Latest / CLEs You Actually Want to Hear / AI in Practice ChatGPT, Claude, and Gemini — 2026 Summer Update
Transcript
- 0:03I am Mark Hoag, chair of the Beverly Hills Bar
- 0:07Association's AI and the Law section, along with
- 0:10my co -chair, Amanda Harris. And I'm also the
- 0:12founder of Axiomic LLC, an AI -first web and
- 0:16iOS app development company. And for my day job,
- 0:19though, I do still run hoagelaw .ai, where I
- 0:22serve as fractional GC to primarily AI -first,
- 0:26OK, I guess that's everybody, SaaS companies
- 0:28and tech startups. Yet another Amazing intro
- 0:32for our speaker today, Stephen Smith. Stephen
- 0:34is the CEO of Intelligence by Intent, where he
- 0:37provides AI strategy advisory and training to
- 0:40law firms and professional services organizations.
- 0:43He has delivered AI training and strategic guidance
- 0:45to over 1 ,000 organizations with a focus on
- 0:48practical adoption, governance, and measurable
- 0:51ROI rather than just theoretical overviews. Stephen
- 0:55brings over 25 years of executive leadership
- 0:57experience, including CFO and SVP rev ops roles
- 1:01at multiple organizations and strategy partner
- 1:04roles at IBM Global Services and TCG Advisors,
- 1:08developing growth strategies for Fortune 100
- 1:10technology companies. Stephen holds an MBA from
- 1:14the Wharton School at the University of Pennsylvania
- 1:16and writes the Intelligence by Intent newsletter
- 1:20at smithsteven .com. That's Smith. S -T -E -P
- 1:26-H -E -N. So thanks, everybody, for being here
- 1:28today. Stephen, take it away. Awesome. Thanks,
- 1:31Mark. Let me go ahead and share my screen here.
- 1:34I'm excited to be back and excited that everyone
- 1:37is here on at least what is a warm day in Los
- 1:41Angeles. So let me get the screen up here, move
- 1:45a couple things out of the way. And I assume
- 1:49you guys can see the slides OK? All right, well,
- 1:53I will assume the answer is yes. Well, look,
- 1:55thank you for having me back. There has been
- 1:57a ton of updates since I last talked about AI
- 2:01with the Beverly Hills Bar Association. There
- 2:04are new models, there are new capabilities, there
- 2:06are new ways of doing work. So we're gonna cover
- 2:09a lot of those different things today. But I
- 2:11always like to start with this slide. And I asked
- 2:14people when I do keynotes and things, you know,
- 2:15when is the last time that you spent three hours
- 2:18on something that AI can do in three minutes?
- 2:20And I hope what you'll see over the course of
- 2:22the next hour is just how much these tools can
- 2:25do now. As I said, it has changed an incredible
- 2:27amount and I'm excited to show you some of the
- 2:29things that are new. So. Over the next 50 minutes,
- 2:32we're going to talk about a couple of things.
- 2:34The first thing is what I call the shift, right?
- 2:37And so in the early days of using AI tools, it
- 2:39was really more like a chat bot, ask a question,
- 2:42get an answer, drop in a paragraph, get some
- 2:44changes. We have really moved now into the work
- 2:47phase where these tools can do real work for
- 2:49you. And you'll see a bunch of examples of that
- 2:51over the course of the day. Then we'll talk about
- 2:53what's new in Claude. Mark and I were talking
- 2:55beforehand about the changes, the Opus 5 model.
- 2:58And we'll talk about some other things here.
- 3:00And then we'll turn our attention to ChatGPT
- 3:03and what they're doing with ChatGPT work and
- 3:05codecs. I was really hoping for more announcements
- 3:08on Gemini. They were supposed to release their
- 3:10new model. They haven't done that yet. So I will
- 3:12talk about a couple of things there. And I'll
- 3:14talk about some of the things that you can do
- 3:16to drive AI adoption going forward. So I wanna
- 3:20start with this point, cause this is one of the
- 3:21most important things. And if you, you know,
- 3:24see any of my writing, if you see me give a keynote,
- 3:27I talk about this a lot. And the big shift that
- 3:29I'm seeing and others are seeing now with, you
- 3:32know, Anthropic, who does Claude and OpenAI that
- 3:34does Chatch BT and even Gemini with Google, Google
- 3:38Gemini, is they're no longer selling answers,
- 3:40they're selling work, right? So all three vendors
- 3:43are now shipping, you know, the same layers,
- 3:44you know, frontier models, firm knowledge, connections,
- 3:47playbooks, agents. You know, both ChatGBT and
- 3:51Claude have office connections. Things can run
- 3:54in the background. They can run in the cloud.
- 3:56They are really changing what the expectation
- 3:59is for what we're doing here. And the way I describe
- 4:01it is I talk about the five modes of AI work.
- 4:05And in the early days, if you're using... Chachi
- 4:08BT or Gemini or, you know, Claude, you know,
- 4:11two years ago, three years ago, we were really
- 4:14more of a replacement for Google, right? It was
- 4:16ask a question, get an answer, tell me about
- 4:18this law, you know, does this memo make sense?
- 4:21You know, how can I make this argument a little
- 4:22bit stronger? Things like that. And it really
- 4:24wasn't ask an answer. And it truly was replacement
- 4:26for Google. Then the phase two was really the
- 4:29drafting phase where it was, Hey, I'm working
- 4:32on this word document. I want you to help me
- 4:34improve it. You know, can you look at these arguments
- 4:36that I miss anything? or I've uploaded a transcript.
- 4:39Can you give me a draft of what was said, things
- 4:41like that. And the vast majority of the work
- 4:43is still happening there to be perfectly honest.
- 4:45And that's how most people use the tools day
- 4:47to day. But where everybody's focused now is
- 4:50called Cowork. And in the early days, Claude
- 4:53was the first one. They launched something in
- 4:55the Claude desktop app. And I'll show it to you
- 4:57shortly called Claude Cowork. And just recently
- 5:00in July, ChachiBT launched something called ChachiBT
- 5:03Work. And really what it is, it's a way for these
- 5:06tools to start working on. files on your computer
- 5:09to be able to process hundreds or thousands of
- 5:12files, to create presentations, to create Word
- 5:14docs, to drive track changes, and to do all sorts
- 5:17of things for you. And again, you'll see some
- 5:19good examples. Where I see things going next,
- 5:22and I'm not gonna focus on this today, but more
- 5:24around delegation. So I've actually got an entire
- 5:26set of agents that run my business for me, and
- 5:29I basically sort of manage those agents. And
- 5:30in the future, we'll see organizations that actually
- 5:33orchestrate sort of... you know, fleets of agents
- 5:35doing things for the business. But what I want
- 5:37to talk about today is really where we are in
- 5:39this cowork phase. And as I said, both companies
- 5:43have taken a very similar approach. So if you
- 5:47have been in the cloud desktop app anytime soon,
- 5:50right? What you've noticed now, you know, in
- 5:52the home button on the upper left that I'm using
- 5:53a Mac version, right? There is the home and there's
- 5:56code, but in home, if you now look in the chat
- 5:59window here, you'll see it says chat, which is
- 6:01your regular chat. I could type something, you
- 6:03know, I could type high in here and I could go
- 6:04get, go get a response to it. And The real power
- 6:08now is in Cowork. And what you'll see here in
- 6:11Cowork and in some of the demos is you can actually
- 6:14point it at a directory. You can give it access
- 6:16to different things. You can have it, give it
- 6:18autonomous rule following to go do and only ask
- 6:21for certain permissions. It is really, really
- 6:23interesting. And I'll show you why this is so
- 6:25powerful now and what you can do with it. And
- 6:28if you go into the ChatGPT app, they've effectively
- 6:31done the same thing. So on the very top up here
- 6:34on the left, it says chat BT or codecs, just
- 6:37like in the other in Claude, it says, you know,
- 6:39chat or. you know, Claude code, but when you're
- 6:42in chat GPT here, what you'll notice on the top
- 6:44of the screen up here, let me move this down
- 6:46a little bit so I can see it is you have regular
- 6:48chat and now you have work and work has the same
- 6:51thing. You can pick a project, you can add a
- 6:53directory, you can access plugins. So they are
- 6:56really shifting what they expect you to do with
- 6:58these tools. And as I said, they're moving away
- 7:01from simple chat now into how do we actively
- 7:04use these tools to go drive things? So again,
- 7:06I'll show you some examples as we go through
- 7:08this today. So one of the questions I get often
- 7:12is how do you know when you use chat or when
- 7:14to use co -work, right? I try to simplify and
- 7:17I try to make it really simple. When you're asking
- 7:20a question, you know, when you just want to drop
- 7:22something in, you want to get an answer or you
- 7:23want to verify it yourself, right? And maybe
- 7:25you want to pace the output into another document,
- 7:27use chat, right? You know, use it the way you've
- 7:29been using it. When you want it to do work, when
- 7:32you want it to create a file for you, when you
- 7:34want it to work on a folder or take action for
- 7:36you, you know, on its own, then you want to use
- 7:38cowork, right? You can define the deliverables.
- 7:41You can identify different sources. You can create
- 7:44agents that run in the background, even when
- 7:46you're laptop is closed now so things can actually
- 7:48run in the cloud and you can actually ask for
- 7:51various levels of approval. where you can actually
- 7:54say, Hey, I want you just to go do everything.
- 7:56Just tell me when you're done or step by step.
- 7:59I want you to show me what's different. I want
- 8:01you to show me what's new. And I want to figure
- 8:03out, you know, how we make this work going forward
- 8:05so you can get permissions. And for a lot of
- 8:07folks, obviously getting permission is very,
- 8:09very important, right? We don't want these things
- 8:11to go off and do things on their own. We often
- 8:13want to say, Hey, here's what we can do. You
- 8:15know, go this far. Then ask me before you continue.
