Concept Quizzes
Test your understanding of key course concepts with interactive quizzes.
These quizzes are for self-assessment. Your answers are saved locally in your browser.
Your Progress
Loading progress...
Week 1: Introduction to Agentic AI
Test your understanding of agents, ReAct, and trajectories.
What does the 'A' in ReAct stand for?
Correct! ReAct stands for 'Reasoning and Acting' - it combines thought processes with actions.
Not quite. ReAct stands for 'Reasoning and Acting' - it combines thought processes with actions.
Which of the following best describes an AI agent?
Correct! Agents are autonomous systems with perception, reasoning, and action capabilities oriented toward goals.
Not quite. Agents are autonomous systems with perception, reasoning, and action capabilities oriented toward goals.
What is a trajectory in the context of agents?
Correct! A trajectory captures the complete sequence of an agent's interactions during task execution.
Not quite. A trajectory captures the complete sequence of an agent's interactions during task execution.
Which architecture type combines reactive and deliberative approaches?
Correct! Hybrid architectures combine fast reactive responses with deliberative planning capabilities.
Not quite. Hybrid architectures combine fast reactive responses with deliberative planning capabilities.
What is the difference between short-term and long-term memory in agents?
Correct! Short-term memory is the current context window, while long-term memory uses external storage like vector DBs.
Not quite. Short-term memory is the current context window, while long-term memory uses external storage like vector DBs.
Quiz Complete!
You scored 0 out of 5
Week 2: LLM Foundations
Test your understanding of prompting strategies.
What is the main purpose of Chain-of-Thought prompting?
Correct! Chain-of-Thought prompting encourages models to show their reasoning steps, improving accuracy on complex tasks.
Not quite. Chain-of-Thought prompting encourages models to show their reasoning steps, improving accuracy on complex tasks.
How does Self-Consistency improve upon standard Chain-of-Thought?
Correct! Self-Consistency samples multiple reasoning paths at higher temperature and selects the most common answer.
Not quite. Self-Consistency samples multiple reasoning paths at higher temperature and selects the most common answer.
Tree-of-Thoughts differs from Chain-of-Thought by:
Correct! Tree-of-Thoughts explores multiple reasoning paths in a tree structure and evaluates each to find optimal solutions.
Not quite. Tree-of-Thoughts explores multiple reasoning paths in a tree structure and evaluates each to find optimal solutions.
What is zero-shot prompting?
Correct! Zero-shot prompting gives the model a task without any examples, relying on its pre-trained knowledge.
Not quite. Zero-shot prompting gives the model a task without any examples, relying on its pre-trained knowledge.
In few-shot prompting, what role do examples serve?
Correct! Few-shot examples show the model what format and style of output is expected.
Not quite. Few-shot examples show the model what format and style of output is expected.
Quiz Complete!
You scored 0 out of 5
Test your understanding of function calling and tool integration.
What is function calling in the context of LLMs?
Correct! Function calling allows LLMs to generate structured JSON calls to external functions/APIs.
Not quite. Function calling allows LLMs to generate structured JSON calls to external functions/APIs.
What does MCP stand for?
Correct! MCP stands for Model Context Protocol - Anthropic's open protocol for connecting LLMs to tools and data.
Not quite. MCP stands for Model Context Protocol - Anthropic's open protocol for connecting LLMs to tools and data.
What is JSON Schema used for in function calling?
Correct! JSON Schema defines the expected structure, types, and constraints for function call parameters.
Not quite. JSON Schema defines the expected structure, types, and constraints for function call parameters.
What is the benefit of tool descriptions in function calling?
Correct! Tool descriptions help the LLM decide which tool to use and what parameters to provide.
Not quite. Tool descriptions help the LLM decide which tool to use and what parameters to provide.
Quiz Complete!
You scored 0 out of 4
Week 4: Planning and Reasoning
Test your understanding of Reflexion, LATS, and agent memory.
