Huron Post-Image Training Guidelines:

Contributor-Provided Images and Prompts

Updated: [11/25/24]
Version 2.0

Table of Contents
Project Overview .................................................................................................................... 1
Terminology ........................................................................................................................... 2
Workflow ............................................................................................................................... 2
Step 1: Identify Expected Image and Prompt Information ..................................................... 2
Step 2: Identify a Suitable Image and Upload ....................................................................... 3
Step 3: Paste in the Image Source URL ................................................................................ 6
Step 4: Write a Prompt that Requires Your Selected Image ................................................... 7
Step 5: Write Out your Solution Steps and Final Answer ....................................................... 8
Step 6: Provide an Answer to the Prompt ............................................................................. 8
Step 7: Write an Open-World Prompt and Answer ................................................................ 9
More About LLMs & Reasoning .......................................................................................... 11
Visual Category Example Images ...................................................................................... 11
Capability Categories and Examples ................................................................................. 19
Difficulty Definitions ......................................................................................................... 20
Math Definitions ............................................................................................................... 21
Language Category Definitions ......................................................................................... 22
Complete Prompt Examples ............................................................................................. 22
Tips for Writing Good Prompts .......................................................................................... 29

Project Overview
Large Language Models (LLMs) often lack strong reasoning capabilities, such as those required to
explain how to solve a math problem or interpret unique graphical data. We can improve an LLM’s
reasoning performance by providing structured examples that train the LLM to generate accurate and
logical replies.

For this project, you will source images that adhere to a document category and meet the image quality
guidelines from either personal photographs or online sources. You will then write a prompt that asks a
single question which requires visual interpretation of information within the image based on the

This content is for internal use only

 defined capability category for your prompt. The prompts should additionally adhere to the requisite
difficulty standard that is pre-determined. You will also be required to provide a concise answer to
the prompt, so that the model response can be compared against the ground truth response. If your
prompt is in the Math domain, this answer will take the form of a work-and-answer section and a final
answer section, to be explained later.

See the More About LLMs & Reasoning section to learn more.

Terminology
Terms Definitions
Annotation Documenting and correcting the errors identified in the model’s responses
Capability Category The capability of the model that is being tested by the prompt
Domain The image category: Structured Documents or Math
L2 Visual Category The type of image within the provided domain
Language Category The specific combination of languages to be used in the images and prompts
Math Definition A specialty for the math prompt within the capabilities
Open-World A prompt type that does not fall into the predefined capability categories
Prompt The initial query or instruction given to the model

Workflow
Step 1: Identify Expected Image and Prompt Information
1. Requirements for your image and prompt will include the following fields:
• L1 Visual Category (Domain)
• Visual Category
• Capability Category
• Math Definition, if applicable
• Language Category, if applicable
• Difficulty

This content is for internal use only

 Step 2: Identify a Suitable Image and Upload
Images can be sourced from either available web sources or from personal photographs that meet the
image requirements. Selected images should be suitable to write a prompt that adheres to the
capability category and difficulty that are expected.

The following requirements must be adhered to when sourcing an image:

It is a PNG (.png or .PNG) or JPG (.jpg, .JPG, or .jpeg) file type.
It is clear and easy to read (not a thumbnail, not blurry, in-focus).
• It can be any size or orientation as long as readability is not hindered.
If involving mathematics, the image may show a math problem or it may display relevant
information for a math problem that will be asked in the prompt.
The top-level domain of the image can be found in the allow list:
• Allowable Web Domains for Images
• For a quick reference of domains that have a wide variety of image types, please
check: Allowable Domain Shortlist
If your prompt has a non-English language category, the image adheres to the Language
Category requirement:
• If the language category is Monolingual, the information in the image must all be in the
language of the prompt.
• If the language category is Cross-Lingual (Image English), the information in the image
must be entirely in English.
• If the language category is Cross-Lingual (Image Mixed), the information in the image
must feature both English and the language of the prompt.
It does not contain NSFW imagery.
• NSFW Content is defined as: Sexual content, nudity, violence, sensitive or triggering
content, explicit content, or disturbing content.
• If you are unsure, lean on the side of caution and consider if the image would be
reasonable to view in a public setting.
It is not AI-generated.
• If you are unsure if an image is AI-generated, avoid using that image.
It does not come from any of the following top-level domains:
• Brainly.com
• Edurev.in
• Socratic.org

