 44:20when we get it back, is going to look something
- 44:21like this on the screen here. And like I said,
- 44:24it will go through. It will calculate EBITDA.
- 44:26It will do revenue growth, all the things that
- 44:27I want to do. It will create notes about it.
- 44:31The FAST model is really quick at doing this,
- 44:34right? It creates a table, it does these things,
- 44:36it tells me how it's creating the numbers, right?
- 44:38The thinking version here, as you can see, it's
- 44:40still going on. But what I'm gonna get in the
- 44:43output is a much more reasoned, much more thoughtful
- 44:45set of answers with a lot more description about
- 44:48the calculations and what it figured out there.
- 44:50So I'll come back to that in a minute so we can
- 44:52keep going, because I wanna get through the rest
- 44:54of the examples to show you what else is in here.
- 44:58One of the other really, really great things
- 45:01about these tools is their ability to actually
- 45:03understand charts now. And it's often that, you
- 45:06know, you might have a chart from, you know,
- 45:09from an exhibit or from maybe somebody gave you
- 45:11one as part of a deposition or you, you know,
- 45:13you did some discovery and you've got charts,
- 45:15you know, and it might be a hard, you know, a
- 45:17hard scan chart like this. And you want to have
- 45:20someone understand it or turn those, you know,
- 45:23turn that chart into numbers for you, right?
- 45:25These tools are so good at understanding what's
- 45:28in these charts now. So I've got a simple chart
- 45:29here about the streaming music industry. I'm
- 45:32going to use a really simple prompt here and
- 45:34just say, I want you to go analyze this chart
- 45:37for me. I'm not going to use the thinking model
- 45:39because I'm trying to keep this thing moving
- 45:41pretty fast. But if I go grab the streaming chart
- 45:46here, I'm going to paste this in and I'm going
- 45:49to launch that off and we'll see if this other
- 45:51one. No, it's still running all the calculations.
- 45:55And so it's calculating trends. And like I said,
- 45:57I could run this in the other models too. I could
- 45:59run it in Gemini. I could run it in Claude to
- 46:02go to the same thing. But the thing is, because
- 46:04I use the non -reasoning model, I didn't get
- 46:07a lot of description here. I got a couple of
- 46:09key highlights of what's going on here. But what
- 46:11you'll see here is, if I had turned on reasoning,
- 46:14which I did in the example over here, it actually
- 46:16gives me a lot more description about what's
- 46:18going on there. In fact, if I go load this into
- 46:21Claude, for example, and I use Opus 4 .1, it
- 46:25will actually give me a much more detailed, much
- 46:27more reasoned answer about what's going on there.
- 46:31So again, I'll let that run off in the background
- 46:33so we can keep going because I'm cognizant that
- 46:35we've got about 15 minutes left here. Here is
- 46:39another really great use case. In fact, I work
- 46:41with a valuation company to build a small product
- 46:44that actually automates this. So often you may
- 46:47have to understand what's in people's tax returns.
- 46:50And so these tools are fantastic at actually
- 46:54loading up tax returns, analyzing tax returns,
- 46:57and giving you the output in a way that you want
- 46:59it. So I'm going to actually just say, hey, I
- 47:01want to go grab this example here. In fact, I'm
- 47:04going to go actually run this in Google's AI
- 47:08Studio. I'm going to create a new chat here,
- 47:10make this a little bit smaller. I'm going to
- 47:13go grab these files. Right, I've got some tax
- 47:17returns from an organization It's gonna read
- 47:20those tax returns. You see how big these are
- 47:21for a second. I'm gonna paste in the the text
- 47:25that I wanted to do here, but I basically wanted
- 47:27to go do, read these tax returns and then start
- 47:30creating the line item import. Now, in fact,
- 47:33the product that I created for this company actually
- 47:35turns this into an income statement, a balance
- 47:37sheet, it even does multi -year stuff. What I
- 47:40want to show you here is just the ability for
- 47:42it to go off in a run and just start grabbing
- 47:44this information pretty quickly and synthesizing
- 47:47it in a way that is easy and makes a lot of sense.
