Question 1

Why are utility-based agents more advanced?

❌ They select any action that completes a goal
❌ They update behavior via trial-and-error
❌ They avoid internal models
✅ They evaluate and rank outcomes using a utility function before selecting an action

Explanation:
Utility-based agents assess trade-offs and choose the best action, not just any goal-satisfying one.

---

Question 2

Advantage of a reflection loop:

✅ It allows the AI to self-critique and iteratively improve responses
❌ Removes human oversight
❌ Prevents incorrect facts completely
❌ Reduces API calls

Explanation:
Reflection improves quality via critique → revision cycles.

---

Question 3

Next step in Reflexion workflow (“foods high in iron”):

❌ Generate new random response
❌ Return initial answer
❌ Start a new agent
✅ Send response to a tool for citations, then revisor updates answer with source-backed information

Explanation:
Reflexion improves reliability by combining tools + revision steps.

---

Question 4

Improving health advice with citations:

❌ Manually add citations
✅ Pass both initial output + tool data to the revisor for a citation-backed revision
❌ Restart responder
❌ Remove constraints

Explanation:
Revisor integrates authoritative references and corrects unsafe advice.

---

Question 5

Purpose of tool_calls in AIMessage:

❌ Log queries
❌ Run revisor’s system message
❌ Define entry/exit nodes
✅ Store key-value outputs like answer, reflection, search queries

Explanation:
tool_calls holds structured tool outputs needed for the Reflexion cycle.

---

Question 6

Correct first step in ReAct:

✅ Think: “I need to check Zurich’s weather”
❌ Recommend clothing immediately
❌ Ask for closet inventory
❌ Query clothing tool first

Explanation:
ReAct always begins with THINK → ACT → OBSERVE.

---

Question 7

Purpose of the “Observation” step:

❌ Log user history
✅ Show the tool’s returned result during graph execution
❌ Display final answer
❌ Evaluate reasoning

Explanation:
Observation = tool output, which the LLM uses for the next reasoning step.

---

🧾 Summary Table