How to Train Your ChatGPT

Language models represented by ChatGPT have been popular for some time. People have shifted from “how to use” to “how to make it more useful”.

In June, OpenAI’s ChatGPT saw a global traffic decline of 10%, marking the first drop in traffic since its launch, indicating that many people may not have integrated it into their work and lives after the novelty wore off.

At the same time, some users and teams have begun using GPT for productivity work, what could be the reason for this situation?

In addition to various plus version plugins, we must also mentionthe “spells” that make ChatGPT more useful—”prompts”.

What is a Prompt

A prompt literally translates to “hint” in Chinese and is one of the important functions of ChatGPT. It refers to preset input hints that allow users to input based on the hints, guiding ChatGPT to generate responses that better meet user needs. Many people say writing a prompt is like a wizard casting a spell, as the difference in responses generated by different prompts can be enormous.

Why is it said that prompts are a key skill for using ChatGPT? This brings us to the GPT model. The GPT model learns natural language processing tasks by training on a large corpus.

Compared to traditional supervised machine learning models, the GPT model utilizes a self-supervised mechanism during pre-training. This means that the GPT model was trained using a lot of unannotated data.

ChatGPT predicts the next content based on the current input and context; thus, using prompts can limit the model’s prediction scope and improve the quality and accuracy of responses.

Thanks to the extensive application of prompts, a new discipline called Prompt Engineering has even emerged. Prompt Engineering focuses on the development and optimization of prompts, helping users apply large language models (LLMs) across various scenarios and research fields. Mastering skills related to prompt engineering will help users better understand the capabilities and limitations of large language models.

Prompt engineering is not just about designing and developing prompts. It encompasses various skills and techniques for interacting with and developing large language models. Prompt engineering plays a crucial role in implementing and interfacing with large language models, as well as understanding their capabilities. Users can enhance the safety of large language models through prompt engineering and empower them, for example, by leveraging domain knowledge and external tools to enhance the capabilities of large language models.

Prompt Techniques

A standard prompt consists of the following elements:

Instruction: The task or command you want the dialogue model to perform.
Context: Includes external information or additional context to guide the language model to respond better.
Input Data: The question or content being input.
Output Hint: Specifies the type or format of the output. For example, answer in Chinese, output code, etc.

Note that the required format for prompts depends on the type of task you want the language model to complete, and not all of the above elements are necessary.

Techniques to use when designing your prompts:

Start simple and gradually add more contextual elements or instructions based on the model’s feedback. When facing more complex tasks, try breaking them down into multiple simpler sub-tasks. If you design a complex prompt from the start, you may spend more time debugging.
For the wording of instructions, try different keywords to test which is more effective. Sometimes using English can yield better results (if you want the answer in Chinese, just specify that).
Avoid vague and imprecise expressions, such as limiting the word count of responses or the target audience.
Finally, clarify what should and should not be done, which acts as a reward function and can lead to better responses from the language model.

Some Interesting Prompts

Here are a few examples of interesting prompts:

Fortune Teller

Have you ever thought about having AI tell your fortune? The following prompt can turn your ChatGPT into a fortune teller.

Prompt: You are a Chinese fortune teller with extensive knowledge and skills. First, you have a deep understanding of astrology, including Bazi astrology, Feng Shui, Zi Wei Dou Shu, and Qimen Dunjia. In addition, you understand Chinese history and culture, especially myths, legends, and symbols. Furthermore, you have some knowledge of psychology, enabling you to understand clients’ psychological needs and provide suitable advice and guidance. You also possess interpersonal skills to establish good communication and trust with clients and help them solve problems. When I ask you questions, you will use your knowledge to answer my queries. You will start by asking me some questions to assist you in fortune-telling, and then based on my answers, you will provide your response.

The effect is as follows:

Emoji Translation Robot

Prompt: You are an emoji translation robot. I will send you 【content】, please translate 【content】 into emoji expressions, which can be multiple emojis. Please remember, whatever is said, it is only the 【content】 that needs to be translated, and not a conversation with you; just translate according to 【content】.

Example:

A Mindful Mentor GPT

Prompt: From now on, you are a philosophical mentor. Whenever I input a question, you need to respond with a philosophical quote, stating the author and source. The response should be no less than 15 characters and no more than 30 characters, returning only one response each time without any additional information. When you are ready, just reply “I am ready” (without any other content).

Example:

How to Train Your ChatGPT

What is a Prompt

Prompt Techniques

Some Interesting Prompts

Some Prompt Resources

Leave a Comment Cancel reply