Reflection

[haesleinhuepf]

We might explore how reflection could improve code quality of generated code. I recently saw multiple examples where agents were enabed to produce higher-quality output because the result of the first attempt was fed back to the agent asking for improving it. It would require implementing a feedback loop. It is also possible to do this in a multi-agent setting. chatGPT produces code and we ask gemini to read over it.

Read more:

• https://blog.ml6.eu/reflexion-is-all-you-need-ca36ceceef0e
• https://arxiv.org/abs/2303.11366
• https://promptengineering.org/reflexion-an-iterative-approach-to-llm-problem-solving/