Why Lawyers Are Struggling With ChatGPT 5.1

In our prior videos, we've been talking a lot about ChatGPT 5.1. We just went from 5.0 to 5.1. There's been some major changes to the way that ChatGPT works that's going to affect you, the way you go about your daily work as a business professional or a lawyer. So, we want to start to dig in on the changes that are built into 5.1 that you need to understand in order to change the way you use this AI tool at work to go from 2x to 5x, 5x to 10x, 10x to 20x. This is a significant change in the way that you need to interact with ChatGPT in order to get the most out of it as a lawyer or business professional. Also, in prior videos, we've talked about who it is that you need to be following as an attorney if you want to learn How to best use ChatGPT and these other AI tools at work in order to become the best fill-in-the-blank lawyer that you can be. And one of the people that I have talked about previously is Nate B. Jones. And Nate is a computer engineer, but he's got this really great approach to AI, which is much more tailored for business people who want to understand how AI works well enough to be able to strategically use it in their workflows. And he does a lot of consulting for companies and business professionals, so he's not just on the tech side. And one of his most recent Substacks deals with 5.1, and hopefully Nate doesn't mind that I'm going to show you some of this content, which is within his Substack, which I'm going to put the link in, which I'm going to encourage you to subscribe to. There's There is a small fee for getting his premium content, totally worth it. Please go subscribe to Nate, and in your free time, look at his videos, watch his videos, read his commentary. It is really spot on for lawyers, and it's where I learn a lot of what I need to know in order to get the most out of these AI tools. So Nate makes the following point. There's been a lot of talk and chatter about the, the changes that occurred in 5.1 and how you can set personality. But his view, and I agree with it, is that the real change in 5.1 that you need to understand in order to modify your workflow is that it is going to follow instructions much more precisely, and that you need to change the way you build your instructions, your prompts, if you want to get the most out of these. Tools. OpenAI has now given you the ability to go from 10x to 20x by doing one simple thing: improving your prompting, improving your prompt engineering. So let's dive into it. Let's get to it. And remember that we're going to talk about the change, but at the end of this video, I'm going to walk through a prompting sequence that you can use In order to make sure that you are building better prompts for 5.1 to work with. So he talks about how sharper instructions has got both an upside and a downside. In older models, you can get away with sloppy conflicting instructions because the model 5.0 would just average them out. So if you said something like, give me a detailed summary of this document, it might view that as a conflict. If it's going to be detailed, then is it going to be a summary? It would be better to say, provide me a summary of the document, or provide me a detailed analysis of this document. But to put detail and summary in the same sentence is conflicting. It would average those out in prior models. In this new model, it is more likely to get confused because it wants to follow your instructions precisely. If you have inconsistency within your prompting, it's going to provide poor results. If you try and tell it too many things in your prompt, it again is more likely that there's going to be some inconsistency or ambiguity in there, and it's not going to do the job as well. You want to think of 5.1 As your— for the first time, as your ability to kind of build agentic flow within the prompt sequence, meaning in our world, you are going to be engineering workflow for an associate. That associate is ChatGPT, and you're going to be building a sequence, building instructions that are very precise, and then turning that new AI associate or paralegal loose to go do the work or to work with you to get to a goal. So instruction following is better, and you want to avoid long prompts with hidden defaults or vague language that can lead to problems. You need to simplify your prompts. You have to treat your prompts and your system messages like specifications for the AI to abide by. This means you need to separate your request for tone, your request for safety. Your workflow can't have 5 things in it. Instead of piling everything into a single prompt, you want to break things up into pieces and act sequentially like a project manager would break down projects by task categories, by tasks, by subtasks, by sequencing. This is the new way in which law will be practiced. So keep instructions simple, non-contradictory, one main goal is preferred. So we're going to dive a lot more into 5.1 and the other nuances of 5.1 in later videos. So make sure you subscribe, make sure you share with a business associate or lawyer that you, that you like and you want to survive the AI age. But for today, we're just going to focus on this one thing. It is how to get better prompts