What are the three main components of the Reflexion framework?
Correct! Reflexion uses an Actor (generates actions), Evaluator (provides success/failure signals), and Self-Reflection (generates verbal feedback).
Not quite. Reflexion uses an Actor (generates actions), Evaluator (provides success/failure signals), and Self-Reflection (generates verbal feedback).
How does Reflexion improve agent performance without traditional RL?
Correct! Reflexion uses verbal self-reflection stored in long-term memory, requiring no gradient updates.
Not quite. Reflexion uses verbal self-reflection stored in long-term memory, requiring no gradient updates.
What is LATS (Language Agent Tree Search)?
Correct! LATS applies Monte Carlo Tree Search to explore multiple ReAct reasoning paths and select the best one.
Not quite. LATS applies Monte Carlo Tree Search to explore multiple ReAct reasoning paths and select the best one.
In agent memory systems, what distinguishes episodic memory from semantic memory?
Correct! Episodic memory stores specific experiences and trajectories, while semantic memory stores general knowledge and facts.
Not quite. Episodic memory stores specific experiences and trajectories, while semantic memory stores general knowledge and facts.
What is the key advantage of Plan-and-Solve prompting?
Correct! Plan-and-Solve explicitly separates the planning phase (decomposing problems) from the execution phase (solving subtasks).
Not quite. Plan-and-Solve explicitly separates the planning phase (decomposing problems) from the execution phase (solving subtasks).
Quiz Complete!
You scored 0 out of 5
Week 5: Multi-Agent Systems
Test your understanding of multi-agent architectures.
In a multi-agent system, what is message passing?
Correct! Message passing is how agents communicate by sending and receiving structured messages.
Not quite. Message passing is how agents communicate by sending and receiving structured messages.
What is an orchestrator in multi-agent systems?
Correct! An orchestrator is a central agent that routes tasks and coordinates between specialized agents.
Not quite. An orchestrator is a central agent that routes tasks and coordinates between specialized agents.
What is the advantage of specialized agents over a single general agent?
Correct! Specialized agents can excel at specific tasks, allowing the system to leverage focused expertise.
Not quite. Specialized agents can excel at specific tasks, allowing the system to leverage focused expertise.
What is emergent behavior in multi-agent systems?
Correct! Emergent behavior arises spontaneously from agent interactions without being explicitly programmed.
Not quite. Emergent behavior arises spontaneously from agent interactions without being explicitly programmed.
What is the blackboard architecture in multi-agent systems?
Correct! Blackboard architecture uses a shared workspace that all agents can read from and write to.
Not quite. Blackboard architecture uses a shared workspace that all agents can read from and write to.
Quiz Complete!
You scored 0 out of 5
Week 6: Agent Frameworks
Test your understanding of LangGraph, AutoGen, and CrewAI.
What is a StateGraph in LangGraph?
Correct! StateGraph is LangGraph's core abstraction for defining agent workflows as directed graphs with state management.
Not quite. StateGraph is LangGraph's core abstraction for defining agent workflows as directed graphs with state management.
What does checkpointing enable in agent frameworks?
Correct! Checkpointing saves agent state at specific points, enabling persistence, debugging, and recovery from failures.
Not quite. Checkpointing saves agent state at specific points, enabling persistence, debugging, and recovery from failures.
In AutoGen, what is a ConversableAgent?
Correct! ConversableAgent is AutoGen's base class for agents capable of participating in conversations.
Not quite. ConversableAgent is AutoGen's base class for agents capable of participating in conversations.
What distinguishes CrewAI from other frameworks?
Correct! CrewAI emphasizes role-based teams where agents have defined roles, goals, and backstories.
Not quite. CrewAI emphasizes role-based teams where agents have defined roles, goals, and backstories.
What is a reducer in LangGraph state management?
Correct! Reducers define how state updates from different graph nodes are combined into the unified state.
Not quite. Reducers define how state updates from different graph nodes are combined into the unified state.