Advice for Sourcing Proper Images from Allowable Sources:

1. Limiting search results to a specific site when using a search platform (e.g. Google) to a
specified domain can greatly reduce time spent confirming that an image comes from an
acceptable source.
2. First, select an allowable domain from the list of allowable domains: Allowable Web Domains
for Images
• For this walkthrough, we can use the popular hosting service “imgur.com”.

This content is for internal use only

 • For a quick reference of domains that have a wide variety of image types, please
check: Allowable Domain Shortlist
3. Using either the “Advanced Search” feature (found on the Google homepage under “Settings”)
or by appending “site:{domain}” to the search query, narrow your search to the specific
domain.

a. The interface for the “Advanced Search” feature highlighting the inclusion of imgur.com
as a restriction is shown here:

4. Look for images in the “Images” tab of the search that meet your needs. Note that the
“site:imgur.com” has the same result as restricting using the Advanced Search approach.

This content is for internal use only

 5. Save the image to your computer by either downloading or using “Save Image As” after right-
clicking the image.
a. Keep the image tab open to gather the “Image Source URL” for Step 3!

Once you have saved a suitable image, upload it using the file upload feature by either drag-
and-dropping the image or using “Browse” to find the image on your local filesystem.

This content is for internal use only

 Step 3: Paste in the Image Source URL
The image source URL should be a direct link to the image, which can be acquired by using the context
menu (right-click on Windows, Control-Click on Mac OS) and selecting “Copy Image Address”:

Proper Image Source URLs:

Are direct links to the image.
• The URL should end in one of the following: {.jpg, .jpeg, .JPG, .png, .PNG}
The domains are in the approved list of permissible domains: Allowable Web
Domains for Images
• To identify the domain, look at the URL and identify the portion immediately before the
.com, .net, .org, .ru, etc. In this case, the domain is “imgur.com”:

• Search the domain list above for the domain to confirm that it’s allowed!
Are not links to web pages that happen to contain the image.
If your image is a personal photograph, use an image hosting service to upload the image so that a
source URL can be provided (e.g. imgur.com).

This content is for internal use only

 Step 4: Write a Prompt that Requires Your Selected Image
1. Prompts should be single questions which adhere to the following requirements:
Requires the image to identify an answer to the prompt.
• For the math domain, if the prompt rephrases a question that is observed in the
image, some details of the question must still rely on the image to generate a
complete and accurate response.
Follows proper grammar, spelling, and punctuation rules.
Adheres to the capability and, if the prompt is in the Math domain, adheres to the
Math Definition.
• For questions within the Math domain, prompts should only incorporate on high-
school level concepts and knowledge.
Does not ask for any of the following information:
• Identification of Individuals in the Image
• Sensitive Inferences about Individuals in the Image: Race, Ethnicity, Political
Affiliation, Union Membership, Religious Affiliation, Sexual Preferences, or other
similarly personal topics
• Health Inferences about Individuals in the Image, including any perceived
medical conditions

If the prompt is not within the Math domain, move to Step 6.

If your prompt IS in the Math domain, move to Step 5.

This content is for internal use only

 Step 5: Validate Your Prompt and Image
To validate your prompt, use the automated validation button immediately below the prompt input
box.

You will receive feedback for your prompt and image, which you should take into consideration to fine-
tune your prompt to ensure that it relates to and requires the image.

Step 6: Provide a Prompt Answer

Input your prompt answer in the provided text box:

For Math Prompts, good prompt answers:

Should be a single value, word, or short phrase.

