Latest / CLEs You Actually Want to Hear / Introductory AI for Hesitant Lawyers: Core Concepts Every Practitioner Should Know
Transcript
- 0:03Hello and welcome to Introductory AI for Hesitant
- 0:07Lawyers, Core Concepts Every Practitioner Should
- 0:11Know. I'm Alapoli Castro, Director of Education
- 0:14and Professional Development at Beverly Hills
- 0:16Bar Association. And today I'm delighted to introduce
- 0:19our speaker, Jessica Moran. Jessica is the founder
- 0:22of AI Edge Partners, an AI consultancy that helps
- 0:26law firms and professional services businesses
- 0:29apply AI to everyday operations. She has spent
- 0:32more than 20 years working at the intersection
- 0:34of law and technology. Her career began in e
- 0:38-discovery and legal technology, where she led
- 0:40large -scale litigation, regulatory investigations,
- 0:43and complex data analysis for AMLaw 200 firms,
- 0:47Fortune 500 companies, and financial institutions.
- 0:51Today, she focuses on helping legal practices
- 0:54identify where AI genuinely fits designing systems
- 0:57that reduce administrative burdens, manage risk,
- 0:59and support better decision making. We're so
- 1:02excited to hear what she has to share with us
- 1:04today. So I'm now going to turn it over to her.
- 1:06Hi, thank you, everyone. Thank you to the Beverly
- 1:08Hills Bar Association for having me today. And
- 1:11thank you, everyone, for being here. My name
- 1:13is Jessica Morin. I have been in technology consulting
- 1:19now for 20 plus years. For a lot of that time,
- 1:22I worked directly with law firms and corporate
- 1:24legal departments. And now I'm focusing on my
- 1:30own consulting practice where I help small businesses
- 1:32and law firms implementing AI in a way that is
- 1:35practical and safe. And that's kind of the lens
- 1:39I'm bringing today. I want to say one thing right
- 1:43at the top. I'm not here to sell you on AI, and
- 1:46I'm not here to tell you, you know, that you
- 1:49have to use it and which tools to use. I am very
- 1:52tech agnostic. I'm not affiliated with any of
- 1:55the products we'll talk about today. My only
- 1:58goal for the next hour is that you leave understanding
- 2:02that this technology is important and it is,
- 2:06you know, being used more and more frequently
- 2:08by everyone and that you have enough information
- 2:11to make informed decisions about it. So let me
- 2:17start sharing our slides. Okay. Hopefully you
- 2:28guys can see that. Okay, so let's move on to
- 2:39you are right to be cautious. That might be a
- 2:44little unexpected, but I know everyone's probably
- 2:46seen the headlines lawyer sanction for citing
- 2:49cases that don't exist because an AI tool invented
- 2:54them concerns about client confidentiality. and
- 2:58underneath it all this general sense that everything
- 3:00is moving so fast and it's super hard to keep
- 3:04up with. Wherever you landed, there's something
- 3:10worth knowing. I think you're not early and you're
- 3:14certainly not behind. Depending on where some
- 3:19surveys I've read recently that roughly four
- 3:23in five lawyers now report using AI in some form,
- 3:28and less than one in five just three years ago
- 3:32were reporting using AI. So I think that's showing
- 3:35that this is no longer a fringe experiment, and
- 3:39it's becoming a standard really. And that's exactly
- 3:43why getting grounded in it matters. So if you're
- 3:47feeling hesitant or cautious, then this is probably
- 3:51the correct instinct. AI is a very powerful tool
- 3:54and can easily be misused and sometimes inadvertently.
- 3:59So what I plan to do today is give you enough
- 4:02grounding that this caution becomes informed
- 4:05instead of just anxious. And so hopefully by
- 4:08the end, you'll know the vocabulary behind all
- 4:11the hype. You'll know the main categories of
- 4:13tools that are out there, and you'll know what
- 4:17these tools can and cannot do for you. And you'll
- 4:20know how to use them without putting client confidentiality
- 4:23at risk. That's the whole goal of this presentation.
- 4:27Nothing more complicated than that. And the best
- 4:31place to start is with the words, the vocabulary.
- 4:36Because a lot of the intimidation around AI is
- 4:40really just that unfamiliar vocabulary. So here's
- 4:45the thing about the AI jargon. Most of it sounds
- 4:48far more technical than the idea underneath it.
- 4:52So let's quickly demystify some of these terms
- 4:55that you're probably keep running into. But first,
- 5:00a little history because AI is not actually new
- 5:05to the legal profession. We have been using forms
- 5:08of it for well over a decade. We just didn't
- 5:12always call it AI. Back around in, I think, 2012,
- 5:16courts began approving what's now called predictive
- 5:20coding or technology assisted review in eDiscovery.
- 5:24And that was machine learning based solely on
- 5:30predictions of which documents were going to
- 5:34be relevant or not. So if you've ever used Westlaw
- 5:38or LexisNexis with natural language search, that
- 5:43was machine learning too. And tools like Lex
- 5:46Machina were doing litigation analytics, predicting
- 5:49outcomes based on a judge or court's history,
- 5:54contract review platforms were using machine
- 5:56learning to flag risky clauses. So the legal
- 6:00profession has quietly been relying on AI for
- 6:04years. What's changed most recently, and the
- 6:08reason we're all in this room, is generative
- 6:11AI, or gen AI. The older tools were able to sort
- 6:15and predict and classify things that already
- 6:18existed, and now generative AI is creating things.
- 6:23It creates and writes. It writes memos, drafts
- 6:27clauses, summarizes depositions. So that's a
- 6:31big leap, right, from sorting information and
- 6:33producing it And now it's creating information
- 6:38which feels very new. And for a lot of people,
- 6:42it is a bit unsettling. So with that history
- 6:47in mind, let's talk about the terms that you'll
- 6:50keep running into. First, you'll hear AI, machine
- 6:53learning, and generative AI or gen AI. Almost
- 6:59interchangeably, they're being used. The piece
- 7:03that is Genuinely new in the last few years is
- 7:07the generative AI part, the part that writes
- 7:09and drafts and can talk back to you now. The
- 7:13engine that's behind all of that is a large language
- 7:18model or an LLM. So this is the only technical
- 7:22thing I'll ask you to remember about it. An LLM
- 7:26works by predicting the most likely next word.
- 7:32Over and over. based on patterns and the enormous
- 7:36amount of text it was trained on. And that sounds
- 7:39very simple, but it actually works. It produces
- 7:44very fluent, confident output. And I'm sure most
- 7:49of you are familiar with working with it a little
- 7:52bit. The prompt is just the instructions that
- 7:57you give. the AI. So we'll spend a little bit
- 8:01more time on prompts later because how you ask
- 8:07genuinely changes the output that you get. Now
- 8:13one of the most important words on the slide
- 8:15for a lawyer is hallucination. And that's the
- 8:19term for when the tool states something false
- 8:22with complete confidence. This is where the fake
- 8:26case citations come from. the tool's not lying
- 8:30to you exactly. It's doing what it always does.
