Here are a few impressive points from youtube (AI prompt engineering: A deep dive) where Anthropic engineers shared their prompt writing tips and experiences.

Impressive Prompt Writing Tips

• When the model makes a mistake, it can be helpful to ask why it was wrong and directly ask the model how to instruct it next time to avoid the mistake.
• When the model faces unexpected input or ambiguous situations, you should provide clear instructions on what to do (e.g., output an "UNCERTAIN" tag) to prevent it from giving a wrong answer.
• Inducing the model to explain its reasoning process before providing an answer (Chain of Thought) improves the model's accuracy.
• You can ask the model to "interview" you. When it's difficult to clearly grasp what you want, you can get help constructing a prompt by asking the model to ask you questions and elicit the necessary information.

Tips for Improving Prompt Engineering Skills

• Read many other people's prompts and model outputs: You can learn a lot by looking at great prompts, analyzing their structure and intent, and by carefully observing the model's outputs.
• Attempt many conversations with the model: You need to practice improving prompts by directly and repeatedly communicating with the model.
• Show your prompts to others: Showing a prompt you've written to someone without prior knowledge of the task can help you discover parts that are not clear.
• Test the model's limits: Trying to do the most difficult tasks you think the model cannot do and exploring its limits leads to the greatest learning.