- 8:19One of the things I want to highlight is how
- 8:23smart these models have gotten. And whenever
- 8:26one of these companies releases a new model,
- 8:28so Opus 5 just released last week, there are
- 8:31a ton of benchmarks that get out there. The one
- 8:33that I really like the best is from this organization
- 8:36called the Artificial Analysis. artificial analysis,
- 8:39and they published something called the Artificial
- 8:41Analysis Intelligence Index. And what this index
- 8:43is, is a conglomeration of the hardest problems,
- 8:47the hardest coding problems, the hardest reasoning
- 8:49problems, and PhD level problems across, you
- 8:52know, the different sciences and mathematics
- 8:54and things like that. And so to give you a sense
- 8:57of where we were, A year ago, the best model
- 9:00out a year ago in August of 2025 was Claude 4
- 9:04.1 Opus. And Claude had the highest benchmark
- 9:07at that time. And it could basically answer about
- 9:10one third of those very, very challenging and
- 9:12very difficult questions. You can see over the
- 9:15course of this year, by February, we had jumped
- 9:18up to a score of 44. And this is a score out
- 9:20of 100. It doesn't mean exactly that 100 or 44
- 9:24% of the problems can be solved, but it's a close
- 9:26enough proxy. So when you think about it, you
- 9:27can see how smart these models are getting. By
- 9:31April, we were in the 50s. By May, we were in
- 9:33the mid to high 50s. And by July, with the release
- 9:36of, again, Claude Opus 5, running on the Max
- 9:39Inking, we're now basically approaching two -thirds
- 9:42of the most challenging questions out there.
- 9:44can be handled by these AI tools, right? They
- 9:47have gotten incredibly smart. In fact, if you
- 9:49saw the announcement, the upcoming ChachiPT 6,
- 9:53I believe the code name is Astra for it, which
- 9:54should get released sometime in the next few
- 9:56weeks, just solved 10 absolutely incredibly difficult
- 9:59math problems that had not been solved throughout
- 10:01history. So these tools are getting better and
- 10:04better. They're getting smarter and smarter.
- 10:06But what really matters for you and what matters
- 10:08for me is this different benchmark. So from the
- 10:12same organization, there's a benchmark called
- 10:14GDPVal. And what they did is you can see over
- 10:17here, I put the description in, right? This is
- 10:20a benchmark designed to evaluate the AI models
- 10:22on real world. economically valuable tasks across
- 10:2544 occupations. So there's 1 ,300 tasks derived
- 10:29from nine major industries, things that actually
- 10:31contribute to GDP. And again, this is the state
- 10:35of the art today. And the average human, by the
- 10:37way, when they benchmark this, has a score of
- 10:401 ,000. So that's the starting point. And it's
- 10:43an ELO score. The best model right now, again,
- 10:46is Claude Opus 5 on Max, which has a score of
- 10:491861. And that doesn't mean it's, you know...
- 10:531 .86 times better than a human or anything like
- 10:56that. The way that ELO scoring works is effectively,
- 10:59if you gave, you know, all 1300 of these tasks
- 11:02to, you know, people to review, if they have
- 11:05a human do the task and the AI do the task effectively
- 11:08with the score 99 .3 % of the time, people are
- 11:11going to want to see the AI output. They're going
- 11:13to rate it more highly than the human output,
- 11:15right? So the ability of these tools to create
- 11:18presentations, to create word docs, to do analysis,
- 11:20to do spreadsheets has gotten absolutely And
- 11:24this has been my benchmark. And I have found
- 11:25this to be absolutely true. These tools create
- 11:28amazing things. So let's jump in. Let's talk
- 11:31about an example here, because I like to make
- 11:33this live. And again, if there are questions,
- 11:35please ask questions real time. I'm happy to
- 11:37answer them. I know Suzanne can interrupt me
- 11:39if there's something pressing. But I'd love to
- 11:41make this as interactive as possible. But one
- 11:44of the things that these tools have gotten great
- 11:46at now are scheduled tasks, right? And so both
- 11:49Claude Cowork has it, ChatGVT Work has it. And
- 11:52in the early days when Claude Cowork came out
- 11:55earlier this year, I think it was originally
- 11:56released in February, you could do a scheduled
- 11:58task, but your laptop or your computer had to
- 12:01be open. It had to be on in order for it to run.
- 12:04Now, most of these tasks actually run in the
- 12:06cloud, which is a great change. And so one of
- 12:09the things we can do is we can say, hey, something
- 12:12like every morning at 5 a .m., I want you to
- 12:14connect to my email, my calendar, check my most
- 12:17recent 24 hours of emails, identify the ones
- 12:20I haven't responded to and also look at my calendar
- 12:22to identify the meetings I've scheduled for the
- 12:24day and any remaining work days in the week.
- 12:26So give me sort of a forward view, provide a
- 12:28detailed summary and recommended responses to
- 12:30anything I need to do in order to prepare for
- 12:33my day. And the way you structure these is, by
- 12:36the way, I have a document here, all these prompts.
- 12:39So all the prompts that you see today. They're
- 12:42not as part they weren't part of the download
- 12:44But you know at the end on the last slide of
- 12:46the deck is my email if you'd like copies of
- 12:48these prompts And this is the simplest one to
- 12:50get far more detailed shoot me an email I'm happy
- 12:53to share these I share these after every event
- 12:54because I'm a huge believer in you know Sharing
- 12:57the learning and making people better at these
- 12:59tools, so this is the prompt I'm gonna go ahead
- 13:01and copy it here And I'm just gonna go say copy
- 13:03and paste and the way you would do this. I'm
- 13:05gonna go into Codecs for a second here, and I'm
- 13:09just gonna start a new chat And actually, we'll
- 13:12make it a chat here. And I'm just going to paste
- 13:13this in. And I'm going to say, every morning
- 13:16at 5 a .m., I want you to do these things, right?
- 13:18And I'm going to go ahead and say go. And it's
- 13:20literally going to... interpret these instructions.
- 13:22It's going to figure out what it has to do, and
- 13:24it's going to actually take it. We won't wait
- 13:26for it to finish, but it will put it over here
- 13:28then in the scheduled tasks. In this case, I
- 13:32created an anonymized one to show at the event
- 13:34here, but I'm going to show you the one that's
- 13:36in Cloud as well. But all you have to do in order
- 13:39to create a task is a prompt like that. It will
- 13:43then create this for you, prepare Steve's daily
- 13:45email and calendar briefing, use a connected
- 13:47Gmail account. If you happen to have If you're
- 13:51connected to Outlook, it connects to your Outlook.
- 13:53It reads your calendars now. I asked it for this
- 13:55example to anonymize the data. But the one I
- 13:58have in Cloddy did the same thing. And this goes
- 14:00through and it runs every day. Here's the daily
- 14:02brief. You can actually do different things.
- 14:05And if I go actually over into Clod, I actually
- 14:08had it created over here. I think this is the
- 14:10version. Let me just click on this here. Yeah,
- 14:13here is what my morning brief looks like. So
- 14:16again, I had it anonymize the data, but it went
- 14:18through every morning. and it says, hey, let's
- 14:21talk about what your day looks like, right? I
- 14:22had a meeting, you know, from 7 to 9 30. I've
- 14:25got some different things on here. It was empty
- 14:27from 9 30 to 12 15. From 12 15 onward, I've got
- 14:30the prep for the CLE event. And then, so again.
- 14:33I disguise the data, but then here it says, hey,
- 14:35here's all the emails that need attention. So
- 14:37these aren't real, but they actually are, but
- 14:39I just changed the names on them. But you can
- 14:40see, uh, you know, Dan Whitfield wants some time
- 14:43today, you know, someone else about whisper and
- 14:45cost, right? Um, there's some things about rules,
- 14:48some things that have been resolved, things like
- 14:49that, and then some draft responses because that's
- 14:51what I asked it to do. So the great part about
- 14:54it is when you have these scheduled tasks, if
- 14:56I go back into here and go look at the scheduled
- 14:58tasks, right? Here's my morning brief. I literally
- 15:02had typed that exact same prompt in here. And
- 15:04because it now runs in the cloud, you can see
- 15:06the little cloud icon here. I don't actually
- 15:08have to have my computer open. And you'll see
- 15:10here, here's one I created a while ago to figure
- 15:13out some weekly Los Angeles and Santa Barbara
- 15:15family events. This one actually has a computer
- 15:17icon. I did this before they made the shift to
- 15:19the cloud. So it actually, my computer has to
- 15:21be open for it to run. But by default now, All
- 15:24of them go run in the cloud. They're great. Any
- 15:26connectors you have, if you're using MCP connectors
- 15:28to your data sources where you have your legal
- 15:30stuff, your email, anything you want to do, you
- 15:33can now automate. And there is no limit to the
- 15:35kind of things that you can automate as you go
- 15:38forward here. So let's keep going. And let's
- 15:41talk about Claude. So most of you know, I spend,
- 15:44if you've seen me present before, I talk about
- 15:46Claude a lot. I do most of my demos in Claude.
- 15:48In fact, if you look behind me right up there,
- 15:50there's a little Claude icon on my bookshelf.