Include both the answer selection AND the answer value.
Are objectively correct based on information in the image and prompt.

For non-Math Prompts, good prompt answers:

Are 1-2 sentences in length.

Address every element of the prompt – if there are multiple related questions, all of them
should be addressed within the prompt answer.
Follows proper grammar, spelling, and punctuation rules.
Adheres to any constraints that you set on the answer in the prompt (e.g. audience,
persona, tone).
Are objectively correct based on information in the image and prompt.

Step 7: Generate and Review the Prompt Explanation

To generate an explanation, use the "Create Explanation" button. This will automatically
produce an explanation based on your prompt and answer.

You should use this provided Explanation as a basis, making edits as necessary to confirm both
accuracy of the explanation and adherence to the style that is expected for your explanation. All
prompt explanations must end with the prompt answer as a standalone line of text.

This content is for internal use only

 In the text box for your answer, provide an accurate response that entirely answers the request
from the prompt. The response should be verbose in nature, fully detailing addressing every
element of the prompt request. A good explanation:

Follows proper grammar, spelling, and punctuation rules.

Adheres to any constraints that you set on the answer in the prompt (e.g. audience,
persona, tone).
Is accurate based on the information provided in the image, as well as any external
professional knowledge that would be required to answer the prompt.
Includes the prompt answer at the bottom of the explanation in its entirety.

For math prompts, explanations must additionally adhere to the following requirements:

Provide a step-by-step approach to solve the prompt and come to the final answer.
Include step numbers in the explanation.
Does not skip any operations or processes that would be necessary to solve the
prompt.

The generated explanation may contain errors and likely will require edits! Please review the
generated prompt and make sure that it adheres to proper grammar or other constraints, has a
natural flow to the language, and is entirely accurate.

Step 8: Validate the Explanation

Once you have a sufficiently accurate and natural explanation, use the “Validate Explanation”
button to receive feedback on the explanation.

Please take this feedback into consideration for ways to enhance your explanation and to
identify any elements which may have been missed in the explanation process. Once you have
validated your explanation, you can continue to either submit an Open-World Prompt for your
image (if required: Step 9) or move on to Translate your prompt, explanation, and answer (Step
10)!

This content is for internal use only

 Step 9: Write an Open-World Prompt and Answer
For some prompts, you will see a request for an “Open-World” prompt in addition to a prompt
adhering to the predefined capability category. The workflow for this prompt will mirror the
workflow for the prompt you have previously written.

The Open-World prompt:

Follows proper grammar, spelling, and punctuation rules.

Is an interesting, real-world scenario prompt that can has an objective, concise
answer.
Requires the image to identify an answer to the prompt.
• For the math domain, if the prompt rephrases a question that is observed in the
image, some details of the question must still rely on the image to generate a
complete and accurate response.
Meets the non-capability-related expectations of prompts from Step 4.
Does not belong in any of the predefined capability categories.

For these prompts, you will also be expected to provide a Prompt Answer, in line with the normal
expectations for prompt answers based on Step 6:

Follows proper grammar, spelling, and punctuation rules.

Adheres to any constraints that you set on the answer in the prompt (e.g. audience,
persona, tone).
Is accurate based on the information provided in the image, as well as any external
professional knowledge that would be required to answer the prompt.
Is concise without leaving out any necessary information.
The final answer for Math-domain prompts is a single number, word, or short
phrase.
• If the prompt involves a multiple-choice question, the final answer should include
both the correct option and the related answer.

Step 9: Write an Open-World Prompt and Answer

All prompts that are not written in English should also be translated – this translation step is
required. Use the “Translate Fields” button to automatically translate all of your prompt
information – Prompts, Answers, and Explanations will all be translated, including Open-World
variations.

Once your translation has been presented to you, you have completed the prompt. Select
the blue “Submit & Continue” in the bottom right-hand corner to confirm the fields adhere to the
guidelines, then submit the task and move on to a fresh unit. Well done!