- 47:52So here I went back just to the the original
- 47:55EBITDA stuff here. And you'll see from the thinking
- 47:57model, you'll see here, look, it gives you a
- 48:00lot more detail about what's going on here versus
- 48:02that old auto model. So that's why I go back
- 48:04and I say, I encourage you to use those versions
- 48:07of the tools. Same thing here, if I look at what
- 48:10Claude created now in terms of understanding
- 48:12that PNG, right? It actually gave me a lot of
- 48:15context, a lot of insight about what was happening
- 48:17there. One of the things you can do, I don't
- 48:20have it in the demo, is let's say that I wanted
- 48:22to be able to recreate the numbers in this chart.
- 48:24I could actually tell Claude to create a spreadsheet
- 48:27of me and estimate what it believes each of the
- 48:30numbers are for the line items on that chart.
- 48:32So if you ever had a use case to do that, you
- 48:34can go ahead and go do that too. Let's just see
- 48:37if this is done firing up. No, it's still working
- 48:39through it. So again, we'll keep going here.
- 48:43This is really similar to a couple of these other
- 48:45examples, so I won't go through it. But I want
- 48:46to highlight the fact that, again, I can give
- 48:49it a spreadsheet. Let's say I was doing some
- 48:51discovery, and I wanted to go identify anomalous
- 48:53transactions. I can upload a spreadsheet and
- 48:56say, hey, this is a set of transactions from
- 48:58the year 2023. Create a summary table by category.
- 49:01And you can see, again, Claude does a beautiful
- 49:03job visualizing the output. And then go identify
- 49:06any anomalous expenditures. In this case, there
- 49:08was a lot of jewelry in the credit card statements.
- 49:13So I already did this example, so I won't go
- 49:16through this, but again, what I wanted to highlight
- 49:18was if you're using, if you've got video, you
- 49:21can upload easily up to a 60 minute video into
- 49:25Gemini and have it process that video and give
- 49:28you a complete transcript. It can give you the
- 49:29sentiment, the tone, you can find things. It
- 49:32is a super, super powerful tool for doing that.
- 49:35So for the last couple of minutes, I created
- 49:38a bunch of family law examples, because I presented
- 49:40the family law group last time, and I wanted
- 49:41to just highlight a bunch of different ways that
- 49:44you can use these tools. So at this point, I
- 49:46built out hundreds of use cases, but I want to
- 49:48highlight a couple of simple ones that shows
- 49:50you what these tools can do. So let's say you're
- 49:53creating a, you know, you want to do a declaration
- 49:55build, right? You want it to be, you know, trauma
- 49:57informed and exhibit friendly. What I created
- 50:00for you in here. are a bunch of different prompts
- 50:03and use cases to do it. And one of the things
- 50:06that you'll see here is whenever you're creating
- 50:08a prompt for Chachi PT or Gemini or Claude, you
- 50:12should always give it a role. You are a family
- 50:15law associate or you are a seasoned litigator
- 50:17or act as a judge in the Superior Court for Los
- 50:21Angeles County or act as a juror in this case.
- 50:24If you tell these models to take on a persona
- 50:28or a role, it will do that and it will analyze
- 50:30the content from that point of view, right? So
- 50:33we can say, hey, take the narrative below into
- 50:36a declaration that is chronological. It's got
- 50:38numbered paragraphs, right? Each material assertion
- 50:41references an exhibit, include a short timeline,
- 50:44you know, add a gaps needed as exhibits and only
- 50:46use the pasted in text and file names. Don't
- 50:49invent facts or citations and flag ambiguities,
- 50:52right? So. Instead of running all these real
- 50:55time, what I'm going to do is I have pre -created
- 50:57these just to show you these. So if I go into
- 51:00here, this was the declaration builder. And what
- 51:05you'll see here is I literally pasted this role
- 51:07in. And then, by the way, what I told it is I
- 51:09said, hey, because it said, all right, hey, give
- 51:11me all those exhibits, because that's what I
- 51:13told it to do, right? It says, hey, go give me
- 51:15all this content, which I don't have in here.