- 15:52I'm a huge fan of the Claude models. The reasoning
- 15:56and writing is fantastic. So in the last couple
- 15:59of months, they've done a number of things. So
- 16:02on May 12th, they updated their Claude for legal,
- 16:05right? So it was a huge industry release. It
- 16:07wasn't just a plugin update. They provided legal
- 16:10connectors to 12 different practices. 20 different
- 16:12legal connectors, 12 practice area plugins. They
- 16:16introduce Claude inside Word. Outlook, Excel,
- 16:19PowerPoint. And I'm gonna show you an example
- 16:21of that in a minute, right? A huge release for
- 16:23them. They have really focused on legal. And
- 16:25that's why so many of the firms that I work with
- 16:27really focus on Claude and use these tools. Then
- 16:31in June, they launched Fable 5 and then Sonnet
- 16:345. And if those of you that use Claude know that
- 16:36they introduced it, then it pulled it back and
- 16:38then it's out again. I will tell you the one
- 16:41hesitancy that I tell all my clients in using
- 16:43Fable. They have different data restrictions
- 16:46on it. here on the commercial plan, they are
- 16:48still holding that data for a while. So if you
- 16:51are conscious and care about enterprise security,
- 16:55I'm hard -pressed to recommend Fable to put legal
- 16:57work in at the moment. But use Sonic, use Opus,
- 16:59use the other models. All the normal non -training
- 17:01stuff applies. So just be careful with Fable
- 17:03until they come back on that stuff. Then this
- 17:06was, I was very excited about this. They introduced
- 17:07the cowork for service that I was just talking
- 17:09about. So you can schedule any task that runs
- 17:12in the cloud, anything that you want to do. It
- 17:14can read documents. It can process things. You
- 17:16can read your calendar. It can go out. It can
- 17:18do internet research for you. Anything you want
- 17:20to do. You can actually fire off now agents.
- 17:22They can run effectively 24 seven. And then on
- 17:26July 23rd and 24th, they launched Opus 5, their
- 17:29biggest flagship model that's not part of Fable,
- 17:31and they introduced new voice capabilities. So
- 17:33you can actually talk to Claude now and have
- 17:35much more interactive sessions than you could
- 17:37before. And I happened to really like the voice
- 17:40model in ChatGPT, which I'll show you an example
- 17:42of in a minute. But the legal stack, as I said,
- 17:45they have really doubled down on legal, right?
- 17:47They built connectors for iManage, NetDocuments,
- 17:50Everlaw, Relativity, Thomson Reuters, DocuSign,
- 17:53more and more. And so I put the link here. This
- 17:56will be part of the presentation. You can go
- 17:57get more about it. You know, it's just part of
- 17:59what they offer. I would encourage you to look
- 18:01at it and look at their practice area plugins,
- 18:03right? They've introduced some really, really
- 18:05great capabilities that help you automatically
- 18:08create skills and do other things. And then my
- 18:10favorite part, honestly, and for most of the
- 18:12attorneys I work with, with their favorite part
- 18:13also was the office surfaces. So they introduced
- 18:16cloud plugins into Word, Outlook, Excel, and
- 18:20PowerPoint. And I'm gonna show you an example
- 18:21of that next. That is an incredible capability
- 18:24and it has really changed how people can use
- 18:27these tools. So again, what I tell people is,
- 18:31you know, co -work because it can run in the
- 18:32cloud can now work while you're in court, right?
- 18:35So back in February, if you're using co -work,
- 18:37you know, it was sort of a cloud code for everybody
- 18:39else like us, but you pointed at a folder, it
- 18:41reads, it edits, it creates, but it ran locally
- 18:44on your machine and your laptop had to be open.
- 18:46Now with these new models, again, as I said,
- 18:48it runs in the cloud, it runs on anthropic servers.
- 18:51You can fire them off from web, from mobile,
- 18:53you can schedule tasks from any time, any place,
- 18:56and the answers land on your phone, you know,
- 18:58anytime. something sense. You can see the results
- 19:00right away. So it's a real change of pace from
- 19:02the way it used to work before. So let's talk
- 19:05about Claude in Microsoft Word because this is
- 19:07a great example. And I'm going to do a quick
- 19:09live example to show you something and I'll show
- 19:10you a real world example of how it works. But
- 19:14if I go into Microsoft Word here, what you'll
- 19:16see is over here I've got a little Claude button
- 19:19up here. So if you are subscribed to Claude,
- 19:22you can download the extension. You can log in.
- 19:24So this is logged into my account. I'm not gonna
- 19:27hit refresh here, I'm just gonna let this go,
- 19:29but I'm gonna give it something really simple.
- 19:30I'm just gonna say, hey, give me a quick three
- 19:34paragraph history on St. Bernard's. So those
- 19:37of you that know me and see my articles, I have
- 19:40a large St. Bernard named Magnus, he's in half
- 19:43my articles, I've got references to him everywhere.
- 19:45So this is gonna go off and it will do a quick
- 19:47change. And so several places, the St. Bernard
- 19:50dog breed, I should have said dogs. So it's gonna
- 19:53go through and it's actually accessing Clonet.
- 19:56Clonet, Sonnet here. You can choose the different
- 19:58models that you wanna work from. You can choose
- 19:59Sonnet, Opus 4 .8, Opus 5, Fable 5. But the nice
- 20:03part about this is it works directly in Word,
- 20:07where we spend a lot of our time, right? So it's
- 20:09doing some work, it's searching the web, and
- 20:11you can see here, it actually created a very,
- 20:13very quick three -paragraph history on St. Bernard's.
- 20:16I could have had it create an NDA, could have
- 20:18it create... You know a prenuptial agreement
- 20:21I could have it create, you know a contract between
- 20:22two software companies But the nice thing about
- 20:25what this does is you can actually have it do
- 20:28track changes for you So I'm gonna look in here.
- 20:30I'm just gonna say Hey in this document change
- 20:34the word Bernard to kitty cat Oh and use track
- 20:39changes So I'm gonna give it that those instructions.
- 20:42So the nice thing is Claude knows how to activate
- 20:46Track Changes in Microsoft Word. So you can see
- 20:49here, it's planning document edits with Track
- 20:50Changes. It's gonna figure out how it handles
- 20:52all the variations of the word Bernard here.
- 20:54And it will now go through and you'll see, you'll
- 20:56actually see the Track Changes on here that it
- 20:58goes through and it'll put my name on it next
- 20:59to the Track Changes. And the beauty of it, by
- 21:02the way, doing it in Word. So if you were to
- 21:04upload a Word document into Claude, which many
- 21:07people do, or ChatchPT, and say, hey, go make
- 21:10these edits and use Track Changes, it will do
- 21:12that for you and give you an updated document.
- 21:14But it will say, edited by Claude, which you
- 21:16probably don't wanna see, right? I'm gonna tell
- 21:18it to go ahead and allow these changes here.
- 21:20But when you do it in your own version of Word,
- 21:23it's actually gonna show up now. You can see
- 21:24here like Steve Smith deleted Bernard, right?
- 21:27So it'll go through and it will do all these
- 21:28things. It's gonna go through really quickly
- 21:30and make all these changes. And it's actually
- 21:32quite fast. So again, normally I would use the
- 21:35Opus model for deeper thinking, but I just wanted
- 21:36to show you that you can do this. It's pretty
- 21:38quick and it makes it work. What I wanted to
- 21:41show you was I actually had a large master vendor
- 21:44manufacturing agreement, right? This is an 18
- 21:47page agreement. You know, it's got 7 ,000 words
- 21:49in it. It's absolutely massive. It's designed
- 21:51between hardware companies. And I went through
- 21:53and I asked Claude to go through and completely
- 21:55transform this from, you know, a B2B. Hardware
- 21:59agreement to basically be a b2c software agreement
- 22:02and it definitely took a while to do it But what
- 22:04you'll see here is it actually went through and
- 22:06it completely transformed this document So when
- 22:09you talk about the ability to identify new clauses
- 22:11to change things to make it happen, right? It
- 22:14is absolutely powerful and fantastic. You know,
- 22:16I use Opus 5 to do this. And it probably took
- 22:1915 or 20 minutes to do because it literally went
- 22:21through clause by clause, piece by piece. And
- 22:24the nice part about it is I can actually decide
- 22:26whether to accept the changes or whether or not
- 22:28to accept the changes because of the way we did
- 22:29it here. But it went all the way through this
- 22:32entire document, all 18 pages. And you can see
- 22:34I basically asked it to change almost every single
- 22:37thing in here. And it has the ability to go do
- 22:39that. So you can work on extremely long documents
- 22:42now. And you can basically have it do anything
- 22:44that you want. One of the other things that they
- 22:48added in is in Excel. And so they built an ad
- 22:51in Excel. And I know many of the attorneys I
- 22:54work with, work in Excel with settlements and
- 22:57analyzing financials and things like that. And
- 22:59so what I did here is I opened up Clot again.
- 23:02I have to move this speaker thing. It's over
- 23:04here, the button and there's a ChatGPT one also.
- 23:06So if you use ChatGPT, ChatGPT has an extension
- 23:10here, a plugin for Excel and they have one for
- 23:12PowerPoint, but they have not released one for
- 23:14Microsoft Word yet. So if you're using... Claude,
- 23:17you know, and ChachiBT, if you wanna use Microsoft
- 23:20Word, only the Word one, only Claude has one
- 23:22for Word at the moment. I would be surprised
- 23:25if ChachiBT didn't release one soon. But what
- 23:28I did, I actually started with a blank canvas
- 23:30and I said, I want you to go create an incredibly
- 23:33detailed financial document for me that shows
- 23:36a P &L for a 50 person law firm, right? Here's
- 23:38the number of partners, here's the number of
- 23:40associates, here's the number of paralegals and
- 23:42admins, and I want you to create an executive
- 23:43dashboard. So from a single prompt, It actually
- 23:46went through, I said, you know, build an assumptions
- 23:48page, you know, how many partners are sorry,
- 23:50it's 23 people, how many partners, how many associates
- 23:52and attorneys, how many paralegals, and it made
- 23:55all the things that aren't assumptions formula
- 23:57driven, right? So it created an incredible thing.
- 24:00It built out the revenue for me, right? You know,
- 24:02and all the different things that built out a
- 24:04detailed income statement, you know, month by
- 24:05month built out a balance sheet, a statement
- 24:07of cash flows. So it has knowledge of anything
- 24:10that you want to do. And then I said, go create
- 24:12an executive dashboard for me, right? Show me
- 24:14what revenue and EBITDA. look like per month,
- 24:16right? What is my monthly ending cash balance
- 24:18look like? And the one thing I will say because
- 24:21I use all these tools all the time is that ChatGPT
- 24:25also, as you'll see in a second, built a fantastic
- 24:27model, but the polish you get out of Claude is
- 24:30just higher than ChatGPT. I mean, I hope ChatGPT
- 24:33fixes that, but as you'll see when I switch over
- 24:35to the ChatGPT one in a second, it has all the
- 24:38functionality, it has all the capabilities, right?