- 8:33It's predicting plausible sounding text, and
- 8:36sometimes plausible and true aren't the same
- 8:39thing. Now, training data is the body of text
- 8:44that the model learned from. The key point is
- 8:48that it is a snapshot. frozen in a point in time.
- 8:52It's not a live, authoritative legal database,
- 8:56for example, and it's not connected to current
- 8:58law unless you're using a specific tool that
- 9:02connects it. A context window is just how much
- 9:07the tool can hold. The amount of context it can
- 9:11reliably keep under the umbrella. And at one
- 9:16time, so it'll be any documents you give it.
- 9:20And plus the back and forth of the conversation
- 9:23that you're having. And there are limits to that
- 9:26as well. You'll see sometimes if you're using
- 9:29a system going back and forth with it, the context
- 9:33will start to slip. And that's because the context
- 9:37window has kind of hit its maximum capacity and
- 9:40can't remember everything going all the way back
- 9:43to the beginning. And then this last one here
- 9:47at the bottom, agentic AI, I've set this one
- 9:50apart on purpose because it's a term that you're
- 9:54probably gonna hear more and more of. It's relatively
- 9:58newer, but it's getting more and more popularity.
- 10:03So agentic AI describes systems that go beyond
- 10:07just answering questions. They are... somewhat
- 10:11autonomous, they can operate based on the instructions
- 10:14that you give it. They take action, usually in
- 10:17steps, and they can use other tools. They can
- 10:20search, draft, they can chain several tasks together
- 10:24to reach a goal. So it's a very capable direction
- 10:29that the technology is heading into. For example,
- 10:35this area, you know, can these agents can Be
- 10:39set up and then run independently and you can
- 10:42have kind of a master agent and sub agents that
- 10:45it supervises and each sub agent has specific
- 10:48tasks and things like that. So it's it's very
- 10:50kind of a more complex higher level use of AI.
- 10:57And it's, you know, for that reason, it's it
- 11:00requires a lot of human supervision. And you'll
- 11:04certainly be hearing this term, I think, more
- 11:06frequently in the months to come. So if you remember
- 11:12one idea from the entire slide, make it this
- 11:14one, these tools complete patterns. They don't
- 11:18look up truths. That single fact explains both
- 11:21why they're so powerful and exactly where they
- 11:24fail. And once you understand that, the hallucinations
- 11:28you know aren't as mysterious and it's a bit
- 11:31easier to predict them. So that's how they work.
- 11:37Let's talk about what's actually available out
- 11:42there. So there are by some counts hundreds if
- 11:50not thousands of different legal AI tools and
- 11:56more launch, I think, every single week. I see
- 12:00if you're paying attention in that area, it's
- 12:04quite overwhelming the amount of different tools
- 12:07that are out there. And it's nearly impossible
- 12:11to track. I'd given up a long time ago. So we're
- 12:16not going to talk specifically about products.
- 12:18We'll focus more on the categories. The first
- 12:22category is the general purpose. assistance.
- 12:27So these are the big names that you've probably
- 12:30all heard of. ChatGPT from OpenAI, Claude from
- 12:35Anthropic, Google Gemini, and Microsoft Copilot.
- 12:41Those are probably the biggest ones here in the
- 12:43US. They're very versatile, they're inexpensive,
- 12:48and they're really pretty good at drafting and
- 12:51summarizing. One thing to know, the free versions
- 12:55and the paid versions often handle your data
- 12:58very differently, which matters enormously in
- 13:03the legal world. And we'll talk a little bit
- 13:06more about that. The second category is legal
- 13:10specific platforms. So in this area, specifically,
- 13:14there's just a ton. coming to market all the
- 13:18time. And these tools are usually built on top
- 13:23of the very same engines as the general assistant
- 13:25ones, but they're fine -tuned towards legal work.
- 13:30So they might have research features, citation
- 13:34tools, and they usually have firmer data protections.
- 13:38So they're really designed more for law firm
- 13:41use. You probably heard of some of the bigger
- 13:44names like Harvey, Spellbook, um, Westlaw has
- 13:48co -counsel, Lexus has protege, um, and there's
- 13:53just so many more. I'm just naming these as examples,
- 13:56not endorsements. Um, and, you know, there's,
- 14:03they are... Even though they're better tuned
- 14:06to the legal environment, they might have better
- 14:09prompts, better training, kind of a more focused
- 14:13data set for legal, but they still aren't bulletproof.
- 14:20I think there was a Stanford study that found
- 14:23that even these legal -specific research tools
- 14:26hallucinated a meaningful share of the time.
- 14:29The error rates were generally lower than the
- 14:32basic chat bots but they weren't zero. So that's
- 14:37one important thing to keep in mind is that even
- 14:42these purpose -built legal tools require you
- 14:45to verify the data that they are presenting.
- 14:51And now the third category is the one that people
- 14:54often forget about and it might be the easiest
- 14:56place to start. It's the AI that is being built
- 15:00directly into tools you might already use. So
- 15:05that's like Microsoft Copilot, which is now inside
- 15:08Word and Outlook. Google Gemini is built inside
- 15:12of Google Workplace. If you use a practice management
- 15:16system like Clio, it has its own AI built in.
- 15:20So you may not have to adopt anything new. The
- 15:25AI is starting to show up in all of our All the
- 15:28platforms that we work on, they're all integrating
- 15:31AI in some way, shape, or form for the most part.
- 15:35So that's typically an easier entry point than
- 15:39maybe trying to bring some outside tool into
- 15:43your IT infrastructure. And the fourth box here
- 15:49is not really a category. It's more of a reality
- 15:51check that whatever you use, you should weigh
- 15:54a few things. These tools predict confident sounding
- 15:58errors. They can give you different answers to
- 16:01the same question. They also cost money, sometimes
- 16:05a lot of money, once you get seriously involved
- 16:08with using them. And there can be a real learning
- 16:11curve. So those are some things to keep in mind.
- 16:15None of those are reasons to avoid AI, but they're
- 16:19simply things to plan for. the same way, you
- 16:24know, you plan for any new tool in your practice.
- 16:29I'm curious if you can, I don't think we have
- 16:32a chat, but if you can maybe raise your hand
- 16:34how many people have heard of some of these tools
- 16:37that we've mentioned, that I mentioned on this
- 16:40slide. If I can see. Okay, well, I know some
- 16:56of the things are locked down in here. But yeah,
- 17:01I think more and more everyone's starting to
- 17:04hear about these names and hear of other people
- 17:07using these tools and some of you may be using
- 17:15some of this without realizing it's. It's actually
- 17:19AI. It's just kind of a new feature that's built
- 17:21into some of the systems you're currently using.
- 17:26So with all of that, the natural question, I
- 17:28think, is where should I actually use this? What
- 17:34should I actually be doing with this new technology?
