ChatGPT is a force to be reckoned with in the world of AI, but only if you know how to use it correctly. If you're thinking, how can someone use it incorrectly? You're right; they cannot. But to master ChatGPT, you need to use it in a way that generates better results. And to generate better results, you need to master the prompts.
In fact, prompt engineering – the process of designing and optimizing the prompts to elicit responses from machine learning models – as it has come to be known, has become a skillset of its own. Some companies are even hiring prompt engineers at insanely good salaries, 6-figure good.
You, too, can become the AI whisperer with practice and the right set of techniques. Here are my two cents on how you can master prompts, with tips and advanced techniques, and become a ChatGPT wizard!
Nailing the Basics of ChatGPT Prompts
A prompt is what you enter in the textbox in ChatGPT to converse with the chatbot. Your whole experience with the bot will be defined by these phrases or questions you put to the AI; it's a classic "what you reap, you sow" situation. There are a few basic tips you need to keep in mind when formulating your prompts to get the best responses possible.
- Be clear and specific: Your prompt should clearly state what you want to discuss or learn about. Without a clear prompt, ChatGPT cannot understand what you're looking for. However, being definitive will help the bot provide a more relevant and helpful response. So, don't be afraid to get into it; there's no word limit for you to get into a detailed prompt. But, of course, keep the prompts focused.
- Use open-ended questions: When you ask open-ended questions that don't have a simple yes or no response, it encourages ChatGPT to provide thoughtful and detailed responses, which in turn can lead to more engaging conversations with the bot instead of your run-of-the-mill conversations that sound robotic in its human mimicry.
- Provide context: Providing background information or context, even a little, can prove helpful in getting a more tailored response when you want to ask about specific topics or situations
These are just a few basic tenets you need to keep in mind before crafting your queries for ChatGPT.
Becoming the Master of ChatGPT Prompts
Getting the basics down to a tee is just the tip of the iceberg. To become a true master of ChatGPT, you can employ these prompt engineering techniques and do much more with your prompts.
Persona/ Role Prompting
Asking ChatGPT to play a role instead of answering with its default behavior can elevate the bot in certain situations.
For example, let's assume you have a job interview and want ChatGPT's help preparing. Simply asking it to help you prepare for the interview will get you standard tips you can get anywhere else on the internet.
But if you ask ChatGPT to act as the hiring manager for the position you're interviewing for and add how you want it to behave, you can simulate a mock interview. Although tips are great, mock interviews are way more helpful.
Here is a sample prompt:
I want you to act like you're an interviewer who is interviewing me for the position of an Editor. You will ask me questions about this position as a prospective candidate and only reply as an interviewer. Don't write the entire conversation at once. Ask me questions and wait for my answers. My first sentence is Hello. Good morning.
Similarly, you can ask ChatGPT to play any role. The key is adding details about how you want it to behave in the specific role to the prompt. By asking it to play a role, ChatGPT can even become your language tutor or help you prepare for certain exams. For example, by asking it to act as an IELTS exam evaluator, the exam format and marking indicators for which are readily available on the internet in detail, I enlisted ChatGPT's help to prepare for the Speaking exam. When I simply asked it to help me prepare for the Speaking section, all it did was give me tips. For a proper simulation, you can use a Chrome extension, like Promptheus, that allows you to dictate your queries to ChatGPT.
Sakshi Garg
Zero-Shot Prompts
Since GPT has zero-shot learning capabilities, zero-shot prompts can be extremely helpful. With zero-shot prompts, you provide ChatGPT with a prompt that includes the desired task or context and any relevant information or constraints. ChatGPT can then use its general knowledge and understanding of language patterns to generate an output pertinent to the prompt.
Zero-shot prompting example:
Generate a paragraph about a new movie that is receiving positive reviews. The movie is a drama about a family dealing with a personal tragedy.
With zero-shot prompting, we don't specifically train ChatGPT to generate text about this particular movie or genre. Instead, we rely on its general knowledge of language patterns and relationships to create a response. The output may be less accurate or relevant than with few-shot prompting, which we'll learn about in a bit.
However, using zero-shot prompts with ChatGPT can be a powerful tool for getting valuable responses in certain situations.
Few-Shot Prompts
Few-shot prompting is a type of machine learning similar to zero-shot prompting but with the added ability to perform a task with only a tiny amount of task-specific training data.
In few-shot learning, ChatGPT is trained on a smaller set of examples for a particular task rather than being pre-trained on a large amount of general data, like in zero-shot learning. Hence, it allows the bot to adapt quickly to new tasks with only a few examples.
The main difference between zero-shot and few-shot prompting is that in zero-shot prompting, the model has not been explicitly trained on the specific task or context but rather leverages its general knowledge to generate a response. In few-shot prompting, the model has been trained on a small set of examples for the specific task, allowing it to generate more accurate and relevant responses.
Few-shot prompting example:
Generate a paragraph about a new movie that is receiving positive reviews. Here are two example reviews:
1. "This movie is a powerful drama that explores the complex dynamics of a family dealing with a personal tragedy. The acting is superb and the script is emotionally engaging."
2. "This movie is a slow-paced character study that delves into the inner lives of its characters. While it may not be for everyone, those who appreciate thoughtful and introspective dramas will find much to admire."
Use these reviews as a basis for your response.
With few-shot prompting, we trained ChatGPT on a small set of examples (the two reviews) for generating text about the specific movie genre and context.
