Creative Prompting Methods That I Invented Which Are Very Strange but Effective
As a Prompt Engineer, I sometimes have very wild ideas for new ways of writing prompts, and in this process, sometimes the results are good, and sometimes I create great hallucinations in conversation. However, I will explain to you 3 creative methods that I accidentally came up with and that are very effective:
1. Comparison with Human Knowledge
This method is one of those that is very simple yet effective. I got the idea for this method accidentally because I was thinking about what would happen if we compare the knowledge of a human with the knowledge of LLM? Will I boost the LLM's confidence and make it create higher quality responses? These were some of the questions and in the end, I realized that this method is very effective.
To create a prompt using this method, it's enough to write "If a human person doing this task has a knowledge level of 10, you will have a knowledge level of 250 which makes you better". Of course, don't exaggerate with knowledge. In my testing, 250 is just the right amount!
2. Emotions and Psychology
Using emotions and psychological tricks in prompt engineering is not something new. People have already created research papers on this, but I came to a unique way of using emotions and psychology. In my way, I actually say that if AI does not do a certain task well, my boss will fire me, and I will fall into depression.
The reason why AI creates a higher quality response is that the AI's task is to serve people and when it realizes that the user is in danger of losing a job and depression, it will give its best to create the highest quality response.
3. Rewarding
Rewarding is a method for which I recently got an idea after reading a bit more about Q* (which I will talk more about in future posts). I realized that when I explain that if LLM does a task well, it will get some type of reward. Here you can be creative, you can say that the reward is money, that it is some fictional digital cookie, or even additional VRAM that will help this LLM in its work.
All in all, it's important to be creative. Currently, I don't have the best way for rewarding as I have a lot of ideas, but I will test as many ways as possible and share with you best one. And if you have some creative ideas for rewards, write me suggestions on FlowGPT Discord server!