- 17:41So I like to think about it in two lanes. And
- 17:47the reason why is, you know, I think when lawyers
- 17:50here, AI for lawyers, they picture handling handing
- 17:53over, you know, all of their legal work to a
- 17:55machine kind of outsourcing all of the decisions
- 17:59and the drafting they do and all that stuff.
- 18:03So which is not at all what it should be. I prefer
- 18:08to think of AI as kind of falling in these two
- 18:10lanes and. you get to choose which ones you want
- 18:16to start in. So lane one is kind of back office
- 18:20and business operations. This is more on the
- 18:24administrative side and business side of the
- 18:26practice. The work that, you know, quietly eats
- 18:30away at your non -billable time. This lane is
- 18:34a bit of a lower risk and is an easier entry
- 18:37point in my opinion. to using AI. So if you are
- 18:41a solo or small firm without much support staff,
- 18:45AI can act like an assistant that you might not
- 18:47have. Now lane two is more of the substantive
- 18:51legal work. This is the part you're probably
- 18:53most wary of and that wariness is appropriate.
- 18:57The rule in this lane is that everything is a
- 19:00supervised draft and never a finished product.
- 19:04You want to make sure that You as an attorney
- 19:07are always the expert, the subject matter expert,
- 19:10and you're signing off and reviewing everything
- 19:13in detail whenever using any sort of AI. So that's
- 19:18really the rule that runs across both of these
- 19:20lanes that you need to really keep in the back
- 19:23of your mind any time you're using AI is that
- 19:27AI can draft, but the lawyer should be deciding.
- 19:30You don't want to give up your judgment. You
- 19:33just want to use it so that you get a faster
- 19:36first draft of work that you were already going
- 19:39to do. So let's make these a little bit more
- 19:43concrete starting with the first one. So this
- 19:52is lane one and I'll walk you through what it
- 19:56actually looks like because this is where most
- 20:01of you can get real value with almost no risk,
- 20:04little to no risk. And there's some pretty cool
- 20:07use cases out there that I've seen. So as I go
- 20:15through these, I'd like for you to do one thing.
- 20:18Pick a single task in your week that you most
- 20:21dread or that eats away a lot of time. Hold it
- 20:26in the back of your mind. And I'm willing to
- 20:29bet, but by the end of this slide, at least one
- 20:32of these examples can map to it. So let's start
- 20:36with client intake and communication. If you
- 20:40imagine a long, rambling intake email, it lands
- 20:43in your inbox. There's maybe three issues tangled
- 20:46together, no structure. You can have AI turn
- 20:50that into a clean summary in the matter of seconds,
- 20:53right? It'll... parse out the issue, the parties,
- 20:57the deadlines, and next steps. You can use it
- 21:00to have a draft of a first pass reply to routine
- 21:04client email that you then review and send. You
- 21:09can even have it build an intake questionnaire
- 21:12tailored to your practice area. I worked with
- 21:15a solo practitioner who was spending the first
- 21:18hour of every morning just triaging the overnight
- 21:21emails. So we set up a simple routine where AI
- 21:24drafted summaries and suggested replies for her
- 21:27to approve. And she got that hour back every
- 21:30single day. On marketing and business development,
- 21:34that's a real gift for small employees that do
- 21:37not have a marketing department. Now it's super
- 21:40easy to take one piece of content and repurpose
- 21:43it into social media posts, newsletter blurb,
- 21:50website FAQ. You can draft attorney bios, practice
- 21:54area descriptions. You can respond to reviews.
- 21:59I've seen a two -attorney firm go from posting
- 22:03nothing on social media for months to a steady,
- 22:06consistent presence simply because the friction
- 22:10of writing from a blank page was gone. I've seen
- 22:14systems where you do a podcast episode or do...
- 22:19one thing that you have a recording of and then
- 22:22you can set up AI to extract the text from it
- 22:27and turn it into multiple social media posts,
- 22:29maybe split up the video and post different excerpts
- 22:33from it, all within a matter of a few minutes.
- 22:39So it's really powerful on that side. And documents
- 22:43and knowledge sharing, you can summarize long
- 22:46documents, email threads, We'll talk about confidentiality
- 22:50on that in a moment. But you can build personal
- 22:55checklists and standard operating procedures
- 22:57just around the way you work already. So that
- 23:01can be really powerful. You can create templates
- 23:04from your own existing documents, and you can
- 23:08create a shared knowledge base that can be queried
- 23:11using AI across team members. So there's a lot
- 23:16you can do kind of on the knowledge management
- 23:18side and templatizing things and just kind of
- 23:22making things run a bit more efficiently by extracting
- 23:25a lot of the information that you already have
- 23:28in your firm and practice area. Now on the administrative
- 23:34side, you can draft internal policies and standard
- 23:38language. You can turn meeting notes into clear
- 23:41action items. Use it to prep agendas. I love
- 23:47to use it for a meeting note taker. That's a
- 23:50really easy way to save time is to just have
- 23:54the AI listening and transcribing. So at the
- 23:59end of your meeting, you have a summary and action
- 24:02items and all of that stuff done for you. You
- 24:09can use it to draft. Summaries of I think your
- 24:15your matters so you can put together like weekly
- 24:20summaries of where all your matters stand and
- 24:24you can use it to help with taking notes and
- 24:27turning them into good time entry narratives.
- 24:31There's just there's just so much you can do
- 24:33here that it's it's really powerful and it's
- 24:36hard to pick one place to start in my opinion.
- 24:40One thing I hear constantly from small firms
- 24:43is how much billable time can leak away through
- 24:46vague or forgotten time entries. So cleaning
- 24:50that up can often be where AI pays for itself
- 24:53within the first month or two. Everything I've
- 25:00just described is you using AI while sitting
- 25:04at your desk, asking it to help with a task,
- 25:07which is great. But there's also an additional
- 25:11next level, and that's where I get generally
- 25:15excited, for a small firm especially, and that's
- 25:19automating and using agentic AI. So instead of
- 25:23you having to prompt AI each time, you can set
- 25:26up workflows. that run on their own in the background.
- 25:30And so you have this kind of machine that you've
- 25:34trained and you've tweaked and set it up. And
- 25:37now you've got repeatable things that it can
- 25:40do on your behalf. And you can still oversee
- 25:43those and verify the final outputs and things
- 25:46like that. But you essentially can create these
- 25:50behind the scenes workers, if you will, to help
- 25:52save a tremendous amount of time and a lot of
- 25:55your admin work. A few examples that we set up
- 26:00recently are intake agents. So this is like a
- 26:04voice AI agent that can answer your phone after
- 26:07hours or handle first touch client intake. A
- 26:12potential client call, you know, at seven at
- 26:14night can be answered instead of going to voicemail,
- 26:17which often means it might be a lost client.
- 26:20So the agent answers, it gathers basic info,
- 26:23and will send you a structured intake summary.
- 26:26For a solo or small firm, that's capturing business
- 26:30that you were potentially losing before. Another
- 26:33one is automated invoice and payment follow up.
