About this guide
In the school context, technology plays a pivotal role for students, educators, and researchers. Generative AI, a subset of artificial intelligence, generates content like text, images, and code in response to prompts. These tools are versatile and can be used for tasks such as summarization, content creation, and revision. This guide provides information about using generative AI tools in an ethical, creative and evaluative way.
What is Artificial Intelligence
“The ability of a digital computer or computer-controlled robot to perform tasks commonly associated with intelligent beings. The term is frequently applied to the project of developing systems endowed with the intellectual processes characteristic of humans, such as the ability to reason, discover meaning, generalize, or learn from past experience.” B.J. Copeland, 1998
Generative AI refers to a type of artificial intelligence that has been trained to perform, replicate, or imitate tasks using enormous amounts of data.
GenAI is built on "Large Language Models", and these LLMs are based on the "Generative Pre-trained Transformer" (GPT) architecture. They recognise patterns in words and sentences and provide responses to prompts by predicting the next likely word in a given context (they are basically very powerful autocorrect).
This technology uses machine learning to mimic human behaviour and generate new content, such as music, art, or text, based on the data on which it was trained.
OpenAI introduced ChatGPT, powered by GPT-3.5, on November 30, 2022, attracting one million users in just a week. This surge in interest spurred the development of other generative AI technologies. Additionally, an enhanced GPT-4 was launched on March 14, 2023, now accessible through ChatGPT Plus and some functionality via Bing Chat.
You may also have heard of DALL-E, which was trained on a large set of image and text pairs to create new images based on textual descriptions
Generative AI tools are evolving incredibly quickly, and their impact on education and research is significant and ongoing.
- Understanding Generative AI, it's impacts and limitations (2021) - eVideo
YouTube video by Analytics Insight, 4 min.
Future of AI which aims to include the cognitive abilities and general intelligence of a human being, including the ability to reason, learn, solve problems, think abstractly, comprehend complex ideas, adapt to variety of situations, be creative and communicate effectively in natural language. Learn more about AGI.”
Types of Generative AI
Text generative AI is a type of generative AI that can produce new text content based on a given prompt, such as a word, a sentence, or a paragraph. Text generative AI can be used for various purposes, such as writing stories, poems, essays, summaries, captions, headlines, and more.
Some examples of generative AI that can create text content include: ChatGPT and Perplexity AI.
Possible uses:
- Provide feedback on the quality of writing, grammar, spelling and coherence. Can also suggest improvements to layout, structure, phrasing and sentence structure.
- Summarize large amounts of information, including class or research notes.
- Produce practice tests with a variety of question styles.
- Help overcome the "fear of a blank page" by providing stimulus or generating ideas to extend and develop.
- Assist with time organisation.
Limitations:
- Sources are not referenced or cited and you are unable to see how it reached the answer.
- Can provide incorrect information with certainty (known as hallucinations).
- May be too generic or contain biases.
- Can misinterpret the given prompt.
Image generative AI works because researchers and developers feed machine learning programs hundreds of millions of images and descriptions that have been scraped from the internet. The model is then able to learn and identify relationships between the images and words. These models inherit the bias of their training datasets.
These tools can produce diverse images in various styles, everything from anime, post-impressionism and watercolours; however, changing or tweaking a produced image can be difficult.
Some examples of generative AI that can create imagery include Dall.E 2, Midjourney and Stable Diffusion. They have also been worked into programs such as Adobe and Canva.
Possible uses:
- Create unique images for presentations.
- Turn concepts into images with little to no artistic ability or experience.
- Can provide creative inspiration.
Limitations:
- Unknown copyright implications if copyrighted works from other artists were used to train the model.
- Quality of images can vary.
- Making changes to an image can be difficult.
Isn't it simple to consume videos and audio? You choose the video or audio you want and press play. But making it is much more difficult and time-consuming. You'll need a script or lyrics, recordings, and editing, among other things. Some of these tasks, if not all, could benefit from generative AI.