- 24:40I mean, everything as you'll see here, you know,
- 24:41you can see it's formula -driven, so it does
- 24:43an amazing job. And if I go look at the ChatGPT
- 24:46one, right, this is the dashboard. So if I compare
- 24:49kind of this dashboard and the graphs to the
- 24:52Claude one, You know, if I go back here, I just
- 24:55happen to think this looks more polished. It
- 24:56looks better. It's more the kind of deliverable
- 24:59I would actually want to send to a client. But
- 25:01that being said, it went out, you know, it build
- 25:03the same kind of assumptions. It builds an operating
- 25:05model. It built the financial statements. Both
- 25:07of these tools can do those things and they're
- 25:09great at it. One of the really great use cases,
- 25:11and I've actually done this with a couple of
- 25:13different attorneys is Let's say you get a settlement
- 25:16agreement or a set of financials from someone
- 25:18and you want to go through and figure out if
- 25:20they're accurate. You can load them up into either
- 25:22Claude or Chachi, sorry, into Excel. open up
- 25:25the clod and the chat should be T button and
- 25:27say, I need you to go through and validate every
- 25:29assumption, look for any errors, look for things
- 25:31that are hard coded that should be formula driven
- 25:32and make this thing happen. It is absolutely
- 25:35fantastic at doing that. And I have found more
- 25:38errors and things that people have sent me than
- 25:40you can possibly imagine. So it is absolutely
- 25:42great when you go there. And I would encourage
- 25:44you to do that. It's a absolutely great use case
- 25:47for doing that. Another great thing that Claude
- 25:51has been really good at updating is they are
- 25:53really good at creating artifacts. And if you've
- 25:55not used an artifact, an artifact is a visualization.
- 26:00So, you know, in ChatGBTs you'll see in a minute
- 26:01you can create images. But Claude can create
- 26:04amazing interactive dashboards and visuals that
- 26:07allow you to show data and represent data in
- 26:10really powerful ways. So one of the things I
- 26:12was doing is I was working with a construction
- 26:14litigation firm. And we wanted to show the conditions
- 26:17on this very detailed contract. We wanted to
- 26:20show the current status of the closing conditions,
- 26:22right? The consents, the diligence risks, the
- 26:24signature status. funds flow dependencies, owner
- 26:27accountability. And when you have hundreds of
- 26:29documents and hundreds of changes, this is a
- 26:31tremendous amount of work. And so one of the
- 26:34reasons I talk about the prompting here is you'll
- 26:36see I use a structure called core, C -O -R -E
- 26:39for context objectives. Oh gosh, I'm drawing
- 26:43a complete bank today. I mean, our role and then
- 26:44examples, right? So this is a very, very detailed
- 26:47prompt. I asked it to go do a tremendous amount
- 26:50of things here, but you can see I wanted to go
- 26:53through and show all these conditions. You know,
- 26:55I've supplied it a bunch of files. I need, you
- 26:56know, confidential professional responsibility
- 26:58limits. I need interactive artifacts. Here's
- 27:01the objectives. You need to confirm the scope
- 27:02and all the blockers. I need to inventory all
- 27:04the sources. I need to extract the normalized
- 27:07fact set, right? Across hundreds of documents.
- 27:09This is a monumental task. And the output that
- 27:12you get out of this, if I go look at this here,
- 27:16is this is an artifact from quad. And you can
- 27:19see here, here's the decision, right? So it's
- 27:21telling me, hey, this packet is not closing ready.
- 27:23Three decisions cancel should focus on first,
- 27:25right? You know, which instrument is operative.
- 27:27whether May 25 displaced May 18th and how the
- 27:3048 ,500 pair is displayed, right? So it then
- 27:34went item by item through here, what the current
- 27:36status is, what the priority is, what the consequence
- 27:38of not doing this is. And you can see here, here's
- 27:41a whole section on conditions and consents, right?
- 27:44Here's the signature status page of everything
- 27:46that's in here. You know, the merger agreement
- 27:47counterpart, okay, execution status is complete.
- 27:50Well, this one's not in the packet. This one's
- 27:51not been started. So you can imagine going through
- 27:54and how long it would take you when you have
- 27:56this level. of documents in a very complex case
- 27:58to do all this stuff. So again, this is a great
- 28:01example of a, this is a very complex visualization.
- 28:04And again, if you would like the prompt, let
- 28:05me know, I'm happy to share it with you. But
- 28:07these tools have gotten so powerful now that
- 28:09they can reason across hundreds of documents,
- 28:12hundreds of objectives and create powerful tools
- 28:14like this. A second one that I was doing recently
- 28:19is I was working with a family law firm and we
- 28:21were talking about, again, the power of Claude
- 28:23to visualize things. And I asked him, I said,
- 28:25hey, you're finalizing this divorce settlement,
- 28:29right? And you've got a financial statement you
- 28:31want to share with one of the spouses. you know,
- 28:33how would you show that to them? How would you
- 28:35explain that to them? And, you know, they pulled
- 28:36up a spreadsheet and I said, well, you know what?
- 28:38I bet Claude can actually help us visualize this
- 28:40in a more interesting way. And I said, look,
- 28:43let's make a really simple prompt here. I said,
- 28:45in this case, the example I'm showing you, because
- 28:47it's fictitious, you know, I uploaded a fictitious
- 28:49document here, but I said, I want you to create
- 28:51a beautiful visual artifact that I can share
- 28:53with the client that summarizes the settlement
- 28:56and gives them a few sliders to show how the
- 28:57values can increase over time and how the values
- 29:00will change. And I want them to be able to allow
- 29:02to and deselect various assets. And so if I go
- 29:05look at this This was, you know, for Avery Bennett,
- 29:09a visualization of her settlement, right? So
- 29:12the net awards to Avery, $8 million. When you
- 29:15look at all the assets, you know, liquid, near
- 29:16liquid, right? Projected value, year 10. So we
- 29:20made this, you know, interactive. So she might
- 29:22say, hey, I want to see what happens if it's
- 29:24seven and a half percent, right? Now I grow up
- 29:26to, you know, $16 million, right? My planning
- 29:29horizon is only seven years. So what does that
- 29:31mean to me? Do I want to include an annual lifestyle
- 29:34draw and take money out of here? much, right?
- 29:36So this became a much better way. And as I was
- 29:39working with this firm, they realized this is
- 29:41a great way to be able to show these things.
- 29:43You know, what's the award made of? You can have
- 29:45it do all these things. And again, All this came
- 29:48from a single uploaded spreadsheet and from the
- 29:51simple prompt I gave. And then we decided to
- 29:53do something fun. We decided to see if we could
- 29:54actually make this a much more interesting visual.
- 29:57So we said, hey, can we make this much more like
- 30:00hand drawn, hand cursive? But you can see the
- 30:03style is different. So you can make it look any
- 30:04way you want. You can have your firms look in
- 30:06your field. But again, you can change it. You
- 30:08can still change the interactive sliders and
- 30:10things like that. But using artifacts, and again,
- 30:13they've gotten better and better. And you can
- 30:14actually publish artifacts and share them with
- 30:16other people. Again, a very, very powerful tool.
- 30:21So to net it out, the big changes are they released
- 30:23the Fable model, they released the Opus model,
- 30:25which is an absolutely incredible model. I find
- 30:27it incredible for coding and for doing detailed
- 30:30analysis and for research. They introduced a
- 30:32voice model, and then they really introduced
- 30:34sort of co -work in the cloud. So let's talk
- 30:37about ChachiBT. A lot of you use ChachiBT. I
- 30:40use ChachiBT. It's one of my favorite tools.
- 30:43And again, their big focus area is they really
- 30:46rebuilt ChachiBT around work, not the chat. And
- 30:50so if you use the ChachiBT app in the... You
- 30:54know, the old app on Windows or on the Mac before
- 30:57they basically deprecated the old one. They integrated
- 31:00their codex, their programming app with their
- 31:03chat, GPT app. And so if I pull this up here
- 31:05now, and again, I showed this earlier. If I do
- 31:07new chat, right? You know what you see on the
- 31:09top. is if I'm in the chat mode, you know, there's
- 31:12chat and there's work. And again, they've designed
- 31:14this to let you go off and do the same kind of
- 31:17things. You know, choose a project, create a
- 31:18directory, you know, use the Microsoft Word plugin,
- 31:21the Excel plugin, the PowerPoint plugin, create
- 31:24content, do whatever you want to do, create scheduled
- 31:26tasks. But they're focused now, as you'll see
- 31:28in the example in a minute, on actually having
- 31:30you go do things. I mean, I'll show a simple
- 31:33example again in a second of how you can do document
- 31:35logs and things like that. But just know that
- 31:37their focus is moving off chat and to actually
- 31:39doing work and if I do a I'm gonna do a new project
- 31:43here I'm going to I will create a new project
- 31:46I'll call it sample downloads I'm gonna add a
- 31:51folder to it here I'm gonna add my downloads
- 31:54folder just to make things really simple here
- 31:56and I'm gonna go ahead and create this project
- 31:59and so one of the things I can do I'm gonna switch
- 32:02the model here because I definitely don't need
- 32:03something this big and powerful speed is fast
- 32:06I'm just gonna ask it Hey, how many files total
- 32:09are there in the downloads directory and how
- 32:11many image files are there in the downloads directory?
- 32:14So one of the things about work, and Claude can
- 32:17do the same thing, is it can actually take action
- 32:19on directories and do things. So you can see,
- 32:22I'm gonna count the files in the downloads directory,
- 32:24then I'm gonna identify images by their extensions.
