Who hasn’t experienced asking generative AI to generate a spreadsheet for a certain purpose and it ends up generating information that has little or nothing to do with what we expected? This is often a lack of appropriate context, or our request is ambiguous and this causes the AI to generate inaccurate, imprecise, erroneous and even responses known as hallucinations (completely meaningless responses).
When we face conversational artificial intelligence such as ChatGPT, Copilot or Gemini, we run the risk of not asking correctly and generating information that can often be of dubious use. We must remember that for these intelligences to function properly, we must provide them with a properly formulated question or request. To help with this, we can use different patterns that will help us better target the requests we make.
Imagine asking a generative AI to tell us what risks we might encounter when using Scrum in a generative AI project. The answer is likely to be adequate, but it may be so general that we can’t use it effectively.
To avoid this, we are going to look at two patterns to generate your prompts more effectively.
RTF prompts
The first pattern for creating prompts is known as RTF. which stands for Role, Task, and Format. Let’s see what each of these acronyms means:
- Role: In this part, we’ll indicate the role the prompt must assume when generating the response to our request. What we do is establish a context for the AI so that it can provide us with a more specific response based on the given role.
- Task: Here, we’ll precisely indicate the task we want it to perform. The generative AI will use the role indicated above to perform this specific task. It’s very important to be as clear as possible about the task and not to assign multiple tasks at once, as we’ll often find that it doesn’t perform one of them.
- Format: With this part, we’ll indicate the format and structure with which we want the response to be generated. In each situation, we may need a different format: table, spreadsheet, image, etc.
Let’s look at a fictitious example applied to management for this pattern:
R – Role
You are an experienced Agile Change Manager specializing in machine learning and artificial intelligence projects.
T-Task
Your task is to identify potential risks when applying Scrum to a chatbot project with generative AI, indicating the different affected artifacts, events where the risk should be addressed, and the role responsible for managing the risk.
F-Format
Create a Risk record with a description, mitigation action, artifact, event, and responsible role in table format.
With this pattern, we will see a significant improvement in the responses we get from generative AI.
If we need something even more efficient and effective, we will need to give it more detail and opt for a more complete pattern.
CREATE prompts
CREATE is the acronym for Character, Request, Examples, Adjustments, Types of output, and Evaluation. It is a pattern that expands on the previous one and provides a much more comprehensive and complete structure for crafting prompts
- Character: As in the previous pattern, this consists of defining the role or persona that the generative AI must assume to generate the response.
- Request: This would consist of the task or tasks that the generative AI must perform.
- Examples: When defining any context, providing examples can greatly clarify the expected result. These examples will make the response obtained much closer to what we want.
- Adjustments & Constraints: In this section, we will indicate any requirements or limitations that must be taken into account when generating the response.
- Types of output: This consists of describing (like the Format in the previous pattern) the type of output, format, and structure we want to obtain.
- Evaluation & steps: In this last part, we provide the prompt with examples or a guide to successfully execute the task.
Let’s see how to rewrite the previous example with this new writing pattern:
C – Character
You are an experienced Agile Change Manager, specialized in machine learning and artificial intelligence projects.
R – Request
You need to identify risks when applying Scrum to a generative AI chatbot development project, considering how these risks affect Scrum artifacts and events, as well as who should manage them.
E – Examples
Some examples of risks would be:
-
- Lack of clarity in acceptance criteria for technical stories.
- Risk of bias in training data affecting chatbot development.
- Difficulties estimating technical tasks related to model training.
A – Adjustments and Constraints
Limit the table to 5-10 relevant risks, including only risks related to the application of Scrum in the context of generative AI, using clear, technical language.
T – Types of Output
Generate a risk report in table format containing:
-
- Risk description
- Affected Scrum artifact
- Scrum event where it can be addressed
- Mitigation action
- Role responsible for managing the risk
E – Evaluation and Steps
Your output must:
-
- Be useful for agile teams working in AI.
- Demonstrate knowledge of Scrum and the specific challenges of generative AI projects.
- Be applicable in real-world contexts.
With this second writing format, we will achieve more specific results tailored to our specific needs.
Of course, in any of these formats and scenarios, the most important thing is to validate the response we’ve obtained and confirm with other people and team members that what was generated is correct.
Although both patterns are valid, the CREATE pattern allows for greater specialization and detail, allowing us to evolve our prompt until we obtain a result that’s suitable and valid for our work.
I encourage you to try out the different prompts shared in this article and see for yourself how much the quality of the output can improve.
Agile Coach en Keepler Data Tech. "I help teams and organizations to take perspective on the problems and needs they have in their day to day, and then draw an action plan with them that fits the goals they want to achieve. I am addicted to continuous improvement, cultural change, teamwork and of course the people who make it happen."
0 Comments