AI Literacy for Librarians and Educators

Covers a bit of history, defines terms like narrow AI versus general AI, machine learning, deep learning, neural networks, probabilistic models, and “black box.”

Covers the 2017 breakthrough known as transformer architecture, OpenAI - the company behind ChatGPT, the meaning of “GPT,” how models are trained, and guardrails.

Focuses on the difference between “discriminative AI” (AI that sorts things into categories), and “generative AI” (AI that generates new content, like text, images, video, or voices). Also mentions models beyond ChatGPT like Copilot, Claude, Gemini, and tools built on OpenAI’s API.

Generative AI systems are hybrid. They combine a language model with a source of facts (like results from a search engine). These models can search the web and use the AI to summarize and link to those results. Some models, like Elicit, are grounded with scholarly papers from Semantic Scholar.

Links to a few short video excerpts that explain embeddings, tokens, and context windows.

Discusses an open source platform called Hugging Face where you can host and train your own AI models. Mentions the importance of open source (or openly-licensed) models for transparency, innovation, education, and ethics.

Learn to describe the technology in a simple way.

Learn about two common AI myths that many people believe - and what the real story is.

Review the differences between discriminative and generative AI.

Try Perplexity, which summarizes results with links to web sources.

Use Elicit to help find scholarly articles.

Covers many of the settings and features of ChatGPT and other models, like sharing your chat, why to start a new chat for each new topic, formats you can ask for in your output, how to turn off “improve the model for everyone” for future training of a model, and setting up Custom Instructions.

How to create a custom GPT (OpenAI), a tour of the GPT store, free tools for creating chatbots, and a walk-through of building a custom GPT called “Chat with Scrooge” (from A Christmas Carol by Dickens). How building these helps you understand the technology.

Best ways to acknowledge use of generative AI, guidelines from publishers and style manuals, AI writing detectors don’t work, students have been falsely accused.

Try some prompting techniques.

Experiment with some chatbots (custom GPTs) made by others.

Build your own chatbot with a free tool like Character AI or Poe.

An overview of two categories of copyright issues: the output of generative AI and the input (training data). Details about different points of view from Creative Commons, American Library Association, and others. Info about fair use, “robots.txt” protocol, web crawlers vs web scrapers, different copyright rulings from countries outside the U.S. Experts say it will take years for the courts to issue final rulings on these.

Training data is biased, what different companies have done to mitigate bias, various reasons for bias with examples, asking for unbiased results in your prompts.

Remember that these issues about low-wage labor for content moderation apply not only to AI, but also to YouTube, Amazon, Meta, and Microsoft -- who also use these services.

Learn to put the climate issues of AI into context, since many stories in the media take statistics out of context.

Try a conversation with Gemini based on particular articles about content moderation. Use the prompts in this activity to generate a fictional debate.

Set up a fictional debate about copyright, based on specific articles.

Use these instructions to have Claude write a prompt you can use to evaluate news stories about the climate impacts of data centers and AI.

Using Google's NotebookLM with a group of documents about generative AI and healthcare.

Uploading files to ChatGPT and Claude. Data visualization, analyzing data from spreadsheets, converting file formats, generating Claude Artifacts, human conversation as interface.

Uploading images and working with them. Getting descriptions, critiques, conversions, advice, practical help, ideas for design improvements, helping people with vision impairments, and generating alt text.

Voice mode is available in several genAI mobile apps. Talk to it and it talks to you. Demo of voice mode in Gemini for getting book recommendations. Various features and settings. Demo of live camera feature in Gemini and integration with Maps in Perplexity. Use cases for accessibility, language translation, role-playing for education, and more.

Using HeyGen to translate speech in videos with lip syncing, real time language translation on mobile phones, differences between Google Translate and generative AI translations, tools for generating transcripts and summaries on YouTube, getting transcripts of audio or video with Descript, reading a transcript while listening to a podcast - good for people with dyslexia

This video shows how students can use AI tools like Claude and NotebookLM to better understand and study academic papers. It walks viewers through uploading documents, asking questions, and generating summaries using Claude. It also introduces NotebookLM as a platform for organizing multiple sources, extracting key points, and creating study aids like outlines and flashcards. Both tools are presented as ways to engage more deeply with scholarly texts and make sense of complex information.

Use Seamless from Meta to translate a recording of your voice to another language.

Use Descript to generate a transcript of an audio recording.

Use computer vision to write alt text for an image.

Using computer vision to write out text from an infographic image.

Experimenting with audio overviews in NotebookLM

Use "stream realtime" from Google to have a live conversation about what you see on your computer's screen.

