Examples of AI

By Jen Deakin, course leader of graphic design and advanced practitioner - Runshaw College and Joel Stead, pedagogy leader - Greenhead Sixth Form College

There are many Large Language Models (LLMs) that can be used to generate content and images. In this chapter we will look at some examples created using OpenAl's ChatGPT, Microsoft CoPilot, Google Gemini, CanvaAl, Midjourney and Adobe FireFly.

The first set of examples shows the use of alternative LLMs. We then explore a range of different use-cases for one LLM - ChatGPT-4.

A comparison of LLMs, using ChatGPT, Microsoft CoPilot and Google Gemini

Using ChatGPT^[1]

ChatGPT can take prompts from users to generate text-based content. Often, the more specific the prompt, the better the output. To produce the output in figure 1, the prompt included a set of notes;^[2] based on those notes ChatGPT produces appropriate questions for the context.

Using Microsoft Copilot^[3]

By inputting an A Level Medieval History question into Microsoft Copilot^[4] (figure 2) we can produce an essay response. Specifying the Assessment Objective which needs to be met produces a higher-quality response. In addition, providing the Al with marking guidance allows it to be used as an assessment tool (figure 3), which could be used by students as a form of self-assessment, or by teachers to reduce workload (although marks and grades should never be given solely by Al). Note that these responses should be treated with caution - responses may not always fully represent an appropriate exam question answer despite additional guidance provided to the Al, and asking students to critique or flesh out an Al response like that in figure 2 may be a useful exercise in both improving essay-writing skills and showing them that Al is not infallible.

Using Canva Al Image Generation^[5]

Image generators such as Midjourney, Adobe FireFly, and CanvaAl can use text prompts to generate visuals including reference to specific styles or details. In the example below, a writing prompt for a GCSE English Language class was used to produce the two images.

Note that these should be checked carefully as Al image generators can produce irregularities - see the dog in the picture on the right.

Using Google Gemini^[7]

Google Gemini has the facility for users to upload images, which the Al will then read. Questions can be uploaded and Gemini asked to provide a solution. In questions which require mathematical notation, asking Gemini to use LaTex notation provides solutions in a mathematical format (although sometimes this fails).

In figure 4 an A Level Maths question^[8] has been uploaded to Gemini. On the face of it, the solution looks good; it is well-explained, and the concepts that Gemini is using are correct. However, Al still frequently makes simple mistakes when solving mathematical problems; it makes slips in calculations, misreads fractions, or makes errors in the method. This solution goes wrong at the factorisation step -Gemini has factorised the equation incorrectly.

Users can correct Al and it will revise its solution, although sometimes it takes many attempts to make the Al correct itself successfully. Again, asking students to check and troubleshoot Al-generated solutions may be a useful activity.

Example use cases - ChatGPT-4

Using ChatGPT, teachers can actually create their own custom GPT for a specific purpose. It is a bespoke GPT that can be more helpful at specific tasks at college. No coding is required to make these, and they can be easily shared internally, for use by anyone. Existing custom GPTs include Khan Academy's 'Tutor Me' tool and tools focused on maths and physics answers, which may be more accurate than the 'base version' of ChatGPT for those subjects. You can browse custom GPs on Chat-GPT by keyword, and these are categorised into different topics such as education.

Teachers can use Al, whether in a custom GPT or in the base version of Chat-GPT, to create custom lesson plans, educational games, and interactive activities for lessons. LLMs can also generate material for classroom discussion or resources based on the curriculum (figure 5), where Chat-GPT is used to create a comprehensive lesson plan.