- 26:37So you can set up a workflow that watches for
- 26:40overdue invoices and sends polite professional
- 26:43payment reminders on a routine schedule with
- 26:48having to remember to write each one. And you
- 26:52don't have to deal with those awkward awkward
- 26:55follow -up emails asking to be paid. Another
- 27:00example are automated matter summaries and reminders.
- 27:05So a workflow can be built to pull together the
- 27:07current state of a matter, generate a short summary,
- 27:11and send that to you or your client on an automated
- 27:14basis, or send reminders about upcoming deadlines.
- 27:19So it's kind of built -in administrative follow
- 27:22through that just happens instead of, you know,
- 27:25you trying to keep all that stuff up in your
- 27:27head and keep track of everything for all your
- 27:30different matters you have going on. So this
- 27:34link can save you a lot of time on tasks. Automation
- 27:39saves you... the task entirely, and that's the
- 27:44direction that I see AI heading in more and more.
- 27:46I think we're graduating from the, you know,
- 27:50let's just use AI, one -off here and there, even
- 27:53as kind of a right -hand assistant, I can ask
- 27:56good things, it'll do it. Now it's sort of, how
- 27:59can I engineer this to repeatedly do this work
- 28:03that I don't have time for anymore? Or, you know,
- 28:07take certain things off my plate. So that's,
- 28:11you know, a very exciting proposition. And I
- 28:15think it's increasingly within reach even for
- 28:17small firms, which is exciting. And I think,
- 28:24you know, I always suggest that lawyers that
- 28:28are hesitant about AI start in this lane because
- 28:34none of this touches legal advice, none of it
- 28:37touches a filing. It's the safer area to build
- 28:41comfort with these tools and to see real time
- 28:44savings quickly. And as always, the AI will draft,
- 28:49but you should decide. We purposely want to keep
- 28:53humans in the loop, approving final things before
- 28:58they go out to clients and things like that.
- 29:01And once you're comfortable there, you can move
- 29:04on to the second lane, which is where the stakes
- 29:12and the value both go up a bit. So the substantive
- 29:16legal work, this is lane two. I wanna keep saying
- 29:21the one phrase as we go through this, supervised
- 29:24draft. So everything here is useful and none
- 29:27of it is ever the final word. You always want
- 29:30to be the expert. You always want to have that
- 29:33final say on anything before it goes out. On
- 29:39research assistance, AI is genuinely helpful
- 29:43for getting oriented in an unfamiliar area and
- 29:47surfacing the issues that you should investigate.
- 29:50But, and this is the big one, you need to verify
- 29:54every source that it gives you. You never ever
- 29:57rely on an AI -generated citation without checking
- 30:01that the case exists and says what the tool claims.
- 30:05So this is where I think we've all seen lawyers
- 30:08get in trouble and continue to get in trouble.
- 30:12This is fascinating to me. We see every week
- 30:16something, I feel like something pop up where
- 30:18there was a hallucinated citation that was submitted
- 30:22to the court. On drafting support, it's excellent
- 30:26for first drafts of routine correspondence, standard
- 30:29clauses, standard agreements, which you then
- 30:32have to revise and make your own. On document
- 30:35review and analysis, you can have it summarize
- 30:38a contract or a set of discovery materials, flag
- 30:41key terms, and compare two versions of an agreement,
- 30:45do a red line. Just remember to always keep confidentiality
- 30:49in mind with this. And on brainstorming, it's
- 30:53surprisingly good. thinking partner. You can
- 30:57outline arguments, ask it to anticipate the other
- 31:00side's positions, and use it to pressure test
- 31:03your own reasoning before committing to it. Now
- 31:08here is what separates firms that dabble from
- 31:12firms that get real leverage out of this lane.
- 31:15They stop starting from scratch every time. And
- 31:18they customize the AI to their own practice.
- 31:22So in practice, that means a few things. They
- 31:25will build a library of saved prompts, the specific
- 31:29detailed instructions that they have refined
- 31:32and that reliably produce the output they want.
- 31:36They create templates for the documents and tasks
- 31:40they do over and over. So the AI starts from
- 31:44the firm's preferred structure and language.
- 31:47instead of a generic one. So some of the legal
- 31:51specific tools let you go further and load in
- 31:54your own playbooks, your standard clauses, your
- 31:57style, even your past work. So the AI's output
- 32:02sounds very close to your firm rather than a
- 32:05generic chat bot. So that's kind of the ultimate
- 32:10goal is to use this in a way that it really begins
- 32:14to sound like you, your firm. It's a very custom
- 32:18level instead of kind of the generic off -the
- 32:21-shelf one. So if you think of it this way, the
- 32:25first time you ask, you are training yourself
- 32:27on how to prompt. So your first question, your
- 32:30first prompt, but then... once you can iterate
- 32:34and get to a good prompt and you can turn that
- 32:37prompt into a template. And at that point, you've
- 32:40effectively built a reusable tool that you and
- 32:44your whole team can use again and again. So that's
- 32:47how kind of a one -off experiment can become
- 32:49a genuine part of the practice. And that's really
- 32:53the difference between being AI, you know, just
- 32:56being a novelty and AI becoming the infrastructure.
- 33:01But same role as lane one, it's just with higher
- 33:04stakes, AI drafts, and the lawyer decides. And
- 33:11that brings us to the part of the talk that matters
- 33:16most, in my opinion, the risks and how you manage
- 33:20them. So I think there's Two key risks we'll
- 33:27get into, and I want to be direct about these
- 33:29risks because pretending they don't exist would
- 33:32not serve anybody, certainly not serve you, but
- 33:35I also want to frame them correctly. So neither
- 33:38of these is a reason to avoid AI, but you have
- 33:42to be deliberate in how you use it. So the first
- 33:45risk is, as we've talked about a lot, then the
- 33:49hallucinated output. As we said earlier, these
- 33:53tools can produce confident fabricated citations
- 33:56and quotes, and courts have started sanctioning
- 34:00lawyers who have filed unverified AI output.
- 34:04The fix, though, is really simple and should
- 34:07be a habit every time. You check everything before
- 34:11it leaves your desks. You check any citations.
- 34:14You check, you know, the cases. You just check
- 34:17the quotes. You just verify all that information
- 34:20or have a human do it for you. And I will say
- 34:25that this is getting a lot better from the tools
- 34:27I've seen recently compared to what they were,
- 34:31you know, two years ago or so. You do see a lot
- 34:35more of the tools showing their thought process
- 34:39and a lot of times they're like, we have to,
- 34:41you know. This requires current information.
- 34:44I have to check this and check that. A lot of
- 34:47times the tools might link to the sources that
- 34:49they found which is really helpful. So you can
- 34:52just click those links and verify the source
- 34:54and you know it's a reminder too that everything
- 34:57on the internet isn't necessarily true. So even
- 35:01just because it's linking to a web page is that
- 35:03a reliable website. Can you trust the data that
- 35:08the AI tool found on that website? So you definitely
- 35:12have to be mindful of all of that when using
- 35:17these tools to help with drafting and researching
- 35:20and putting this stuff together. The second risk
- 35:25is confidentiality and privilege. how you use
- 35:30AI can affect whether your material stays protected.