- 32:26So you can see how fast this was, right? In the
- 32:28main directory, there were 582 files, including
- 32:31200 images, including all subfolders, there are
- 32:342 ,700 files and 360 images, right? So it goes
- 32:37through and it does things. really, really fast,
- 32:40right? So it really is powerful. I could have
- 32:43it change things. I could have it do things.
- 32:44I can have it move all those files to a different
- 32:46directory. If they were legal documents, I could
- 32:49have it analyze those as I've seen the example,
- 32:51but it really lets you do other things really,
- 32:54really powerfully. And so one of the things that
- 32:57they've done is they're really focused on skills
- 32:59and really driving those things. So they've added
- 33:01a ton of new sources. So you can connect your
- 33:03sources. You can connect your knowledge bases.
- 33:05You can connect Dropbox. If you use box or Google
- 33:08drive, you can add those things. You can create
- 33:11automated skills in all these things. But the
- 33:13biggest thing for them is the ability to create
- 33:15these agents. Like I showed you that agent earlier
- 33:17to run on a schedule. But what I could do, I
- 33:20could literally say in this task here, if I go
- 33:22back in. my tasks. Let's see. I'm back in here.
- 33:26I went into, I'm going to go back in here. I
- 33:28could say, Hey, I want to turn this into a scheduled
- 33:32task. I'd like this to run once every, you know,
- 33:35once a week, every Friday at 3pm where you count
- 33:38the number of files and give me an update and
- 33:39then give me the Delta change from the prior
- 33:41run. So turn that into a scheduled task for me.
- 33:43And then by the way, I want you to turn that
- 33:45into a skill that I can call anytime, call it,
- 33:47check my files. Right. So I'm not going to spend
- 33:52a lot of time going through skills, but skills
- 33:54are incredibly useful. It allows you to basically
- 33:56have routine things that you do again and again,
- 33:59like the way you process the document, the way
- 34:00you write, the way you do things. But it's going
- 34:02to go through here now. It's going to create
- 34:03a scheduled task, and it's going to create a
- 34:05new skill reusable called Check My Files that
- 34:08I can then call at any time as we go forward.
- 34:11But like I said, the big shift here is from simply
- 34:13chatting to working and doing things. And again,
- 34:16I'm going to show you an example in a second.
- 34:19This might be my favorite new function. So voice
- 34:23has become a control service in ChatGPT, not
- 34:25just a Dictaphone. So it is a full duplex tool
- 34:29now. It listens while it speaks. You can fire
- 34:31off other things. I'll show you some examples.
- 34:33It drives work in Codex agents, either from the
- 34:37phone or from the web interface, or sorry, it's
- 34:39desktop only at the moment. I forgot they haven't
- 34:41put it on the phone yet, but you can launch different
- 34:43agents. You can do things. Whereas Claude is
- 34:45really just sort of an interactive, you know,
- 34:47kind of discussion back and forth. Chat GPT's
- 34:49voice mode lets you go do things. So let me show
- 34:52you here. And one of the things that's great
- 34:54is, I'm gonna get a move the video here. They've
- 34:57actually introduced browsers. So there's a built
- 35:00-in browser here. So I can say, hey, go to my
- 35:02Intelligence by Intent website. You know, go
- 35:05to CNBC. And you can interact when you're chatting
- 35:09with an open browser here. And it introduced
- 35:11computer use. But what I'm gonna do here is I'm
- 35:14gonna basically fire off this mode here. I'm
- 35:16just gonna go to the voice mode. Hey chat, can
- 35:21you look at my open browser tabs and tell me
- 35:22what's open right now at the moment it for the
- 35:24in -app browser I Don't know if I turned on sound
- 35:31when I was sharing actually, let me stop share
- 35:34for a second. Hold on I want to make sure that
- 35:35I'm sharing audio as well. I'm not sure. Oh,
- 35:38I did not so There I'm now resharing this with
- 35:42audio so you can hear it because it basically
- 35:44said checking Yeah, I check you got it. You're
- 35:47okay So just give me the browser tabs. And by
- 35:50the way, in the background, go tell me how many
- 35:53new emails I got in the last 24 hours that are
- 35:55unread. And then fire off another process to,
- 35:57by the way, go look and open up the application
- 36:00clearly on my Mac. And then you'll create a new
- 36:03document in there based on the Google Docs file
- 36:06that's in Chrome. And then I want you to go spin
- 36:08off another agent and go do some research for
- 36:10me around the latest offerings from the Beverly
- 36:13Hills Bar Association. Then send me an email
- 36:14about that. Yep, I've heard on it. Let me pause
- 36:22it here. Hold on. Let me go over here and Here
- 36:26is my it's down here. I'm gonna I'm just gonna
- 36:29pause browser is still empty So there are no
- 36:32tabs open there. You can see that you see unread
- 36:35emails from the last day I'm still working on
- 36:38the document and research pieces. Yeah, so one
- 36:41of the things that's nice I've muted the thing
- 36:43here is I can open up something else like if
- 36:45I open this up Hey, Chad, what screen are you
- 36:49seeing right now? Are you seeing my Excel or
- 36:52are you seeing my document that's the BHBA summer
- 36:55AI update? What do you have visibility to on
- 36:56the screen? Checking that now. I'm looking at
- 37:04your screen now to distinguish which one is open.
- 37:07I see Google Slides with the BHBA summer AI update
- 37:12deck on the opening slides. So I'm going to stop
- 37:15this. But one of the things, you know, I mean,
- 37:17I could spend an entire hour giving you guys
- 37:19demos of what this tool can do, but it will go
- 37:22off and it will, you know, it will open up browsers.
- 37:24It will connect to email. You can control things
- 37:26on your computer. I can actually have it open
- 37:28up Excel, create a, you know, create a brand
- 37:29new sheet, populate data, copy things from one
- 37:31to another. It really opens up a tremendous amount
- 37:34of use cases for the way that you use these tools
- 37:36and their voice mode. Like I said, it's actually
- 37:39much better than the in the quad one, it lets
- 37:41you just talk to it, interact, and you can keep
- 37:43going as it's doing things and fire off new processes.
- 37:46So it really, for me, what I do, it has absolutely
- 37:49changed the game in the way that I work. So something
- 37:53else that Claude is fantastic at is it is amazing
- 37:56at image generation. You can create detailed
- 37:58settlement. Let me make this a little bit bigger
- 38:00just so you can see these images a little bit
- 38:01better. You can create detailed settlement documents.
- 38:04You can create descriptions of how business entities
- 38:07are tied together. I use it for image generation
- 38:10for a lot of the articles I write and things
- 38:11like that. But the image generation tool in ChatGPT
- 38:16is absolutely incredibly powerful and I would
- 38:18encourage you to use it if you need it for different
- 38:20things. So let me show you kind of work in process
- 38:25and how this goes So one of the things and I
- 38:28do work with a lot of family law firms But you
- 38:29can get examples across almost any type of firm
- 38:31when they get a new matter in a new client And
- 38:34they'll often get a ton of financial documents
- 38:36in right and need to understand what's in those
- 38:38documents So they need to they need to open those
- 38:40up they need to OCR them if they haven't been
- 38:42OCR'd They need to put Bates numbers on them.
- 38:45They need to copy them into a new directory.
- 38:46They want to update a document log and so Chachi
- 38:51BT work or Claude co -work and actually now go
- 38:53do this automatically. And so I have a sample
- 38:56prompt in here. And what you'll see here, it
- 38:59says, hey, you're helping a family law paralegal
- 39:01build a document log from a financial disclosure
- 39:03production. I've attached two things, the sample
- 39:05document log spreadsheet, a set of produced documents,
- 39:09bank statements, brokerage and your retirement
- 39:10accounts, things like that. Here's what I want
- 39:12you to do. I want you to go create a new subdirectory
- 39:15called output, right? I want you to copy the,
- 39:17create a new file called the new output. I want
- 39:20you to then go into the PDF, analyze it carefully,
- 39:23you know, break it apart because there's multiple
- 39:24PDFs in here. I didn't put the OCR instruction
- 39:27in this particular sample, but you can do that
- 39:28as well. You can have it OCR things. Then I want
- 39:31you to go through and actually put Bates numbers
- 39:33on it and then update the document log when you're
- 39:36all done. So I actually ran this earlier today.
- 39:39in co -work here. And if I go up into my pin
- 39:42stuff, the financial disclosure document log
- 39:44here, let me just make this smaller here. I mean,
- 39:47I literally, I dumped this prompt in, as you'll
- 39:50see here, it went through and it analyzed, it
- 39:53broke out, it says, hey, here's your new output.
- 39:55Here's the new spreadsheet that I've created
- 39:56for you. And it actually, it's open here in the
- 39:59browser, but it actually put it in the file,
- 40:00right? But it actually has all the information,
- 40:02you know, that I asked for in all the different
- 40:04spreadsheet. And then it went through and it
- 40:06created, you know, every different document.
- 40:08Every PDF, it broke it apart, it put the Bates
- 40:11numbers on it. You can see here, here's the grant
- 40:14deed, here's SMS 002, here's SMS 003, SMS 004.
- 40:19And if I were to open up the... use cases over
- 40:22here. And if I look at this is where it was in
- 40:25document log over here, it actually created this
- 40:27whole set of output. So it took that one PDF,
- 40:29it broke all these things up. It created this
- 40:31new output spreadsheet for me. It really is a
- 40:34powerful tool. And that's why I talk about it
- 40:36actually doing work for you. So you can point
- 40:38it at entire directories. You can have it go
- 40:40do really, really powerful and incredible things.
- 40:43And it's a, again, it's a terrific use case in
- 40:46terms of the way to use these tools. I'm going
- 40:49to show you one more example here, an exhibit
- 40:51builder. This is actually one that I actually
- 40:54happened to do in Claude, although I could have
- 40:55done it in Chachapiti also. But there's a great
- 40:58way to use these tools to do exhibits. And again,
- 41:00this is another good visualization for artifacts.
