Tree of Thoughts Quiz

Answers and explanations appear as you take the quiz.

1. How does Tree of Thoughts differ fundamentally from chain-of-thought prompting?
   • It explores multiple reasoning branches and can backtrack
   • It retrains the model on new data
   • It removes all intermediate reasoning steps
   • It only works for image generation
2. Which three components make up a Tree of Thoughts system?
   • Tokenizer, embedder, and decoder
   • Thought generator, state evaluator, and search algorithm
   • Encoder, attention head, and softmax
   • Dataset, optimizer, and loss function
3. In the original Tree of Thoughts Game of 24 experiment, what result did breadth-five search report for its selected test set?
   • 100% success on every possible puzzle
   • 74% for every model and search budget
   • 74% success for GPT-4 with ToT, versus 4% for a single chain-of-thought baseline
   • 4% for ToT and 74% for chain-of-thought
4. How does a Tree of Thoughts system typically evaluate partial solutions?
   • By querying an external database
   • By measuring GPU temperature
   • By counting the number of tokens used
   • By prompting the model to rate states as sure, maybe, or impossible
5. What is true about implementing Tree of Thoughts on an existing model?
   • It is a prompting and search wrapper, not retraining
   • It requires retraining the model from scratch
   • It can only run on specialized hardware
   • It permanently changes the model's weights
6. Which related framework lets reasoning paths merge and reuse intermediate results?
   • Linear regression
   • Graph of Thoughts
   • Batch normalization
   • Beam width pruning
7. Which components are commonly specified in a Tree of Thoughts setup?
   • A new tokenizer, an optimizer, and a training corpus
   • A database, a browser, and a GPU driver
   • Thought proposals, an intermediate-state evaluator, and a search procedure
   • A label encoder, a loss scaler, and a checkpoint
8. Why can Tree of Thoughts increase inference cost?
   • It always retrains the base model
   • It requires a separate model for every token
   • It eliminates all model calls after the first thought
   • It may generate and evaluate multiple candidate states across search steps