56min · about 9,800 wordsDownload Transcript
0:00Foreign.
0:13Welcome everyone. I see we're still getting signed on. There's tons of people in this class today. That is quite all right. We'll get everyone a chance to get signed on and get settled here and just go over a few houses before we start here today. First housekeeping item is when do you get that CLE code for your CLE credit? Your CLE code comes at the end of today's seminar. Within the last 10 to 15 minutes of the seminar you will get the code auto populated into the chat feature. So the chat box here on Zoom Lexvig comes on that lasts 10 to 15 minutes and populate your CLE code into there. In addition to that, Lexvid puts instructions of what to do with that CLE code. Once you do get it, you will need to write it down on a separate piece of paper or take a screenshot of it or you know, take a picture with your iPhone.
1:13I have that because after the seminar concludes you will exit out of today's seminar, enter into your individual lexfed account, open up the seminars and there should be a box there for you to put that code next to today's seminar. You will hit enter. Your attendance will be submitted back to LexFID. LexVid will then report it to your respective jurisdictions. Now Please note that LexVid has 30 days to get your attendance here today reported to your respective jurisdictions and but it's not an instantaneous process so keep that in mind if you are trying to meet your deadline for the end of December. None of us are here doing that now are we? We're all we're not procrastinating at all. Not in the least.
2:01If you are trying to meet that December deadline, do not fear if you attend here today you submit your attendance. Just because you don't see it on your audit report with your jurisdiction before the end of the month doesn't mean you won't get credit for it. If you complete it within December before your deadline, you will get credit for it. When it does get populated onto your report with your respective jurisdiction. Just know there is that little bit of time lapse just because of how many days LexBid has to return back to your jurisdiction. If you do not see it on your jurisdictional report within the next 30 to 45 days then please do reach out to LEXD or your individual jurisdictions to figure out where it's at in the process to make sure we do get your attendance recorded at the seminar. And don't worry either. LexVid records this and records the attendance of everyone here today. And how long you stay in the seminar so they can pull that right back up and make sure it gets reported if for some reason there is a hiccup in the system. Aside from that, if you have any additional questions, please do not hesitate to ask those questions. You can put the questions in the chat feature and I will try to get the get to them either simultaneously with the presentation or I'll get to them when the presentation concludes or answer them independently after through email with Lex and you the person submitting a question. So if you have a question, please do not hesitate to ask it. Just put it in that chat feature so we can
3:35make sure it gets answered with that. If you have not taken a seminar from me before, My name is Carrie Sheehan, Assistant General Counsel of Taft, Dentonius and Hollister law firm. I'm in the Indianapolis, Indiana location here in the Midwest and in that role I do anything ethics compliance and internal ethics compliance for TAS and all of its offices and lawyers. So I deal with ethics compliance 24 hours a day, seven days a week in my role as Assistant General Counsel. One of the areas in which I particularly enjoy and I work on our AI committee within our firm. I'm also on the Indiana Task Force Committee for Generative AI.
4:18In the state of Generative AI, you can say this is kind of one of my fun little research things that I do that I really looked into AI and the different platforms that attorneys use. And so today here we are going to review some of this stuff. If you've taken other Gen AI seminars, going to review some of the ethics, some of the compliance, talk about some new cases that have come out, look at a few new cases that have come out as well, and then talk about basically how do we choose a platform, what platform is going to be best for us and what are some of the functionalities of these platforms that can help you in the legal profession and in your job. Then if we have time too, where you're going to bounce out to do a live demonstration.
5:04So when I say bounce out, I mean bounce out of the PowerPoint and go to the Internet and do a live demonstration of how you can create your own Genai agent, meaning your own little chatgpt within Big chatgpt or if you have Perplexity or Claude, it's the same process and how you can use your little agents to specialize in different areas to help you in your legal practice. We're going to try to do some live hands on demonstrations as well as go over all the risks and concerns with generative AI, case law we've seen out there and a lot of stuff all in this hour. So please bear with me if I talk fast.
5:47It's just so we can get it all in hopefully and get it in in a good period of time. So we're not going over, but we're also not going short on our time either. So you get all of your credit here today with that. Let's just kind of talk about if you've never had a Genai seminar this will be for you. Or if you have had one and just don't know the differences of generative AI, this is for you too. Just so we're all on the same page with our definitions and what terms mean. There is actually three kinds of AI. There's regular artificial intelligence. This is kind of our virtual assistant speech recognition, you know, spelling, grammar checks, predictive text, or you know, the fill in the blank, you know, when it autocorrects. Those things are artificial intelligence. It's also can be when you call the doctor's office, it says press 1 or say what you want and then it says I don't understand, I have to hang up now.
