The Perpetual Loop Prompt: A New Approach for Training AI Language Models for Better Results

As someone who works extensively with large language models like ChatGPT, I've been exploring new prompting techniques to improve their performance on complex tasks. For the last eight months, I've explored multiple knowledge bases on Prompt Engineering and experimented with dozens of different prompt patterns.

I gathered one insightful thing about prompting LLMs from all my experiences. This insight helped me write better prompts that resulted in highly professional outputs than most of you could get from an average prompt you write for Chat GPT or Gemini.

First, do you know what's the biggest weakness of the prompts written by novice users of Chat GPT?

No, not the structure. It's an easy pick for anyone to understand that they have to structure their prompts better in order to get a good outcome. Even with a properly structured prompt, you will not get exceptional results, due to this one most important weakness.

That weakness is the "lack of a learning loop".?

Now, what is a learning loop? Look at how humans understand things better and solves complex problems. We understand situations better when we are given a chance to ask clarification questions, learn more context about the problem, and then keep indefinitely improving our solutions until we find the near perfect solution for the problem. It's an endless loop of learning that improves human decision-making, isn't it?

The basis for the 'perpetual loop prompt pattern' is exactly the same.

The core idea behind the Perpetual Loop Prompt is to facilitate an iterative feedback loop between the human and the AI model. Instead of just providing a single prompt and getting a one-off response, this approach treats the AI like a persistent "competent intern" who can be continually trained and course-corrected through constructive feedback.

To get the best out of generative AI, like Chat GPT, it's important to guide and nurture it like an eager intern. Effective training involves giving feedback, clear direction, and iterative learning. The Perpetual Loop Prompt Pattern recommends using a consistent, feedback-driven approach to help Chat GPT reach its full potential and evolve over time. By using plain and easy-to-understand language, we can make sure that everyone can understand and follow along with this approach.

Here's how the Perpetual Loop Prompt works:

1. You should assign the AI model a role or persona through an initial prompt, just like you would when onboarding a new employee.
2. Instruct the AI to produce an initial draft output for a given task or objective.
3. Include instructions for the AI to self-evaluate its initial output and pose follow-up questions to get clarification from the human on how to improve.
4. Have the AI implement the feedback, update its output, and repeat the self-evaluation and feedback loop until the human is satisfied.

This approach draws inspiration from how we develop high-performing employees in the workplace. The most capable workers possess wide knowledge, seek clarity on assignments, and aren't afraid to ask critical questions to produce better results.

By employing the Perpetual Loop Prompt method, AI language models can be trained to exhibit those same traits over multiple iterations. The human's role shifts from being just a prompter to an interactive coach and evaluator.

I've already seen promising results when using this technique, with AI models demonstrating improved understanding, generation quality, and task performance compared to one-off prompting. However, I'm still exploring and expanding the potential applications.

Interested in Learning More About the Perpetual Loop Prompt?

If you're keen on exploring the Perpetual Loop Prompt pattern or other innovative ways of utilizing generative AI for marketing and business, I recommend enrolling in the Generative AI for Modern Marketing course at the Asia Pacific Institute of Digital Marketing (APIDM). This comprehensive program covers advanced AI prompting techniques, practical applications, and ethical considerations in the real world.

The era of simply treating AI as a black-box question-answering tool is evolving. With interactive prompting methods like the Perpetual Loop, we can take a more hands-on approach to shaping and refining AI's capabilities. I'm excited to see what other innovative prompt engineering techniques emerge as this field progresses rapidly.

NOTE:

Perpetual Loop Prompt is an LLM prompt pattern conceptualized by the author (Amitha Amarasinghe) as part of a Generative AI training program for marketers. If you use this prompt pattern in any of your training programs, lectures, conference talks, podcasts, or other content types, please credit the original creator of the prompt pattern.