Quiz 9 - Practical Lessons from Deploying Real-World AI Agents [11/10]

JavaScript isn't enabled in your browser, so this file can't be opened. Enable and reload.

Email *

1. What is the primary challenge in moving AI agents from simple demos to real-world, enterprise-grade products, according to the lecture?

1 point

(A) Finding a powerful enough foundation model that can reason.

(B) Developing effective RAG (Retrieval-Augmented Generation) techniques.

(C) Building the massive, complex infrastructure for reliability, safety, and observability that goes far beyond the core AI components.

(D) Designing a compelling user interface for the agent.

2. Why are traditional software testing methods (like unit tests) insufficient for AI agents?

1 point

(A) Because agents are too slow to run in a standard CI/CD pipeline.

(B) Because traditional tests cannot handle the Python-based tools agents use.

(C) Because agents are non-deterministic, requiring new testing methods like user simulation (τ-bench) to check for reliability and policy adherence in dynamic conversations.

(D) Because agents can only be tested by human users in real-world scenarios.

3. What is a key technical challenge in building production-level voice agents?

1 point

(A) Finding an end-to-end, audio-to-audio model that is fast enough for real-time conversation.

(B) Standard metrics like Word Error Rate (WER) are "terrible" because they incorrectly penalize successful filtering of background noise, requiring new quality measurements.

(D) It is technically impossible to handle multiple speakers or background noise.

4. The lecture contrasts "transactional" agents with more advanced "relational" agents. What is the key technical capability that enables this shift?

1 point

(A) The ability to use more complex and varied tools from a larger library.

(B) A more sophisticated RAG system that can query larger document stores.

(C) An Agent Data Platform that provides long-term memory and integrates with Customer Data Platforms (CDPs) for proactive, context-aware engagement.

(D) A more "human-like" and empathetic voice model.

5. How does the security risk for a deployed AI agent differ from traditional software?

1 point

(A) AI agents are inherently safer because they can reason about and prevent security threats.

(B) The security risks are identical to traditional software, focusing only on issues like SQL injection.

(D) Agents face compounded vulnerabilities: all "classic software vulnerabilities" plus new "AI vulnerabilities" like prompt injection, requiring a layered defense of guardrails and supervisors.

A copy of your responses will be emailed to the address you provided.

Submit

Clear form

reCAPTCHA

Privacy Terms

This form was created inside of UC Berkeley.

Does this form look suspicious? Report

Forms