6:46Because it's one input in for one input out. If you don't get exactly the right input, it doesn't know what to do. That's the old artificial intelligence or the general artificial intelligence we've been so used to seeing, even back to the 1950s, 1960s and on forward now. In 2022 we had the emergence of our generative AI, really hit the market when ChatGPT really went national and came out with, you know, the features that it has. The generative AI artificial intelligence or gen AI as it's commonly known, it produces new text, images, audio and content that makes it think or feel like it's a human being. It can pivot, it can take an open ended question and answer it just like a human being could. But the thing we have to Remember with generative AI and these large language models, those are your ChatGPTs, your Google, Gemini, your Claude, your Perplexity.
7:51They're known as large language models. The thing we have to remember with them is that they are pre trained transformers. It's a machine, that's all it is. We should never mistake it for a human being. We should never mistake it for anything other than what it is, which is a pre trained transformer. It is a machine that is programmed to pull data from a certain data pool that you make a big swimming pool of all this data floating around. And then this Pre trained transformer is programmed for, when you ask it a question, to search the swimming pool for connections for same words, for similarities between bath and what data is floating in the swimming pool.
8:40Then it pulls that data out and gives you an answer, whether it's right or wrong. Because our pre trained transformers are programmed to give you an answer. They are programmed to give you some response, even if it's the wrong response. We always have to think of these as machines. We should never cross over into thinking of a generative AI as a human being because it's not. It's not someone who can draw on past experiences. It's not something that can critically think or have critical analysis. It is only as good as that swimming pool of data that is programmed to pull from. And each generative AI model is programmed to pull from a different subset of data. Some overlap, some doesn't, but is programmed to pull.
9:38Each platform can give you a different result. Each platform can have its own hallucinations or lies or false information coming out of it. And each platform is different or better for different things. It's not a one size fits all. It's kind of like humans in that sense too. Because if each platform is different, it's just like each human being is different. Y' all have different experiences, different knowledge, different gifts, different talents. Generative AI, it's a pre trained transformer from a swimming pool of information just trying to make logistical connections, give you a response.
10:21That's also why it's self learning, self training. Because say for instance, I prompted to say something, prompted with a question saying write me a motion for summary judgment. It gives me a response back and I tell it no, this is wrong. This is a plaintiff face or this is xyz, I need you to change this and this around. Well, most large language models are self learning. So now it's getting smarter. It's using my input to make new connections to give an output that it thinks is going to be closer to what I'm looking for. And then the next time maybe an unknown third party or someone else in the world asks the same generative AI the same or similar question that I did, it's going to have learned from me prompting it and re prompting it to maybe give that person a better answer initially than it gave me because now my inputs are part of its swimming pool. And then it can use my inputs as an output to an unknown third party who asks the same or similar question.
11:33Generative AI is it's making connections from a data subset and then giving out those answers or giving out the connections that's made to us. The end user saying this is what I've decided or this is what I found. Now you can start to see when you break it down like that how many things can go wrong in here, how we get our false responses, how we can get inaccurate information and different things. And there's so many more risks too, which we'll talk about here in just a second. That's how generative AI, those large language models that are all the craze right now really work and down to the simplistic format of where they're getting information from. Now there's also agentic A AI.
12:21The last thing on your screen agentic AI is think about it as the future is what we're striving towards. Those self driving cars, the AI agents and assistants, the workflow automation kind of replacing your assistant type of information here. That is those proactive, you know, goal setting multitask to take what some human beings do, those very simple administrative functions and now giving them to a machine to complete. That's more our agentic AI which is a developing area. But as for our purposes as lawyers, it's really this generative AI, this individual section that is going to be the most concern and what everyone is worried about when it comes to figuring out is it going to replace this?
13:11How does this affect the legal profession? Now some other definitions here too as we go through hallucinations. If you've ever heard that term that means lies, false information, it means a generative AI product gave you false information, a hallucination, it was tripping, if you want to think of it that way and made up stuff and gave it to you. They can also produce these generative AI models bias or discriminatory content. Think of that swimming pool. Now think where that data in that swimming pool came from. It's not all from user input, it's from the Internet. The Internet is not unbiased. The Internet is not non discriminatory. It is probably the most biased and discriminatory place in the world is the Internet. And this is where these gen AI models are pulling from. That's the swimming pool. That's 75% of the swimming pool data is from that. You can also get biased discriminatory content just like any technology too. They have the risk of confidentiality breaches. The platform itself you're using could be breached.
14:27That self learning, self training can also be breached because again anything you input into them, it produces out or could produce out to a third party if it's self learning and self training. That's that last part where it takes our inputs, stores them, teaches itself off of them, then uses them to answer other people's questions. Any gen AI model that self learns self trained is not safe for substantive legal work. That means if you are putting your client's name in there, if you are putting clients confidential information in there, if you're doing any substantive type of legal work with your client's information details included in it, then you shouldn't be using one that self learns and self trained because you have no control over the data, you have no control over what it does with it, and you have no control over who it sends it out to once it's in its swimming pool. That is a breach, as we'll see in the next slide of Rule 1.6 confidentiality.