- 35:35If you paste confidential client information
- 35:37into a public consumer grade tool, you may be
- 35:41breaking confidentiality and putting privilege
- 35:43at risk. So the practical answer here is to favor
- 35:48a tool that offers real data protections when
- 35:52you're doing client work and to be thoughtful
- 35:55about what you are putting in it. You know there's
- 36:00new case law on AI coming out almost every week
- 36:03right now and the courts are not in agreement
- 36:07so it's very tricky. There are very few clear
- 36:11rules about almost anything in this space yet
- 36:14so it's a very new very quickly evolving space.
- 36:18We could probably talk on this topic you know
- 36:20this is a topic for a whole other panel and I'm
- 36:23not a lawyer. I'm not giving any legal advice,
- 36:26so I want to caveat that. You definitely have
- 36:28to do your own diligence here and look at your
- 36:32local bar requirements and things like that.
- 36:36But I think based on what we've been seeing just
- 36:39in the past few months that courts, you should
- 36:44be aware that courts have started treating what
- 36:47you type into these tools, the prompts, the input,
- 36:51the output. There have been instances of the
- 36:54courts treating that data as potentially discoverable.
- 36:58And they require those, they end up requiring
- 37:00those materials to be handed over during discovery.
- 37:04There's also been instances where courts have
- 37:06protected this information. So it's certainly
- 37:09not, it's not clear. But they're, you know, the
- 37:14safe assumption until it's all settled is that
- 37:17your AI conversations could end up being discoverable.
- 37:20So treat them the way you would treat an email
- 37:22that you might have to produce one day and you
- 37:26know just keep that instinct to be mindful of
- 37:30what you put in these tools and it might not
- 37:33always be private. So here I have on the right
- 37:38just a practical safe use checklist to keep in
- 37:44mind and I think these materials will be available
- 37:50so you don't have to jot anything down. But number
- 37:53one, you want to verify every citation and every
- 37:55fact before it leaves your desk, non -negotiable.
- 37:59You want to anonymize or simply leave out sensitive
- 38:03client details in your prompts, you know, if
- 38:05you don't need to put in anything about names
- 38:07or date of birth or any PII or anything that,
- 38:11you know, just take it out. It doesn't need to
- 38:14be in these prompts. That usually doesn't...
- 38:17It doesn't matter one way or the other. You want
- 38:21to check the tools that you're using, their data
- 38:23retention and training terms, before you trust
- 38:25it with anything nonpublic, really. You want
- 38:30to disclose your use of AI, where your rules
- 38:32and your client require it. I'm seeing more and
- 38:36more that the AI use is being requested to be
- 38:42made known in advance and discovery and things
- 38:45like that. People are, you know, courts are aware
- 38:49of it and more and more they're asking to talk
- 38:53about it upfront. So just know what the rules
- 38:57are when you're using it for your situation.
- 39:02And keep a human review on anything substantive.
- 39:06We call this human in the loop. So that's a good
- 39:10step to build into any AI workflow. to have a
- 39:14human set of eyes on something before it gets
- 39:17finalized or sent out. So that's it. Those five
- 39:20habits I think cover the vast majority of the
- 39:23risk that you'll find when using these tools.
- 39:30So assuming that you're using these tools carefully,
- 39:33let's talk about how to actually get good results
- 39:37out of them. because this is a skill all in of
- 39:45itself. So the six parts of a strong legal prompt.
- 39:50The quality of what you get out of an AI tool
- 39:52depends enormously on how you ask. I will say
- 39:58that the most recent chat models I've been using,
- 40:04It seems to require a little less. They seem
- 40:07to be a little bit more user friendly in the
- 40:09prompting area, where even if you don't enter
- 40:11like a super detailed, well -structured prompt,
- 40:15it will kind of figure things out and work its
- 40:18way around it. But this is kind of the gold standard
- 40:24in terms of how to structure a prompt and how
- 40:26to think about it. And it will really affect
- 40:29kind of your output. These are six parts that
- 40:35I tend to fall back on. They're not always necessary,
- 40:39but they typically will get you very good, better
- 40:44output than if you just start with something
- 40:46generic. So the first part is a role. You want
- 40:49to tell the tool who to be. You know, say you
- 40:53are a litigation associate who specializes in
- 40:57California employment law. That will produce
- 41:00a much sharper answer than not having any role
- 41:03at all. It shapes the tone, the depth, and how
- 41:07closely the output will stay in your area. It
- 41:10needs an objective. So, you know, using a precise
- 41:14verb, you can say draft, summarize, compare,
- 41:17analyze, outline. Those are all better than vague
- 41:21verbs like review or check, which can, you know,
- 41:25have the tool kind of guessing what you meant
- 41:28and its answer is going to reflect that guessing.
- 41:33Context, you want to give it good context. Give
- 41:35it a lot of context that is detailed. Give it
- 41:39the jurisdiction, relevant facts, posture, governing
- 41:43law, your client's goals. This is a really important
- 41:48one because the less context you give, the more
- 41:51the tool will fill in on its own. And that can
- 41:54be where some of this invented stuff can creep
- 41:57in. The format, you want to tell it the format
- 42:02that you want back, a table, a numbered list,
- 42:05a memo, something, a short summary that a client
- 42:09could read. This is a good habit to get into
- 42:14to make sure you aren't using unnecessary steps
- 42:17back and forth to clean things up and reformat
- 42:20things with the tool. um constraints if there's
- 42:23any requirements any boundaries you want to set
- 42:27those give it you know a word count or sources
- 42:29to include and exclude issues you want to prioritize
- 42:33um it's you know having limits will really help
- 42:38it understand what to do versus covering trying
- 42:42to cover everything very broadly um and iteration
- 42:46is really the key, you want to have it, you know,
- 42:53iterate, try multiple versions, tweak things,
- 42:56flag gaps or uncertainties or places where information
- 42:59would improve the answer. It turns it more from
- 43:03a single request into an actual conversation.
- 43:07And use the AI to help you sometimes. If you
- 43:10aren't sure how to best prompt it, you can ask
- 43:12it, what's the best way to prompt for this thing?
- 43:15And you can ask it, too, to ask questions. Ask
- 43:18me up to three questions if you need to clarify
- 43:23anything. So you can work with it to improve
- 43:26your prompt and help iterate things to a good
- 43:31outcome. Now, as I said, you don't need all six
- 43:35every time. A quick question might just need
- 43:39an objective, a little context, and a format.
