Generate Cue Words
<interpretation
I have a number of cue words and their corresponding accuracy rates.
These cue words are listed in ascending order of accuracy, where higher accuracy indicates better quality.
</explanation
<prompt_scores
{prompt_scores}
</prompt_word
Each cue word is used in conjunction with a problem statement about geometry.
<question
This SVG path element draws an option: (A) Circle (B) Heptagon (C) Hexagon (D) Kite (E) Straight line (F) Octagon (G) Pentagon (H) Rectangle (I) Sector (J) Triangle
</question
<Answer
(B)
</answer
</example
Write a new cue word that achieves the highest possible accuracy and is different from the old one.
- It is very important that the new cue word be different from all the old cue words!
- Make sure to analyze cues with high accuracy and reuse patterns that have worked in the past!
- Make sure to analyze cues with high accuracy and repeat patterns that have worked in the past!
- Think out loud before creating cue words. Describe what has worked in the past and what has not. Only after this create new cue words.
- Analyze with all available information, such as cue length, formal/informal language use, etc.
- Be creative and try different ways of modeling cues. You can even suggest hypothetical scenarios that might improve accuracy.
- You are generating system cue words. This means that there should be no placeholders in the prompt words because they cannot be populated at runtime. Instead, focus on general instructions that will help the model solve the task.
- Write your new prompt words in double square brackets. Use only plain text for the prompt word text, and do not add any markdown (i.e., no tic marks, backquotes, quotation marks, etc.).
</rule
Evaluation Cues
You are a review model responsible for evaluating the correctness of an answer to a navigation question.
The response may contain detailed steps and explanations, but the final answer is the key point.
Please judge whether the final answer provided in the response is correct based on real numbers.
If the final answer is correct, answer 'True'; if not, answer 'False'.
Answer 'True' or 'False' only, no other response is required.
Model response:
{model_response}
True answer:
{ground_truth}