15:36So any substantive legal work where you're inputting client confidential information, which is so huge, including that client's name, including anything on the public docket, anything relating to the client is confidential under Rule 1.6. If you are putting 1.6 information into your generative AI model, you're breaching Rule 1.6 and you should not be using it for substantive legal work. Now you may think, okay, great, then what's this seminar about? Because that's all the platforms, that's perplexity, that's Claude, that's ChatGPT. Well, we will learn there are versions of these platforms that come at a nice price per user, per email, called enterprise or closed circuit models of our large language models where basically you purchase a firm subscription to the platform and then you're in the four walls of your firm. Now when I say within the four walls of your firm, your subscription that you purchase and it's not the one you go online and pay 20 bucks a month for, it's more like $450 a user. So if you think you have the version for 20 bucks a month, you may want to think again. This is one where you have to call the platform, call the person who owns the platform and get a firm subscription. Now what the firm subscription does is it still has the swimming pool of information that it pulls from. So all the same places it pulls from. But anything you input into does not store, it does not self learn, and it does not give it out past the four walls of your law firm, meaning it will not go out to unknown third parties.
17:27It's kind of like a client file. It's just going to stay floating around in the four walls of your firm. This protects confidentiality this allows us to use these generative AI large language models for client substantive work. But you have to have the enterprise or subscription model that guarantees that it's not self learning, self training, it's not storing your information and it's staying within the four walls of your law firm. Otherwise we'll see in the next slide you're breaching confidentiality. Something to think about that self learning, self training is the key. That definition of what that means is this key for ethics and compliance that we're going to learn about. Some other things to think about here too though, before we get to that ethics and compliance in full. These large language models do not recognize copyright, patent, trademark, and they don't tell you sometimes where the information comes from. Or if they do tell you where it comes from, it could be false citations or false information. So we have to be very careful with the outputs we get from these along with the confidentiality aspect of these things, because that's what's really going to hurt its lawyers now, the outputs of these. If you think generative AI is going to replace human beings, replace lawyers, replace your first year associate, replace us in the legal profession, you're wrong. And that is because anything that comes out of a generative AI model, even those enterprise versions that I was talking about, cannot be trusted.
19:11It has to be double checked against reliable sources like Westlaw, LexisNexis to make sure if it says case law, the cases are correctly named, the citations match up with the case name, one of them is not false, or you know, that hallucination is never going to replace critical thinking, is never going to replace intuition and it's never going to replace kind of that experience that we have as lawyers. Meaning maybe you've been in front of a judge 10 times and you know, at certain times of the month or you know, at a certain date of the year, the judge has a personal thing going on and you know, it's just a bad time to get a ruling. Gen AI model won't know that they're not going to have that intuition, that human insight into things.
20:02So it's never going to replace that. And then we always have to have that human oversight, that double checking because again, it's a machine, it's a pre trained transformer that produces false results. No one false results. 450 lawyers used it to their detriment, turned it into briefings in the court and been sanctioned for false results, but they didn't Double check or 450 lawyers across the country and those are only the Ones that have been reported, that are known, there's probably way more than that. We have to think about these things when we are using generative AI. This comes in with our ethical rules, competency rule 1.1. You have to be knowledgeable in the things, including technology that touches their concerns, the practice of law, our clients and our cases.
20:58Generative AI is one of those things. It doesn't mean you have to know how it's coded, it doesn't mean you have to know every single detail of every single platform. But that basic explanation I just gave you of how it pulls, what self learning means, how confidentiality is affected, those bare bone basics are what you need to know because that's what's going to affect your client, affect your law practice. So if you're one of those that think you would retire before this really hit, you should have retired in 2022 if that was the case. Because this is here and it's moving forward and we are going to have to learn it. Even if you never use it, you need to learn it. Because nowadays you're not going to come up against the Google client. When I say that, I mean the client who Googles every legal question they have before they come see you and then you give them a legal response and they say, well that's not what Google said. Now it's the. That's not what ChatGPT told me. We have our ChatGPT lawyers and our ChatGPT process and so we have to be knowledgeable and competent because this affects our clients. Our clients are probably using it if you represent companies or individuals or anyone.
22:13So it's going to affect our law practice. Biggest thing there is too kind of skipping to that rule 1.6 in the upper right hand corner of your screen, that confidentiality, like I was saying, they don't protect confidentiality unless you have an enterprise version, a paid firm subscription. Just like you would go out and get a Nexus subscription or Westlaw subscription. You know, you get a representative, you get one for your firm, you get user accounts on your firm email address, you're not protected. If it's self learning, self training, you're not protected. You should never use it for legal work, substance of legal work, or put any confidential client information. Again, that confidential information is very broad. Anything that touches and concerns the representation, no matter the source, is the definition of confidentiality, which includes anything on the public docket, anything you learn about the client or the client's case, or anything ancillary to that, the client's name. That's all confidential potential. A lot of you will say, yeah, but it's not privilege. It doesn't matter. We can't release confidential information.
