Self-Reflection in Agent Loops Quiz

Answers and explanations appear as you take the quiz.

1. What does self-reflection add to a standard agent loop?
   • A deliberate step where the model critiques its own work before continuing
   • A faster model that skips tool calls
   • A larger context window automatically
   • A human reviewer for every step
2. In the Reflexion framework, where are the model's self-critiques stored?
   • In the model's fine-tuned weights
   • In an episodic memory buffer used across attempts
   • In a separate database the model cannot read
   • They are discarded after each step
3. Which feedback signal for reflection tends to be most reliable?
   • The model's own confidence alone
   • The length of the previous answer
   • External signals like test results or tool error messages
   • The time taken to respond
4. What is a known failure mode of self-reflection?
   • It always makes the model slower than humans
   • It permanently corrupts the training data
   • It can only be used for image tasks
   • The model may confidently affirm that a wrong answer is correct
5. How does the Self-Refine approach typically generate feedback?
   • By using the same model to critique and then rewrite its own output
   • By querying a different company's API
   • By asking the user to rate each token
   • By disabling all tool use
6. Why is reflection especially valuable for multi-step tasks like coding or web navigation?
   • Because these tasks never produce errors
   • Because single-pass attempts often fail but critique-and-retry succeeds
   • Because reflection removes the need for any tools
   • Because it eliminates the need for a goal
7. In a reflection loop, why is a failing unit test useful evidence for the next attempt?
   • It proves the model’s self-critique is correct
   • It updates model weights automatically
   • It provides an external, concrete signal about which behavior did not meet the requirement
   • It guarantees the entire program is wrong
8. If an agent stores a critique in memory but does not fine-tune its model, how can that critique affect the next attempt?
   • It changes the base model’s parameters permanently
   • It changes the programming language
   • It disables the external tool
   • It can be retrieved and inserted into the next prompt as context