- 43:42But for something bigger, like a memo, a brief,
- 43:44or anything where the stakes are a bit higher,
- 43:47you're going to want to work through the full
- 43:48list, and you'll get consistently stronger first
- 43:52draft to build from that way. So you understand
- 43:57the tools, the risks, and how to prompt. The
- 44:00natural next question is where do I actually
- 44:03start? So my advice for getting started is deliberately
- 44:09modest and low stakes. because I find that the
- 44:14firms that succeed almost never try to do everything
- 44:17at once. That's a recipe for overwhelm. So first,
- 44:22my advice is pick one low stakes task, choose
- 44:25a single lane one use, like summarizing or drafting
- 44:29routine emails, and just try it this week, one
- 44:33task. Second, when you move to anything involving
- 44:37real client work, use a paid tier that comes
- 44:40with data use protections. It's worth the cost
- 44:44the moment client information is involved. And
- 44:47that doesn't mean that, you know, you can willy
- 44:49-nilly paste whatever you want in these tools.
- 44:51There's a lot of different levels of data protections.
- 44:54There's things that, tools that still put data
- 44:57in their cloud. There's ways to keep it boxed
- 45:01in just to your own working environment. a bazillion
- 45:05different options out there. So that's a whole
- 45:06world in and of itself. But make sure you understand
- 45:09the tools they're using and their data use agreements.
- 45:16And so you understand clearly where that data
- 45:18is going before you put it in. That's really
- 45:20the point to drive home there. And third, you
- 45:24want to set a few simple ground rules for yourself.
- 45:27Write it down about what you will and will not
- 45:30put into these tools. And I think the thing to
- 45:37remember in the reframe that I think matters
- 45:40most is the lawyers at real risk here are not
- 45:46the ones experimenting carefully and verifying
- 45:50their work. The ones that are at risk are the
- 45:53ones that are using AI with no guardrails at
- 45:56all. So careful curiosity is the safe path. doing
- 46:01nothing or using it recklessly are the two real
- 46:05risks, in my opinion. So before we open it up,
- 46:10I want to spend just one minute on what this
- 46:13looks like when it actually works over time.
- 46:17So what does successful adoption look like? So
- 46:23knowing about AI, which is most of what we've
- 46:26done today, is honestly the easy part. The firms
- 46:30that actually benefit from it tend to share a
- 46:33handful of habits, and I want to leave you with
- 46:38a list of those habits that you can start thinking
- 46:40about, maybe using some of these on your own.
- 46:44So they start with one workflow, and they get
- 46:47it genuinely working before expanding. I think
- 46:54is really key. Start with one task, one goal.
- 46:58What is your goal that you want to achieve? What
- 47:02does that end state look like? Any sort of measurable
- 47:06things you can do around it. But start with that
- 47:10one thing. Don't try to tackle everything all
- 47:13at once and boil the ocean. The successful firms
- 47:18choose their tools deliberately. They match the
- 47:22tool to the tasks and the data protections to
- 47:25the sensitivity rather than defaulting to whatever
- 47:28is, you know, free or has the best marketing
- 47:32or something like that. And they put a simple
- 47:38policy in writing that I think is key as well
- 47:45to especially if you are more than one person
- 47:49in your firm. You want everyone to be on board
- 47:53and on the same page with how to use AI safely.
- 47:59So a short, clear policy on confidentiality,
- 48:02verification, and disclosure, that will turn
- 48:05this ad hoc experimenting into a safe, consistent
- 48:09practice across the firm. Firms that are successful,
- 48:16they will build repeatable workflows. So that's
- 48:19where you start to actually see a lot of the
- 48:22efficiency and this compound, you know, efficiency
- 48:26and interest building up over time, repeatable
- 48:31workflows. When something, a prompt works well,
- 48:34they save it as a template that the whole team
- 48:36can reuse, you know, instead of reinventing things
- 48:38every time. And measuring, they measure real
- 48:42time saved. So, you can make decisions based
- 48:46on actual results rather than just, you know,
- 48:49the hype. And it's amazing how few people do
- 48:52this. So having those metrics, having, okay,
- 48:55well, how much time does it take me right now
- 48:58to do this? And if I can get that off my plate,
- 49:01how much time do I save? You know, you can quickly
- 49:04quantify a lot of these things by starting to
- 49:09measure from the beginning what you're doing
- 49:11and training. they train their team and they
- 49:15revisit it as we all know this stuff is changing
- 49:18rapidly and it's not just a one and done you
- 49:22know you're trained and now you're AI literate
- 49:24for the rest of your life um no it needs to be
- 49:27revisited it needs to be an ongoing thing that
- 49:32you guys have to constantly get up to speed in
- 49:34we all do it's just it's changing so fast that
- 49:37unfortunately it's it's um it takes a lot of
- 49:41repeat time to keep on top of the training and
- 49:45what's new out there. So that is, I think, the
- 49:51end of what I had prepared. Thank you so much
- 49:57for spending time with me today. And I hope that
- 50:02this combination of caution and curiosity is
- 50:08a helpful way to think about approaching AI.
- 50:12And I have my contact information. If you ever
- 50:15want to reach out and talk more or have any questions,
- 50:18I'm happy to chat via email or if you want to
- 50:22LinkedIn, that's fine. Would love to connect
- 50:25and hear what you guys are using and any questions
- 50:27you might have. So yeah, let's open it up for
- 50:31questions. Right, Jessica. It looks like we have
- 50:37a question here. Are AI searches done with Lexis
- 50:41or similar programs still privileged attorney
- 50:44client? So I'm not qualified to answer that.
- 50:49I'm not an attorney, but I. There's it's just
- 50:56there's no, there's no barometer for this stuff
- 51:00yet. It's not clear what is. Considered privileged
- 51:04or not. You know, it's just there's there's very
- 51:07little out direction out there. Normally. I think
- 51:13I've seen maybe one instance in the news recently
- 51:16where privilege was attorney client privilege
- 51:19was waived for the use of AI. The judge ruled
- 51:23that it was waived by the use of AI where a lawyer
- 51:27I think used a third party chat AI chat. and
- 51:35use that to draft the document. And they said
- 51:38that that was waived because the privilege was
- 51:41waived because of their third party use in preparing
- 51:45a material. So that certainly doesn't mean that
- 51:51anything you type into AI means you're waiving
- 51:52privilege or it's now discoverable. It's a very
- 51:55nuanced area. What I have seen more on more case
- 52:01law about is or more decisions out there being
- 52:04made is on the defendant side, you know, using
- 52:09AI and that AI information being asked to be
- 52:15produced. So their chat transcripts with the
- 52:18AI being discoverable and that type of thing.
- 52:22But it's definitely a gray area and it's, I think,
- 52:27definitely warrants its own session, I would
- 52:28say. Great, thank you. It looks like we have
- 52:35another one here. Have you worked with mediation
- 52:37practices? Obviously confidentiality is the cornerstone
- 52:41of the process, the violation of which can end
- 52:43someone's practice. Tips, for example, using
- 52:45a tool to record a mediation to yield notes for
- 52:47the attorney is a major red flag for the mediation.