23:25Everything concerning the client is confidential. We cannot give it over to a third party without their consent. The attorney client privilege, those communications directly between the attorney and the client is a subsection of confidentiality and a privilege argument for when counsel is trying to get out of conversations with the client. Protecting that basic privilege, the confidentiality is everything guarding your representation, no matter the source. And the rules say that it's that everything we have to protect and that we can't give over to third parties without informed consent from our clients. And unless you communicate with the clients for 1.4 and they say yes, you can use these public models.
24:11Put my private information in there, put my case information in there. I don't care if it goes out to third parties, we can't be using them. You either need to get client permission or get an enterprise version to protect that confidentiality. They're your options. You can use the free public, the ones that are $20 a month, but get your client permission before you put client information in there because that's the key to consent or disclosure of confidential information. The client holds the key confidentiality. If the client consents, then you can do it read ABA formal opinion 512. It goes through what should be in that consent. When you get it through from your client, that consent needs to be informed, describing the risks and benefits of using the generative AI model and how that's going to affect the client's data and the risk associated with that. Again, that goes back to competency too. But we have to communicate that with the client and get the informed consent because it's affecting confidential information. Either get the enterprise versions that can protect confidentiality or have that hard conversation with the client and ask tell them you want to use ChatGPT on their case. My response as a client would be, well then why am I paying you? I can go out and use the free myself. Why am I paying you to do it again?
25:37You have to think about that communication, what it's going to look like on top of that too. Fees if you're an hourly attorney. ChatGPT, these large language models are not your best friend. Even if you have the enterprise version. They're not your best friend until you switch kind of your billing model. That's because say these large language models can do in 30 seconds to a minute what would have taken you three hours. You just cut your own billable hour down because you can only charge for the time you actually spend on something, not the time you would have spent absent the technology. So if ChatGPT is doing it in 30 seconds or a minute, it takes you within an hour to review it.
26:25That human oversight. So now you're at an hour and maybe six minutes. So 1.1 on the billable hour when that would have taken you normally four to five hours. You just cut your own time because you can only charge the time you actually spend, not the time you would have spent. That technology. We got to think about that. Maybe it's time to go to more alternative fee arrangements, flat fee value based billing, those different things. Work with your departments and firm. What would be best for that. We also have to remember too that if it does produce those false results or hallucinations, we can violate all of our truthfulness rules, our 3.3, 3.4, 4.1, 7.1, anything regarding truthfulness and candor to the tribunal, third parties and advertisements.
27:15If we use it to help us with our advertising, use false results out to that third party and we violated a truthfulness rule, it's no defense to blame your law clerk, it's no defense to blame the first year associate, it's no defense to blame the pre trained transformer. And you having no knowledge that it could produce false results because it's a competency tells you you have to have knowledge, you have to know that these produce a false result. You have to know you have to double check and triple check everything it gives you. You can't just take it, print it and send it off. Otherwise you violate the candor rule, the competency rule and maybe confidentiality rule, who knows, depending on what platform you're using.
28:03The other thing with this too is when I say you can't blame the law clerk, you can't blame your assistant, you can't blame your paralegal because under rules 5.1 and 5.3, if it's the associate, if it's the law clerk, if it's your non attorney assistant one under your direct supervision, they violate the rule, you violate the rules, your duty to supervise them, your duty to train them to make sure they comply with the rules of professional conduct. So claiming down the line of liability is not going to help you in a disciplinary complaint and it's not going to help you with sanctions up in front of the judge. You need to take ownership of it and know that you have to be competent, you have to do the human oversight.
28:50Do not rely on your associates or law Clerks, if they don't know how to use it, you may think, well, it's supposed to be this great machine, this great thing. Again, it's a machine. You should think of that machine as a machine. And if you can't conceptualize that, think of ChatGPT or Claude or Perplexity, Google, Gemini. A first year associate. Would you trust the product of a first year associate on their first day at the firm? Would you blindly take what they wrote you and turn it into the court? The answer to that is no. Why are we blindly taking stuff from a machine and turning it into the court when we know it has proven to have false results, it's proven to not be accurate.
29:38Again, thank you. Good. As the first year associate, that's easier to get your mind around. Know that it can produce those false results and the blame game doesn't work. You are ultimately responsible. If you sign it, turn it in, you're responsible for the crime. You are the one violating the rules. Keep that in mind. Those biased or discriminatory results that can come from it. That's where we draw our 8.4 D as in dog and 8.4 G violation. Those rules go to prejudicial to the administration of justice, kitchen sink rule. Kind of throw everything in it and see what sticks in that rule. That's 8.4D then 8.4G. Bias and discrimination.
30:21It produces biased content. If it produces discriminatory content, you take it and turn around. You can violate 8.4g as in girls, the anti bias, anti discrimination rule while practicing law. Again, there are multiple rules here that could be violated. If we don't get confident in generative AI. If you think it's not affecting your practice, it's affecting your practice. I almost guarantee you you have at least one client using it. You may not know it, they may not tell you, but I guarantee you have at least one client using it. I guarantee you someone on your staff has used it or one of your attorneys has used it, maybe without your knowledge. It is here and we need to get competent and we need to have that conversation about it because too many lawyers are getting in trouble for it. Too many lawyers are getting sanctioned and the case law is starting to come out, the sanctions are starting to come down.