Music tracks and metadata (artist name, album name, genre, year song was released, etc.) are analysed by AI music generators to detect patterns and features. They can also be trained on song lyrics, and some can even produce sheet music.
Suno, AIVA, Soundful, and Murf.ai are some instances of generative AI that can create audio material.
Creating a video typically requires the use of audio, visual and text elements. There are generative AI video programs that have been trained on some or all of these elements allowing you to control what is created.
Some examples of generative AI that can create videos include Sora, Gen-1 Runway and Invideo.
Learning to code is similar to learning a language, and can be just as hard.
Instead of searching the internet for help when coders get stuck, generative AI models can now be used to help with generating and improving code or even finding errors.
This means that you could use a tool like GitHub Actions to reformat new code to match old, instead of spending hours manually formatting it yourself; or even build a complete website with no coding experience. But beware, it'll likely come at a cost. check out this useful resource: A Beginner's Guide to Prompt Engineering with GitHub Copilot
Some examples of generative AI that can create code include ChatGPT, CodeT5 and Tabnine.
There are many generative AI tools that claim to automate parts of the research process and make long, complex texts easier to decipher. This type of AI often analyses research papers that users upload to extract key information or summarise a paper. they can also help with literature review mapping and citation.
Some examples include: ChatPDF, Elicit, Raxter, Research Rabbit and Scite
Possible uses:
- Compare and evaluate how you have interpreted a research paper.
- Locate other research papers that might relate to your topic.
- Summarise papers and extract keywords.
Limitations:
- Can provide incorrect information with certainty (known as hallucinations).
- Can provide inaccurate analysis or evaluation of a research paper.
- Might miss relevant papers or words.
Get ready for the future: everything you need to know about AI
Series: Crash Course AI
Crash Course (2019, Aug 10. What is Artificial Intelligence? [Video]. YouTube https://www.youtube.com/watch?v=a0_lo_GDcFw&t=55s
Crash Course (2019, Aug 17). Supervised Learning [Video]. YouTube https://www.youtube.com/watch?v=4qVRBYAdLAo
Crash Course (2019, Aug 24). Neural Networks and Deep Learning [Video]. YouTube https://www.youtube.com/watch?v=oV3ZY6tJiA0&t=36s
Crash Course (2019, Aug 31). Training Neural Networks [Video]. YouTube https://www.youtube.com/watch?v=lgKrup5oi_A&t=5s
Series: Practical AI for Instructors and Students
Wharton School (2023, July 31). Practical AI for Instructors and Students Part 1: Introduction to AI for Teachers and Students [Video]. YouTube https://www.youtube.com/watch?v=t9gmyvf7JYo
Wharton School (2023, August 1). Practical AI for Instructors and Students Part 2: Large Language Models [Video]. https://www.youtube.com/watch?v=ZRf2BfDLlIA&t=592s
Wharton School (2023, August 2). Practical AI for Instructors and Students Part 3: Prompting for AI. [Video]. YouTube https://www.youtube.com/watch?v=wbGKfAPlZVA&t=15s
Wharton School (2023, August 3). Practical AI for Instructors and Students Part 4: AI for Teachers. [Video]. YouTube https://www.youtube.com/watch?v=SBxb5xW7qFo&t=21s
Wharton School (2023, August 4). Practical AI for Instructors and Students Part 5: AI for Students. [Video]. YouTube https://www.youtube.com/watch?v=ZorvXYUZtRg&t=48s
What is ChatGPT
The AI Iceberg - Understanding ChatGPT
In his blog post, Leon Furze uses the analogy of an iceberg to explain some of the features of ChatGPT, a chatbot application built on top of a large language model (LLM) like GPT-3 or 4.
He compares the vast dataset that the LLM is trained on to the underwater bulk of the iceberg, which forms the foundation of the model’s knowledge and capabilities. He then compares the LLM itself, which is the result of the training process, to the visible part of the iceberg that we interact with. He also discusses some of the challenges and limitations of using LLMs, such as misinformation, bias, inappropriate and toxic content, and data privacy issues.