- 52:51Yeah, there's definitely scenarios that you cannot
- 52:54record. You know, you cannot record meetings
- 52:58and things like that if you are going to record
- 53:01anything, of course, needs to be disclosed up
- 53:03front. I think that's pretty common in every
- 53:06every state. But yeah, there's definitely cases
- 53:11where you can't just have an AI join your meeting
- 53:13and record everything. There's some. Tools that
- 53:17you can use to do it that a lot of the tools
- 53:19by default are recording things and putting them
- 53:22in a cloud somewhere and you want to be careful
- 53:23about that. There are some tools that do it more
- 53:27local, so it keeps it in your environment. It's
- 53:30not sending recordings and transcripts out into
- 53:33a third party hosted solution. So I think the
- 53:37key is to know. What meetings you know, know
- 53:41the know the rules of the road for me and this.
- 53:45Recording meetings is nothing new. I mean, Zoom
- 53:47has been able to do that for a long time. So
- 53:51it's kind of similar to that. The AI just adds
- 53:54an extra layer of usefulness to those recordings
- 53:57by preparing you the summary and the notes and
- 54:01the follow -up action items, that type of thing.
- 54:04So you have to be mindful of whether recording
- 54:06is appropriate and allowed. And then... There
- 54:10are tools that you can use locally versus in
- 54:13the cloud. Of course, there's usually you pay
- 54:17a bit more for them, but they are better for
- 54:20certain types of confidential work. I hope that
- 54:27answered your question. Yeah, it looks like we
- 54:30have another one. Great presentation. Thank you.
- 54:32Do you know of any database or list of legal
- 54:34tools available on the Internet? Ooh, I don't.
- 54:39I'm sure they are. I can't, I haven't vetted
- 54:42them, so I wouldn't want to recommend anything.
- 54:45If you just start searching, I know there's people
- 54:48on LinkedIn that post about this stuff all the
- 54:51time. There are probably people in your... you
- 54:56know, in your local area that are keeping lists
- 54:59of this type of stuff. If you just start like
- 55:01looking at LinkedIn and looking online for some
- 55:04of these things, I bet you can find people that
- 55:06are tracking some of this stuff. I personally
- 55:09don't. And I think it's very overwhelming to
- 55:14start at that broad list level of here's all
- 55:18the different AI, here's all the different tech,
- 55:21legal tech solutions out there. That's a massive
- 55:24undertaking to try to even understand that where
- 55:27I like to have people focus is let's look at
- 55:30the problem. Let's look at what you do day in
- 55:32and day out where there is for improvement where
- 55:35there you know, what's the low -hanging fruit
- 55:37of things that could be automated or AI can come
- 55:41and help out and focus on point of one problem.
- 55:44Then you can look at the solutions just for that
- 55:47specific problem and that will automatically
- 55:49narrow it down to a list of five to 10 things
- 55:53versus hundreds. I don't have a good resource
- 56:00for a database of all the legal AI tools out
- 56:03there, unfortunately. There's a lot of them.
- 56:08Thank you. And it looks like I have another question
- 56:11here and creating a billing template. Is it enough
- 56:13to delete the names of the client from which
- 56:15the template is drawn and deleting any identifiable
- 56:19matter discussion? Yeah, I think so. If you're
- 56:28using a template in your own system. Like if
- 56:32you're using Microsoft Copilot, for example,
- 56:34and you have it set up so it stays within your,
- 56:38there's different tiers, there's different licensing
- 56:40and whatever. If you have it set up so that it's
- 56:43within your own network, then you can use it
- 56:46pretty safely with your, you know, to say, here's
- 56:50a template. Enter client name, enter company
- 56:53name, and it'll customize it and send it out
- 56:56the way you need it with all those details in
- 57:00there. If you don't have a system that's working
- 57:02just within your own private network, then yes,
- 57:06you're going to want to anonymize any of that.
- 57:09You can put in a placeholder for... Names and
- 57:14business names and just fill those in yourself
- 57:16when you before you send them. That's what I
- 57:19would recommend, but yeah, it really depends
- 57:22on. Whether you're using a closed system or something.
- 57:26Out there on the cloud and how you handle that.
- 57:34Hey, let's see someone's also mentioning billing.
- 57:39I'm experimenting to use AI for billing to take
- 57:41my emails. Uh, to convert to billing entries,
- 57:44do you have a suggestion? Convert my emails,
- 57:51like, so it kind of. The emails that you do is
- 57:55a work, I'm guessing send and receive throughout
- 57:57the day and then summer and then puts together
- 57:59like a narrative for your. For your time entry.
- 58:07This is kind of how I'm reading that. So yeah,
- 58:09I think that's a great use case. If you're using
- 58:14Microsoft Office and Outlook and you have Copilot,
- 58:17it can scan your inbox and tell you what you
- 58:19sent and give you kind of a summary based on
- 58:22matter type and things. It's very good at that.
- 58:25It can give you like a list of tasks and you
- 58:27can train it on the type of format that you want.
- 58:31And you can use that as like a repeatable a repeatable
- 58:34thing it does, you know, every day or every week
- 58:37and give you kind of a report like that. So yeah,
- 58:40I think that's a great, pretty simple use case.
- 58:43Again, if it's making sure it's within your environment
- 58:47and you aren't sending, you know, you aren't
- 58:51giving access to your internal email to a third
- 58:53party cloud system with very little data protection.
- 59:01Great, thank you. And then let's see. Another
- 59:04question in your presentation, you emphasize
- 59:06human involvement in terms to setting standards,
- 59:09verification and quality control. It seems that
- 59:11these products are promoted as a form of substitution
- 59:15for humans, not augmentation of humans. Is that
- 59:17your experience? Sometimes? Yeah, I've definitely
- 59:21talked to people that are promoting and selling
- 59:23tools and marketing things, and they kind of,
- 59:27you know, position it as this easy button where
- 59:29you just click the button and it all gets done
- 59:31for you. And that is, I think, very dangerous
- 59:36and very misleading. I mean, you can click a
- 59:39button, but the result that you get is rarely
- 59:41going to be something that is very useful. So
- 59:47in everything, it needs to be iterated upon.
- 59:50You need to train it. You need to work with it.
- 59:52And eventually, you might be able to get it to
- 59:54a point where there is this easy button. But
- 59:56you always need to verify. sample the output
- 1:00:02you know test make sure yeah it's it's like working
- 1:00:05with another person you know a junior associate
- 1:00:08if you will you know you need to you can give
- 1:00:10them all a good input but you're gonna have to
- 1:00:12QC their work so yeah I don't think it's at a
- 1:00:15system yet and I don't know that it ever will
- 1:00:17be where it fully replaces everything we do and
- 1:00:20can do it autonomously I think will always have
- 1:00:23to be Giving it the instructions, iterating it,
- 1:00:28training it and verifying at the end of the day.
