Loading...
🔄
Reflexion(RX)
Agents learn from experience through verbal self-reflection and episodic memory
Complexity: highReflection
🎯 30-Second Overview
Pattern: Agents learn from experience through verbal self-reflection
Why: Converts failures into linguistic insights, enabling learning without fine-tuning
Key Insight: Actor → Evaluator → Self-Reflection → Memory → Improved Performance
⚡ Quick Implementation
1Actor Executes:Generate actions/text for task
2Environment Responds:Get feedback/results
3Evaluator Scores:Assess performance objectively
4Self-Reflect:Generate verbal reinforcement
5Update Memory:Store reflection for next trial
Example: Actor(task) → Result → Evaluate → Reflect(verbal) → Memory → Actor(improved)
📋 Do's & Don'ts
✅Convert feedback into actionable linguistic hints
✅Maintain episodic memory of past reflections
✅Use specific, concrete reflection prompts
✅Implement clear evaluation metrics
✅Limit reflection depth to avoid overthinking
❌Store raw trajectories without reflection
❌Use vague reflections like "do better"
❌Skip environment feedback integration
❌Allow unlimited reflection cycles
❌Ignore memory constraints in long tasks
🚦 When to Use
Use When
- • Multi-step reasoning tasks
- • Programming challenges
- • Sequential decision making
- • Learning from failures
Avoid When
- • Single-shot tasks
- • Time-critical applications
- • Simple Q&A
- • Deterministic problems
📊 Key Metrics
Success Rate
Task completion over trials
Learning Curve
Performance improvement rate
Reflection Quality
Actionability of insights
Memory Efficiency
Relevant recall vs storage
Convergence Speed
Trials to optimal performance
Generalization
Transfer to similar tasks
💡 Top Use Cases
Code Generation: Write → Test → Reflect on errors → Fix bugs → Improve
Game Playing: Move → Observe → Evaluate → Reflect on strategy → Adapt
Problem Solving: Attempt → Check → Analyze failure → Learn → Retry
Writing Tasks: Draft → Review → Reflect on weaknesses → Revise → Polish
Agent Training: Act → Get reward → Reflect on decisions → Update policy
References & Further Reading
Deepen your understanding with these curated resources
Academic Papers
Contribute to this collection
Know a great resource? Submit a pull request to add it.
🔄
Reflexion(RX)
Agents learn from experience through verbal self-reflection and episodic memory
Complexity: highReflection
🎯 30-Second Overview
Pattern: Agents learn from experience through verbal self-reflection
Why: Converts failures into linguistic insights, enabling learning without fine-tuning
Key Insight: Actor → Evaluator → Self-Reflection → Memory → Improved Performance
⚡ Quick Implementation
1Actor Executes:Generate actions/text for task
2Environment Responds:Get feedback/results
3Evaluator Scores:Assess performance objectively
4Self-Reflect:Generate verbal reinforcement
5Update Memory:Store reflection for next trial
Example: Actor(task) → Result → Evaluate → Reflect(verbal) → Memory → Actor(improved)
📋 Do's & Don'ts
✅Convert feedback into actionable linguistic hints
✅Maintain episodic memory of past reflections
✅Use specific, concrete reflection prompts
✅Implement clear evaluation metrics
✅Limit reflection depth to avoid overthinking
❌Store raw trajectories without reflection
❌Use vague reflections like "do better"
❌Skip environment feedback integration
❌Allow unlimited reflection cycles
❌Ignore memory constraints in long tasks
🚦 When to Use
Use When
- • Multi-step reasoning tasks
- • Programming challenges
- • Sequential decision making
- • Learning from failures
Avoid When
- • Single-shot tasks
- • Time-critical applications
- • Simple Q&A
- • Deterministic problems
📊 Key Metrics
Success Rate
Task completion over trials
Learning Curve
Performance improvement rate
Reflection Quality
Actionability of insights
Memory Efficiency
Relevant recall vs storage
Convergence Speed
Trials to optimal performance
Generalization
Transfer to similar tasks
💡 Top Use Cases
Code Generation: Write → Test → Reflect on errors → Fix bugs → Improve
Game Playing: Move → Observe → Evaluate → Reflect on strategy → Adapt
Problem Solving: Attempt → Check → Analyze failure → Learn → Retry
Writing Tasks: Draft → Review → Reflect on weaknesses → Revise → Polish
Agent Training: Act → Get reward → Reflect on decisions → Update policy
References & Further Reading
Deepen your understanding with these curated resources
Academic Papers
Contribute to this collection
Know a great resource? Submit a pull request to add it.