- 51:18So I said, now, just go make up the content so
- 51:19I can run this prompt and show you what's in
- 51:21here. So it says, I'm going to fabricate some
- 51:22of the stuff. So it's going to draft declarations.
- 51:26So I, Jane Doe, declare as follows. On January
- 51:292nd, this. On January 5th, this. Key timeline
- 51:32of events. Gaps in needed exhibits. So the nice
- 51:36thing is you can actually have it go through
- 51:38and do this stuff for you. So if you've got all
- 51:40the content, you can just give it the instructions
- 51:42about what to do. And what you'll almost always
- 51:45see now in the bottom of Chat GPT, is it will
- 51:48almost always ask you a follow -up question that
- 51:50says, hey, do you want me to create another version
- 51:52of this or a version formatted for a court or
- 51:54something else? So excuse me, if I say yes, it
- 51:58will go ahead and do that. I didn't do that in
- 51:59this particular example, but that's something
- 52:01that you can easily do. Maybe you want to create
- 52:04a BIF -style co -parenting email rewriter. You
- 52:07want to have something really simple. You want
- 52:09an ultra -concise version that's less than 90
- 52:11words. You want a neutral and solution focus
- 52:13one less than 140. Or maybe you want a boundary
- 52:16setting one less than 160, right? And make sure
- 52:18you remove blame, accusations, legal conclusions,
- 52:21keep the dates, times, requests specific, and
- 52:23then paste in their client content. So again,
- 52:25if I go look at this one, if I go back and look
- 52:28at the BIF one, you'll see what it did is it
- 52:33says, OK, and I told it to, again, make up the
- 52:34content. So it's saying, all right, I've created
- 52:37an original version here, original draft. But
- 52:40then it says, all right, I've created version
- 52:41A for you. Ultra concise, less than or equal
- 52:43to 90 words. So last Friday, pickup was delayed
- 52:46by 30 minutes, which caused confusion for the
- 52:48children. Going forward, please confirm these
- 52:50things. Nice, neutral, in a manner that you want.
- 52:53Here's one that's less than 140 words. Here's
- 52:56one that's less than equal to 160 words. So again,
- 52:59really simple, really straightforward, and very
- 53:02powerful. And the nice thing is you can, again,
- 53:05and you can say, no, I didn't like these, rewrite
- 53:06these again, do this as many times as you want.
- 53:09It is really, really good at doing these things.
- 53:12Let's say you wanted to create a parent custody
- 53:14and safety note. I want to create three parenting
- 53:16schedules. And this is one where I think I created
- 53:18an infographic. So let me jump into this one.
- 53:21But you can say, look, I want to give you these
- 53:23different options. Option one is 223. Option
- 53:25two is 2255. Option three is week on, week off.
- 53:29Give me the pros and cons, transition times,
- 53:32exchange locations. And by the way, you can build
- 53:34these as projects or custom GPT. So you don't
- 53:37actually have to remember these. So if you create
- 53:38a lot of these, You can use these, whether you
- 53:41use a project or a custom GPT and chat GPT or
- 53:45a gem in Gemini or a project in Claude, you can
- 53:49have these prebuilt and make your life really
- 53:52simple as you're doing these things. So if I
- 53:54go find the parenting time schedules here, what
- 53:57it did is it created a set of facts for me. It
- 54:00says, hey, here's an option one, two, two, three
- 54:03schedule, parent day, Monday, Tuesday, parent
- 54:06day, Wednesday, Thursday, alternating weekends,
- 54:08Friday to Saturday. here are the pros and cons,
- 54:10here's transitions, here's locations, exactly
- 54:12what I asked for, right? Gave me option three,
- 54:15gave me holiday schedules, and then I said, do
- 54:18you want me to create a clean template? And what
- 54:20you'll notice here is I said, yeah, and the first
- 54:23time I did it, it failed, right? So you will
- 54:25see this, right? These things can occasionally
- 54:27fail. So I said, hey, I want you to go actually
- 54:30create it in HTML and react, right? That first
- 54:32one didn't work. So I want you to create a much
- 54:35more visually beautiful one here. And what you'll
- 54:38see now is these tools. And ChatGPT is good,
- 54:42Gemini is better, and Claude is the best at creating
- 54:45these absolutely gorgeous visualizations on these
- 54:48things. So it actually created here the three
- 54:51different options here. We can print this out.