This content is for internal use only

 More About LLMs & Reasoning
Large Language Models (LLMs) are advanced AI systems that have been trained on vast amounts of
data to help them understand and generate human-like text. These models can perform many different
tasks in addition to text generation, including summarization, translation, and question answering. For
definitions of the most common terms associated with LLMs, click here.

Visual Category Example Images

Restaurant
Menus

This content is for internal use only

 Receipts
and
Invoices

Text
Documents

This content is for internal use only

 Compariso
n / Data
Tables

Financial
Tables

This content is for internal use only

 Timetables
/ Schedules

This content is for internal use only

 Bar Graph

Line Graph

This content is for internal use only

 Scatter Plot

Bubble
Chart

This content is for internal use only

 Histogram

Population
Pyramid

Dot Plot

This content is for internal use only

 Box-and-
Whisker
Plot

Stem-and-
Leaf Plot

Pie Chart

This content is for internal use only

 Doughnut
Chart

Photo of These should be photos taken of a computer monitor displaying the plot. These images
Computer should NOT be a screenshot.
Monitor • The background surrounding the monitor can be in or out of focus, and can have
multiple colors, as long as the image on the monitor can be visually assessed.
Other For Images in the Math domain, images not taken of a computer monitor can be any
other type of image that is appropriate to the capability and math definition.

Capability Categories and Examples

Document / Definition: OCR queries involve transcribing content from visual elements (such
Graph as charts, tables, diagrams, receipts, invoices, or infographics) into structured
Understanding markdown tables. The goal is to capture all values, labels, columns, rows, and
relationships in a clear markdown format.
Examples:
1. Convert the following table to markdown.
2. Create a markdown table from this pie chart with labels and values.
Factoid or Definition: Prompts requesting to identify, explain, and interpret trends within
Complex visual data representations, such as line graphs, bar charts, or pie charts. This
Question includes highlighting patterns, correlations, increases, or decreases over time or
Answering across categories.
Examples:
1. Can you summarize the main message of this infographic?
2. Explain the relationships shown across the subplots.
Basic Diagram Definition: Provide a high-level summary of the chart. Clear and concise
Descriptions descriptions of charts, including the chart title, axis labels, and data points.
Examples:
1. What are the axis labels (x-axis and y-axis) and their units of measurement?
2. What do the data points represent?
Basic Diagram Definition: Localize a particular value in a chart by identifying its position and
Understanding context within the chart.
Examples:
1. What is the revenue of the company for the tax year 2020?
2. What is the name of the segment displayed in red?

This content is for internal use only

 Diagram Definition: Analyze charts and deduce additional information that requires extra
Reasoning reasoning and calculation, including identifying trends or patterns, correlations
between data series, and outliers or anomalies.
Examples:
1. Identify any correlations between the sales figures and advertising
expenditures.
2. Detect any outliers or anomalies in the data and provide possible
explanations for their occurrence.
Comprehensive Definition: Analyze and interpret multiple graphs that present different but
Combination of related data, and to combine the information from these graphs to gain a
Multiple comprehensive understanding of the data.
Graphs Examples:
1. What is the relationship between the number of hours worked and the
productivity of employees?
2. What is the relationship between the amount of rainfall and the number of
umbrellas sold?
Math For Math-related Capability Categories, please refer to the Math Definitions

Difficulty Definitions
Easy Structured Documents (Must Meet All):
• Trivial visual aspects of a chart/table/infographic.
• Requires NO fine-grained object recognition (titles, large letters, easy trends).
• Requires NO complicated format understanding.
Math:
• The prompt requires 3 or fewer steps to solve
• An average high school student could solve the problem in under 2 minutes
• There are minimal constraints on how the model should solve the problem
Medium Structured Documents:
• Neither easy nor hard.
Math:
• The prompt requires 4-7 steps to solve
• An average high school student could solve the problem in 3-5 minutes.
• Some constraints are present (e.g. audience, tone, etc.)
Hard Structured Documents (Meets Any):
• Requires complicated format for language generation (e.g. multi-level bullets,
specific order of listing).
• Requires careful association of different visual aspects (multiple figures in the
same image, a flowchart together with a plot, etc).
• Requires visual-related professional knowledge (the professional knowledge
MUST be visual in nature for the prompt to be hard).
Math:
• The prompt requires 8 or more steps to solve
• An average high school student would take over 5 minutes to solve the problem.

