Skip to Main Content

Generative AI at UVA

This guide features links and information about generative AI, including ethical use, citations, considerations for use, and more.

LLM Chatbots

Large Language Model (LLM) chatbots are likely the most widely known AI tools. 

One of the challenges of keeping up with this technology is the pace at which it is changing. The librarians will work to keep this page updated but keep in mind changes may happen frequently. 

This page was last updated on March 25, 2024. 

 

ChatGPT is a natural language processing chatbot driven by generative AI technology that allows you to have human-like conversations and much more. ChatGPT was created by Open AI and was launched November 2022. 

  • Pricing Information: ChatGPT's has various tiers of service. There is a free tier of service for individuals, which allows access to the ChatGPT 3.5 model. The free tier's information is up to date only as far as January 2022. ChatGPT plus is available for $20/month and allows access to ChatGPT 4 model. ChatGPT 4.0 is up to date as of April 2023. There are additional tiers of service that are more expensive as well. 

    ChatGPT 3.5 inputs are text. This text can take the form of normal text or code snippets. ChatGPT 3.5 output are the same text or code snippets, though slightly better formatted.
    ChatGPT 4 inputs are also include text and code snippets but also allows for file upload. Users can also upload word documents, spreadsheets, and images. ChatGPT 4 outputs include text, code snippets, images, and files. 

  • Source: OpenAI's large language models, including the models that power ChatGPT are developed using three primary sources of information: Information that is publicly available on the internet, Information that is licensed from third parties, information that users or human trainers provide. ChatGPT's model "reads" a large amount of existing text and learns how words tend to appear in context with other words. Then it uses what it has learned to predict the next most likely word that might appear in response to a user request. This is similar to auto-complete features on search engines, smartphones, and email programs. 

  • Corpus and training data: GPT includes publicly available information on the internet, information licensed from a third party, information provided by users or human trainers. 
    For the first type, GPT uses information that is freely and openly available on the Internet. This does not include information behind paywalls or on the dark web. GPT applies filters to remove information that it does not want the model to learn from or output, such as hate speech, adult content, personal information, or spam. GPT does not copy or store training information in a database. Instead, it learns about association between words and those learnings help the model up date its numbers/weights. The model then uses those weights to predict and generate new words in response to a user request. 

  • Benchmarks and guardrails: As stated by OpenAI, a large amount of data on the internet relates to people. OpenAI uses training information only to help our models learn about language and how to understand and respond to it. OpenAI does not and will not user personal information in training information to build profiles about people, to contact them, to advertise to them, to try to sell them anything, or sell the information itself. 

  • Support/Documentation

  • Accessibility: Browser extensions to make ChatGPT more accessible include:
    Read the Web - Text to Speech
    Natural Readers - Text to Speech
    Be My Eyes - for the vision impaired
    JAWS - for the vision impaired

  • Privacy Policy
    Statement on ChatGPT Compliance

Claude is an artificial intelligence, created by Anthropic using Constitutional AI. Anthropic claims to develop artificial intelligence in a safe, responsible way that benefits the public, not only shareholders. 

  • Claude has a free tier of service which is similar in functionality to ChatGPT. Claude Pro is $20/month for additional functionality. Inputs include text, code snippets, word documents, and spreadsheets such as excel or .csv files. Outputs are basically textual including standard text, code snippets, etc. 

  • Foundation Model: In the link cited here, Claude has an extensive policy/statement on what lengths Anthropic went to in order to ensure a responsible process of training the AI model. 

  • Corpus and training data: Claude's training data is current as of August 2023. According to Claude, Claude 2.1 was trained on a diverse range of public domain and scraped data that was carefully filtered and processed. This includes web content, expertly curated datasets, human dialogue data from movies and TV shows. Content was moderated and scrubbed of explicit or harmful language, conversations promoting hate or violence, misinformation, or biases. 

  • Benchmarks and guardrails

  • Support/Documentation

  • Accessibility: As of March 2024, there are no specific accessibility documentation

  • Privacy/Security

Microsoft Copilot is a chatbot developed by Microsoft that launched February 7, 2023. The service was introduced as Bing Chat but rebranded as Copilot. Copilot uses the Microsoft Prometheus language model, built upon OpenAI's ChatGPT-4 large language model, and has a similar conversational style interface to ChatGPT. Copilot is integrated into Microsoft's suite of softwares and services including Microsoft Office and Microsoft 365.

