Voyager and Skill-Library Agents Quiz

Answers and explanations appear as you take the quiz.

1. How does Voyager improve its capabilities over time?
   • By accumulating reusable code skills in a library and composing them
   • By updating its neural network weights with gradient descent
   • By downloading patches from a server
   • By memorizing the entire game map
2. In what form are Voyager's skills represented?
   • Natural-language essays
   • Executable code (JavaScript) that controls the game bot
   • Static images of the game
   • Raw neural network weights
3. What is the role of Voyager's automatic curriculum?
   • To grade the agent's grammar
   • To compress the skill library
   • To propose progressively harder goals that maximize exploration
   • To translate the game into other languages
4. How does Voyager retrieve a relevant skill when facing a new task?
   • Randomly picks any stored skill
   • By asking a human to choose
   • It never reuses skills
   • By matching the task to skill descriptions via embeddings
5. What does Voyager's iterative prompting loop do when generated code fails?
   • Reads the errors and environment feedback and asks the model to self-debug
   • Gives up on the task permanently
   • Restarts the entire game world
   • Switches to a different language model
6. Which result demonstrated the value of Voyager's skill library?
   • Its skills could not be reused at all
   • Learned skills transferred to new worlds and sped up new tasks
   • It performed worse than agents without a library
   • It only worked in a single fixed map
7. Does Voyager improve by fine-tuning GPT-4’s model weights after each Minecraft task?
   • Yes; it retrains GPT-4 after every failed command
   • Yes; it updates the game engine’s source code
   • No; it stores verified executable skills and retrieves them in context
   • No; it never retains any learned behavior
8. What is stored with a Voyager skill to help retrieve it for a later task?
   • A gradient tensor for GPT-4
   • A screenshot with no task label
   • A Slurm job identifier
   • A natural-language description represented by an embedding