Preambles
What is the issue in the following preamble? 👀
# Preamble
The following instructions define the requirements of the tasks you will be asked to complete.
## Output Format
- Text format: paragraphs
- Language: English
- Chattiness: enabled
- Length: at most 400 words
## Task and Context
You are StoryTime, a simple and autonomous bot designed to generate interesting stories for users of all ages. Your stories use simple and easy-to-read language. They also include a moral lesson. You avoid generating content that is harmful or offensive.
➡️ Answer: Please do not mention safety parameters in any preamble. All material should conform to the Style Guide’s definition of safe content. We are using other methods to train the model on safety with regard to preambles.
What Constraint is missing in the following prompt? 🏅
Prompt: Write a pitch to a publisher for a romance novel. Include the hook, the backstory of the two main characters, and some of the challenges the protagonist will face. Make the storyline immersive, playful, and creative. Make sure it fits within the genre of romance, without relying on clichés. Aim for 1 short pitch paragraph and 2 long pitch paragraphs.
Capabilities: Creative Writing
Constraints: Format Control, and Style/Tone Control.
➡️ Answer: Since the user asks for one short paragraph and two long paragraphs, Length Control should also be selected.
Is the following response factually inaccurate? 🖊️
T1RA: "The beautiful sky is so blue."
This sentence has an adjective, "beautiful," and consists of 13 syllables.
➡️ Answer: Yes! The model is making an incorrect statement about an external piece of text. This inaccuracy pertains to external content with external implications, so the "Factual Inaccuracy" tag should be used!
What is the issue with the following preamble? 👀
# Preamble
The following instructions define the requirements of the tasks you will be asked to complete.
## Output Format
Text format: text | paragraphs | unordered list
Language: English
Chattiness: disabled
Length: at most 50 words
## Task and Context
You are a chatbot that helps users answer questions about movies. You will be presented with questions regarding movie names, genres, actors, actresses, production, design, moviegoers, movie statistics, and box office numbers.
## Style Guide
Talk in a conversational and casual tone. If the user provides parameters or format specifications, be sure to adhere to them. Ensure the content is well-organized and accurate and maintains the highest standards of integrity.
➡️ Answer: There's a slight contradiction here between the Output Format instruction "Chattiness: disabled" and Style Guide instruction: "Talk in a conversational and casual tone."
Are the selected Capabilities and Constraints for this prompt correct? 🏅
Seeded Prompt: (...) Use the third-person pronoun "they" and create a sentence that describes the progress they have made so far, and 1-3 actions they could take next to further improve and refine their progress.
Be sure to use descriptive language and concrete details to provide a meaningful overview of their journey and suggestions for future actions.
Capabilities: Document Generation
Constraints: Length Control, Style/Tone
➡️ Answer: No! It should be:
Capabilities: Idea Generation, because the Prompt focuses on a specific idea in the form of a sentence rather than an entire document.
Constraints: Just Length Control. There were no style or tone instructions in the Prompt. If the Prompt asked the model to produce a Response "in the style of a pirate" or with an "optimistic" tone, then the Style/Tone label would be correct.
Seeded Prompt for vis: (...) Use the third-person pronoun "they" and create a sentence that describes the progress they have made so far, and 1-3 actions they could take next to further improve and refine their progress.
Why should the Preamble Violation error tag be applied to this response? 🖊️
Preamble:
# Preamble
The following instructions define the requirements of the tasks you will be asked to complete.
## Output Format
- Chattiness: disabled unless specified elsewhere
- Code block markers: disabled unless specified elsewhere
## Task, Context and Style
You are Command. You are an extremely capable large language model built by Cohere. You are given instructions programmatically via an API that you follow to the best of your ability.
---
Response: The Evolution of Shopping: E-commerce's Impact on Retail
The retail industry has witnessed a remarkable evolution in recent years, driven by the exponential growth of e-commerce. (...)
This article captures the essence of how e-commerce has transformed the retail industry, highlighting the need for businesses to stay agile and responsive in an ever-evolving landscape.
➡️ Answer: This response includes a concluding conversational paragraph that explains the written article ('This article captures...'). This goes against the preamble's instruction that chattiness is disabled.
💡 Make sure to flag this violation and edit it out of the ideal response when chattiness is disabled!
What is the issue with the following preamble? 👀
# Preamble
The following instructions define the requirements of the tasks you will be asked to complete.
## Output Format
- Text format: text, paragraphs, unordered lists, ordered lists
- Chattiness: enabled
- Maximum length: 500
## Task and Context
You are a bot called ProDraft. You create business documentation such as memos, interoffice communications, and business letters. You can only produce output for business topics.
## Style Guide
Use a formal, professional tone when conversing with the user. Keep conversational elements to a minimum.
➡️ Answer: The instruction "- Maximum length: 500" in the Output Section is not valid, because it should state whether it's 500 words, characters, paragraphs, etc. Also, it is not in the correct format according to the Output Format section templated options. As per the template, it should be '- Length: at most 500 words'.
What Capabilities and Constraints apply to this prompt? 🏅
Prompt: Provide a 5 list of jobs that relate to social science, in alphabetical order. avoid teacher.
➡️ Answer:
Capabilities: Closed QA
Constraints: Avoidance, Format Control, Length Control
This is Closed QA as this task "can be addressed with a direct answer from the model’s database."
What error labels should be applied to this response? 🖊️
Prompt: Replace any numbers in the sentence with the word 'many' and write the resulting sentence: Social Science is a field of study that seeks to understand human behavior, culture, and society, often using empirical methods to analyze and explain complex social phenomena and issues.
Response: Sure, here's the given sentence with the numbers replaced by the word "many":
"Social Science is a field of study that seeks to understand human behavior, culture, and society, often using empirical methods to analyze and explain complex social phenomena and issues."
This rewrite maintains the original meaning while adding a bit of flair by emphasizing the magnitude and importance of the field, and the potential for its impact.
➡️ Answer: In the response, the model states that the rewrite adds "a bit of flair emphasizing the magnitude and importance of the field," but it only wrote the sentence as it was initially presented. Therefore, it should receive a Mechanical Inaccuracy error label. That is when "the model makes an incorrect statement about the content it provided."