Navigating Generative AI Tools at Harvard: A Comprehensive Guide

Generative AI is rapidly transforming various fields, and Harvard University is at the forefront of exploring and utilizing these powerful tools. This article provides an in-depth look at the generative AI tools available at Harvard, focusing on their capabilities, approved data classification levels, and how to access them. We'll explore the different categories of tools and provide guidance on responsible AI usage, adhering to Harvard University's AI guidelines.

HUIT's Role in Generative AI

Harvard University Information Technology (HUIT) plays a crucial role in providing access to and guidance on using Generative AI tools within the Harvard community. They carefully evaluate and curate a selection of tools, ensuring they meet the University's security and data privacy standards.

Generative AI Tools Categories

HUIT categorizes Generative AI tools into two primary groups:

1. AI Tools for General Use: Designed for broad applications and productivity enhancements, these tools are accessible and user-friendly.
2. AI Developer Tools: Targeted towards developers, these tools offer APIs and functionalities for integrating LLMs into custom applications and services.

AI Tools for General Use: Enhancing Productivity and Creativity

This category includes versatile AI assistants ("chatbots") that understand and generate human-like responses to text prompts. These tools can:

• Generate text, code, and images
• Translate languages
• Write creative content
• Summarize information
• Integrate with productivity and collaboration tools

Specific tools in this category include:

• Harvard AI Sandbox: A secure environment to experiment with multiple Large Language Models (LLMs). It includes features like code generation, creative writing, data analysis, summarizing, text generation and editing, image generation, and translation. Access is coordinated by Schools and Units; contact HUIT for details. Approved for: Level 3 data and below.
• OpenAI ChatGPT Edu: A versatile chatbot capable of generating text, code, images, and more. It offers chatbot customization, code generation, creative writing, data analysis, image generation, summarizing, text generation and editing, and translation. Access is provided and coordinated by Schools and Units; contact your local IT department for details. Approved for: Level 3 data and below.
• Adobe Firefly: Generate images and text effects by simply typing keywords or a description. Trained on stock images, openly licensed and public domain content. Also integrated into Adobe apps.. It offers image generation and image editing. Available to Harvard faculty, staff, students, and researchers as part of Harvard Adobe Creative Cloud license. Approved for: Level 3 data and below.
• AI Meeting Assistants: While generally discouraged due to privacy concerns, HUIT conducts limited pilot programs with approved tools to evaluate their potential within the Harvard environment. Refer to AI assistant guidelines for more information.

AI Developer Tools: Building Custom Solutions

These tools are geared towards developers who want to integrate Large Language Models (LLMs) into their applications, products, or services. They provide functionalities for:

• Chatbot creation and customization
• Building and testing applications
• Accessing model training and deployment
• Coding
• Predictive analytics

Current limited access offerings include:

• Amazon Q
• Amazon Bedrock
• Amazon Sagemaker
• Azure AI Studio
• Copilot Studio
• Google Vertex AI
• OpenAI APIs

These tools are available by request from HUIT. Approved for: Level 3 data and below.

Data Classification Levels

Understanding data classification is crucial when using Generative AI tools. Harvard University categorizes data based on its sensitivity, and it's essential to use tools that are approved for the appropriate data classification level.

• Level 3 Data and Below: These tools are only approved for non-confidential data. Refer to the data classification table for detailed information on each data level. Avoid entering data classified as confidential (Level 2 and above) into publicly-available Generative AI tools.

Important Considerations for Using Generative AI

• University Guidelines: Always adhere to Harvard's initial guidelines for using Generative AI.
• Vendor Risk Assessment: If planning to use a private Generative AI tool, ensure the vendor undergoes a risk assessment by Harvard's Information Security and Data Privacy office before using it with any non-public data. Request a vendor risk assessment.
• Privacy and Security: Be mindful of privacy and security risks, especially when using AI meeting assistants. Refer to the AI assistant guidelines to learn more information about risks and availability.

Staying Informed and Getting Help

AI is continually evolving, so staying updated is key.

• HUIT Website: Regularly check the HUIT website for updates on available tools, guidelines, and training opportunities.
• Local IT Department: Contact your School's local IT department for additional tools and support tailored to your specific needs.
• Contact HUIT: If you need advice, have questions, or have specific AI tool requirements, Contact HUIT to help clarify your needs.

Conclusion

Generative AI offers exciting possibilities for enhancing productivity, fostering creativity, and developing innovative solutions. By understanding the available tools, adhering to University guidelines, and staying informed, Harvard community members can harness the power of AI responsibly and effectively.