Andrew Bolis Shares Expert Prompting Guide for ChatGPT

In a recent LinkedIn post, Andrew Bolis discusses the intricacies of crafting effective prompts for AI models like ChatGPT, highlighting OpenAI’s newly released Prompting Guide. Bolis emphasizes that the way a user phrases a prompt and what they choose to emphasize can significantly influence the AI’s output, suggesting that ignoring certain aspects can limit results.

Bolis breaks down the art of prompt engineering into nine actionable tips, underscoring the importance of structured and precise communication with AI. He begins by stressing the fundamental elements of a strong prompt:

“Use Role + Goal + Boundaries
↳ Clearly tell the model who it should act as, the outcome you want, and the limits or boundaries it must not cross.”

This foundational advice, according to Bolis, sets the stage for more complex interactions. He further elaborates on providing context, suggesting that information should be presented in a logical sequence.

Structuring AI Interactions for Optimal Results

Andrew Bolis advocates for a methodical approach to prompt design, which he argues is key to unlocking the full potential of AI tools. He points out that the placement of instructions within a prompt can dramatically affect how the AI processes them.

Prioritizing Key Instructions

Bolis highlights a specific technique for ensuring the AI focuses on critical directives:

“Put Key Instructions at the End
↳ The model pays strong attention to the last lines, so save your most important directions for the end of your prompt.”

This strategy, as Bolis explains, leverages the AI’s tendency to give more weight to information presented later in the prompt. He also introduces the concept of validation through a ‘Chain of Verification,’ where tasks are broken down into smaller, manageable steps that are checked for accuracy before proceeding. This methodical approach, Bolis suggests, is crucial for complex tasks that require a high degree of precision.

Enhancing AI Accuracy and Reliability

Further delving into techniques for improving AI output quality, Andrew Bolis presents methods designed to mitigate errors and biases. He champions the ‘Dual-Pass Response’ method, where an initial draft is followed by a self-assessment phase.

“Use Dual-Pass Responses
↳ Have the model draft a first version, then refine it against a self-check rubric to improve accuracy and quality.”

In Bolis’s view, this iterative refinement process is vital for achieving polished and accurate results. He also stresses the importance of ‘I Don’t Know’ Honesty, advising users to instruct the AI to admit uncertainty rather than generating potentially incorrect information. This, he argues, prevents the creation of plausible but factually wrong outputs. Bolis encourages adopting ‘Multiple Viewpoints,’ prompting the AI to explore a problem from various perspectives before synthesizing the best elements. Additionally, he recommends using ‘Clear Structural Delimiters,’ such as code fences, to separate instructions and data, thereby enhancing clarity and preventing misinterpretation.

Guiding AI with Examples

Finally, Andrew Bolis emphasizes the power of exemplars in prompt engineering. He suggests using examples to set clear expectations for the AI.

“Guide with Examples
↳ Provide the model with two good examples and one poor example before asking for output to set quality expectations.”

This technique, according to Bolis, helps the AI understand the desired quality and format for its response, leading to more consistent and relevant outcomes. Through these detailed tips, Bolis aims to empower users to interact more effectively with AI, transforming their engagement from basic queries to expert-level prompt engineering.