Prompt Engineering via Prompt Patterns

Summary

The article discusses the use of the Game Play Pattern in prompt engineering to create interactive, problem-solving scenarios with large language models (LLMs) like ChatGPT, leveraging the model's ability to generate content and follow rules to engage users in a game-like learning experience.

Abstract

The "Game Play Pattern" is a method within prompt engineering that utilizes the concept of games to engage with large language models (LLMs). This pattern is characterized by two main components: rules and content. It is likened to educational activities such as mathematics examinations, where students apply learned rules to solve problems. In the context of LLMs, users define the rules and scope of the game, and the model generates relevant content or scenarios. This interactive approach allows users to test their knowledge and skills in various subjects, including prompt engineering itself, by crafting prompts for the LLM to respond to and evaluate. The article suggests that this pattern can be applied to any known or unknown area for skill improvement or testing and encourages readers to engage with the content by clapping, sharing, and subscribing to related channels for more information.

Opinions

The author believes that the Game Play Pattern is a practical and engaging way to interact with LLMs, making learning and problem-solving more dynamic.
The author emphasizes the importance of specificity in the game's topic to enhance the interest and complexity of the game play.
The author values the LLM's capability to generate diverse and relevant scenarios, which can be used to simulate real-world challenges, such as diagnosing a compromised Linux terminal.
The author appreciates the feedback mechanism inherent in this pattern, where the LLM not only presents problems but also evaluates the user's responses, providing a comprehensive learning experience.
The author sees the potential for continuous learning and skill refinement through the use of the Game Play Pattern in conjunction with other patterns like the Infinite Generation Pattern.
The author encourages readers to participate actively in the learning process by applying the discussed patterns and to support the educational content by engaging with it on various platforms.

Prompt Engineering via Prompt Patterns — The Game Play Pattern

You can switch to video version of this article

If you have learnt mathematics in high school, you have used this pattern in real life. Game is a very broad term with tons of meanings. For this video, I would pick the Collins English dictionary definition of game which goes like ‘ A game is an activity or sport usually involving skill, knowledge, or chance, in which you follow fixed rules and try to win against an opponent or to solve a puzzle.’ We can exclude the sport or physical activity part out since we are not going to play soccer with a large language model. This brings the definition down to activity where you create a game by defining the rule, and then try to play with the opponent, which in this case is the LLM, and try to win. The game usually is a text based game.

A game usually has two main constituent components. The rules and the content. This is how you apply game play pattern. When I said you have applied this pattern in real life if you studied mathematics, what I meant was your teacher taught you all the rules for a particular activity lets say multiplication or division, and when you are ready, you are given a set of questions where you try to apply those rules and solve the questions. This is usually called examination or test, but actually it fits the formal definition of game. Here the rules are the method or procedure of doing multiplication or division, and content is the questions where different numbers or logic questions are used to test your ability to apply those rules effectively to solve the problem. This is what we would be doing or playing with ChatGPT while applying this pattern.

LLM has a vast training set, and thus is excellent in coming up with content related to any topic. For this pattern, you would like the LLM to generate scenarios or questions around a specific topic, and then try to apply problem solving or other skills to accomplish a task related to the scenario. You would select the topic or spell out the rules of the game, and LLM would be responsible to generate content using its vast knowledge pool.

The key contextual statements for this pattern are

Create a game for me around X

One or more fundamental rules of the game.

The first statement instructs LLM to create a game and provide the scoping of game to topic area X. As a general rule, the more specific the topic, the more interesting is the game play.

Next user can specify certain rules for the game. The only important bit is the rules should fit within the capabilities of the LLM. The game can be as simple as create a game around multiplication or it can be made more complex like LLM like ChatGPT assuming role of a linux terminal which has been compromised by an attacker, and you, as an IT expert is going to use commands to try and figure out how it was compromised. Then you specify the scope of the attack to a few types which can be diagnosed with commands. You can instruct it to print a scenario of what symptoms led to investigation containing clues to get you started.

You can effectively test your knowledge of a subject with questions about the subject being created by the LLM. The beauty of this pattern is, it doesn’t just stop there. Since LLM is the game master, and you are responding to its generated questions, you can even ask it to evaluate your answers. A very interesting example I came across was using this pattern to test your skills of a prompt pattern. For that you tell LLM to create a game about prompt engineering, where it would give you a task that can be accomplished by prompting it. The task should have a reasoning component to it. You (the user) would try to write a prompt for LLM to solve the task. LLM would produce the output, and explain how well my prompt solved the task. You then include the infinite generation pattern to instruct LLM to continue creating tasks until you tell it to stop. Ask the first question.

How amazing and beautiful is that. You are playing a game with the model, where it is giving you problems to solve, and evaluating you on tasks that can be solved by proper use of prompt engineering. It would give you proper feedback on your prompt whether it was clear, to the point, and led to properly solving the task.

You can add further instructions to further scope the questions about a particular pattern and test your skills on applying that. You can try out the patterns we are discussing in this video series to test your knowledge and skills using the game play pattern.

You can use this pattern to improve or test your skills in any known or unknown area. I hope you enjoy doing so, and if you do, don’t forget to clap for and share this article. You can also consider subscribing to our YouTube channel as well. Thank you!!!.