LLM Sandbox Response Generation Instructions
To complete the Sandbox project successfully please review the following module:
Sandbox Orientation Module
([Link]
password: beach24
You can refer to this module at any time.
Task Categories
Creative Writing & Content Creation: A prompt request asking for the creation of
written work. This may be a story, essay, poem, fact, or fiction etc.
Summarization: Entails condensing provided texts while retaining their essence, with
requirements for length, format, and the exclusion of non-original details. Variants
include concise summaries, multi-document summarization, and using different
methods like lists or tables.
Rewriting & Text Transformation: In rewriting and text transformation prompts you
may be asking the model to proofread and correct or transform the text within set
parameters. This type of prompt may also include asking the model to compose text
based on a given text. As with summarization, text can be used from any open
source (not from behind a paywall) and you may need to provide the URL. Content
can be anything from pop culture to law - as long as it is applicable to your location.
Brainstorming & Ideas Inspiration: A brainstorm prompt created to generate ideas
on a specific topic. It encourages discussion, creative thinking and the possibility for
different perspectives, possibilities, and solutions.
Open Q&A: An open question is like those below are where the model is asked a
question and expected to answer. The question may be direct and require a factual
answer, or it may be more open-ended and philosophical.
Closed Q&A: A closed question is one which asks the model to provide an answer
from text or information also provided with the prompt.
Classification: A prompt requesting the AI model to classify or categorize specific
content based on the given topic.
Chit-Chat/Chatbot: The purpose of a chit-chat prompt is to engage in conversations
with the model or ask for specific conversational tasks.
Extraction: Extraction prompts ask the model to retrieve specific details or data from
a document, website, or other sources. For this sandbox, you must also provide the
text/source from which the information is extracted.
DO’s
• Please check the Guidelines and Module provided for this task and read
them carefully.
• All entries should be in the same locale noted in the task name (en_US should
be in American English, en_AU should be in Australian English. etc.).
• All entries should include content that is related to the specific locale
whenever possible. Please consider cultural, linguistic, and societal
differences.
• We expect flawless language with correct grammar and spelling according to
the rules of the locale.
• Formulate your prompts and answers based on the selected task category
and quality requirements (see the orientation module for details).
• Both prompts and answers must be clear, concise, and to the point.
• No online content should be copied and pasted without modifications by the
writer. Do NOT copy/paste answers from any AI tools, especially Generative
AI tools like GPTs, or online sources.
• If you are copying text for task type such as summarization, this text should
be from an open source (i.e., not from behind a paywall) and you must
include the URL to the source in the corresponding field provided.
• Before submitting your entry, you will be required to click on a button to run
a plagiarism and AI detection check on the webapp. You will be unable to
submit your entry if those checks are not run or if they fail.
• Besides prompt- and response-specific requirements mentioned in the
guidelines, please follow the below:
o Prompt:
▪ Should be at least 15 words long.
▪ The token count, a measure of input size equivalent to
approximately 3 characters, will appear below the prompt box.
▪ Should include references to the local culture.
▪ Should follow all requirements for the task category.
o Response:
▪ Should be at least 35 words long.
▪ Must be helpful, truthful, and harmless.
▪ Answers must completely and accurately answer the prompt.
▪ Should be clear and concise.
▪ Should not contain personal or subjective opinions and/or
bias.
▪ Should use reliable sources for your locale (or global).
Cite your sources.
▪ In most cases, answers should include:
1. A short introduction using natural language.
2. A chain of though logic for computation.
• Make sure to be aligned with the prompt characteristics of a single
turn (prompt + response) or multiple turn (a conversation with several single
turns)
DON’Ts
• Although you can use inspirations from the internet and support yourself
with external resources, plagiarized text or prompts from Generative AI tools
(e.g., ChatGPT, Brat) are prohibited and will be rejected once detected.
• Avoid oversimplifications. Please make sure that a short introduction in the
response is always included.
• Avoid chaotic responses with no structure. Please make sure that your chain
of thoughts is logical in each answer and can be easily understood.
WEBAPP UI
Here’s a quick overview of the webapp UI:
• Prompt: Input box for the prompt.
• Answer: Input box in which to write an acceptable answer to the human
prompt.
• Add turn and label task category accordingly (only available in Multi-
Turn tasks): After writing your first prompt and response turn based on the
task category, keep the conversation going by clicking “Add Turn” for the
next query and answer input fields to appear. Remember, a turn is counted
as a prompt and an answer. A multi-turn conversation should include a
minimum of three turns. Please don’t forget to label task category types as
well, as for turn 2 onwards you could add turn of different task categories.
• Source: If you have used a source to inform your answer, please click and
enter the source URL into the box that will appear. If you have used more
than one source enter them all.
• Submit: When you have completed your conversation and are happy with the
content, click submit.