- 1:00:34Great it looks like we, I know we're at time,
- 1:00:36but we have 2 questions left. Would you like
- 1:00:38to tackle those Jessica? Sure. All right, so
- 1:00:42first question being I've heard that the world
- 1:00:45will change soon and become expensive. Not sure
- 1:00:48why, but that is the route as to why the Mac
- 1:00:50studios are sold out all over the US businesses
- 1:00:53are getting ready to keep their own intellectual
- 1:00:55property. Do you know or understand what's going
- 1:00:57on? I'm not too sure. Yeah, that's an interesting
- 1:01:01one. There's a few things that I think are playing
- 1:01:04into this sudden recent feeling like it's. becoming
- 1:01:10very expensive. One is the token costs have gone
- 1:01:14up a little bit. So a lot of the tools, so tokens
- 1:01:18are just kind of the context, right? Like the
- 1:01:20amount of text you're basically putting in and
- 1:01:24you get charged. Like if you start using these
- 1:01:26tools, if you notice you'll get, like if you
- 1:01:28use the free tier, if you use the $20 a month
- 1:01:30tier, you are starting to get. hit depending
- 1:01:33on what tool you're using. Like, I use Claude,
- 1:01:36for example, a lot. And Claude, this past month,
- 1:01:39started putting harder limits on its tiers of
- 1:01:43data use really. So people using it on a day
- 1:01:47-to -day basis were hitting their limit and saying
- 1:01:49you've used all of your allotted data for the
- 1:01:51day, you know, check back in five hours to finish
- 1:01:54using it or pay a la carte. So Claude, I think
- 1:01:58a lot of people started using Claude this past
- 1:02:00year and they started specifically bringing in
- 1:02:05their data limits a bit harder with people and
- 1:02:09So I think that's one contribution. We've seen
- 1:02:13some stories in the news of companies that gave
- 1:02:16AI to all their employees. And then, you know,
- 1:02:18six months later got hit with like multi -million
- 1:02:20dollar bills for all of their AI usage. So I
- 1:02:24think that's been in the media a little bit more.
- 1:02:27And then, yeah, there's the Mac, the Mac, the
- 1:02:31Mac servers, the Mac minis that are, that's kind
- 1:02:34of a whole other level of autonomous AI. um open
- 1:02:40claw and these other systems that really aren't
- 1:02:43used in the legal world but they're kind of more
- 1:02:45of autonomous agents where you can just they
- 1:02:48run on their on your computer and they need powerful
- 1:02:51computers so people are buying mac minis to run
- 1:02:54these autonomous ai systems um and the mac minis
- 1:02:59are sold out everywhere because of it um so that's
- 1:03:01another kind of aspect i think it's all converging
- 1:03:03to Bring to mind that yeah, the stuff is not
- 1:03:07free. It's far from it and a lot of tools are
- 1:03:11being given out for free to get people to use
- 1:03:14them. But the people are starting to have to
- 1:03:16pay for them and it's something to keep in mind
- 1:03:18that. You might not be thinking of upfront is
- 1:03:22the cost down the road for implementing some
- 1:03:24of this stuff. So a good point. All right, and
- 1:03:29I have another question. Do you have to. Do you
- 1:03:32have a guide to have emails summarized daily?
- 1:03:39Not a guide. It's really, we would just really
- 1:03:42be a prompt for your, it depends on what you
- 1:03:46have, you know, the systems that you're using
- 1:03:48and the AI solutions that you're using. If you
- 1:03:52have one that's integrated that you use that
- 1:03:54can access your inbox, it's a simple prompt of,
- 1:03:57you know, summarize my email. From the past day
- 1:04:00or the past week or organize it triage it based
- 1:04:04on things that you know, I need to reply to things
- 1:04:07that are urgent, you know, or give it give it
- 1:04:09the custom instructions. Pretend it's a person,
- 1:04:11you know, like a new intern that you got is my
- 1:04:13advice. If you have a new intern that just showed
- 1:04:16up at your office today and you wanted it to
- 1:04:18do this task, what would you tell? How would
- 1:04:20you, you know, what instructions would you give
- 1:04:22that person? So just. you know, start with something
- 1:04:25like that and then fine tune it based on the
- 1:04:27results you're getting. And if you have a system
- 1:04:30that can set up scheduling scheduled tasks, depending
- 1:04:34on what AI you use, it can, you know, once you
- 1:04:37get it fine tuned, you can set it up to be a
- 1:04:39daily task that does every morning at 8am and
- 1:04:43sends you a summary and kind of takes that off
- 1:04:46your plate. So that's another cool trick to do.
- 1:04:49Once you start doing these daily administrative
- 1:04:52things, it's If you can schedule them to run
- 1:04:55automatically, that can save you some time too.
- 1:05:03It looks like we did get another question right
- 1:05:06now. Jessica, do you have the time to answer?
- 1:05:09Yeah, I've got time. All right. What type of
- 1:05:12guardrails are in place to prevent the hacking
- 1:05:14of law firm AI that we have seen with law firm
- 1:05:16contacts or other files? That is an evolving
- 1:05:26area, very evolving. There have been some known
- 1:05:31hacks. I think a lot of these systems, again,
- 1:05:36it's gonna depend on the tools you have, the
- 1:05:38access they have, your own internal IT setup.
- 1:05:42So I really can't get into any specifics and
- 1:05:45I'm not an IT expert, but there are definitely
- 1:05:53risks out there. There are kind of, we've seen
- 1:05:58some backdoor ways of using AI nefariously to
- 1:06:02access some of this stuff. So I think the IT
- 1:06:08issues are being If they're found to be, if there's
- 1:06:13found to be an issue, the systems seem to be
- 1:06:16pretty quick to patch them and, you know, fix
- 1:06:19them and fix any holes or openings in their own
- 1:06:23internal setups. And then it's definitely something
- 1:06:27that your own IT, if you've got an IT person
- 1:06:30or, you know, depending on the size of your firm,
- 1:06:32if you have someone that that's something they
- 1:06:34should be looking into. But I think if you're
- 1:06:38using just kind of off the shelf products, You
- 1:06:41just want to be careful about the data that you're
- 1:06:42putting into them. If you are putting, you know,
- 1:06:45if they're hosting data and you are uploading
- 1:06:48documents into it, then that it's like any third
- 1:06:52party software, right? You need to verify how
- 1:06:56that third party is handling. their security
- 1:06:59and protecting your data. Ideally, you want to
- 1:07:03avoid putting data in their system if it's not
- 1:07:06going to be kept on their environment that you
- 1:07:08don't necessarily have control over. That's a
- 1:07:11very nuanced area. It depends a lot on your own
- 1:07:18setup and what tools you're using. The more you
- 1:07:23can keep the confidential information and the,
- 1:07:26you know, use the client information and all
- 1:07:28of that stuff out of these systems, I think the
- 1:07:31better it is unless you're in a walled -in environment
- 1:07:34with a reputable provider that's within your
- 1:07:37network and it's not being shared out on the
- 1:07:39cloud at all. All righty, that's it for questions
- 1:07:45today. Thank you so much, Jessica, for a fantastic
- 1:07:47presentation today. And thank you so much to
- 1:07:50our audience for joining us. Have a great day,
- 1:07:52everybody. Thank you, everyone.