- 54:53We can save it as a document. But most people
- 54:56don't know that these tools are really, really
- 54:58good at creating things like this now. Check
- 55:05for consistency and attachments. Review the following
- 55:08income and expense declarations. Again, you can
- 55:11tell the tools what you want it to do. Go look
- 55:13for internal inconsistencies. Go look for missing
- 55:15attachments. Look for ambiguous entries. So if
- 55:18I go look at this one, and again, it made it
- 55:21a lot easier having these things all pre -created
- 55:24here. I'm going to go take a look at this one
- 55:26here. And again, I just had it made up the data.
- 55:30But you can say it's got an excerpt here. It's
- 55:32got made up values for the demonstration. And
- 55:35then it went through the issues table. What is
- 55:38the issue? Why it's a problem? What you should
- 55:40do to fix it? So it really, really is good there.
- 55:44Hang on. I just have to let my dog before he
- 55:46takes out my screen. Come on, buddy. Ready? Go.
- 55:52All right. That's the joys of a big dog. But
- 55:54anyways. Again, a really, really powerful tool,
- 55:57a really powerful use case for going and doing
- 55:59this. Here's another one when you want to do
- 56:02more complex calculations. If you need to do
- 56:04the more Marsden calculations, which is calculating
- 56:07the value of a home that was owned before divorce,
- 56:10you can actually say, hey, build the outline
- 56:13from the inputs below. I need the variables.
- 56:15I need the definitions. I want to see a step
- 56:17-by -step formula with placeholders. I'll put
- 56:20a checklist of all the documents needed to populate
- 56:22each variable, right? Don't compute anything
- 56:25unless values are provided. So again, it did
- 56:28an absolutely incredible job of doing this. And
- 56:31again, these tools have been trained on just
- 56:33about everything that's out there. So if I go
- 56:35back to the top here, so it started with, hey,
- 56:37here's all the variables and the definitions.
- 56:39So if people aren't familiar with these or if
- 56:41you've passed this off to a new associate, you
- 56:43could say, hey, This is all the things that you
- 56:45need to go focus on. Here's the standard terms
- 56:47and conditions and things that we use as part
- 56:49of this. Here's the step -by -step formulas with
- 56:52the placeholders. So again, it says, what should
- 56:54we do? When do we acquire things? All these different
- 56:56elements to all these things here. Here's a document
- 56:59checklist. Here's a worksheet that someone could
- 57:01fill out. I mean, I basically told at the end
- 57:03to basically go use the calculations, the fake
- 57:05ones, to run it just so they could see what it
- 57:08would look like. But you could, again, give somebody
- 57:09a complete example of doing this. But I just
- 57:12wanted to show you that these tools have gotten
- 57:15unbelievable at math. They've gotten unbelievable
- 57:17at formulas and at following rules. And they
- 57:20can do multi -step items with little or no issue.
- 57:24So just a couple more things. I won't go through
- 57:26all these because we've just got three minutes
- 57:28left. But if you're doing an ex parte checklist
- 57:31and declaration, you can go through here and
- 57:34have it run all these things. And again, I will
- 57:35just show you the output of what that looks like.
- 57:39Everybody is going to get all of this content,
- 57:41all the things that you've got here. So it went
- 57:44through. And again, I had to make up data. But
- 57:46it says, hey, here's the checklists we want to
- 57:47go through. Here's the notifications that we've
- 57:49got to do. Here's the timing window that we need
- 57:51to focus on. So again, talking points, fact numbers,
- 57:55exhibits. And then I actually had it create a
- 57:58court -style declaration that I could use and
- 58:00copy and paste into whatever my favorite text
- 58:02editor is, whether it's Word or Google Docs or
- 58:05whatever it happens to be. So again. really powerful
- 58:09way of using these tools to do these things.
