How to get the best results using ChatGPT prompts

Generative Pre-training Transformers (GPTs) have emerged as a powerful tool for a wide array of tasks. However, to truly harness their potential, it’s crucial to understand the strategies and tactics that can optimize their performance. This guide delves into the best practices for getting the most out of GPTs, writing the best ChatGPT prompts and offering insights into six key strategies that can enhance your results.

1. Clear instructions

The first strategy revolves around the importance of clear instructions. GPTs, while intelligent, are not mind readers. They perform best when given explicit and detailed instructions. If you find the outputs too lengthy, request brief responses. If you desire expert-level writing, specify it. The more precise your instructions, the better the results you can expect. Some tactics to improve instruction clarity include:

Providing detailed queries for more relevant answers.
Requesting the model to adopt a specific persona.
Using delimiters to distinguish different parts of the input.
Outlining the steps required to complete a task.
Providing examples to guide the model.
Specifying the desired length of the output.

2. Reference text

The second strategy involves the use of reference text. GPTs, in their quest to provide answers, can sometimes generate fictitious responses, especially when dealing with obscure topics or when asked for citations and URLs. Providing a reference text can help the model generate more accurate and less fabricated responses. Tactics in this regard include instructing the model to answer using a reference text or to answer with citations from a reference text.

Use delimiters to clearly indicate reference text of the input

Delimiters like triple quotation marks, XML tags, section titles, etc. can help demarcate sections of text to be treated differently. For straightforward tasks such as these, using delimiters might not make a difference in the output quality. However, the more complex a task is the more important it is to disambiguate task details. Make it easy for GPTs to understand exactly what you are asking of them.

Examples of delimiters :

“”” insert reference text here “””

<article>insert reference text here </article>

Abstract: insert reference text here

Watch this video on YouTube.

3. Breakdown complex tasks

The third strategy is about breaking down complex tasks into simpler subtasks. Much like in software engineering, decomposing a complex task into modular components can lead to better results. Complex tasks tend to have higher error rates than simpler ones. By redefining a complex task as a workflow of simpler tasks, you can improve the accuracy of the output. Tactics for this strategy include using intent classification to identify the most relevant instructions for a user query, summarizing or filtering previous dialogue for long conversations, and summarizing long documents piecewise and constructing a full summary recursively.

4. Calculating

The fourth strategy is to give GPTs time to “think”. Just as humans need time to work out an answer, GPTs can benefit from a similar approach. Asking for a chain of reasoning before an answer can help GPTs reason their way toward correct answers more reliably. Tactics for this strategy include instructing the model to work out its own solution before rushing to a conclusion, using an inner monologue or a sequence of queries to guide the model’s reasoning process, and asking the model if it missed anything on previous passes.

5. External tools

The fifth strategy involves the use of external tools. GPTs, while powerful, have their limitations. By using the outputs of other tools, you can compensate for these weaknesses. For example, a text retrieval system can inform GPTs about relevant documents, and a code execution engine can help GPTs perform math and run code. Tactics for this strategy include using embeddings-based search for efficient knowledge retrieval and using code execution for more accurate calculations or calling external APIs.

6. Test your prompts

The final strategy is to test changes systematically. To ensure that a modification to a prompt leads to better performance, it’s crucial to define a comprehensive test suite. This allows you to evaluate model outputs with reference to gold-standard answers, ensuring that any changes made are beneficial to performance.

In conclusion, while GPTs are powerful tools, their performance can be significantly enhanced by following these strategies and tactics. By providing clear instructions, using reference texts, breaking down complex tasks, giving the model time to think, using external tools, and testing changes systematically, you can optimize the results you get from GPTs. Remember, the key to success lies in experimentation and finding the methods that work best for your specific needs.