LLM Data Needs

This template guides you through identifying data needs for your Large Language Model (LLM) and identifying how AI might guide you to achieve your goals.

‍

Materials:

1. This template
2. Markers or pens
3. Sticky notes

Time: 60 minutes

‍

‍

Steps:

Part 1: Open Ideation

1. Brainstorm ideas about what you want your AI solution to achieve
2. Vote on the tool that you want to develop.

‍

Part 2: Identifying Data Needs

1. Analyse your data needs, differentiating between data that you have, want and need and where you can source this data.

‍

Public Data:

1. Identify existing data: List all publicly available data sources you currently use (e.g., news articles, open-source code).
2. Needed data: Specify additional publicly available data that would enhance your LLM (e.g., specific datasets, research papers, domain-specific knowledge bases).
3. Unavailable data: Note any publicly available data that is relevant but inaccessible (e.g., paywalled data, restricted datasets).

Private Data:

1. Identify existing data: List all private data you currently use (e.g., user logs, internal documents, customer feedback).
2. Needed data: Specify additional private data that would improve your LLM (e.g., anonymised user interactions, clickstream data, sentiment analysis).
3. Unavailable data: Note any private data that is relevant but unavailable due to privacy concerns or technical limitations.

User Data:

1. Identify existing data: List all user data you currently use (e.g., search queries, preferences, interactions with the AI tool).
2. Needed data: Specify additional user data that would personalise the LLM's responses (e.g., user profiles, demographics, feedback on specific outputs).
3. Unavailable data: Note any user data that is relevant but unavailable due to privacy regulations or user consent limitations.
   

Part 3: How Might AI

‍

1. Use the sentence structure proposed to guide you in writing solution statements and fill in the blanks for the following questions:

• What can the AI achieve?
• What data is critical to exploit?
• What process needs to happen?