- 41:04But again, you could actually have Chachapiti
- 41:06do the exact same thing. It will build out the
- 41:08same file. And what I'd asked it to do is I said,
- 41:10hey, I want to demonstrate a very complex Gantt
- 41:14chart or a process for this to go through. to
- 41:16be a demonstrative exhibit. I want to have it
- 41:18as an HTML file that I can send around and send
- 41:20clients. So again, here's the full context. I'm
- 41:23helping this construction litigation firm. Here's
- 41:25a whole bunch of facts about this case. Here's
- 41:27my objectives, right? I want a self -contained
- 41:29interactive HTML file that's done in Gantt style,
- 41:33critical delay timelines and things like that.
- 41:35You can see all the things that are in here.
- 41:37And again, very, very detailed prompts. Six months
- 41:40ago, you could not have done a prompt like this.
- 41:41These tools couldn't have followed it, but now
- 41:43they can. They are so powerful in terms of doing
- 41:46that. And what you get out of this, if I go find
- 41:49the Meridian Tower here is, and again, I could
- 41:52have done this in ChatGPD or Claude, but you
- 41:54can actually create a very, very powerful ant
- 41:57chart here that is interactive. I can drag things,
- 41:59I can move things around. But again, it really
- 42:01lets you visualize this data and do things that
- 42:04were challenging to do in the past. So you can
- 42:07build amazing things. One thing that's not in
- 42:09the presentation I want to highlight is I was
- 42:12working with a client recently. We had to analyze
- 42:15thousands and thousands and thousands of documents.
- 42:17There was north of 50 ,000 documents and we were
- 42:19getting ready for a deposition. We wanted to
- 42:20be able to ask questions about, hey, you know,
- 42:23if we go to pose this person has something been
- 42:25asked there, right? So I use Chachi BT to start,
- 42:29you know, from a basis of 50 ,000 documents to
- 42:31get it down to, you know, the 5 ,000 pages that
- 42:34matter in the real world. One in this particular
- 42:36example, there are 400 documents here and 1200
- 42:38pages, but when it allows them to do, they can
- 42:41go in. I'm going to just type the word Thanksgiving
- 42:43here. And in their real world one that we built
- 42:46it for sub one second response, but it goes through
- 42:48and identifies. all the documents, all the 400
- 42:51documents here where Thanksgiving is referenced,
- 42:54it pulls up all the passages here, and then it
- 42:57pulls up the actual document and they highlight
- 42:58it. So if they're sitting in a meeting or they're
- 43:01in a deposition and they're positive that something
- 43:03was asked and answered maybe a year or two ago,
- 43:05we can point this at the entire corpus of data
- 43:08and it would instantly find that document, highlight
- 43:10those phrases for them. So you can see here,
- 43:12it found every document, holiday parenting proposal,
- 43:15holiday parenting email chain, another one of
- 43:17the email chain. for holiday rotation plan. So
- 43:20everything that it found Thanksgiving or the
- 43:22word holiday, because it really uses semantic
- 43:24search, it pulls all these things up. So these
- 43:26are the kind of tools you can actually build
- 43:28now and deploy really, really easily across clients
- 43:33and organizations. So let me talk about Gemini
- 43:35for a minute. I want to make sure there's time
- 43:37for Q &A at the end because I love the questions
- 43:39and what we're doing. So as I said, Google has
- 43:42not launched their new 3 .5 Pro model. They did
- 43:45launch an updated 3 .5 Flash model. The big change
- 43:48they made was to, if you've used a tool called
- 43:51Notebook LM, they now call it Gemini Notebook
- 43:53on July 16th. It's still standalone, but they
- 43:56updated the model in there, which is fantastic.
- 43:58It's much faster. It's much more powerful. There's
- 44:01a lot more use cases in it. So I'm gonna show
- 44:03you an actual matter from... from a notebook
- 44:06that I have here and it's a funny one it's a
- 44:09fictitious one but the advantage of using these
- 44:11tools is you can actually upload hundreds of
- 44:15documents in fact if you've got the paid version
- 44:16you can upload it's either 500 or 600 documents
- 44:19so I did a funny one I like dogs And so I uploaded
- 44:23in here 300 depositions of different dogs, right?
- 44:27So depositions of, in this case, you know, here's
- 44:29the Dachshund Abby and what the information is,
- 44:31right? So my name is Abby. I'm a Dachshund. I
- 44:33have lived with the Wilson family for eight wonderful
- 44:35years. So notebook LM lets you put sources in
- 44:39and then you can ask questions about it. So I
- 44:41can just, I can ask something like how many documents
- 44:45contain golden retrievers, right? And the nice
- 44:48thing about what's now Gemini notebook, what
- 44:50used to be called notebook LM is it will only
- 44:53look in the sources you provided. It's not going
- 44:55to go search the web. It's not going to look
- 44:56at anything else that's been trained on. It's
- 44:58only going to go work on this. So if you've got
- 45:00a, you know, a corpus of documents, it's under
- 45:03about 500 documents. You can upload that into
- 45:05a single notebook and then go ask questions.
- 45:09You can also do all sorts of other things. So
- 45:12while that's going off and thinking, by the way,
- 45:14if you see the text here, right, you'll see it
- 45:16looks like it's speaking German. I actually asked
- 45:18it to speak back. to me as if it was a German
- 45:20dog trainer to be funny. So I was asking questions
- 45:22about, you know, what do you know about Magnus
- 45:24is like, Hey, I don't know anything about Magnus.
- 45:26There's no dog named Magnus in here. Right. I
- 45:28asked it to be kind of snarky and kind of funny.
- 45:30You can make it do anything you want there. But
- 45:32what I want to show you is you can actually have
- 45:34it go create, you know, a detailed roster of
- 45:37everything that's in there. You can do all sorts
- 45:38of great analysis. You can have it create presentations.
- 45:41If I go open this up here, it created an entire
- 45:44presentation. You know about the the content
- 45:47that's in here, right? You can have it go do
- 45:49it You can edit these presentations. You can
- 45:51have it create diagrams and videos and all sorts
- 45:53of things So this was one of their big changes
- 45:55they continue to add to it. It is an absolutely
- 45:58fantastic tool So you see here it got back to
- 46:00the question about golden retrievers It says
- 46:02there's nine of these and the nice part is what
- 46:04you'll see here is the golden retriever training
- 46:06files It'll actually have a citation here So
- 46:09when you're searching for a document that you've
- 46:11been working on it will actually then allow you
- 46:13to go it'll show you exactly where it pulled
- 46:16that information when you ask a question, and
- 46:17then if you click on it, it'll actually pull
- 46:19up the original source document. So while we
- 46:22still wait for them to update the models, that's
- 46:24a big change. The other thing that they released,
- 46:27and this is unfortunately only available in their
- 46:29consumer tools, and I typically tell clients,
- 46:31please don't use the consumer tools that translate
- 46:33your data, but they've introduced something called
- 46:35Spark. And Spark is their agentic tool where
- 46:39you can actually have it go connect to all your
- 46:41different data sources. You can have it go do
- 46:43different things. You can have it declutter your
- 46:45inbox. You can have it search for the things.
- 46:47I've had it search for very complex travel things.
- 46:50So I use it on a personal basis. I'm waiting
- 46:52for them to roll that out in the Google workspace,
- 46:55the business edition. They have not done that
- 46:57yet, but know that they are very, very focused
- 46:59on agents, just like Claude work, Claude cowork
- 47:02and Chachi PT work. So I would say coming soon,
- 47:05watch for it. It's really, really good. So before
- 47:09we get to some Q &A, what I tell people is look,
- 47:11you know, pick the workflow first, right? When
- 47:13you're thinking about doing something, understand
- 47:15what does your workflow look like today? So one
- 47:18of the clients that I've worked with, I actually
- 47:20had one of the paralegals go through and videotape
- 47:23the screen to understand, okay, when they get
- 47:25in new documents, right? Whether they get them
- 47:26in through email or through the court or through
- 47:29a Dropbox link. How do they look at those? How
- 47:31do they process those? What decisions do they
- 47:33make, right? So we started to look at what does
- 47:35the end to end output look like and what does
- 47:37the process look like? So we can actually start
- 47:39to automate those. We pick one and we built these
- 47:42incredibly complex skills so that now we can
- 47:44save, you know, paralegals and admins and attorneys,
- 47:47tons and tons of time to go through and to use
- 47:50these tools and to do really different things,
- 47:52right? So pick something first, focus on it.
- 47:55And I would say, you know, pick what you're doing,
- 47:57write it out, you know, have a partner sign off
- 47:59that, yep, this is what we want to test. And
- 48:01this is exactly what we did. We want to build
- 48:03skills to do this because we want to automate
- 48:05this process. Right. And then we run it supervised.
- 48:08We run the existing process to make sure that
- 48:10the AI is doing it right. We run the manual process.
- 48:12You're on the AI process kind of side by side.
- 48:15And we look at then. how much time saved, how
- 48:17much rework do we have to do, right? And what
- 48:19we've found again and again is that the time
- 48:22savings when you start using, you know, Claude
- 48:24co -work or ChatGPT work drives huge increases
- 48:28in productivity, huge increases in quality, and
- 48:31it really allows us to save things and do things
- 48:33in ways that we couldn't before. And then we
- 48:36take the first process, then we start scaling
- 48:38it, fixing it, or if it didn't work, we kill
- 48:39it, right? We go on to the next one. But I am
- 48:42a big believer that look, ROI is incredibly important.