and build a foundation for workflow within 5.1. Let's dive into a real-world example of how to prompt 5.1 in order to get the best results. So, we know that it is important to be precise. We know that it's important to be short. We know that creating a set of instructions is the key to 5.1. Why wouldn't you use the AI tool to build those instructions? A lot of lawyers, a lot of business professionals, they just start out by asking the AI tool to do something. That is not the way to use AI. You're going to get very average or sub-average results if that's your process, if that's your workflow. Your workflow should be on most tasks to take a deep breath and go slow. Start with building the instructions using the AI tool. Let me give you an example, show you what I mean. So we're going to want to do a project. My son has got a landlord-tenant matter. His landlord will not give him back the deposit, so I've been helping him So this is going to be a good example. We're going to build some instructions to help him navigate through the small claims court action, which is coming. So I've told ChatGPT, I want you to help me build a set of instructions for this project. That's what we're doing first. We're not asking it to draft the complaint for small claims. We're not asking it to develop arguments or to analyze case law. No, we're building the foundation upon which we're going to do the work. So we're going to do some instruction building, which is also known as meta-prompting, using the AI tools to improve your prompt, to improve your instructions. So I'm going to tell it a little bit about what this project is all about so it can help build the prompt. Our client is a tenant. Great. That's important. They're trying to get their deposit back. Short and to the point. After the conclusion of tenancy, the landlord refuses to provide the deposit. Straight and to the point. We are operating under Michigan law. Straight and to the point. These are the essential things that it needs to know in order to start building the instructions. Before beginning, I want you to ask me questions which will help build a set of instructions for this project and operate on all inputs and outputs. This is really important, so let's just take a pause before you do anything, ChatGPT. We're going to build instructions, but I want you to ask me the questions, get the information that you need, understanding what our goal here is: get the deposit back for a tenant from a landlord who refuses to provide the deposit under Michigan law. I want you to ask me the questions that you need to know in order to build really good instructions. And I tell it, give it some more context. In essence, we're going to go through meta prompting exercise. Now pause. You can use this approach on many different workflows within ChatGPT. And I always, when I'm building a project folder for a client, I always start by having a, uh, you starting a thread which is going to be about building the custom instructions for the GPT. Now I haven't done any here. We're going to go right into the thread, but normally my first thread would be building the instructions, and I'd be able to actually take what I'm going to show you here in a few minutes and paste it into the instructions so that for this project folder, for this client, it's following the instructions, these high-level instructions. In all instances. So I'll go back up to the top and it's now going to say, in order to build great instructions, I need to pin down your goals, constraints, and preferred working style so that the meta prompt we build will consistently produce exactly what you want for every interaction. And it's now going to give you a whole bunch of questions. And I say, okay, that's a lot of questions, 12 of them. I want to answer them one at a time. Let me know if you have any follow-up questions at each point, starting with number 1. Here's the answer. Good. It's saying good. A couple of clarifying questions. So now you have an interactive tool that is not just doing what you tell it, but is collaborating with you. Key point, power tip. You want to use your AI in a way to where it's working with you and challenging you. So another prompt strategy that I use is to always tell it to ask me questions in order to be able to provide improved results. If there is a gap in information that you think would be helpful, please let me know. Ask me to provide that information. I'll let you know the answer, or if there is not any information to provide. Again, creating an interactive assistant, just like a good associate would do. A good paralegal would come back to you and say, hey, I need more information. So it's asking me what the different tasks are that's going to be required of this project. And I provided answers. And I say, but look at, however, because we're creating instructions that will cover all necessary activities for pursuing small claims actions from start to finish, we don't want to leave anything out of the parameters. So it's trying to say, well, exactly what things do you want me to do? And I've said, okay, we can have examples, but I may need you to do lots of things. So don't get too constrained. If we were to say you're only helping me draft the