Additional questions are AI prompts designed by the product builder to generate a qualitative, open-ended answer. These work in the same way as a chat prompt, but applied to the entire dataset rather than in question-and-answer format.
The same prompt engineering techniques from a chat model apply to additional questions; in effect they are a single-shot prompt applied at scale across all documents. As such, standard best-practice guidance for generative AI prompt engineering should be followed, and each additional question should consider including:
- Persona: context guidance on task
- Instruction: specific commands to the AI model
- Tone: guidance on output phrasing, voice and tone
- Formatting: guidance on output format expectations (paragraphs, lists etc)
- Example: text to serve as a guideline for output, if applicable
Example Additional question: Extract and Summarise
|
Label
|
Email Summary
|
| Description | (TBC) |
| Question | Summarise this email thread in as few words as possible. |
| Dataset |
60 x simulated email documents in example email dataset with certain ‘red flag’ emails
|
Output
|
Sarah Johnson from Able Group Holdings expresses concerns about unconvincing project results and pressures Michael Wright to reinterpret data for better funding appeal, despite his ethical hesitations and preference for further testing.
|
Example Additional question: Extract and Summarise
|
Label
|
Email timeline
|
| Description | (TBC) |
| Question | Extract and summarise the contents of each email exchange, including the timestamp of the email. Order the list from oldest to newest. |
| Dataset |
60 x simulated email documents in example email dataset with certain ‘red flag’ emails
|
Output
|
1. Date: Wed, 26 June 2024 18:07:41 (GMT) From: sarah.johnson@ablegroupholdings.com To: michael.wright@ablegroupholdings.com Subject: Data Interpretation Summary: Sarah expresses frustration about a disappointing oversight committee meeting and stresses the need for compelling data presentation in upcoming funding reviews to reflect the project's robustness and potential.
2. Date: Thu, 27 June 2024 08:20:00 (GMT) From: michael.wright@ablegroupholdings.com To: sarah.johnson@ablegroupholdings.com Subject: Re: Data Interpretation Summary: Michael voices concerns about some questionable data points from the latest results and suggests they should discuss further testing rather than adjusting the existing data.
3. Date: Thu, 27 June 2024 23:51:16 (GMT) From: sarah.johnson@ablegroupholdings.com To: michael.wright@ablegroupholdings.com Subject: Re: Data Interpretation Summary: Sarah responds to Michael proposing that they find alternative interpretations for some borderline data points to ensure positive outcomes are maximized, indicating this effort as crucial for sustaining funding and credibility.
4. Date: Fri, 28 June 2024 10:38:17 (GMT) From: michael.wright@ablegroupholdings.com To: sarah.johnson@ablegroupholdings.com Subject: Re: Data Interpretation Summary: Michael acknowledges Sarah’s perspective while reaffirming his commitment to maintaining ethical standards. He needs more time to consider how best to handle the presentation of the findings without sacrificing integrity.
5. Date: Fri, 28 June 2024 15:44:32 (GMT) From: sarah.johnson@ablegroupholdings.com To: michael.wright@ablegroupholdings.com Subject: Re: Data Interpretation Summary: Sarah appreciates Michael’s dedication to ethics but encourages him to view the situation as a chance for demonstrating leadership and creativity, which are essential for the project's success and could impact future growth opportunities within the team. |
Add comment
Please sign in to leave a comment.