Reflexion – how AI models critique their own answers

Reflexion

This technique lets an AI model analyze and improve its own answers. After producing an initial answer, the model "reflects" on it, judging its quality, completeness and relevance, and then suggests improvements.

AdvantagesDrawbacks
Raises the quality and accuracy of answersCan increase the time it takes to generate an answer
Lets the model fix its own mistakesPotentially more resource-hungry
Encourages a deeper understanding of the taskCan complicate debugging and control

Example of a Reflexion prompt:

Answer the following question: 'What factors influence global warming?'
Once you have given your answer, analyze it against these criteria:
1. Completeness of the information
2. Scientific accuracy
3. Clarity of presentation

Then suggest improvements to your initial answer based on that analysis.

Try it in GPTunneL