Further examples of specific educational uses are below:

1. Educational games and activities - Al can help generate ideas for gamification. It can create game ideas tailored to the specific needs and interests of a class, helping teachers to vary their practice through generating relevant and engaging games. In figure 6, Chat-GPT is given the prompt "Please help me to create an educational game", and the Al responds with a series of questions to help better tailor the output. Once these questions are answered, the Al continues to design the game structure, rules and how it serves learning goals. The conversation continues as Chat-GPT is asked to write the HTML code for the game; the Al provides a simple HTML-5 skeleton with some interactive elements. The code would then need to be brought into an HTML editor, saved, and opened with an internet browser.
2. Providing feedback - Al can be a helpful tool for providing students with feedback. Some models can provide another perspective on students' work and can help them improve their drafts. We acknowledge that when giving feedback there needs to be a "human in the loop;" however, teachers can use it as an aide for giving robust and detailed feedback. Also, Al can give an additional insight into how an entire class is doing as a whole: you can give Al all of a class's responses and ask it to identify common strengths, weaknesses and misconceptions. You can see feedback in action in figure 7 where Chat-GPT is given a zip file containing student assessment sheets and asked to create a list of common strengths, weaknesses and next steps. The prompt asks that Chat-GPT not refer to any file names or student names in the output. Chat-GPT creates an extensive list of common themes which could then be used for delivering whole-class feedback. The conversation is extended to ask the Al to provide a checklist based on these common themes, so that students can use this to check their work before submitting it for assessment.
3. Language learning - For students that are learning new languages, Al can serve as a practice partner that simulates real-life conversations. It can converse in different languages, offering corrections and suggestions to improve language skills. This can be seen in figure 8, where the Al converses in Spanish and offers assistance.
4. Accessibility - Al can help make education more accessible to students with disabilities. It can convert text to speech for visually impaired students, or assist those with learning differences. In figure 9, a Level 3 tourism student asks for text to be converted to speech. The Al responds with an MP3 file with the audio.
5. Personalised learning assistance - Students can use Al as a personal learning tutor. It can help clarify complex topics through providing detailed explanations in different ways, as in figure 10. It can also offer practice quizzes tailored to the student's learning pace and preferences. There are many Al platforms that can act as a personal tutor, or an LLM like Chat-GPT can also be used. In figure 10, a Level 3 Science student asks the Al to explain quantum theory, and later asks it to explain it in a different way to help them understand.
6. Administrative assistance - Al can automate tasks such as answering common queries from students and parents, scheduling meetings and managing communications. For example, you can use the Al to generate letters and consent forms for trips, as in figure 11.
7. Professional development - For teachers who are wanting to expand their knowledge and skills, Al can serve as a resource. It can also keep teachers up to speed on educational trends, teaching tools and professional courses on offer. As Chat-GPT is currently confined to a fixed knowledge base, it does not search the internet for new information, and therefore some content could be out of date. Figure 12 shows Al suggesting SEN CPD opportunities for a further education teacher within their local area and online.
8. Image generation - Teachers can use Al image generators to produce custom illustrations, icons, photographs and infographics that could be used in lesson presentations or in dual-coded resources to explain complex concepts. We used Canva earlier in this chapter, but ChatGPT can also use the image generator DALL-E to create almost instantaneous images from simple descriptive text prompts. In figure 13, we prompt Al to "Create a simple illustration of a human heart, with spaces for students to label the parts". A simple diagram is the output, with letters for labelling. After further inspection, it is unclear whether the letters on the labels are random or relate to the first letter of the answer. This highlights the importance of checking the output, as it could make sense visually but not make sense in learning terms.

Notes

1. ChatGPT (2024). ChatGPT.
2. Study Rocket (2024). Royalties and Revenue - Level 3 Music BTEC Revision, Study Rocket.
3. Microsoft Copilot (2024). Microsoft Copilot.
4. OCR (2017). SAM and Markscheme - OCR A Level History A: Popular Culture and the Witchcraze of the 16th and 17th Centuries, OCR.
5. Canva (2024). Canva.
6. Griffin Teaching (2023). Insider GCSE creative writing tips + 106 prompts from past papers, Griffin Teaching.
7. Google Gemini (2024). Google Gemini.
8. PMT (2024). Physics & Maths Tutor.