31:19The highest sanction so far in the United States, and I say the United States because there's been one in Australia that's been higher, is $30,000, $30,000 sanctioned for turning in false results to a court based on what they got from generative AI model. It is Getting real. It is getting serious. We're going to start seeing more disciplinary complaints coming down as they work their way through the disciplinary commission. We've got to start learning it. If you want to use it, get that enterprise version like I talked about that enterprise safe version versus the public gen AI models. Public models don't keep our confidences, violate the rules, self learn self trained and can send your inputs out to unknown third party enterprise version and keep it within the four walls of your firm. The most or the most used are Perplexity, Cloud, CoPilot, Westlaw and LexisNexis have their own AI now within their databases.
32:24You do have to have the highest subscription to get Westlaw AI or LexisNexis protege, but they actually have AI within the search tools now where it can give you. You type in what you want just like you would into a large language model and it gives you a summary of cases. That's a starting point. Even these models, you should not take that stuff verbatim, copy it over as a quote and turn into the court. They're summaries of the cases they found. Again, a starting point. I will say that I have the luxury, the privilege to have both Westlaw AI and LexisNexis protege. Even when you put the same input into both of them, outputs are not the same.
33:08Each site similar cases, but also different cases. That's because sometimes some cases in LexisNexis are in Westlaw or vice versa. Or maybe one engine made a connection that the other engine did not. If you do have the luxury of having both running the same identical query or prompt through both of them to compare the results, because they're both going to be slightly different or very different depending on your prompt. And that's the same with any Genai model, not just Westlaw or LexisNexis. But if I put the same prompt into Claude and Perplexity, I get two different results. And that's because they have two different swimming pools, two different pre trained transformers programmed by two different people and they draw connections differently which gives out different results.
34:04Even if I change one word in my prompt, just one word, I can get a different result than I did the first time. These are not perfect machine. These are not the end. All the glory Hallelujah. We have to do no more work in the legal profession. These are things that can aid us as the starting point but not the ending point. So we have to think of them as that first year associate, that law clerk, the pre trained transformer. If you want to think of it like that too, that is not a one size fits all and that each of these are different. Even if you prompt it the same way, you get varied results through all of them. So how do you pick which is the most appropriate for you? Well, first you have to figure out what information you're going to be putting into it. It's client confidential information.
34:53We need an enterprise version. If you're just doing generic stuff like draft me a set of plaintiffs and derogatories, we can use a free public version because there's no client confidential information. The key, the key main consideration is is their client confidential information. If there is, you can't use the public versions that self learn, self trained and give out your information to third parties. You can only use generic stuff in those. If there is client confidential information, we need an enterprise version. Then you have to figure out, okay, which platform do I want to purchase with the enterprise version? There's different ones.
35:34Copilot, Plexity, Blog Chat, GPT, they all have enterprise versions. Which one's best you have to look at too. Okay, which one fits? Where are they pulling from? This gets you down into analyzing their terms and conditions, how they store their data, how they authenticate their data, what they do with their data, and managing those things. Again, we're going to have to do some research here. We're also going to have to look at, you know, making sure that it does protect that client confidentiality. So we don't have to get that consent or we still get the consent if we're, you know, want to make sure we cover all of our dot, our I's and cross our T's. If we're using any platform that risks the disclosure, like those large language models, the public ones, we need the client consent if client confidential information is going into it. You should also consider the cost associated with each tool, pricing structure and what your firm can afford.
36:32And then two, what's appropriate for effective use? Say for instance, maybe your firm wants people to use it, but they want people to know how to use it. So instead of just blindly passing out passwords and user logins like we do with LexisNexis and Westlaw, sometimes maybe forcing people to go through AI training in your firm before they can have access to these models, before they get that enterprise safe version of it for use in client services. So again, internal trainings, internal management of the appropriate tools, having teams, having maybe a policy go a long way as well with it too. Let's look here at a few cases like I said there's over 400 and some cases out there regarding generative AI and people getting in trouble.
37:25And I am going to switch screens here real quick just so I can show you. This is actually this AI Hallucination in Red is actually a website and the website is right here on the screen in front of you, which I'm just going to pull it up live here. So I'm going to stop sharing for just a second. Switch screens over. Let me find the right one. This one I shared the wrong screen. I apologize. So this is her live Internet version of the Hallucination website. So you can see here, there is 468 cases identified so far. You can see it right there on the screen in front of you as being fake cases or hallucinations. This website was recreated by a lawyer who just has cases reported to him, researches the cases and tracks them. So in this website it is by country, you can see there on the left hand side, 468 cases reported in the United States. It breaks it down by who cited the false AI. And as I look to the left, I'm looking at the screen, I have dual screens up here. The parties are the people you know who cited the false AI or who were the ones being sanctioned for its use. It goes through too is it fabricated false quotes, misrepresentations, outdated advice. Some of these can only be good up and say till 2020 and they don't have anything past 2020.
