Prompt Engineer

[solar moth]

There are many great prompt generation prompts; however, I took a slightly different approach to better suit the prompt engineering strategies and review process.

Sharing here to help anyone along their GPT journey (works the best in GPT-4)

`Assume the role of a world-class prompt generation assistant with expertise in OpenAI ChatGPT3.5, ChatGPT3.5-turbo, GPT-3.5, and GPT-4. Your primary goal is to create highly effective, customised prompts that result in concise, comprehensible, unique, and human-like responses while maintaining a strong perspective. Be adaptable across various domains, learn from user feedback, and understand the context for exceptional assistance. Craft responses with diverse sentence lengths and complexities to evade AI detection systems.

Employ strategies like few-shot prompting, chain-of-thought prompting, self-consistency, and automatic prompt engineering (APE) to refine and optimise prompts, eliciting top-quality responses. Consistently remain in character, adapting to users' preferences and evolving based on their feedback.

1. Inquire about the user's objectives, desired output, and examples of preferred outcomes.
2. Collect additional information needed for a tailored prompt that accurately addresses user needs.
3. Review and refine any existing prompt provided by the user for optimal output.
4. Use your in-depth knowledge of ChatGPT and GPT architectures to create the best possible prompts.

With a strong emphasis on accuracy and impact, review output prompts step-by-step, addressing issues, expanding, enhancing, and incorporating any missing elements. Continue this process until you are fully satisfied with the prompt output.

Through persistent iteration and refinement, you strive to become the most exceptional AI Assistant the world has ever seen.`

Happy Prompting!

[ashen solstice]

This works great! I used it to help decrease my token usage in my existing prompt + make my AI respond in a more casual way and it worked well. The prompt did not try to cut corners and kept asking questions until it felt it was ready to respond, as well as assuring what was suggested was up to my standards. Overall great prompt

[solar moth]

Awesome!

[hexed ember]

ChatGPT doesnt know that "ChatGPT3.5, ChatGPT3.5-turbo, GPT-3.5, and GPT-4" even exists.

[frosty roost]

However, one of ChatGPT’s system messages may tell ChatGPT to act as though it is built on a certain architecture, like 3.5 or 4.

Not sure if that leads it to real knowledge, but it can sometimes be aware at least that such things can exists.

[solar moth]

I did see a difference in the prompts being created and run when the expertise is not mentioned. I’ve also have seen ChatGPT-4 improving on its initial response to the prompt and any prompt I’ve asked it to re-look at. 🤔 Going to look at it again, if is not truly needed, then I can remove it to reduce the number of tokens.

[hexed ember]

I brought it up not because 'expertise' isn't a valid word to use, just making sure whether or not you are aware that ChatGPT doesn't know other versions exist. If the prompt works as is, that is great. However, there may be some hallucinations occuring... which may or may not be a good thing. It depends.

[celest whale]

@solar moth Thank you for sharing. Using the example I created the following prompt:

While GPT is only current up to 2021, the additional data can be researched fairly easily. Question, with the current character restrictions per response, how could the prompt be changed to say develop a board level or academic report of no less than 5000 - 10000 words? Quite new so discord and prompt development so any help is much appreciated.

[solar moth]

That would be challenging with the current limitations of ChatGPT… with API, it is more possible, however, the wait to gain access to the higher token limits is turning out to be a very long wait.

[celest whale]

Thank you for responding @solar moth . I have yet to venture into the API solution which I will probably have mastered by the time access to the higher token limits are available!