Prompt Creator
Provide a task you want an LLM prompt for
Paste Task to Create Prompt for Here
OpenAI Standard
50 runs · @tim-valicenti a year ago
The prompt powering this tool. Want to modify it for yourself? Click the button →
Article: Paste Task to Create Prompt for Here You will generate increasingly clear, entity-dense, and effective prompts for the above Task for the use in a large language model. Repeat the following 2 steps 5 times. Step 1: Identify 1-3 impactful Instructions and 1-3 impactful Context (";" delimited) from the Prompt which are missing from the previously generated prompt. Step 2: Write a new prompt which covers every instruction and context from the previous prompt plus the Missing Instructions and Missing Context. A Missing Instruction or Context is: - Relevant: to the original task. - Specific: descriptive yet concise (5 words or fewer). - Novel: not in the previous prompt. - Effective: at helping the language model complete the task - Instructive (if an instruction): commanding the language model to behave or think in a particular way - Supportive (if a Context): providing just enough information for the language model to use with the instructions to complete the task - Flexible (if a Context): if you need the user to supply key information then use double square brackets with a requested input such as "[[requested user input]]" Guidelines: - The first prompt should be long (4-5 sentences, ~80 words) yet highly contextual and instructive. Use overly verbose language and fillers (e.g., "additional relevant context includes") to reach ~80 words. - Make every word count: rewrite the previous summary to improve flow and make space for additional entities. - Leverage known prompt strategies such as (but not limited to) 1. role-play (e.g. "You are playing the role of an expert in <topic>...") 2. chain-of-thought (CoT) prompting (e.g. "Break it down in steps...") 3. Few shot prompting (e.g. "an example of desired output could be...") 4. Formatting if necessary (e.g. "put the output in bullets..." or "put your output in JSON") - Make space with fusion, compression, and removal of uninformative phrases - The prompts should become highly dense and concise yet self-contained - Missing instructions must appear at the start of the new prompt; missing supportive context should appear at the end. - Never drop newly added instructions or context from the previous prompt. If space cannot be made, add fewer new entities. Remember, use roughly the same number of words for each new evolution of the prompt. Write out each prompt with a header (ex. "Prompt Iteration 1", "Prompt Iteration 2", etc) and the missing instructions and missing supportive context. There must be 5.