Why it’s important to know what can be generated. Examples of generated images: realistic animals and people, historical-looking photos, art styles of the past, abstract patterns, cultural commentary, objects that don’t exist. Image creation in ChatGPT. Restyling images (no styles of living artists allowed), why style isn’t covered under copyright. Adding or replacing objects in images, comic pages, generating infographics. Realistic photos of public figures can be generated in various settings (but not in dangerous settings). Understanding multimodal generation - a unified model for text and images.

Other image models, their features and methods. Microsoft Designer, Firefly, Ideogram and Midjourney. Image generation features built-in to other tools. Image to image, inpainting, outpainting, image to 3D, text to 3D environment, sketch to image. How the technology works to generate images (autoregressive models and diffusion models). Popular datasets for image training, image tutorial from University of Arizona.

Examples of stereotypical and biased images. System prompts, the DALL-E system card, ways that companies (OpenAI, Adobe, Runway) are trying to mitigate bias, Google blocking requests for images of people, how prompt to avoid bias. “Inclusive Prompting Glossary” from Dove.

This video looks at AI video creativity, from polished “talking head” generators to artful Midjourney‑plus‑Pika mashups, and closes by raising ethical questions about verifying what we watch.

Eleven Labs for voice generation, with examples. Cloning your own voice and getting paid when it’s used. Generating a new voice for a patient who lost speech due to a brain tumor. Speechify for turning any text to audio. Generating music with Suno, with examples. Idea for prompting for music. Fears and unethical uses of voice and music generation. Beneficial uses of these technologies, with quotes from musicians and music professors.

Issues in two areas: output (art or music you create, can it be copyrighted?), and the input (training AI). Updates on some of the cases from independent artists, parts thrown out, parts that continue. Lawsuits against music generators from big labels like Sony. Response from Udio and Suno. Artists fighting back with poisoning tools and why experts say it won’t work. Tools for artists to train AI on their own work.

Predictions and harms of deepfakes. AI deepfakes and elections. Misperceptions about deepfakes from important studies. How to spot AI-generated content - make it a standard part of information literacy. Tips for what to look for in generated images. How this may not work in the future as the technologies get more realistic. “Content credentials” metadata for tagging AI-generated images. Ideas for watermarking non-AI images from Nikon, Sony, and Canon. It’s easy to remove watermarks. Using reverse image search to investigate the origin of an image. Ideas for assignments and activities for teaching about deepfakes.

News stories are mostly about artists against AI, but some artists are embracing generative AI. Many examples of artists, architects, and designers who use generative AI. Artists in South Africa using generative AI. “Disrupting the narrative” by using AI images for cultural commentary and activism. AIArtists.org website - recommended. Artists with disabilities using generative AI. How art schools are dealing with this. Tips for what you can do if you’re worried about the ethical issues. Ideas for educational uses of generative AI multimedia.

Use specific prompting techniques to generate images of diverse people.

Use specific prompting ideas for generating images in various art styles.

Complete a guessing activity for a set of images (which ones are AI-generated?)

Use Suno to generate music.

Use Eleven Labs to generate voices.

Use Hedra to generate a lip-synced video clip from an image.

Common misconceptions about AI’s societal effects. Historical examples of technology-related media scares and flawed predictions. AI related news stories often lack crucial context, especially about data center usage and carbon footprints. Looking at both legitimate concerns and positive developments, including AI's role in climate change mitigation and advancing renewable energy. Ideas for practial actions to take and for using Mike Caulfield’s SIFT method to evaluate news coverage.

Explore the new “deep research” feature in many different models, including ChatGPT, Gemini, Elicit, and others. See example of research reports, look at benchmarking reports that evaluate these tools, and look at specific use cases.

Use this exercise to understand information about synthetic data that became misleading hype in the news.

Learn about solutions journalism as a possible way to mitigate hype-filled negative news stories.

Use a custom bot that detects misleading language in news stories.

Use a fact-checking bot that gives context and background to claims in order to help you fact-check.

Use Deep Research in ChatGPT or Gemini to generate a report.

How to stay current. Cast a wide net and follow different types of publications. Look outside librarianship to all sorts of professions. Look at sources from around the world. Follow different types of sources, like newsletters, social media, podcasts, videos, automate with alert services. A few recommended books. Tips for people who don’t have a lot of time. Tips for people who like to experiment with new technologies and want to spend more time keeping up. Recommended guides and tutorials from University of Arizona Libraries. How to follow the author of these videos.

A short list of recommended sources for staying current.