Typically, when we talk about AI we are referring to large language models (LLMs) like ChatGPT that generate human-like text. It is important to remember that LLMs do not "think", they are trained to predict the next logical word(s) in a sentence based off of patterns and probabilities in their training data. 

The different LLMs are built on different neural networks with different training data. Some power applications that are built for specific purposes.  We use an Application Programming Interface (API) that is being powered by a specific LLM. For example OpenAI (LLM) is used to power APIs like ChatGPT.

Artificial Intelligence: the ability of computer systems to mimic human intelligence 

Generative Artificial Intelligence (GAI): Any AI that uses deep learning models to generate or create new content. LMMs and Diffusion Models are both examples of GAI

Artificial General Intelligence (AGI): AI that can think, make decisions, and feel like humans can.

• Expert Systems: uses rules and logic to anticipate a wide range of possible scenarios
• Machine Learning: uses probability and statistics to recognize patterns and generalize
• Neural Networks: a type of machine learning model - computing systems modeled like the neural connections in the human brain.

Foundational Models: deep neural networks trained with large data sets using machine learning techniques that mimic human trial and error

     Large Language Models (LLMs): foundational models that focus on language

     Diffusion Models: foundational models used to create images and videos

GPT: Generative Pre-Trained Transformers - this architecture is used by Large Language Models and Diffusion Models. 

• Generative: produces text or images by sorting probabilities of the next word or pixel
• Pre-trained: requires and enormous data set
  • Supervised Training - data is labeled so the model learns associations between inputs and desired outputs. Ex: Email Spam Filter
  • Unsupervised Training: data is not labeled and the model learns to associate unseen trends or relationships
  • Reinforcement Learning: unlabeled data and algorithms in an environment where specific outputs are rewarded
• Transformers: technology that allows the tokens (parts of words/characters/pixels found in the code) to be compared with each other simultaneously.
• Parameters: internal variables in a neural network that can be adjusted to change the output

Reference

Definitions adapted from Bowen, José Antonio, and C. Edward Watson. Teaching with AI : a Practical Guide to a New Era of Human Learning. Johns Hopkins UP, 2024.

Inputs

The input that we give to a LLM is called a prompt. A prompt is typically more literal and conversational than a Google search as AI is trained on natural human language. A common piece of advice is to treat the AI you are using as very bright intern on their first day of working for you. Prompting is an iterative process - meaning most tasks will take several rounds to get AI to produce the outcome that you want. 

Common Prompting Advice:

• Bland/generic verbs produce more bland content
• Variety is confusing - avoid synonyms when referring to the same thing
• Negative commands are confusing - try to use positive instructions - ‘do this’ instead of ‘don’t do this’
• Responds well to models - provide of examples of what you want the response to be like
• Asking it to slow down or to work in stages can produce better results
• Asking it to identify any need for clarification before it starts can help improve the prompt

Reference:
Prompting advice adapted from Bowen, José Antonio, and C. Edward Watson. Teaching with AI : a Practical Guide to a New Era of Human Learning. Johns Hopkins UP, 2024.

CLEAR Framework

Developed by Leo S. Lo, the CLEAR Framework for Prompt Engineering is designed to build critical thinking skills into prompt engineering. CLEAR stand for:

• Concise: Use clear and focused language in your prompt
• Logical: Use a logical flow and order with your prompt
• Explicit: Specify any desired output format, content, or scope
• Adaptive: Be flexible and try new approaches based off of the output you are getting
• Reflective: Continually evaluate and improve your prompts 

Reference:  Lo, Leo S. "The CLEAR Path: A Framework for Enhancing Information Literacy through Prompt Engineering." .Journal of Academic Librarianship , 4 (2023)

Graphic created with Napkin.ai based off the work of: Lo, L. S. "The CLEAR Path: A Framework for Enhancing Information Literacy through Prompt Engineering." .Journal of Academic Librarianship , 4 (2023) .-CC:BY-NC 4.0