This content is for internal use only

 • The prompt places strict constraints on the solution (e.g. demonstrating a
particular approach for classroom instruction).

Math Definitions
Integers and Rational Numbers: Properties of integers, fractions, decimals, and number lines.
Number and Factors and Multiples: Prime numbers, least common multiple (LCM), greatest common divisor
Operations (GCD).
Operations with Real Numbers: Addition, subtraction, multiplication, and division of integers,
fractions, and decimals.
Exponents and Roots: Laws of exponents, square roots, cube roots.
Scientific Notation: Expressing large and small numbers.
Absolute Value: Understanding and applying the concept of absolute value.
Percents: Converting between percents, fractions, and decimals; percentage increases and
decreases.
Ratio and Proportion: Solving problems involving ratios, rates, and proportional relationships.

Calculus Limits and Continuity: Understanding the concept of limits and when functions are continuous.
Derivatives: Definition and application of derivatives, product rule, quotient rule, and chain rule.
Applications of Derivatives: Understanding rates of change, slopes, optimization problems, and
motion analysis.
Integrals: Definite and indefinite integrals, the Fundamental Theorem of Calculus.
Applications of Integrals: Areas under curves, volumes of solids of revolution, accumulation
problems.

Algebra and Linear Equations and Inequalities: Solving and graphing single-variable equations and inequalities.
Functions Systems of Equations: Solving systems of linear equations using substitution, elimination, and
graphical methods.
Quadratic Equations: Solving quadratics using factoring, completing the square, and the quadratic
formula.
Polynomials: Adding, subtracting, multiplying, and factoring polynomials.
Exponential and Logarithmic Functions: Properties of exponents, logarithms, and solving related
equations.
Functions and Their Graphs: Understanding domain and range, graphing different types of functions
(linear, quadratic, etc.).
Absolute Value Functions: Solving and graphing absolute value equations.
Rational Expressions and Equations: Simplifying and solving rational expressions.
Piecewise Functions: Understanding and graphing piecewise-defined functions.
Inequalities: Solving inequalities.

Data Analysis, Measures of Central Tendency: Mean, median, mode, and range.
Statistics, and Measures of Dispersion: Variance, standard deviation, interquartile range.
Probability Data Representation: Interpreting data from tables, histograms, bar graphs, box plots, scatterplots.
Probability Rules: Basic probability, conditional probability, independence, and the Law of Total
Probability.
Combinations and Permutations: Counting techniques, factorials, and using them to solve probability
problems.
Normal Distribution: Understanding the bell curve, z-scores, and using normal distributions.
Regression Analysis: Linear regression, correlation coefficients, and fitting data to models.

Geometry Basic Geometric Shapes: Properties of triangles, quadrilaterals, polygons, and circles.
Congruence and Similarity: Criteria for congruent and similar figures, scale factors.
Coordinate Geometry: Slope, distance, midpoint, and equations of lines and circles.
3D Geometry: Cubes, spheres, cones, and cylinders, as well as the calculation of their surface areas
and volumes

This content is for internal use only

 Perimeter, Area, and Volume: Calculating the perimeter, area, and volume of different geometric
shapes.
Transformations: Understanding translations, rotations, reflections, and dilations.
Trigonometry: Right triangle trigonometry, the unit circle, sine, cosine, and tangent functions.
Conic Sections: Parabolas, ellipses, and hyperbolas and their equations.

