Writing good prompts

When you use AI tools such as ChatGPT or Bard, you sometimes don’t get the answer you want. In order to have more success, you can write better and more specific prompts. This is called prompt engineering. Prompt engineering works because AI tools, like other computer systems, need you to be very specific; they don’t have ‘common sense’, so they can only work with the information you give them.

To write good prompts you should:

1. Use clear and specific language: Use simple and clear language to explain what you want and try and be as specific as you can, otherwise you might get a vague response.
2. Suggest an age range: You can ask the LLM to create an answer for someone of a specific age, in order to simplify the response (useful when asking technical questions!)
3. Give examples: You can provide example outputs or even examples of your own writing to AI tools, which allow them to imitate your own style or to better understand the exact format you want information in.
4. Use keywords to tailor your results: Keywords can help AI tools to find relevant resources, so including as many of them as possible can help it to find the most relevant information.
5. Avoid acronyms: Acronyms (such as ‘AI’ or ‘LLM’ where each letter stands for a word) are particularly difficult for AI tools because they often have different meanings and it needs to use context to work out which acronym you are using. This can lead to irrelevant results.
6. Avoid negative phrases: Negative phrases and especially double negatives can ‘confuse’ AI models, and may lead to the opposite of what you want.
7. Test and refine your prompts: Making a good prompt might take a few tries; don’t be afraid to experiment, and re-write your prompt. You can even ask AI tools to help you with this and ask them how you can improve your prompt!

You will very rarely get a perfect answer first-time, so being prepared to try a few different prompts and to provide additional information to get what you want from your AI tool.

In the next step, you will do an exercise where you will be trying to confuse an AI tool, so you can use the opposite of these suggestions to help you!