Quiz Complete!
You scored 0 out of 5
Week 7: Advanced RAG
Test your understanding of Self-RAG, CRAG, and adaptive retrieval.
What are reflection tokens in Self-RAG?
Correct! Self-RAG generates special reflection tokens to decide when to retrieve and assess relevance.
Not quite. Self-RAG generates special reflection tokens to decide when to retrieve and assess relevance.
How does CRAG (Corrective RAG) improve retrieval?
Correct! CRAG evaluates if retrieved documents are correct/ambiguous/incorrect and takes corrective action accordingly.
Not quite. CRAG evaluates if retrieved documents are correct/ambiguous/incorrect and takes corrective action accordingly.
What is adaptive retrieval in RAG systems?
Correct! Adaptive retrieval decides on-the-fly whether retrieval is needed based on query characteristics.
Not quite. Adaptive retrieval decides on-the-fly whether retrieval is needed based on query characteristics.
What does RAPTOR stand for in RAG?
Correct! RAPTOR creates hierarchical summaries of documents organized in a tree structure for multi-level retrieval.
Not quite. RAPTOR creates hierarchical summaries of documents organized in a tree structure for multi-level retrieval.
What is query decomposition in RAG?
Correct! Query decomposition breaks complex questions into simpler sub-queries that can be answered independently.
Not quite. Query decomposition breaks complex questions into simpler sub-queries that can be answered independently.
Quiz Complete!
You scored 0 out of 5
Week 8: GraphRAG and Knowledge
Test your understanding of knowledge graphs and graph-based retrieval.
What is the key insight of GraphRAG?
Correct! GraphRAG builds knowledge graphs to support both local (entity-specific) and global (thematic) queries.
Not quite. GraphRAG builds knowledge graphs to support both local (entity-specific) and global (thematic) queries.
What is community detection used for in GraphRAG?
Correct! Community detection groups related entities, enabling global summarization at different hierarchy levels.
Not quite. Community detection groups related entities, enabling global summarization at different hierarchy levels.
What is entity extraction in knowledge graph construction?
Correct! Entity extraction identifies named entities (people, organizations, concepts) and relationships between them.
Not quite. Entity extraction identifies named entities (people, organizations, concepts) and relationships between them.
How does HippoRAG differ from standard RAG?
Correct! HippoRAG is inspired by hippocampal memory systems, using pattern separation and completion for retrieval.
Not quite. HippoRAG is inspired by hippocampal memory systems, using pattern separation and completion for retrieval.
What is the difference between local and global search in GraphRAG?
Correct! Local search targets specific entities/relationships, while global search leverages community summaries for themes.
Not quite. Local search targets specific entities/relationships, while global search leverages community summaries for themes.
Quiz Complete!
You scored 0 out of 5
Week 9: Hallucination Prevention
Test your understanding of verification and factual accuracy.
What is an intrinsic hallucination?
Correct! Intrinsic hallucinations contradict information provided in the source/input documents.
Not quite. Intrinsic hallucinations contradict information provided in the source/input documents.
How does Chain-of-Verification (CoVe) reduce hallucinations?
Correct! CoVe generates verification questions about the output and answers them independently to check consistency.
Not quite. CoVe generates verification questions about the output and answers them independently to check consistency.
What does FActScore measure?
Correct! FActScore breaks text into atomic facts and measures what fraction are supported by a knowledge source.
Not quite. FActScore breaks text into atomic facts and measures what fraction are supported by a knowledge source.
What is claim decomposition in hallucination detection?
Correct! Claim decomposition splits complex statements into atomic facts that can be individually verified.
Not quite. Claim decomposition splits complex statements into atomic facts that can be individually verified.
Why might self-correction be less effective than independent verification?
Correct! Models often confirm their initial errors. Independent verification with fresh context is more reliable.
Not quite. Models often confirm their initial errors. Independent verification with fresh context is more reliable.
Quiz Complete!