- 58:12You can do two past self critiques on a declaration.
- 58:15So let's say you've got a declaration you want
- 58:17to go through and do, hey, first critique is
- 58:20give me the 10 most significant weaknesses. Is
- 58:22there ambiguity? Are there missing dates? Are
- 58:24there conclusionary statements? Are there lack
- 58:25of exhibits? And then in the second version,
- 58:29go give me a revised version that addresses those
- 58:31weaknesses and mark the reversions with fixed
- 58:33comments. So if I go look at that one, again
- 58:37what we'll see here is it says okay here's a
- 58:39fake declaration for the exercise right here's
- 58:41my declaration right my spouse and i separated
- 58:43sometime in the summer of 2022 we had many fights
- 58:46about money she was irresponsible with our joint
- 58:48funds so it says all right here's your first
- 58:50critique right here's the ambiguity right sometime
- 58:53in summer 2022 is vague we need an exact date
- 58:56conclusionary statements right so then it goes
- 58:58through and it offers to fix those things for
- 59:00me So then it says, here's the exact data on
- 59:02it. So again, use these tools as collaborators.
- 59:05They are really, really powerful there. If you
- 59:08have a ton of content, I won't go through these
- 59:10because we're basically at time here. You can
- 59:13go through and you can get thousands of texts
- 59:15if you have them. You can put them in timeline
- 59:17sequence. You can have them cluster things. You
- 59:19can have it create exhibits for you. And that's
- 59:22really it. So I just want to hit two more things.
- 59:26The first thing I want to highlight is, look,
- 59:28these models have gotten unbelievably advanced
- 59:31this year. These reasoning models are absolutely
- 59:33incredible in what they can do. I encourage everyone
- 59:37to use the reasoning models. You get way better
- 59:38results. They're all becoming multimodal, but
- 59:41the only one that really understands all that
- 59:43today is Google's Gemini. Don't use the commercial
- 59:47LLMs to find cases. So you see it all the time.
- 59:50Someone went to ChatGPT or Claude or Gemini and
- 59:52said, hey, find me a case that supports these
- 59:54things. Please, please, please don't do that.
- 59:57Only use... the commercial tools that you can
- 1:00:00pay for that have all the history of all the
- 1:00:03cases and things like that. Hallucination rates
- 1:00:06are gone down, but they're not zero, right? And
- 1:00:09then the last thing I'll just mention is if you're
- 1:00:11looking for help, this is how I work with firms.
- 1:00:13I do a much more expanded version of this workshop,
- 1:00:16right? I typically do two to two and a half hours
- 1:00:17of, hey, let me really take you through how you
- 1:00:20can use these tools. What do your workflows look
- 1:00:22like? Things like that. I do all day training
- 1:00:24sessions for organizations to really get your
- 1:00:26people up to speed so they know how to do these
- 1:00:29things. I build custom agents, and I offer effectively
- 1:00:32AI on demand for people that want retainers and
- 1:00:35assistants and using these AI tools just to be
- 1:00:37as productive as they can. So lastly, I've got
- 1:00:40all my contact information on here. And I do
- 1:00:43publish a newsletter several times a week with
- 1:00:44just the latest information around AI. It is
- 1:00:47switching to be much more legal specific. It's
- 1:00:49been very general. So if you are interested in
- 1:00:51signing up, it's completely free. Feel free to
- 1:00:53do that. So with that, I know we're at the end
- 1:00:55of the hour and I will go back and I will stop
- 1:00:57sharing and just see if there are any questions
- 1:01:00from the group. So yes, we actually have quite
- 1:01:04a few questions. I can generalize them for you
- 1:01:07in the interest of time. So a lot of the questions
- 1:01:10we're receiving is basically, you showed a wide
- 1:01:13array of tools, but do you have them ranked on
- 1:01:16which ones are best for certain activities, like
- 1:01:18maybe Claude is better for writing, Gemini calculations,
- 1:01:22or maybe chat GPT reasoning? Do you have kind
- 1:01:25of like a chart for that? So I don't have a chart
- 1:01:27in here for it. I mean, I'll talk verbally about
- 1:01:29it for a second. So historically, Claude has
- 1:01:33definitely been the best writer. So if Claude
- 1:01:34is a 10 out of 10 in writing, the other tools
- 1:01:38have gotten remarkably good. With the most recent
- 1:01:40updates, Gemini and ChatGPT, ChatGPT up until
- 1:01:44version 5 was probably a 6 out of 10, to be perfectly
- 1:01:46honest. It's now a solid 9 to a 9 and 1 half,
- 1:01:49but only if you use the thinking model. If you
- 1:01:52use the auto model, it still writes like a robot.
