PLAN.md

Infinite Joke Agent Plan

Objective

To create a system that continuously generates, evaluates, and adds new one-liner jokes to this repository, with the ultimate goal of discovering the funniest joke in the world.

Phases

Phase 1: Joke Generation

• Use a large language model (LLM) to generate new one-liner jokes.
• The jokes should be in the style of the existing jokes in src/jokes.js.
• The output should be a single string.

Phase 2: Joke Integration

• Create a script that takes a new joke as input.
• The script will read src/jokes.js.
• It will parse the JavaScript array of jokes.
• It will add the new joke to the array.
• It will write the updated array back to src/jokes.js, maintaining the existing formatting.

Phase 3: Joke Evaluation

• Develop a rating system for the jokes.
• The rating system will be based on an LLM evaluation.
• The LLM will be prompted to rate the joke on a scale of 1 to 10 for funniness, originality, and cleverness.
• The ratings will be stored in a separate file, perhaps jokes-ratings.json, mapping each joke to its rating.

Phase 4: The Infinite Loop

• Create a main script that orchestrates the process.
• The script will:
  1. Generate a new joke.
  2. Integrate the joke into src/jokes.js.
  3. Evaluate the joke and store the rating.
  4. Repeat.
• The process will run until stopped manually.

Finding the Funniest Joke

• The "funniest joke in the world" is a moving target.
• We will define "funniest" as the joke with the highest rating in our system.
• The agent will continuously strive to generate a joke with a higher rating than the current best.

Future Enhancements

• User feedback mechanism for rating jokes.
• A web interface to display the jokes and their ratings.
• More sophisticated joke generation techniques.