AI Tools and Resources

The current generation of artificial intelligence (AI) algorithms and tools are descendants of pioneering work on cognitive science, computer science, economics, game theory, and mathematics going back to the 1950s. 

What do the recent developments in AI platforms and their public availability mean for us as students and teachers? The technology is evolving rapidly in its potential uses and ethical issues. This guide provides background and resources to help you get a handle on it! 

Modern AI platforms are composed of multiple groups of algorithms with different goals. At their simplest, these platforms take training data, use machine learning algorithms to "learn" from this data, and then pass on what it has learned to a model which uses this knowledge to generate some output. Below are some simple definitions for key ideas related to modern AI platforms.  

• Algorithms are a process or set of rules to be followed in calculations or other problem-solving operations, especially by a computer.
• Generative AI is a type of AI system that generates text, images, or other media in response to user prompts.
• Generative Pre-Trained Transformer (GPT) is a LLM developed by OpenAI using unsupervised learning.
• Large Language Models (LLMs) such as ChatGPT apply deep neural networks to analyze huge amounts of text and use this information to respond to prompts from users.
• Machine learning is a sub-field of AI focused on the problems of designing recursive algorithms capable of learning.
• Natural Language Processing refers to a branch of artificial intelligence concerned with giving computers the ability to understand and generate text and spoken word in the same way humans can.
• Neural Networks are an approach to machine learning using many simple, but densely connected algorithms to solve complex problems. Deep Neural Networks employ many layers of neural networks. 
• Supervised learning is a machine learning technique where the authors of the model tell the machine learning algorithm how to handle the training data in order to generate the desired output.
• Training data are the information that is digested by a machine learning algorithm. 
• Unsupervised learning is a machine learning technique where the machine learning algorithm creates its own labels for variables within the training data.

This is a brief overview of better known Large Language Models (LLMs) as of September 2023. Initially, they were available as free, beta versions but many are moving towards paid plans for improved versions. 

Ethical questions:

• As these tools are monetized, are they creating more social inequities by favoring those who can afford to pay for access?
• Do these tools protect privacy and intellectual property? For individuals, what do the Terms of Use or FAQs say? At a minimum, default settings in these systems use what you enter for their ongoing training and might factor into the responses they provide to other users.

ChatGPT - free 
Developed by Open AI and built on GPT-3.5 which responds to text prompts.
Not connected to the Internet. Trained on Internet data only through September 2021.

ChatGPT Plus - monthly fee
Developed by Open AI and built on GPT-4 which can respond to image and text prompts.
Connected to the Internet 

Bing Chat - free with a Microsoft account
Developed by Microsoft and built on GPT-4 
Connected to the Internet 
Use with Microsoft Edge or with other browsers using the New Bing Anywhere extension.

Bard - free with a Google account
Developed by Google and built on PaLM2
Connected to the Internet 

Claude 2 - free with account sign-up; pricing for premium version just released 
Developed by Anthropic using their own model. 
Claims it is focused on reliability and safety. Not connected to the Internet.

Perplexity AI - free with account sign-up needed to use GPT-4 model; also has a paid Pro version
Developed by Perplexity and built on GPT-3.5 and GPT-4
Connected to the Internet