38:55In Genai model. You also got to look at that too. And then what it does here on the right it goes, tells you the case name, the jurisdiction of which it's in and you can actually search by these two, you can sort these by date. Are you using it? So if it's a pro se that got in trouble, the lawyer and so forth, so on. If they have the tool that was being used that gave the false results that you put it there as well, then the nature of the hallucination and the outcome. So again if there's monetary damages, attorney's de sanction, disciplinary complaint, dismissal, whatever it is. Here's that that was our first case in 2023 that came down $5,000 fine that they got plus the letters to the judges you can see there.
39:42And I have spot checked some of these and they are accurate over 468 reported cases. This is updated regularly as cases come in. Lawyers producing false fabricator results from Genai to court. Like I said, the biggest one, let's see if I can find it on here. It's a California case that has a $30,000. I don't think it's going to allow me to search do word find up at the top there. I tried to search for it, but it's a California case that was decided for the $30,000 monetary sanction. I don't want to take the time to scroll all the way through it. But you can do it on your own too. As you can see there www.DamienCarluton.com hallucinations and you should be able to find it again.
40:34This is something that we should really take a look at and realize that lawyers are getting in trouble for this and it's not the good in trouble. We're getting in trouble for it because again, we're not thinking we're doing stuff we probably shouldn't be doing doing stuff without being competent in what we're doing. We just need to stop and think about what we're doing before we do it and know the risk of everything Again, these are just some of the cases here that I pulled out. Mata versus Avianca. That's that first one in 23 that came down there was a case Smith versus Farwell. Three separate legal memoranda all cited and relied on fictitious cases blaming the ignorance of AI. Two recent law grabs.
41:18Like I said, I mean lower levels is no defense. Cruz versus Carlin. 22 out of 24 cases were fictitious. In the briefing from Gen, they even had names like ADC Corp. Jones, XYZ Corp. You can tell the attorney didn't read it, finally took it, turn it in. That should be a red flag to anyone when you see that generic of case bonding. Gates Year versus Goodyear two cases that did not exist as well as quotes admitted to using AI and failure to double check. Lawyers are getting in trouble for this and so we have to really stop and think when it comes to it. There is a question that came in the term AI hallucination. Does it apply only to false sites as in Luvik or false information in general?
42:05It can be to both false information and false sites. Genai can give you false cases like the one here in front of you like Finders versus Keyboard and then a citation. Now with this the name can be wrong or the citation can be wrong or both can be wrong like the name here. Finders vs Keepers is the false information but it's a correct citation. It just cited it to the wrong case. The actual case was Johnson vs. Caldwell with that citation and a different year. But it can produce one thing wrong. Everything wrong or Vice Versa. It can also give you false quotes, which is that false information that you're talking about as well. So we can say this case stated this and it really didn't. And how it does that is if it's reading these cases, say from articles from a swimming pool or from the Internet, it finds a case name. Maybe up here at the top of the article, the next paragraph, there's a quote and it logistically thinks, okay, they're close and probably let's put them together, they're false. And so that's what it's trying to make connections because it's not handing data and double checking it. It's just trying to make connections.
43:21Result. And so we can have the falsification information. Another question came in. What is the remedy when a pro se party falsely cites cases? Remedy is one to ask them, you know, is this real? Did you use Genai motion to compel if they won't answer you, and then motion for sanctions after that. Now, a lot of the cases that have gone up when the opposing counsel cites the false cases, they do get sanctioned. Sometimes their brief gets stricken. There was one case out of the Eastern District of Michigan where both parties in the same briefing cited false cases. The first attorney turned them in. The second attorney must not have read the first brief. And then the second attorney turned in a brief. They both had false cases in it, both used gen improperly.
44:13The judge struck both, the brief sanctioned them. Both didn't impose fines though. So it just depends on the court you're in. That case was in the Eastern district actually in September 2025. It's a relatively recent one. But the appropriate remedy and if a pro se does is just like a lawyer does too. If a pro se is doing it, they are held to the same standard as a lawyer, the judge. You can move for sanctions against them, you can move to have their brief strip. And now it's going to be, how does the judge handle that? Because I know sometimes judges with pro ses are a little more lenient or a little more kind of educating more than anything else. And so. But again, you can move for sanctions against them. You can move to have the brief.
45:01Now, obviously we can't turn them into the disciplinary commission necessarily. Or you could try, but you're not going to go anywhere really with that. But it's going to be more your court sanctions, your Rule 11 sanctions that are going to come into play when it comes to that. Another question that came in, it seems that Relativity or similar document review programs aren't equipped to catch these false sites. That is correct. Relativity will not catch false sites. And I know that because I've used relativity, we have relativity, you know, we use it on a regular basis for document review, document management, but it is not, it's not programmed to catch false sites. There is. I can't think of the name of it off the top of my head, but when I teach over at the law school, there is a program where we can run student papers through to find plagiarism. Those plagiarism trackers are starting to incorporate into them as if generative AI or they use the large language model to write the paper.
