by Kyle Triska
Quick Summary
Keeping track of the rapid pace of development of artificial intelligence is difficult. This article is a great place to start. It provides simple descriptions and links to several of the tools freely available today.
Minitex staff have begun discussing Large Language Models (LLMs) and the current generation of artificial intelligence (AI). So far, our conversations have centered on what is out there, what is possible, and what is ethical. Although LLMs, neural networks, and machine learning have existed for some time, the possibilities that have emerged in the wake of OpenAI’s release of ChatGPT have made us wonder how we can leverage this technology. We are not currently using AI in any of our products or services, although staff have tested Github Copilot and other pair-programming tools. As a library service, we take privacy, security, and attribution seriously. If, in the future, Minitex uses any of these technologies, those tenets will continue to be our priorities.
Below, you will find descriptions and links to many of the websites and apps that we have been learning about and discussing. This list focuses mainly on practical and educational implementations of generative AI, as opposed to those that are creative or artistic.
Things to Know and Watch For
Do not use any AI or LLM product or service before understanding the following.
- Terms of Service should be read. The data you are providing by interacting with a LLM most likely will be used to train it.
- Copyright ownership around generative AI has not been resolved. This is true both for content generated and the content used to train models.
- Due to their design, LLMs are prone to making errors in accuracy and can even completely fabricate information, known euphemistically as "hallucination."
- These tools are trained on data produced by humans and can be subject to human biases/prejudices.
Understanding LLMs
Catching up on the weird world of LLMs
Both a blog entry and video presentation. A moderately easy-to-understand overview of LLMs by one of the creators of Django, a Python-based web architecture.
What Is ChatGPT Doing … and Why Does It Work?
A more robust dive, specifically focusing on ChatGPT, by Stephen Wolfram, the creator of Mathematica and WolframAlpha. The article focuses on math and technical concepts, but can be glossed over to reach the core information.
A Survey of Large Language Models
A long, in-depth, academic history and survey of LLMs. Great for understanding how we got to ChatGPT4 and all of the other LLMs produced through research.
Bulletpapers
A hub for papers about AI, driven by AI. All of the papers on the website are fed to their LLM. From this, a summary is generated as well as semantic links to related content. And most impressively, they offer a chatbot that can help explain and discuss the article. I chatted with their bot about an article that discuses how colonialism affects natural language processing, a key feature of all LLMs.
Models
OpenAI Whisper
Whisper is OpenAI’s speech recognition, translation, and identification model. It can take video and audio files or streams and output subtitles. You could feed it a home movie in Japanese and have it generate subtitles in English. Certain languages work better than others, and LLMs do not translate and understand idioms or cultural context yet. But it can still accurately do much of the original translation work. Many tools use this model. The easiest to get up and running being Buzz.
Phind
Phind is a coding-centric search engine LLM. You can ask it questions about code, and it will use its training data and search the web for supplementary information to answer. One of the best things about this model is that it cites its sources and provides links to them. This is lacking from most other models.
Learning and Personal Use
Biblos (GitHub)
A semantic Bible search engine with AI summarization. A neat tool to show how viable a LLM is when trained on a publicly available dataset (the Bible).
ChatPDF and PDF.AI
Similar to BulletPapers (see above). These sites use LLMs to chat with PDFs you upload. They can summarize, answer questions, and help explain complicated concepts and jargon. These sites should only be used with open licenses that allow for sharing of content, such as the University of Minnesota’s Open Textbook Library.
PrettyPolly
A language-learning app that uses Whisper for speech recognition, and ChatGPT for the conversation agent.
Quivr (GitHub)
Promoted as a “second brain,” Quivr is a tool to chat with your data. You create buckets (“brains”) of information that Quivr stores and analyzes. You can then ask a brain questions about its contents.
Fun and Interesting
AI Town (GitHub)
A simulated town filled with chatbots, inspired by a paper about “generative agents - computational software agents that simulate believable human behavior.” The agents go about their daily activities and converse with each other. You can view their past conversations, or log in and talk to them about their experiences in the town. Someone also made a cat version.
Gandalf
A site that went viral earlier this year, that gamifies prompt injection. The point of the game is to try to force the LLM to tell you its secret password. It starts off easy, then levels up each time you get it to reveal the password. Not only is this a game, it is a tool used by Lakera.AI to out-source prompt-injection testing. They can use successful attempts at revealing the password to better guard their AI from attacks.
TrickAI
In a way, this is the opposite of Gandalf. You are given puzzles that an AI can not answer. These are often puzzles that require the creativity and understanding of wordplay that LLMs do not yet grasp. You can also write puzzles and attempt to fool the LLM. If you succeed in tricking it, your puzzle can become playable by other users on the site.
Want to know more?
Artificial Intelligence and Libraries Bibliography
A seemingly close-to-exhaustive bibliography on AI and libraries.
Impromptu: Amplifying Our Humanity Through AI
A “travelog” of conversations with GPT-4 and what its impact will be on various fields across education, business, and creativity. Written by Reid Hoffman.
Free Development Courses from DeepLearning and OpenAI
Learn how to work programmatically with LLMs like ChatGPT. Courses range from beginner to expert.
Want Even MORE?
If you found this article interesting or useful, please fill out the following survey to let us know what related contented you would like to see.
