Overview
U-M GPT initially comes with prompt limits of approximately 25 prompts per hour for text-based models (GPT-3.5, GPT-4, etc.) and approximately 5 prompts per hour for image-based models (DALL-E 3). Each query asked is considered one prompt. When initiating a chat with U-M GPT:
- Access the Prompt: Log in to U-M GPT with your uniqname and UMICH (Level-1) password, and complete Duo two-factor authentication.
- Start the Conversation: Enter your query, request, or message in the U-M GPT interface. Be clear and concise.
- Interact with the Responses: U-M GPT will process your input and generate a response. Read the response and interact with it as needed. Provide additional information when answers are unclear or incomplete.
- Clarify or Correct: If U-M GPT's response doesn't address your query accurately, or if you encounter any misunderstandings, you can clarify or correct the information by providing more context.
- Fully Review Output: When using U-M GPT, or any other AI technology, it is important to fact-check to ensure the information is accurate and from trusted sources. Cross-reference with established knowledge.
Effective Prompts
Constructing effective prompts for U-M GPT can significantly enhance the quality and relevance of the responses. Refer to Improving U-M GPT Prompts for information on practicing deliberate prompt construction. Consider enrolling in the online course Generative AI Prompt Literacy for a deeper dive into mastering the art of prompting.
AI Model Descriptions
When using U-M GPT, it is also important to understand the Artificial Intelligence (AI) model offerings. AI models are tools or algorithms trained on huge datasets. They can then make predictions, recognize patterns, reach conclusions, and generate content in response to a prompt or query.
We currently offer multiple AI models in U-M GPT. While you are encouraged to experiment with the different models depending on your needs, there are advantages to using specific models for particular tasks.
Unlike some commercially available AI chat tools, ITS Generative AI tools are hosted by the University of Michigan or controlled in a U-M Cloud environment, secure, and initially provided at no cost for faculty, staff, and students. Interactions with ITS Generative AI tools are only being used for the purposes outlined in the Privacy Notice.
- GPT-4 Turbo is an updated version of GPT-4 with a knowledge-awareness cutoff of April 2023. It can be better for complex tasks, such as generating code, teaching, or explaining complex topics. However, at times it can perform more slowly than other models, so you may find GPT-3.5 or Llama 2 are better suited for less-complex tasks.
- GPT-3.5 has become one of the most popular AI models due to its speed. Many commercially available AI chat and summarization tools are built using this model. GPT 3.5 is an excellent choice for most general use cases. Training data: Up to September 2021.
- Llama 2 is an open-source model that has been trained to generate "human-like" responses. Training data: Up to September 2022.
- DALL-E 3 is an image-generation model designed to generate accurate, highly-detailed, photorealistic images based on text descriptions. Users are limited to five images per hour. Each prompt entered is completely new; you cannot ask U-M GPT to refine previously-generated images. Only standard definition (SD) 1024 x 1024 images are supported at this time.