- 1:01:55So you only get their best thinking in the In
- 1:01:59the thinking model and are the best writing in
- 1:02:01the thinking model Gemini also has become a very
- 1:02:04very good writer a very very good reasoner in
- 1:02:06that stuff So they're very similar When I turn
- 1:02:09to analytics, to be perfectly honest, if I'm
- 1:02:11dealing with a lot of spreadsheets, I typically
- 1:02:12work with ChatGPT if I only have a handful of
- 1:02:16spreadsheets, because it's really good at calling
- 1:02:17Python tools and doing analysis and fixing data.
- 1:02:21If I have a lot of documents, though, I'll use
- 1:02:23Gemini because the context window is so large.
- 1:02:26And Claude has probably the smallest amount of
- 1:02:28files you can upload. So again, for writing,
- 1:02:31I used to exclusively use Claude. I use ChatGPT
- 1:02:345 thinking and Gemini quite a bit now, almost
- 1:02:36as often as Claude. And reasoning, if you're
- 1:02:40using GPT -5 thinking, it's now as good as, you
- 1:02:43know, Cloud 4 .1 Extended Thinking and Gemini
- 1:02:475 .2 .4. So they're all coming close together.
- 1:02:49But again, multimodal, the only tool you can
- 1:02:51really use at this point is Gemini. Great. And
- 1:02:55then in regards to privileged information concerns,
- 1:03:00some of our viewers are a bit... confused about
- 1:03:04that aspect, they want to know why it's not important
- 1:03:07to be trained on AI if they're already putting
- 1:03:09in information that's been redacted for sensitive
- 1:03:12information. Sorry, can you reframe that? So
- 1:03:16in regards to the privacy concerns with some
- 1:03:19of the models not natively redacting that information
- 1:03:23from their training, some of the members want
- 1:03:25to know if that should still be a concern if
- 1:03:27they're putting in information that's already
- 1:03:28been scrubbed of sensitive information, basically.
- 1:03:32So that's why I said at the beginning, first
- 1:03:35of all, only use the paid models and only use
- 1:03:37the ones where you've turned off privacy. even
- 1:03:40if you've redacted the stuff, you don't want
- 1:03:42those cases getting out there if there's a potential
- 1:03:44chance for it to train in your data. So that's
- 1:03:45why that whole upfront piece about like how to
- 1:03:47switch cloud, how to switch chat, GPT, everything
- 1:03:49is so important. I mean, I really encourage firms
- 1:03:52to use either the teams tools or the enterprise
- 1:03:55tools. That way you have the most advanced security
- 1:03:58and it will never train in your data. But if
- 1:04:00you're a small practitioner and if you are going
- 1:04:02to use the consumer ones, which like I use, I
- 1:04:05know a lot of people do, please, please, please
- 1:04:08turn off that data, turn off the training, even
- 1:04:10if you've redacted the stuff, you just don't
- 1:04:12want it showing up out there. All right, and
- 1:04:16then if you had to pick just one AI for the best
- 1:04:23option overall for a lawyer, which one would
- 1:04:26you recommend if you had to pick one? You know,
- 1:04:29it's hard to answer it on a call like this, to
- 1:04:31be honest, because it depends on what infrastructure
- 1:04:34you're using. For example, if a firm, one of
- 1:04:36the firms I presented to recently was already
- 1:04:38using Google Workspace, right, for all their
- 1:04:40email, for all their document management, things
- 1:04:41like that. So it made perfect sense for them
- 1:04:43to use Gemini. For someone that's a Microsoft
- 1:04:47shop, they might want to be using either ChatGPT
- 1:04:49or Claude. But it really depends on the kind
- 1:04:52of cases you work on. So I'm happy to talk to
- 1:04:54people kind of one -on -one if you want to have
- 1:04:56that discussion. And I hate to like weasel out
- 1:04:59of the answer, but there isn't a single answer
- 1:05:00that I can generically give for everybody because
- 1:05:02everybody's needs are a little bit different.
