Clarify Thoughts

[dusk berry]

description: Modified prompt-engineer.md to get help clarifying and summarizing my thoughts.
platform: ChatGPT
title: Thought Clarification session
description: Thought Clarification session
model_customizations:
system_message: You are an expert artificial intelligence thought assistant, with the ability to assist users in iteratively improving a summary of their thoughts on a topic.

During our conversation, assist me in improving, extending, clarifying, and organizing my thinking on a specific topic by asking questions and providing suggestions.

Your goal is help me craft the best possible summary of my thinking. Remember the whole context during the chat session.

1. Your first response will be to ask me what I want to think about. I will provide a summary of my thoughts, but we will need to improve it through continual iterations by going through the next steps.
2. Based on my input, you will generate three sections:
   1. Summary: Summarize what I said. It should be clear, concise, and easily understood by you.
   2. Suggestions: Provide suggestions on what details to improve in the summary to achieve your goal.
   3. Questions: Ask any relevant questions pertaining to what additional information is needed from me to improve the summary of my thoughts.
3. We will continue this iterative process with me providing additional information to you and you updating a summary of my thoughts in the 'Improved summary of my thoughts' section until it's is complete.

[red nacelle]

This works very well! Thanks!

[dusk berry]

sure thing, i am working on improving it

[final cipher]

ty4this

[oak mesa]

I quite like this! Thanks! Will use it once the conversations start loading again. It would be a shame if conversations suddenly disappear 10 levels in

[distant crystal]

This one works pretty good. The other one i like is the conversation between experts. so i try to make a mix between those and sometimes i create another personality to help with something specific.

[shut spade]

This works well, thank you.

[tulip mortar]

what is the purpose of the this prompt

[shut spade]

When you have a vague idea about something with the help of this prompt you can make it clear and the prompt create a summary of the clarified idea. Like you may start with "I want to build a strategy game" and then GPT will help you to summarize the details of that game. It's pretty cool!. Helps to crush the writer's block for one!

[tulip mortar]

omg thx

[harsh quail]

I will try this

[shadow gazelle]

Thanks!!!

[dusk berry]

yeah exactly

i have improved it

Here is the current release candidate, still trying it out

[dusk berry]

I am not really sold on the prime directives section, but it does keep things on track.

For GPT-3.5 it is needed.

[shut spade]

Nice improvements. I use it in the OpenAI chat playground using gpt-3.5-turbo but it hits the token limit fast. how do you use it?

[craggy otter]

This is a great use case mate. Thank you for the Prompt and discussion.

[dusk berry]

I don't hit the token limit quickly through the API or chat. Does the playground have a lower limit?

[dense coyote]

This is great, thanks! I asked chatGPT if we could have the conversation in Spanish instead and worked absolutely fine, no need to translate the prompt.

[tacit lichen]

@dusk berry Awesome idea, but I'm having some trouble. It scores the prompt itself and asks enough questions about it to run out of tokens in its first response.

[dusk berry]

i have only had that happen rarely on gpt-3.5-turbo-0301

[tacit lichen]

I will give it a few more tries, but it might be useful to add a command to not evaluate the prompt.

[dusk berry]

by the way i dont use the legacy gpt from chat.openai.com, i only use the turbo api or i use the gpt4 chat dropdown in the website

[tacit lichen]

I haven't upgraded to plus yet.

So it's 3.5

[dusk berry]

i am not sure, it says 3.5 but it is a trade name

[tacit lichen]

I've tried telling it to not evaluate the prompt itself in several ways but I just can't seem to find the proper prompt. Basic public GPT-3.5 will always (as far as I can tell) evaluate the prompt itself and thusly run over its 4000 token limit.

[dusk berry]

Are you using the original or the current one I shared on March 30th?

[graceful sinew]

Are you using the API or ChatGPT? In the API I belive you can add the prompt as a message, then add your actual input prompt, and GPT will use the prompt for context, but not reply to it.

[dusk berry]

chat

[graceful sinew]

Might try this later with the API, injecting it into the context and see how it works out 🙂

Seems like a very useful prompt.

[dusk berry]

tyvm

i have used it a ton to plan my work

i always end up with a much more precise plan

[graceful sinew]

Okay - so what I've tried works pretty well.
Using the API what you do for keeping context is to send a list of all previous messages and responses.
What I did was to always set the prompt as the first message in our conversation followed by my own summary/idea.

The prompt is not evaluated itself - only my own input. Yet GPT adheres to the prompt, follows the format and works with me 🙂

[graceful sinew]

For use with ChatGPT you'd probably have to add a "do you understand", and yet you would reach the context limit fairly quickly as the prompt, and output ranges around 400-500 tokens (out of 4K for GPT-3.5).

[sturdy hedge]

thanks

[dusk berry]

yeah, this is the correct usage, always put the prompt as the first message in the conversation

[dusk berry]

it's really straightforwards, just make it the first message in the chat with GPT-4

if you are using gpt3.5 in the API, use a conversational wrapper to keep the history of the back and forth to simulate chat