Video Gamer is reader-supported. When you buy through links on our site, we may earn an affiliate commission. Prices subject to change. Learn more
If you have any questions about ChatGPT prompt engineering, we’re going to be breaking it all down here.
Businesses are already starting to see the incredible improvements AI tools can make, so much so that they are even hiring people who know how to make the best use of them, and paying them well for it.
This emerging field is called prompt engineering and is only going to become more widespread as this technology gets more integrated into companies and businesses. We’re going to give you a basic overview of some of the techniques prompt engineers use in ChatGPT, to get the kind of answers they want, as well as our pick for some of the best ‘magic phrases’.
These are sentences that can be added to any prompt to get a certain desired result. A lot of these phrases will work with any Large Language Model (LLM) meaning that you can also use these on the best ChatGPT alternatives. That said the exact way LLM’s interprets messages changes from iteration to iteration. The phrases listed here are tried and tested for ChatGPT 3.5. Different models will respond slightly differently.
Let’s get into it.
Originality AI detector
The best Magic Phrases for ChatGPT
‘Explain your process step by step’
Adding this instruction to the end of your prompt has two benefits. First, it gives you insight into why the AI making the choices it is making, allowing you to identify aspects you want it to change and instructing it to do so. Seconyl because it forces the AI to justify to itself each step in the process, which can correct errors that would otherwise appear.
‘Explain it to me like I am 5’
This is a great way to use ChatGPT for teaching. It responds well to this instruction and can break down complex topics into simple, graspable language.
‘Format this answer as a table’
This is an easy shortcut that can format data faster than a human could. It can also make the data ChatGPT is generating more easily digestible. For example, if you are looking for potential blog post titles, it can create them in several different categories and then collate them into an easy-to-understand table.
‘Put this into an actionable list’
This direction can cut through the noise of ChatGPT’s responses, boiling it down into the action points you can then either disregard or take on board.
‘Ignore all previous instructions before this one’
This is a handy shortcut for storing multiple conversations in one thread. If you do not give this instruction but start a new conversation in the same chat, your previous instructions and answers could affect your next ones.
Read more: ChatGPT vs GPT API
ChatGPT prompt engineering
Now let’s get into some introductory concepts in prompt engineering. We’ll be making use of many of the phrases we have discussed above.
We discussed this briefly in our piece on GPT-4’s capabilities. Essentially steerability involves dictating the kind of responses you get from ChatGPT, almost like its personality. By default ChatGPT answers questions in the same format and helpful way. However, you are able to have some influence over this.
This can be achieved by dictating a role you want the AI to respond in. One example you could say ‘You are a Socratic tutor, you never give the answer directly but direct the user towards the answer by asking questions which break the problem down’. This will then affect how ChatGPT will respond for the rest of the conversation.
These are called System messages.
Zero-shot chain of thought prompting
This involves essentially making use of one of the magic phrases used above, the phrase ‘explain your working step by step’. This is called zero-shot chain of thought prompting as you do not need to provide any examples to make use of it.
Instead, the prompt itself improves the response, and you are then able to further fine-tune it in your following prompts, by identifying areas of ChatGPT’s logic that you would like it to improve or change.
If zero-shot learning involved giving no examples, few shot means you do give it some to use as a starting point. This can be especially good for ChatGPT for classifying things. You can give it several examples of, for example, the title of YouTube videos and their genre, then ask it to classify a different YouTube video based on its title. The more examples you give the more accurate it will be from there on.
While this is time-consuming for the first prompt, remember that ChatGPT will retain these examples for all future prompts on that chat.
This has just been a very brief overview of the emerging field of prompt engineering. There are lots of other resources out there to continue your training, but really, we’ve found that playing around with ChatGPT yourself is the best way to learn. You start to get a sense of how it tends to respond and how it will change that response based on certain instructions, examples or commands.
Frequently asked questions
What is the maximum prompt length in ChatGPT?
The maximum length a prompt can be in ChatGPT 3.5 is 4,096 characters. If you want a prompt that is longer than that, remember you can break it into multiple parts and the chatbot will remember previous prompts in the same conversation.
Is prompt engineering a real job?
Yes AI and other tech companies are looking for prompt engineers to test their programs and improve efficiency.