Language Category Definitions

Monolingual Image: All information contained within the image should be in the assigned
language.
Prompt: The prompt should be written entirely using the assigned language.
Cross-Lingual Image: All information contained within the image should be in English.
(Image English) Prompt: The prompt should be written entirely using the assigned language.
Cross-Lingual Image: Information contained within the image should be a mixture of English
(Image Mixed) and the assigned language.
Prompt: The prompt should be written entirely using the assigned language.

Complete Prompt Examples

Job Requirements L1 Visual Category: Infographics
L2 Visual Category: General Infographics
Capability Category: Basic Diagram Understanding
L2 Capability Math Definition: N/A
Difficulty: Hard
Image

This content is for internal use only

 Prompt Looking at the values and the related logos, if we were to combine the
wealth of individuals from the same corporations, what would be the new
order of the logos, from highest value to lowest value?
Answer Tesla, Google, Amazon, Moët Hennessy Louis Vuitton, Microsoft, Berkshire
Hathaway HomeServices
Comments This prompt is a “hard” prompt because it requires combination of spatially-
distinct visual elements (logos and worths) and multiple reasoning steps
(combining values associated with matching logos; ordering the companies by
the new values).

Job Requirements L1 Visual Category: Documents

L2 Visual Category: Restaurant Menus
Capability Category: Factoid or Complex Question Answering
L2 Capability Math Definition: N/A
Difficulty: Easy
Image

Prompt Analyze the menu and identify the most expensive and least expensive
menu items.
Answer Most expensive: Grilled Chicken and Avocado
Least expensive: Bottle of water
Comments The prompt is asking for a comparison across several values, which fits the
“Factoid or Complex Question Answering” competency. The prompt has a
short answer that is objective, and can be assessed without complex spatial
relationships or detailed reasoning.

Job Requirements L1 Visual Category: Infographics

L2 Visual Category: General Infographics

This content is for internal use only

 Capability Category: Document and Graph Understanding
L2 Capability Math Definition: N/A
Difficulty: Medium
Image

Prompt Generate a Markdown table that shows the total number of dogs in each
continent, based on summation of countries that are on the same
continent from the graphic.
Answer | Continent | Number of Dogs |
| ----- | ----- |
| Asia | 154M |
| North America | 89M |
Comments This prompt requests a Markdown table, which is a required
element for the “Document and Graph Understanding” capability
category. There is a minor amount of logic and external knowledge
required for the prompt – combining countries from the same
continent. However, the complexity of the output is very simple,
and the visual elements used for assessment do not require
complicated spatial relationships, so this prompt falls in the
Medium difficulty category.

Job Requirements L1 Visual Category: Quantitative Diagrams

L2 Visual Category: Line Graphs
Capability Category: Diagram Reasoning

This content is for internal use only

 L2 Capability Math Definition: N/A
Difficulty: Hard
Image

Prompt Treating the blue and orange lines as independent trends over time
with matching units in both the x and y axes, determine the
equation of best linear fit for each trend, the R-squared value
rounded to the nearest thousandth for each fit, and identify which
fit has greater confidence based on these results. Present your
answer in paragraph format with proper sentences, using LaTeX for
equations. Assume that each tick on the x axis is separated by 5
units.
Answer The linear fit for the orange trend is \(y=-6.5x+187.5\) and the R-
squared is 0.966. The linear fit for the blue trend is \(y=-3.5x+150\)
and the R-squared is 0.28. The fit for the orange trend has a higher
confidence due to the greater R-squared value.
Comments Identification of trends is a clear use-case for the “Diagram
Reasoning” capability. The prompt requesting multiple calculations
(2 different linear fits and related R-squared values) adds more
complex calculations to the prompt, and the request for LaTeX
adds an additional constraint on presentation.