- 48:45We wanna focus on these things. We wanna make
- 48:46this stuff happen. We want it to go do all these
- 48:49things. So a couple of things I say in every
- 48:52presentation. And this is really important because
- 48:55I know there's still a lot of concern about security
- 48:57and where we're going and you know These models
- 48:59are getting better and better. Look there is
- 49:01still risk, right? You know, one of the original
- 49:03cases, you know motto versus Avianca You know
- 49:05was not the last case is getting sanctioned.
- 49:07I write about sanctions on my blog You can see
- 49:10what the latest stats are I follow the guy that
- 49:12there's a French attorney that that tracks all
- 49:14the different things You know, there are real
- 49:16attorneys to get sanctioned. So know that at
- 49:19the end of the day you have to verify everything
- 49:21that goes out. Before you say anything to a judge,
- 49:23before you send it to the opposing counsel, before
- 49:25you send it to a client, make sure the tool is
- 49:28not hallucinating. I will say that especially
- 49:30Claude Opus 5 is much, much better now at not
- 49:33hallucinating and it will tell you when it doesn't
- 49:35know something, right? You know, avoidance is
- 49:38not safety. I mean, Every attorney I know is
- 49:41now using these AI tools. Your clients are expecting
- 49:43it. They're saying, hey, why aren't you getting
- 49:45time savings out of this? What tools are you
- 49:46using? And then if you need help with governance,
- 49:48reach out to me. I can help you with doing that
- 49:50thing. The bar is moving. The technology competence
- 49:53is now part of the duty of competence. The standard
- 49:55is no longer whether you can use AI. It's whether
- 49:58you use it well. So learning how to use it well,
- 50:00learning how to do these things is really, really
- 50:03important. I tell people, look, there are four
- 50:06rules that I think are super important for using
- 50:08AI in practice. The first thing, and if you've
- 50:11seen me present before, is incredibly important.
- 50:14only use the commercial tools, right? Use the
- 50:17team or enterprise version of Claude, use the
- 50:20business or enterprise version of ChatGPT. They
- 50:23have made it, both Claude now and ChatGPT have
- 50:26made it very inexpensive to use their business
- 50:28version, which has all the commercial controls,
- 50:31all the privacy, all the no training on your
- 50:33data. You can start with as little as two licenses
- 50:35per month. I think it's $25 a month per seat.
- 50:38So effectively for $50 a month, you can use a
- 50:41commercial tool now, either from ChatGPT or Claude.
- 50:43and have access to all these advanced capabilities,
- 50:47right? Please do that. Do not use a consumer
- 50:49one, even if you turn training off, right? There
- 50:52are too many risks, you know, and you really
- 50:54risk, you know, waiving, you know, your work
- 50:56product rights and things like that. So please
- 50:57use the commercial ones. Again, match the right
- 51:00tool to the right surface, right? Use cowork
- 51:02to go drive and do things. Use it to create code.
- 51:05Use these, you know, these Excel add -ins and
- 51:08the word add -ins, they are fantastic. You know,
- 51:09you can use it with your commercial subscription
- 51:11so that anything you do there, is not being trained
- 51:14on. This one is really important and I tell everybody
- 51:17this, you should treat every client sort of into
- 51:20their own project. So when you go into Claude,
- 51:22and actually I'll just do this real time here,
- 51:24right? So if I were to go into Claude, and if
- 51:26I had a new client, I would actually create a
- 51:29new project. for this client, right? Because
- 51:32I want all the matter, all the content about
- 51:34this client in one place. And I do the same thing
- 51:36in ChatGBT. Because these tools have memory,
- 51:40right? When you design a project properly, right?
- 51:42The memory stays in there. It doesn't bleed out
- 51:44into other things. And actually, that's a really
- 51:46important point. So the default is, I'm gonna
- 51:47call this client X, and I'm gonna just go ahead
- 51:50and say, create a project. So if I'm in Cloud
- 51:52and I create a project here, everything in here,
- 51:55every chat I have, everything I upload is only
- 51:58going to be in this project. So if I have another
- 51:59project called Client Y, it's not going to go
- 52:02there. Nothing I talk about here could go there.
- 52:04If I just do general chats, there's a risk that
- 52:06you might have said something about Client X.
- 52:08And if you're having a different chat about Client
- 52:10Y in general chats outside a project, there's
- 52:13a risk that... These AI tools, if you've turned
- 52:16on memory, might understand and think about,
- 52:18oh, hey, yeah, we already talked about this,
- 52:20right? I'm going to bring this over here. So
- 52:21please use projects for each of your different
- 52:23clients and each of your matters. Something really
- 52:25important in ChatGPT, and I just want to highlight
- 52:27this. If I go create a new project here, I'm
- 52:31going to call this client X. This is super important.
- 52:35I'm going to say default memory. No. See the
- 52:38drop down here? Use project only memory. By default,
- 52:41if you create a project, you know, in ChatGPT,
- 52:44it will not only have memories of things that
- 52:46are in that project, it'll have memories of things
- 52:47that are outside that project also. So you want
- 52:50to make sure in order to make sure that things
- 52:51for this client in this matter are only here,
- 52:53make sure you have project only memory and then
- 52:55create that project. So super, super important.
- 52:58And then intent, right? You know, I talk about
- 53:01prompting with intent. I've got these detailed
- 53:03prompts, you know, context, objective, role,
- 53:05examples, do all that stuff. So let me end with
- 53:08my five second commercial here. Look If you need
- 53:11help, I'd love to help you. I've worked with
- 53:13over 3 ,000 attorneys to this point across training
- 53:161 ,000 different firms. I do detailed executive
- 53:18workshops across all these tools. And I'm not
- 53:20paid to sell any of these. I have no allegiance
- 53:22to any of them. I try to help people pick the
- 53:24best tool for them. I do training and ongoing
- 53:26education. I've helped organizations develop
- 53:29their AI governance and usage policies. And for
- 53:32select firms, I'll do AI case support. If you've
- 53:34got a very complex case and you're trying to
- 53:36do really advanced AI work, reach out to me.
- 53:40As Mark said when he started, I run a large sub
- 53:43stack. It's got about 2000 subscribers now. It's
- 53:45at smithsteven .com. It's right here. I publish
- 53:48articles almost every day. There'll be a new
- 53:49one out here tonight, but go take a look at it.
- 53:52It's completely free. It will always be free.
- 53:54And my email is here. My phone number is here.
- 53:57I'm happy to reach out and help. And just the
- 53:59last point I want to make is, as I said earlier,
- 54:02you've got to verify everything that comes out
- 54:04of AI, right? AI doesn't have a bar number, you
- 54:07do. So please, you know, pick the workflow. Your
- 54:10name goes in the work. Verify that it's accurate.
- 54:12Make sure that it's true. So I know we've covered
- 54:15a ton of stuff. I'm going to stop sharing. First,
- 54:17I want to say thank you for the time. And I'm
- 54:19going to go in and I'll ask and see if there
- 54:21are any questions in the chat here. So actually,
- 54:26I can open up the Q &A now. And let me start
- 54:29going through some of these. So can I recommend
- 54:32co -console? So I don't actually work with co
- 54:34-console. So it's, I don't, I can't recommend
- 54:37or not recommend. It's not one that I work with
- 54:39on a regular basis. I really focus on the general
- 54:41tools, the chat CPTs, the clods, the Geminis
- 54:43of the world. I know a lot of people have used
- 54:45it and I've honestly heard a lot of mixed reactions.
- 54:49How to avoid AI hallucinations? And so AI hallucinations,
- 54:54so. Like I said, use the more advanced models.
- 54:56And by the way, turn on the most advanced reasoning.
- 54:59So when you select, I'll share my screen for
- 55:02one more second here, because this is actually
- 55:04another really important point. If I go back
- 55:07into Claude here, let me just move this Q &A
- 55:11out of the way so I can see here. If I go into
- 55:13Claude, one of the things you can do when you
- 55:16pick your model, so I'm on Opus 5, and then they
- 55:18have effort level, right? Low, medium, high,
- 55:20extra, and max. The more, first of all, the more
- 55:24advanced you choose the model, right? Or the
- 55:26more, the higher level of effort you put in,
- 55:28the more thinking it does, the less likely it
- 55:30is to make a hallucination. It also takes longer
- 55:32to answer and it takes more of your subscription,
- 55:34more of your tokens. So I typically start for
- 55:36basic tasks that, you know, it's sort of like
- 55:38low and medium, but when I'm doing final presentations,
- 55:40I basically switch over to high because it's
- 55:43great. I've also built a detailed fact checker.
- 55:45I didn't share it on the screen, but I have a
- 55:47detailed fact checker that I run things through
- 55:48at the end that will actually go search the web,
- 55:50search every document. uploaded and look at every
- 55:53single thing and compare all the facts all the
- 55:56assertions and so if you want a copy of that
- 55:58email me I'll email it to you and you can add
- 56:00it as a skill it's really easy to do. How can
- 56:03a lawyer or using cloud or Chachi BD avoid having
- 56:06history name on AI chats and use becoming discoverable
- 56:08in event of litigation? So again, if you're using
- 56:11the commercial ones, it's not training on your
- 56:13data, right? So it's, you know, that's your stuff.
- 56:16It's not out there shared publicly. You know,
- 56:18it's, you know, it's really not discoverable.
- 56:19I mean, they, they make it very difficult to
- 56:21get to that stuff. And that's again, why talk
- 56:23about using the commercial ones. If you go back
- 56:25in my sub stack and search on the, Search for
- 56:28$200. I wrote an article a couple months ago
- 56:31about the latest privacy policies and all these
- 56:32things. And it talks about how to make sure that
- 56:36you're using these tools safely and confidently.
- 56:39What level of cloud subscription is necessary
- 56:41desired? So again, you need to have a commercial
- 56:44one. So they have consumer ones and the consumer
- 56:46ones are basically called pro and they're called
- 56:48max. And again, even though you can technically
- 56:51turn off training, I advise all my clients, please
- 56:53don't do that, right? Then they have commercial
- 56:55ones, right? And they're ones for Claude, they're
- 56:57called team or enterprise. So subscribe to that.