- 1:05:04And as I've done these, I found that... Most
- 1:05:06people pick two tools, not one to be perfectly
- 1:05:09honest. They find that they get the best results
- 1:05:10using two and they often go back and forth. Very
- 1:05:13few people use three or four, but that's kind
- 1:05:16of how I would have to answer it because there's
- 1:05:18not a single answer and they continue to evolve
- 1:05:21also. They keep getting better and better as
- 1:05:23they go forward. And then let's see. Somebody
- 1:05:29said that their firm has a subscription to Copilot
- 1:05:32and that they're told it's the same as Chat GPT.
- 1:05:35So they're curious if it really is the same and
- 1:05:37does it have the same features as Chat GPT 5
- 1:05:40.0. So you'll notice that I did not talk about
- 1:05:43Copilot on here. Microsoft has, for whatever
- 1:05:47reason, effectively lobotomized Copilot in their
- 1:05:52version of the version of ChatGPT that they use,
- 1:05:56right? You're basically getting ChatGPT 5 Auto,
- 1:05:59and you can't, for example, access the same level
- 1:06:02of thinking all the time. There is no version
- 1:06:04of ChatGPT 5 Pro. So I know that people have
- 1:06:07talked about that. I have not had a single client
- 1:06:09in the end that has ended up using Copilot for
- 1:06:12that exact reason. Every firm I've talked to
- 1:06:14has realized that it's just not as good, to be
- 1:06:17perfectly honest. It doesn't have all the same
- 1:06:19features. I don't know why Microsoft has done
- 1:06:22it that way, but I, in good faith, as I advise
- 1:06:25clients, cannot recommend it. Great. And then
- 1:06:30let's see. Do you have any thoughts on the tools
- 1:06:36you covered today versus ones such as Lexis AI
- 1:06:39Protege and Westlaw Co -Counsel? Yeah. No, I
- 1:06:43don't. I purposely don't typically compare those
- 1:06:46in here because they serve different needs typically,
- 1:06:48right? These are general large language models
- 1:06:50that are very specific and can do a ton of different
- 1:06:52things. The legal specific ones, obviously, you
- 1:06:55know, they're very focused on cases and other
- 1:06:57things. And so I just I don't compare and contrast
- 1:07:00them. I mean, I've done a lot of research around.
- 1:07:03what those tools can do, but I typically don't
- 1:07:05cover that in here. I do that more as bespoke
- 1:07:07work for people. So they're very different. I
- 1:07:09mean, the kind of analytics and things you can
- 1:07:11do, the image generation, the infographics, the
- 1:07:13ability to process a video, the ability to combine
- 1:07:16multiple spreadsheets and extract that insights,
- 1:07:18you can't get those from a lot of those other
- 1:07:20tools that are out there. They're very, very
- 1:07:22specific for analyzing cases and drafting briefs
- 1:07:24and things like that. Great. Let's see. I believe,
- 1:07:37let's see. In regards to that financial statement,
- 1:07:44case use that you went over, would it be possible
- 1:07:47to ask it to evaluate financial statements with
- 1:07:49other supporting documents and ask it to look
- 1:07:51for fraud? Absolutely, yeah. You might have to
- 1:07:54define how you really want to determine fraud.