  • UVA Copilot is available to UVA Users!
    • Audience: UVA Copilot is specifically designed for faculty and staff at the University of Virginia (UVA).
    • Functionality: It is a chat-based service that accepts text, image, and voice-to-text prompts. Users can engage in conversations and receive AI-generated responses.
    • Security: UVA Copilot is maintained by Microsoft and operates within a secure platform where UVA-related information can be entered.
    • Access: It is available to those with a central Microsoft 365 account at UVA.
    • Generative AI: UVA Copilot leverages AI technology to provide helpful information and engage in dialogue.
    • Development: UVA Chat+ (powered by GPT-4) is also in development, offering early access through a pilot program in McIntire. Feedback from this pilot will inform its general availability
       
  • Pricing Information: Copilot has various tiers of service. There is a free tier allows use of a GPT-4 model during non-peak times. Copilot pro is $20/month and allows for more functionality.

  • Copilot is based on OpenAI's GPT-4 language model.

  • Corpus and training data: GPT includes publicly available information on the internet, information licensed from a third party, information provided by users or human trainers. 
    For the first type, GPT uses information that is freely and openly available on the Internet. This does not include information behind paywalls or on the dark web. GPT applies filters to remove information that it does not want the model to learn from or output, such as hate speech, adult content, personal information, or spam. GPT does not copy or store training information in a database. Instead, it learns about association between words and those learnings help the model up date its numbers/weights. The model then uses those weights to predict and generate new words in response to a user request. 

  • Benchmarks and guardrails: As stated by OpenAI, a large amount of data on the internet relates to people. OpenAI uses training information only to help our models learn about language and how to understand and respond to it. OpenAI does not and will not user personal information in training information to build profiles about people, to contact them, to advertise to them, to try to sell them anything, or sell the information itself. 

  • Support/Documentation

  • Accessibility

  • Privacy/Security

Gemini (formerly Bard) is Google AI. It offers a number of input formats including text, audio, and images. It offers a variety of support on areas ranging from working as a virtual assistant, writing, code, explaining topics. 
  • Large Language Model

  • Corpus and training data includes a "diverse dataset that is both multimodal and multilingual, incorporating web documents, books, code, and media data." Source

  • Google's AI Principals

  • FAQs

  • Gemini Help

  • Accessibility - As of March 2024, there are no specific accessibility documentation for Gemini

  • Privacy Hub

Further Reading

Still in beta, Latimer is developed by FutureSum.AI and designed to address the common failings of other AI tools. FutureSum describes the goal of their AI tools are "crafted to align with and amplify the distinctive voices and experiences of Black and Brown communities, ensuring their unique perspectives are not only heard but celebrated."
Further Reading

Created by Inflection AI, Pi (which stands for Personal Intelligence) is marketing as a friendly support AI personal assistant. 
Further Reading

Other AI Tools

One of the challenges of AI has been keeping track of the tools that are available. Included are only a few options that offer interesting ways of using AI. 

If you'd like to learn about other tools, Ithaka S+R is maintaining a list of General AI Tools. 

goblin.tools is a collection of small, simple, single-task tools, mostly designed to help neurodivergent people. Most tools will use AI technologies - currently they're using OpenAI's models. 

This includes:

  • Magic ToDo
    • Help break down tasks into small steps in a to-do list
  • Formalizer
    • Rewrites your text based on tone - this includes: more professional, more polite, less snarky, easier to read, etc 
  • Judge
    • Evaluates the tone of a sample text
  • Estimator
    • Offers a time estimate for an activity
  • Compiler
    • Create an organized to-do list from a brain dump
  • Chef
    • Create recipes based on preferences, ingredients, allergies, and more

goblin.tools is free and the creator plans that they will stay free without ads or paywalls. Mobile apps are available on Android and iOS

TrevorAI is a planner and task scheduling app that offers the ability to connect to your calendars and create a plan of your tasks, due dates, and appointments. It offers AI tools to help with scheduling and with creating action plans related to specific tasks.