- 56:59The minimum is basically 50 bucks a month for
- 57:01Claude. And it's about 50 bucks a month for ChachiBT.
- 57:04ChachiBT's entry level one is called business.
- 57:06You can find them on the website. They're easy
- 57:07to do. For solos and small firms that don't have
- 57:11access to closed enterprise models, how do we
- 57:13avoid violating client confidentiality? Oh, again,
- 57:15the same thing. Use the commercial ones, right?
- 57:18You can, I mean, look, the bar association has
- 57:20said if you've done your due diligence and you're
- 57:21using tools that have the security and privacy,
- 57:24you can put things in there. Now, a lot of my
- 57:26clients still want to sanitize the documents.
- 57:28They want to take things out and you can do that.
- 57:30In fact, you can write tools that actually do
- 57:32that as well. But you can actually, but, you
- 57:35know, so you can sanitize it before you put it
- 57:36in there if you're worried about that. But the
- 57:38tools, because they're not training and they're
- 57:40contractually obligated not to, and I can send
- 57:41you their data processing agreements and everything
- 57:44else. I look at this stuff all the time because
- 57:46this stuff's really important. Does Microsoft
- 57:48365 Copilot have the ability to schedule tasks?
- 57:51Yes, they have added that into Copilot 365. You
- 57:55have to subscribe to it. That's the extra fee.
- 57:57And there's extra, sorry, that's the extra subscription
- 58:00fee. And there are now extra fees for Microsoft
- 58:02to run it. So check the pricing on that. They
- 58:05actually charge when your agents run additional
- 58:07fee. Whereas today, ChachiBT and Claude don't
- 58:10do that. I love chat GPT and Claude, but last
- 58:14I checked only Gemini has had relative inexpensive
- 58:16version that maintains sufficient confidentiality
- 58:18for legal work. So I use notebook at LM. Has
- 58:21anything changed so that Claude or chat? Yeah,
- 58:23so I just talked about this. Look, you can go
- 58:25read the subset article I wrote about search
- 58:27for a dollar sign 200. But so if you're using
- 58:30the commercial ones, it has the exact same kind
- 58:32of confidentiality and privacy that the Gemini
- 58:35has. And again, if you're using the Gemini consumer
- 58:37version, you do not have that same level of confidentiality.
- 58:41Please, please, please only use Gemini Workspace.
- 58:45If you're using Notebook LM, make sure you're
- 58:47using the Workspace version of that, because
- 58:48there's a consumer one. And there's ways that
- 58:51that data can be trained on if you're not careful.
- 58:54Are any of these connected yet to legal databases
- 58:56for reliable legal research? So if you go look
- 58:58at what Claude introduced in their legal release
- 59:02back in, I think it was May, with the 20 connectors,
- 59:04they're adding more and more stuff to it. The
- 59:07firms are, you know, Thomson Reuters and all
- 59:10the, and Westall and stuff, you know, they're,
- 59:11they are limiting what can be done there, but
- 59:14they are slowly adding different things into
- 59:15there. And there are other public sources out
- 59:17there. You can connect Claude to, or Chachapi
- 59:19T2, where you can actually get very good legal
- 59:21research. So it's slowly changing over time,
- 59:24but the incumbents are trying to lock down and
- 59:26make sure you have a subscription and do things
- 59:27like that. Do all the office plugins utilize
- 59:30the same privacy data structure as your business
- 59:32enterprise account? Yes, great question. As long
- 59:34as you are logged into like your Claude version
- 59:37on your commercial account and you're using Word
- 59:40or Excel with that account or chat should be
- 59:42the same thing, it inherits the exact same data
- 59:44privacy stuff so you can use it exactly the same
- 59:46way. I'm so glad you asked that question. That
- 59:48was a great question. Do you need to first have
- 59:50a Claude annual subscription before you can use
- 59:52Claude in Word or is it paid, limited? No, you
- 59:55can do monthly. So you can do a commercial monthly
- 59:57subscription to either Claude or ChatGPT. And
- 1:00:00as part of that, you can access the downloads
- 1:00:03as well. So definitely start monthly, figure
- 1:00:05out which tool you like, and go from there. Malwarebytes
- 1:00:09reported that Claude Chastain in Google search.
- 1:00:11Yeah, look, this is a super misleading headline,
- 1:00:15OK? People had to physically. go into a chat,
- 1:00:18they had to share it, they had to make a public
- 1:00:20link available, and then someone had to actually
- 1:00:22share that on the website that Google was indexing.
- 1:00:25Your searches, your artifacts, they are not Google
- 1:00:28-able by default. I mean, I've written about
- 1:00:30this in the past. There's an article about this,
- 1:00:32too. It's a completely misleading headline that
- 1:00:35your chats are showing up there. Again, you had
- 1:00:36to go and you had to physically share it. You
- 1:00:38had to say, this is available to anyone on the
- 1:00:40planet with this link. And then someone had to
- 1:00:42post that link. on a website that Google was
- 1:00:44indexing, right? Nothing that you're doing in
- 1:00:47this as long as you haven't shared it publicly
- 1:00:48is going to show up by default. How can we integrate
- 1:00:52AI into our case manager systems? Oh, so if you
- 1:00:54go. Look what Google is doing. So a lot of these
- 1:00:57companies, you know, Clio is doing this too.
- 1:00:59They're adding something called MCP servers,
- 1:01:01these connectors that you can actually start
- 1:01:04integrating and connecting documents and getting
- 1:01:05them out and doing things like that. So I didn't
- 1:01:07cover that in here, but there's a lot of updates
- 1:01:09happening in there to start making this integration
- 1:01:11more seamless and easier to do. There's a bunch
- 1:01:15more questions. I'm going to let me just scroll
- 1:01:18to see if there's anything that's sort of really
- 1:01:19new here, because I know we're past time and
- 1:01:21I do want to need to wrap up. This is a good
- 1:01:25question. Does Claude permit you to have an agent
- 1:01:27log into your Westlaw account or perform legal
- 1:01:29research to develop case summaries? So look,
- 1:01:32Claude or ChachiBD can actually control a Chrome
- 1:01:34browser and can log in and can do things. However,
- 1:01:37You know, Westlaw and others have basically sort
- 1:01:39of in their terms and conditions and said they
- 1:01:41don't want agents automatically connecting in.
- 1:01:44So you have to go check the terms of service
- 1:01:46very carefully about whatever legal research
- 1:01:48tool you're using. These tools can technically
- 1:01:50do it. Look, Claude and ChatGPT can log into
- 1:01:53all my paid accounts. It can actually navigate
- 1:01:55things. I mean, it can do research for me. It
- 1:01:57can pull things down. It can actually do downloads.
- 1:01:59It can get files. But you've got to check the
- 1:02:01terms of service of what you're accessing. So
- 1:02:02just be careful around that stuff. So yeah, so
- 1:02:08someone asked about a great webinar. Thank you.
- 1:02:10I appreciate the comment. Again, Westlaw, LexisNexis,
- 1:02:13I don't natively compare those. So I try to be
- 1:02:15very careful about that stuff. So I don't generally
- 1:02:18cover those. I try to leave that to other people
- 1:02:20that are in the space. How did you add your chats
- 1:02:23for Gemini? Can you upload chats from GPT and
- 1:02:26Claude easily? Actually, I'm not sure I understand
- 1:02:29the question from Jay. So if, Jay, if you send
- 1:02:33me a follow -up email, I'll try and answer that
- 1:02:34question. I'm just not exactly sure what that
- 1:02:36meant here. And I'm going to skip the thing about
- 1:02:40autonomous weapons and things like that. And
- 1:02:44transcripts, the potential problem with analysis
- 1:02:45and research is that any requirement to first
- 1:02:47load all documents and transcripts for analysis
- 1:02:49and research, are you saying you can do a search
- 1:02:51of your entire hard drive? Oh. So great question.
- 1:02:55I'll end with this one. So there is a limit if
- 1:02:58you're doing a chat as to how many files you
- 1:03:00can upload and how much each one of these chats
- 1:03:01can handle, right? So it can actually manage,
- 1:03:05you know, I think you can upload, you know, depending
- 1:03:07on the tool in the day, you know, 15 or 20 documents
- 1:03:09in a given chat and that's sort of all can look
- 1:03:11at. When you're using a tool like Claude Cowork
- 1:03:13or using you can actually point it in a directory.
- 1:03:18And even if it has, you know, 50, a hundred,
- 1:03:20a thousand files, you can tell it to start going
- 1:03:22and opening up those files one by one and looking
- 1:03:24for things and then creating an index and finding
- 1:03:26those things. So that that's like the document
- 1:03:28log use case I did. I mean, when I showed you
- 1:03:30that example of the deposition search, we started
- 1:03:33with 50 ,000 documents and we created indexes.
- 1:03:35We knew which documents to go grab. Then we grab
- 1:03:37those and we put them in a special search. So
- 1:03:39that's the power of these tools that allows you
- 1:03:41to do things that you cannot do in a regular
- 1:03:43chat. That's the power of work and what you're
- 1:03:45doing here. So look, I will try to get some of
- 1:03:49the other questions. I'll answer them by email
- 1:03:50if I can get copies of these from Suzanne and
- 1:03:53the team. Again, I want to thank you for the
- 1:03:55time. There are so many great things going on.
- 1:03:57These tools have gotten so good. And if you've
- 1:03:59got questions, reach out. And if you're looking
- 1:04:01for assistance, again, please reach out to me.
- 1:04:04But thank you again. I really appreciate it.
- 1:04:06you know, look happy using these tools. These
- 1:04:08tools are amazing and they get better and better
- 1:04:11every day. So I love talking about it and using
- 1:04:13them and hopefully you can tell from that. So
- 1:04:15again, thanks everybody and have a great rest
- 1:04:16of your day.