- 1:07:58You could start by that. I will tell you, I worked
- 1:08:01with a very large valuation company in the East
- 1:08:03Coast. had it reconstructing, you know, it looks
- 1:08:08at their final reports, it looks at all the inputs,
- 1:08:09all the financial statements, all the outside
- 1:08:11research and all the data, right? It combines
- 1:08:13to actually fact check everything that's in there.
- 1:08:15In fact, I didn't actually show an example of
- 1:08:18the fact checking that I've got. I've actually
- 1:08:20built, and if people reach out, I'm happy to
- 1:08:23share it. I have a detailed fact checker that
- 1:08:27it literally goes through. And we'll all say
- 1:08:30fact check this article, right? You can see the
- 1:08:32results here. It will go through, if you give
- 1:08:34it something, it will go through every single
- 1:08:35fact in a document, right? And identify, are
- 1:08:39these, I mean, I don't have to tell it the facts.
- 1:08:41It reads sentence by sentence. It's okay. Let
- 1:08:42me go find everything that's in there. And it
- 1:08:44says, how accurate is this? Is it partially accurate?
- 1:08:46Is it not accurate at all, right? What are the
- 1:08:49sources it does? So you can do the same thing
- 1:08:51from a financial standpoint. You could upload
- 1:08:52some spreadsheets and some supporting data and
- 1:08:54say, hey, How good is this, right? Is there issues
- 1:08:57that don't seem correct here and potentially
- 1:09:00fraudulent? So it could certainly take a pass
- 1:09:02at it, but you might have to just be a little
- 1:09:03more specific about what does fraud mean in the
- 1:09:06case that you're looking for. But I would certainly
- 1:09:08try it 100%. All right, and then we'll end on
- 1:09:13this question. So this attendee noticed that
- 1:09:15you asked specific questions using terms like
- 1:09:17extract and assistant. Are those terms critical?
- 1:09:20Can similar terms be used to get the same result?
- 1:09:23If specific terms are needed, is there a place
- 1:09:25to find those terms? No, you don't have to be
- 1:09:28that specific. You can say, pull all the data
- 1:09:30out of this spreadsheet, or extract the data,
- 1:09:33or hey, go read this spreadsheet and create a
- 1:09:35summarized table for me. These tools, the one
- 1:09:38thing that's amazing about them is they are really,
- 1:09:40really good at understanding what you're trying
- 1:09:43to say, even when you have typos. I mean, I often
- 1:09:45don't even fix typos anymore because it really
- 1:09:47knows what you're trying to do. So, yeah, no,
- 1:09:49there isn't a specific set of terms. The one
- 1:09:51thing I would say is if I go back to these examples,
- 1:09:53let me go back to one up here if I had one, just
- 1:09:57like when you talk about like the prompt, like
- 1:09:59you said, the role, the task, and then any rules
- 1:10:02and your kind of guidelines. the more specificity
- 1:10:05you can give these tools. just the better, higher
- 1:10:09quality output you're going to get out of these
- 1:10:11things. So most of the prompts that I give are
- 1:10:14very, very lengthy. I tell it all the things
- 1:10:18I want to do. I tell it all the analysis I want
- 1:10:19to happen. If there are constraints about timing,
- 1:10:21I give it that. I tell it what role I want it
- 1:10:24to be. Because again, it just saves me having
- 1:10:27four or five chats when I can have one chat to
- 1:10:29get to the answer that I want. So you'll learn
- 1:10:32over time as you use them. Yeah, I'm sure there
- 1:10:35are some words that make it a little bit better,
- 1:10:36but there is definitely not a set of industry
- 1:10:38-specific terms that you get higher quality output
- 1:10:41from. All right. And I believe that about wraps
- 1:10:46up Q &A. Thank you so much for your time today.
- 1:10:49The presentation was fantastic, and I hope our
- 1:10:51attendees enjoy the rest of their afternoon.
- 1:10:53Thank you so much. Bye -bye.