46:03But there's no perfect one and there's no one that I would say use this one. It's going to be the sit down and grind through the citations. Look each one up in Westlaw, look each one up in LexisNexis. There is not a catcher for brief citations yet. And I know that's what we're all wanting, because who wants to sit down and look up every single case? Well, that's why we have associates, that's why we have law clerks. Give it to them to look at the cases and make sure they're accurate. I have a feeling that we will see that technology coming soon to catch those plagiarisms, to catch those false sites and false citations, but we just don't have it yet. We are starting to see, though guidance from the court in different jurisdictions.
46:49Some of the guidance we're seeing. I know if you are in Illinois, they came out with like a 35 to 40 page generative AI manual for lawyers. So if you are an Illinois practicing lawyer, I would go out and look for that on the bar association's website. It's the Generative AI manual for lawyers in Illinois and Illinois courts and practice. And in these, it's a lot of times it's saying, if you're going to use generative AI, that's fine, just cite it. We're probably going to see an update to Bluebook of how to cite a ChatGPT eventually. Because they're just saying cite it. If you use it, cite it. If you use it, don't lie about it, tell us about it. And that is so opposing counsel and the judges know where the information's come from. And it's not this hide the ball like we first saw in Nada vs Avianca where they lied about using it, lied about using it, then finally admitted to using it. We're not doubling down anymore. If you're going to Use it, write it and then two it goes through all the safeguards except ABA formal opinion 512 went through all the ethics we discussed here today. So jurisdictions are saying follow that guidance from the aba, make sure what you're using complies with the ethics and incite it and just tell us you're using it.
48:05Other jurisdictions have forbid it outright. Judges and the judges preferences, the local rules or the judges, you know, court rules have said no. Genai, in my courtroom you need to look jurisdiction to jurisdiction, judge to judge, preference to preference. Right now it's so all over the board since we don't have clear guidance really from the Supreme Court from anywhere. Some states are more ahead of other states but again it kind of goes down to those judges preferences. If the judge says you're not using it in my courtroom, you're not using it in their courtroom. If you do, you're probably going to get a harder sanction.
48:42Again look to the judge's preferences, see what your jurisdiction is doing if there's guidance from your Supreme Court or from your bar association or any of those things. Indiana, unfortunately if you're on from Indiana you don't have that yet from our Supreme Court or from our bar association. Is it coming? I think it's coming. Don't know 100% for sure but I feel it's coming. If you're in other jurisdictions you may have guidance. I know, I think Delaware does right now. Florida, California, Illinois, there are some with clear guidance or clear, I won't say clear but clearer guidance on how to handle generative AI in those jurisdictions. As we are winding down here on time, I do want to bounce out kind of show you.
49:26Let me share a different screen here with you just how you can use generative AI if you have say the enterprise version or not even just an enterprise version but even if you're just using it public Gen AI, what some things you can do with it. So on the screen in front of you here is my personal one. This is not my firm one. This is the public facing ask everything I say self learn Public facing Generative AI. We're not going to do any client related work in it obviously but this is just kind of that chatgpt that anyone can get for free or for the $20 a month. What I want to show you here is if you look on the side of my screen, kind of where my cursor is going, you see I have something called birthday party finder.
50:12Then underneath that professional responsibility. These are my Gen AI agents. These are Agents that I've created myself to do certain tasks and certain things and they're specific to the task and things that I have basically prompted them to do. So what they are is this birthday finder one is I have a six year old daughter and I didn't know what kind of birthday I wanted to throw her. So I created, you can see here it says created by Carrie Shan creates personalized birthday ideas, visuals and reminders. So I can say here and it's geared towards Carmel, Indiana birthday Ideas for a 7 year old girls birthday party in Indiana.
51:03I type that in there and I have programmed this to give me a list of different places. I've also programmed it to give me costs associated with it, a breakdown, citations or links out to their websites which you can see there. These are all real places. I been to all these with my child at some point. But again these are all just birthday ideas. Now I'm using this generic again you can kind of take this into the law context here in a minute. It also gives me themes, flexible outdoors. Again these are real venues, these are real things, creative ideas, you know how much time, you know, what food and snacks. If I want to budget, I can ask for a budget. I can tell the number of guests. It even gives me locations around Indianapolis. This is all accurate. I know it's all accurate because again I've been to all these places but I created this agent to do this certain thing to give birthday recommended ideas. And how I do that is I go here, this chat GPT down the side and there's this tab that everyone should have called explore. When you hit explore, these are all the specialty DPT or chat GPT that do certain functions. These are created by individual people or individual companies. You can create your own to do certain functions. You can scroll down here, you can see there's one for image generator humanizer, right for me you can click more Logo creator. You can see they're created by various different platforms, some by ChatGPT itself, some by individuals and you can create your own. To create your own you go up here in the right hand corner,
52:45you hit Create, it's going to give you this screen. Now you can either create it with ChatGPT's help if you don't know how to program it, or you can go to the configure tab, create the name, description, give it the instructions you want. Instructions are going to be like I want you to give me a list of party ideas. I want you to give me the locations, I want you to give me the cost. I want you to give a budget, I want you to give a theme. You got to tell it what output you want. Think of it like a first year associate. More details you give it, the better output you're going to get out. Less details you give it, more frustration you're going to have and it's not going to do what you want. The only thing about this is unlike a first year associate, this doesn't talk back, doesn't roll its eyes and it's 24 hours a day, seven days a week. So program it to do what you want, then you will enjoy it.
