Comment 8: Talking to Models

Communication is a fundamental skill. It doesn’t matter what you are doing: communication is central to whatever it is. It is taught in schools, it’s part of work trainings, and we’ll even engage therapists to understand how to communicate in our personal lives. When we are using a search engine, querying a database, or using an AI model we are still engaging in communication. In case it wasn’t obvious, the models I’m talking about are AI tools.

Now that a lot of people got disappointed and left, let’s get into this.

You might be a little concerned that this is going to become a hyper-complicated discussion on prompt engineering. I promise that isn’t going to happen. We’re going to discuss some of the same concepts, but the goal isn’t to become AI prompt ninjas. That sounds pretty cool, though. No, we’re going to talk about how to communicate with the model: something I like to call prompt formulation.

Syntax

This word has a lot of meanings with some only existing in specific use cases but, broadly, syntax is communicating in a way that can be understood. While search engines, databases, and AI tools aren’t people we still need to communicate with them in a way they understand.

In Google when we use “exact phrase” we are telling the machine that we are searching for those two words in that exact order.

You could do the same thing in WestLaw with quotation marks, but if you didn’t know that you could use exact +s phrase. Two ways to get the same result!

AI tools, however, largely rely on semantic statements. That is to say: we are supposed to engage with them using natural language.

Prompt Formula

The literature of how to write a prompt gives all kinds of suggestions and examples. Honestly, there is a lot of great advice out there. There are two that I’ve come across that people seem to find very useful. I’ll introduce one but spend most of my time on the second, which I think is useful in almost all use cases.

The first is Intent, Context, and Instruction which is often called the ICI Method.

Intent: What you want

Context: Why you want it

Instruction: What you want the AI model to do for you.

It’s a useful construction to use for thinking about how to communicate with AI and it gets the job done, but there is another formula that better encompasses what we should be thinking about when prompting.

While it doesn’t have a name that I’m aware of, I’ll call this the CTCF Method.

Context: Who you are and why you’re asking

Task: What you want the model to do

Constraints: What to avoid (tone, scope, etc.)

Format: How you want the response delivered

Let’s consider a prompt and then break it down into the components of the formula.

I am a second-year law student. I am writing a Motion to Compel Discovery regarding a slip-and-fall accident in Ohio. Create a detailed outline for this motion, including standard headings, potential arguments for relevance, and a tone that is firm but professional.

Context: I am a second-year law student. I am writing a Motion to Compel Discovery regarding a slip-and-fall accident in Ohio.

Task: Create a detailed outline for this motion

Constraints: a tone that is firm but professional

Format: including standard headings, potential arguments for relevance

The reason why I prefer the CTCF Model is that it reminds you to think about what you don’t want AI to return. This is important for several reasons:

First, since you use AI models to help you complete a task (Never! Ever! To do the task for you!), it’s vital to constrain it to only work on what you need.

Second, without constraints a model may start assuming what you want which gives it room to move away from your request. This could make more work for you as you’re now removing stuff you never needed.

Finally, think about formulating your prompt like ordering lunch. If you hate tomatoes and don’t request no tomatoes, you’re going to end up with tomatoes…which you hate. So many tomatoes!

TLDR

AI can be a helpful tool for accomplishing a task, but you need to put some thought into what you want to accomplish and what you need the AI model to do. Using a formula like CTCF can help you better communicate your needs and, hopefully, get better responses.

——————————

The ABA Rules of Professional Conduct, Model Rule 1.1 Comment 8 requires, “To maintain the requisite knowledge and skill, a lawyer shall keep abreast of changes in the law and its practice, including the benefits and risks associated with relevant technology.” To that end, we have developed this regular series to develop the competence and skills necessary to responsibly choose and use the best technologies for your educational and professional lives. If you have any questions, concerns, or topics you would like to see discussed, please reach out to e.koltonski@csuohio.edu  with “Comment 8” in the subject.