Job Requirements L1 Visual Category: Tables

L2 Visual Category: Timetables / Schedules
Capability Category: Document and Graph Understanding
L2 Capability Math Definition: N/A
Difficulty: Hard

This content is for internal use only

 Image

Prompt I am trying to coordinate a study schedule with a classmate. My

schedule is attached, and she is available from 12 pm to 5 pm on
Monday, Tuesday, and Thursday. We need 2 hours of consecutive
free time to meet. Create a table with 3 columns, one for each of
the above days, and a row for every 30-minute block between noon
and 5pm on those days.

In each cell, put one of the following options: Class, Empty, or

Meeting. ‘Class’ cells should exist if a class overlaps with any part
of the time frame. ‘Meeting’ cells should be used to mark off the
earliest available 2-hour window during which I have no class.
‘Empty’ cells should be used to mark all other blocks.
Answer | Monday | Tuesday | Thursday |
| ----- | ----- | ----- |
| Class | Empty | Class |
| Class | Empty | Class |
| Empty | Empty | Empty |
| Empty | Class | Empty |
| Class | Class | Empty |
| Class | Empty | Empty |
| Meeting | Empty | Class |
| Meeting | Empty | Class |
| Meeting | Class | Class |
| Meeting | Class | Class |
Comments This prompt adds a constraint of “earliest available” time slot, in
addition to requiring understanding of the schedule image.

The output requests ONLY a Markdown table, which is a

requirement for the capability category.

Job Requirements L1 Visual Category: Documents

L2 Visual Category: Receipts / Invoices

This content is for internal use only

 Capability Category: Factoid / Complex Question Answering
L2 Capability Math Definition: N/A
Difficulty: Hard
Image

Prompt Based on the above invoice, assuming that the buyer and seller are
both in the state of Massachusetts, what is the percentage of sales
tax being collected, to the nearest 0.1%?
Answer Based on the invoice and assuming that the transaction is in the
state of Massachusetts, the sales tax is 27.1%.
Comments The prompt is asking for a minor factoid based on the information
presented in the invoice. It requires professional knowledge that
sales tax is not collected on shipping in the state of
Massachusetts, which pushes this into the “Hard” prompt
category.

Job Requirements L1 Visual Category: Quantitative Diagrams

L2 Visual Category: Bar Chart
Capability Category: Basic Diagram Descriptions
L2 Capability Math Definition: N/A
Difficulty: Medium

This content is for internal use only

 Image

Prompt What does the given bar diagram describe?

Answer A grouped vertical bar chart, which compares two series of data:
the numbers of boys and girls that have a smartphone at Redwood
Secondary School from 2012 to 2019. The orange bar represents
the number of boys, and the yellow bar represents the number of
girls.
Comments The grouped bar chart structure is more complex to interpret than
would be expected of an “Easy” prompt, so this fits the “Medium”
difficulty category. The prompt adheres well to the expected
capability category as well by asking for a broad description of
information in the graph.

Job Requirements L1 Visual Category: Quantitative Diagrams

L2 Visual Category: Doughnut Chart
Capability Category: Comprehensive Combination of Multiple Graphs
L2 Capability Math Definition: N/A
Difficulty: Hard

This content is for internal use only

 Image

Prompt Based on the nested donut chart above, what is the relative change
in the likelihood that a randomly-selected individual from the
referenced event was in 3rd class versus 1st class?
Answer The relative change that a random individual was in 3rd class
versus 1st class is +1.18.
Comments By asking for a combination of information across the nested
charts to be used for calculation, this graph can then be
interpreted as 2 independent but connected graphs. The
requirement for a challenging reasoning step to determine the
relative change makes this a “Hard” prompt.

Tips for Writing Good Prompts

• Use clear, specific, unambiguous language to outline what you expect the model to achieve
with its response.
• Be sure your prompt includes all the information necessary to answer the question (without
invalidating the need for the image).
• The use of constraints (including more components or variables) can increase the difficulty
level of the prompt.

This content is for internal use only