53:35Now what you can do here too is you can also tell it to where to pull information from. If you wanted to just pull from your HR policies and create a system where anyone can go on and say do we have an HR policy about this? And it only pulls from your HR PDF. You upload your HR PDF here, you take off these web canvas and image generation tabs down here telling it not to pull from the web, not to pull from any of those things and it will only pull information from the document you uploaded. You get to control that swimming pool of information that it pulls from uploading stuff or giving it instructions to only pull from a certain website.
54:18Or you can even tell it to pull from the world wide web from different places. You're telling it what to do, you're telling it how you want your output to look like, you're telling it where to pull from and you get to name it. If you don't know like say how to write. Great instruction. You can ask it here because it's going to give you different prompts. What would you like to make? I can say I would like to make a program that answers attorney, legal ethics.
54:55And then I hit enter and it's going to say group and it's going to, it's going to think and then it's going to start producing this for me. Now it may take a minute here because it does have to create different things and name. Of course my Internet's always slow when I'm giving demonstrations, but that's life of it. But it's going to start to think and it's going to start configuring a name, it's going to start configuring instructions, it's going to start configuring all of those things. And here it goes. So here what it does help analyze. It's going to give you behavior, it answers questions, it gives safety, reliability, the suggestion of the name. Now let's Name it. My suggestion is this. Do you want legal guide or you prefer a different name? I'm going to say Legal Ethics guide is a find name. It's going to tell us it's a find name.
55:43We're going to go on. So it's going to say, okay, great. And then it's going to prompt me to the next thing it needs. Next we're going to create the profile picture. You can see over here on the right. Just saw something. Change it, change the name. It's creating it here on the right and answering questions on the left. So now it's saying, let's do a profile image. What would you like? Let's put scales of justice. Just do something. I spelled justice wrong, of course. Okay, so we just hit that. You're going to see. It's going to keep creating. It's going to give me images. It's going to keep creating it on the right and answering its questions here on the left. The other thing it's doing as I'm answering these questions in this create feature because maybe I wasn't a great programmer, know how to do the language and see, you can put it, put the picture on the right there. This is great.
56:34What it's also doing is in this configure button, it's asking me the question as it goes down and starting to fill them in for me. So it's creating this for me. And here now it's saying, okay, here's some prompts. It's starting to create it like that Birthday finder. I'm creating an agent to do what I want. Eventually I'll tell it where to pull the information. Yeah. Any limitations, I get to control it. I get to control where it's from. Actually, the one I was just creating for you, I have created myself called Professional Responsibility Model Rules. I created it to pull just from the Model Rules of Professional Conduct, the annotation and the comment and to answer questions.
57:20I created this for law students who are taking professional responsibility. I deployed it yet? No, I'm still testing it out and honing it in. I created one that can answer questions and give students a practice in pre question when they are studying for it. So like here I can say and you represent both plaintiff and defendant. And so we hit enter. It gives you the model rule. It answers the question, gives you the model rule. Here it's not. It's not designed to provide content. So being difficult tonight. That's all right. But anyway, it usually goes down and gives the model rule. It gives different things too when it comes to it and answers the questionable. Again, like I say, I'm playing with the instructions on this because once you create it, you can actually go back in and edit it and do different things. So like here I can delete the conversation, go back to this, but I can also go in and edit it or change instructions, change up where it pulls from. I go up here to the top, drop down, there's an edit feature. Then I can go in and edit how it's configured. So you can see here how I configured it, the model rules, answered questions. I gave it a website to pull from conversation starters. Those are my instructions that I keep modifying from these are things you can do internally within your firm. Really help. Maybe search HR policies, maybe search different things. Create it to answer common questions only from certain data or documents. If it's client confidential information, make sure you create them in the enterprise version. If it's
58:59not client confidential information, you can use the regular platform. These are just fun. There's no substantive information in here. I'm not using it for client work, just using it for fun. You can use these to your benefit in your practice and to help, you know, answer questions across the firm as well. With that, we are right at time at 06:00 Eastern Standard Time. Your code is Y as in yellow, Z as in Zebra, the number 9B as in Paul and B as in boy. Again that's Y as in yellow, Z is in Zebra. The number 9B as in Paul and B as in boy. You'll want to write that down. Go out to your LexBid account, enter that in where prompted to get your CLE credit here for day.
59:46I hope you got some information. Maybe you're gonna go test out some Gen AI, create your own agents, have some fun with it, and hope to see you again on another Lex Bid seminar. Thank you everyone.