You scored 0 out of 5
Week 10: Agent Evaluation
Test your understanding of benchmarks and evaluation methods.
What does AgentBench evaluate?
Correct! AgentBench evaluates LLMs as agents across 8 different environments including OS, web, and games.
Not quite. AgentBench evaluates LLMs as agents across 8 different environments including OS, web, and games.
What is pass@k in agent evaluation?
Correct! Pass@k measures if any of k independent attempts successfully completes the task.
Not quite. Pass@k measures if any of k independent attempts successfully completes the task.
What does SWE-bench evaluate?
Correct! SWE-bench tests if agents can fix real software engineering issues from GitHub repositories.
Not quite. SWE-bench tests if agents can fix real software engineering issues from GitHub repositories.
What is LLM-as-Judge?
Correct! LLM-as-Judge uses a language model to automatically score and evaluate agent performance.
Not quite. LLM-as-Judge uses a language model to automatically score and evaluate agent performance.
Why are human baselines important in agent evaluation?
Correct! Human baselines help calibrate what performance levels are achievable and meaningful.
Not quite. Human baselines help calibrate what performance levels are achievable and meaningful.
Quiz Complete!
You scored 0 out of 5
Week 11: Domain Applications
Test your understanding of code agents, financial agents, and safety.
What enables Devin to handle long-horizon coding tasks?
Correct! Devin combines multiple tools (shell, browser, editor) with persistent state for complex, multi-step tasks.
Not quite. Devin combines multiple tools (shell, browser, editor) with persistent state for complex, multi-step tasks.
What is flow engineering in AlphaCodium?
Correct! Flow engineering in AlphaCodium designs iterative code generation flows with test-driven feedback.
Not quite. Flow engineering in AlphaCodium designs iterative code generation flows with test-driven feedback.
Why are regulatory constraints critical for financial agents?
Correct! Financial agents must comply with regulations like SEC rules to avoid unauthorized financial advice.
Not quite. Financial agents must comply with regulations like SEC rules to avoid unauthorized financial advice.
What is a key safety concern for healthcare agents?
Correct! Healthcare agents must protect patient privacy (HIPAA) and provide accurate medical information.
Not quite. Healthcare agents must protect patient privacy (HIPAA) and provide accurate medical information.
What is code sandboxing in agent systems?
Correct! Sandboxing isolates code execution in a secure environment to prevent malicious or erroneous code from causing harm.
Not quite. Sandboxing isolates code execution in a secure environment to prevent malicious or erroneous code from causing harm.
Quiz Complete!
You scored 0 out of 5
Week 12: Research Frontiers
Test your understanding of Generative Agents, Voyager, and future directions.
What enables emergent behaviors in Generative Agents?
Correct! Generative Agents use memory streams, reflection, and planning to enable emergent social behaviors.
Not quite. Generative Agents use memory streams, reflection, and planning to enable emergent social behaviors.
What is the skill library in Voyager?
Correct! Voyager builds a skill library of code functions it learns, enabling continual skill acquisition.
Not quite. Voyager builds a skill library of code functions it learns, enabling continual skill acquisition.
What is curriculum learning in open-ended agents like Voyager?
Correct! Curriculum learning has the agent propose and pursue increasingly difficult tasks to expand capabilities.
Not quite. Curriculum learning has the agent propose and pursue increasingly difficult tasks to expand capabilities.
What is Constitutional AI?
Correct! Constitutional AI trains models to follow explicit principles through AI-generated feedback and revision.
Not quite. Constitutional AI trains models to follow explicit principles through AI-generated feedback and revision.
What is an open research problem for agent memory systems?
Correct! Long-term memory consolidation and efficient retrieval over extended time horizons remain challenging.
Not quite. Long-term memory consolidation and efficient retrieval over extended time horizons remain challenging.
Quiz Complete!
You scored 0 out of 5
Reset Progress
Clear all saved quiz answers and start fresh.