Overall, few-shot prompting can be a more powerful tool than zero-shot prompting for tasks that require a more specific or nuanced understanding of language patterns. Still, it may need more task-specific training data.
So, depending on the complexity of your task, you can adapt one of these techniques. If the task is more complex, few-shot prompting will get you to desired results quicker, but for lesser complex tasks, zero-shot prompting would suffice.
Few-shot prompting can be extremely useful when you want ChatGPT to solve complex math problems. While ChatGPT-4 is already good at solving complex mathematical problems, especially with its Code Interpreter plugin model, ChatGPT-3.5 has problems with it. But few-shot prompting can even get GPT-3.5 to solve them correctly when you give it an example of how to solve a similar problem.
Chain-of-Thought Prompting
Chain-of-thought prompting involves generating a series of prompts that build on each other to guide ChatGPT's output toward a specific topic or line of thought. Each subsequent prompt is designed to build on the previous one.
This type of prompting can be useful in scenarios where the desired output requires a more complex or multi-step thought process, such as generating a detailed response to a complex question or providing step-by-step instructions for a task.
The following prompt, for example, guides ChatGPT through the problem step-by-step, building upon previous responses to arrive at the final answer:
A store sells apples for $0.50 each and oranges for $0.75 each. If John buys 6 apples and 3 oranges, how much does he pay in total? Response options (after each step):
1. John buys 6 apples for $____
2. John buys 3 oranges for $____
3. John pays a total of $____
ChatGPT would accurately give the following answer:
1. John buys 6 apples for $3.00
2. John buys 3 oranges for $2.25
3. John pays a total of $5.25
Users can get confused between the chain-of-thought and few-shot prompting since they seem similar. The main difference between the two is that few-shot prompting is used to improve a model's ability to generate output by providing training data. In contrast, chain-of-thought prompting guides a model's thought process toward generating a specific output.
Moreover, both techniques can be effective for solving math problems, but the choice of technique may depend on the specific problem and the individual using it. For simpler math problems, few-shot prompting may be sufficient. However, chain-of-thought prompting may be more effective for more complex problems requiring a deeper understanding of mathematical concepts and relationships between variables.
Add Personality and Knowledge
You can add personality to your prompts by using language and phrasing that reflects a particular tone, style, or attitude. This can make the outputs more engaging or relatable for the user. Adding personality can involve using humor, sarcasm, or other stylistic elements. To add more personality to it, you can add adjectives like clever, witty, humorous, lighthearted, etc., that'll make the response drastically different from its standard formal and robotic responses.
An example of such prompts can be:
Write a satirical 500-word post explaining why robots will never replace humans in the kitchen. Write as an AI expert who has worked with robot chefs for over a decade and use humorous examples to support your argument.
In addition to adding personality, you can augment ChatGPT's responses with generated knowledge. Generating knowledge refers to using ChatGPT to generate new knowledge or insights based on available data. For example, you can ask it to analyze a large corpus of text and generate summaries, key insights, or even predictions based on the patterns and relationships found in the data. You can even provide it with links to articles on the web and ask it to summarize them. An example of knowledge generation can look like this:
Generate 5 facts about why we need humans in the kitchen.
Then, you can ask ChatGPT to use the knowledge it generated in the previous response in its subsequent response.
Although they're great by themselves, by combining these two approaches, we can create prompts that not only guide ChatGPT toward generating a specific output but also infuse it with a personality to make it more engaging.
Use the above facts and write a satirical 500-word post explaining why robots will never replace humans in the kitchen. Write as an AI expert who has worked with robot chefs for over a decade and use humorous examples to support your argument.
ChatGPT does generate a blog post that's satiric, humorous and isn't boring to read.
Additional Tips
While prompt engineering techniques are particularly important in tasks that require a high degree of precision or accuracy, such as generating medical diagnoses or legal briefs, there are other (less technical) tips that can help you make ChatGPT responses more fun and productive.
Get Answers in Tabular Form
You can ask ChatGPT to give answers in the form of a table which can be extremely helpful when you're asking for a lot of information.
For example, if you want ChatGPT's help to generate a meal plan for the week, asking it to do so in tabular form can keep the information better organized.
ChatGPT can even make changes to the table if you ask it to, or you can get it in a format that you can use in other apps.
Set Limits or Constraints
Setting limits on ChatGPT's outputs can make them more interesting or efficient. Typical constraints would like asking it to write a blog post under 500 words. This would ensure that the bot does not unnecessarily fluffs the piece.
But you can even get creative. For example, you can ask it to write a paragraph in alphabetical order or where each word starts with the letter that the previous word ended with.
Ask ChatGPT to Answer in the Style of an Author
Similar to giving ChatGPT a personality, you can ask it to speak in the style of your favorite author. Of course, it can't mimic their genius, but it can mimic their style rather closely.
Use Promptpal
If you still need inspiration in crafting prompts for ChatGPT, you can check out Promptpal. It is a platform where users can share and discover prompts for AI platforms like ChatGPT, Bing, Bard, Midjourney, Dall-E, etc.
It has collections of prompts for different categories, like Legal, Marketing, Design, Art, Real Estate, Development, etc. And most of the prompts are available freely. Moreover, since it is a sharing platform, new prompts are being added constantly.
Mastering ChatGPT Prompts can help you leverage the abilities of this amazing bot far better than users who don't put any time into learning this skill. And yes, it is a skill that will undoubtedly have a massive impact in the future.
Member discussion