complaint, then it's not going to be able to help me develop the arguments or prepare for the final hearing. Keep in mind, 5.1 is going to follow your instructions. So you want them to be specific and narrow. But understand that if you go outside that narrow scope, it's now going to provide Less than optimal results. Now it's going into some more information, some more clarifiers, and I'm responding here. The instructions should cover all the items in number 1, but again, we want to make sure that instructions broadly cover all various activities. So again, I'm scaling it back from trying to develop a list of all activities. Any activities that are going to be required for this project Entering small claims court, and I'm telling you we should work within the statutory framework for security deposits in landlord-tenant law. I already told it Michigan law. We do want strict compliance with both sides in terms of what is required, and assume the judge is going to work strictly within the statute. So again, providing it information. This is my expertise in landlord-tenant court. My belief is that the judge isn't going to get off into the. into the sides of the highway, they're going to be looking at the statute and following the statute. Understood. And so now it's asking me about tone. The tone should be that of a Michigan landlord-tenant from the tenant's perspective. This doesn't need to be overly legalistic. We don't want them to sound formalistic. We want to have the tenant sound rational, reasonable, objective. And here's some information about tone, which is really important. Do we want the AI to develop very histrionic arguments or to be very aggressive? No, we want to appear reasonable because in this case, the landlord we know is going to sound histrionic, is going to sound retaliatory. We want to juxtaposition this. This is where my expertise as an attorney comes in to be able to say, this is what is going to win the case and help the AI understand that. As we work through this process, You're going to see that finally we get to the point, let's go ahead and start drafting the meta prompt. And it gives me this really solid prompt that I could not come up with on my own that we will paste in the instruct— into the instructions in order to do the work. This model supports a tenant bringing a small claims action in Grand Traverse County seeking return of security deposit under the specific statute. This model helps from the complaint through preparing for the hearing. General approach: always adopt the perspective of a well-prepared, calm, rational tenant who understands the rules. Reasonable, objective, and fact-focused. Avoid hostility, emotional, legalistic language. Use short, clean sentences. Focus only on what a Michigan small claims judge will care about: facts, timelines, documents, statutory compliance, and credibility. That is a really strong prompt that you could not likely develop on your own. And this becomes really important in 5.1, where you are designing, in a sense, software instructions, very strict instructions as to how this software, this AI is going to act, what its goal is, how it's going to be constrained. And the better those instructions, the better results you're going to get. So communication style, procedural guidance, evidence and preparation, exclusions and boundaries. You can edit this within the instructions. I'm going to copy this. I'm going to go to the project folder. I'm going to go to the 3 dots on the right, which you can see here. I'm going to edit the instructions and I'm going to paste them Here, get rid of the beginning and the end, and I've got my instructions for this project. I can follow this same protocol when I get to the next thread that I start, which is to draft the the complaint, the claims in small claims court. So I'm going to tell it. Let's go ahead and use the the template. Required in Michigan for small claims actions. We're going to draft an addendum where we are going to state our claim with specificity. It's going to follow those instructions. But now I might say, I want you to help me develop a prompt for this thread, for this specific task of drafting the demand or the complaint in small claims For the various different things that we're going to want to allege. So one of the things we're going to have to do is figure out what are we going to want to allege by way of facts and by way of law. So help me develop a prompt for this thread that breaks our effort into components, develop the legal theories so that then second component we can develop the facts so that third component we can do the drafting. This. This is the workflow and the strategy that you need to have. This is the mindset that you need to have when working with 5.1. I'm going to be doing a lot more videos because I believe that this is foundational as to what is going to be happening in the future and how you will need to level up your expertise in these AI tools in order to get the most out of them. My name is Enrico. Go ahead and subscribe. Share with someone you care about that you want to survive the AI age. Go ahead and comment, ask any questions below, and I'll answer them. We are all in this together. We'll see you in the next video.