Skip to main content
PromptWizard is a feedback-driven optimizer that improves prompts through a multi-stage process. It mutates prompts with different “meta-styles,” critiques failures, and then refines the prompts based on that feedback, using beam search to select the best candidates.

When to Use PromptWizard

✅ Best For

  • Task-aware prompt refinement
  • Creative domains and content generation
  • Improving prompt style and meta-instructions
  • Tasks requiring reasoning

❌ Not Ideal For

  • Quick, simple optimizations
  • When teacher model quality is low
  • Projects with tight computational budgets

How It Works

PromptWizard uses a cycle of mutation, critique, and refinement:
  1. Mutation: A teacher model generates prompt variations by mixing in different meta-prompts and styles (e.g., “Think step-by-step”).
  2. Critique: For prompts that perform poorly, the teacher generates detailed reasons for the failure.
  3. Refinement: The critiques are used to generate improved versions of the prompts.
  4. Selection: Beam search is used to evaluate all candidates and select the top performers for the next round.
1

Mutate

Generate creative variations of the prompt with different styles.
2

Critique

Analyze low-scoring prompts and generate failure critiques.
3

Refine

Use critiques to generate improved prompts.
4

Update Beam

Score all candidates and select the top N to continue.

Underlying Research

PromptWizard is based on the concept of self-evolving prompts, where an LLM iteratively improves its own instructions.

Next Steps