I totally agree with what you said about checking the validity of Ai being the hardest thing. I had a similar experience where I was asking ChatGPT for information and it just completely lied to me and I had to correct it, which is crazy!
I didn’t know how wrong Ai could be with computer science questions, but I guess it makes sense with what we learned this week that Ai is really good at creative human things and not at these high tech type questions. At least not always. Even with the shark tank game you played it’s super interesting how Ai can imitate and handle these scenarios full of cultural knowledge so well.
Do you think if someone asked Ai to write a script for a show like shark tank that you would be able to tell the real script from the Ai one?
Before this module, I knew very general things about Ai and had only played around with ChatGPT and various free image generators. I learned through this week’s readings much more about Ai and how best to utilize it. I’ve always been a bit weary of generative Ai tools but I think through this process I became a bit less scared of using these programs available to me.
What I learned:
LLMs are large language models, these represent a sub-section of generative Ai tools like ChatGPT, Bing Chat, and Gemini Google that produce human-like text based on prompts and in some cases are connected to the internet.
This is what ChatGPT gives as a definition of generative Ai:
“Generative Ai is a type of artificial intelligence that creates new content—like text, images, music, or video—by learning from existing data. It uses models like GPT for text or GANs for images to generate realistic, creative outputs, mimicking human-like creativity. It’s used in various fields like art, content creation, and entertainment.”
I learned about how Ai is basically undetectable, how it is based on prediction and how even it has its limitations. For example, 10-20% of what they produce is hallucination, it’s not real and they are not good at understanding the text they are putting out, there is also a large environmental impact of using so much power.
I learned about SAMR and TPACK, two models of evaluating the use of technology in learning.
I chose to use Gemini Google because unlike ChatGPT which I am more familiar with. This LLM is connected to the internet. I made sure to listen to the advice given to me for talking to an Ai like giving the Ai a role to play. Telling it who it was. So I told it it was a student at university. I broke my request down into steps like I was suggested to do to help it process what I wanted from it and not have it get confused along the way.
This is what I asked of Gemini Google:
You are an education student in university writing about the pros and cons of the SAMR model of learning. First, give 3 examples of substitution in the classroom and secondly describe their pros and cons in terms of learning.
This first response was lengthy but pretty accurate. It understood what SAMR model was and what I wanted and assumed the role of an education student. It delivered correct information on the types of substitutions and their cons in the classroom. My one complaint with this was that it was too lengthy so I tried again:
Could you shorten this information down to be more bite sized?
This was much better. The information was made more digestible but I didn’t like that the pros and cons weren’t specific to the examples of substitution it gave me, so I tried again:
Could you write that again but with the specific cons attached to each example of substitution.
With my third try I finally got the result I wanted. It definitely helps to pick at the Ai to get it to do what you want. Doing this did make me feel a bit bossy though.
Then I repeated the same process but with a slighlty more specific prompt.
You are an education student in university. You will write a short TPACK analysis of the use of a Generative AI tool for learning, specifically gemini google.
Then I felt stuck in a loop trying to get different formats of the answer. I kept getting the same format and undergoing the same shortening process. I found it hard and a bit tiring to type such direct instructions to the AI. I know that is the easiest way for it to understand our requests, but it makes me feel bad.
A lot of the answers and things Ai generates can be very cookie-cutter perfect, in a way that becomes bland. I found it fascinating how the term Ai generated is associated with things that aren’t very good. It is true that Ai allows for creativity to be accessed by more people and can make creation faster. However, there are still so many things about Ai, like using artists’ content from the internet without consent and a lack of critical thinking that limits it and makes it sometimes a questionable source of help. This module did show me different ways however in which I could use Ai to benefit my learning experiences. Ai is not perfect but it is a growing tool that cannot be ignored and we should all learn how to use effectively.
“Define Generative Artificial Intelligence, a short and sweet version plz.” prompt. ChatGPT, OpenAI, 10 Oct. 2024, https://chatgpt.com/?temporary-chat=true.
“You are an education student in university writing about the pros and cons of the SAMR model of learning. First give 3 examples of substitution in the classroom and secondly describe their pros and cons in terms of learning.” prompt. Gemini Google, 1.5 Flash, 10 Oct. 2024, https://gemini.google.com/app/e1f6fea957ad0cb2.
“Could you write that again but with the specific cons attached to each example of substation. ” follow-up prompt. Gemini Google, 1.5 Flash, 10 Oct. 2024, https://gemini.google.com/app/e1f6fea957ad0cb2.
Hi Markus, I like that you accommodated your screencast content to be more non-technical for those who might not be familiar with what you are talking about. Because I think the intrinsic load of what you are trying to teach is fairly high already you were able to minimize that. Saying that though I still had no idea what you were talking about (I am not in computer science) haha nice try though. Your screencast was very relaxing.
I like that you went into your personal experiences to describe examples of load theory. To answer your question: I had experiences with extraneous load when it came to teachers who began to teach things that weren’t a part of our learning curriculum. They would slip up and start to teach ahead when the unit or whatever we were working on didn’t include that.
Now, I have a question for you. How would you label your video perhaps? What would you title it so students could decide if this was at their level or not?
As I am an art history student, not an education student, I am very new to all this language and ideas of learning. For example, I had never heard of cognitive load theory or Richard Mayer’s principles of multimedia learning or Allan Pavio’s dual coding theory.
So for this week, I learned how to use Screencastify to record a short educational video with Google Slides. The video might look simple but creating it was a frustrating experience and took me some time. Keeping in mind that humans are dual-channelled and that I need to mix various media, I began creating my Screencast. I made slides of different colours and fruits with the corresponding colours. Then I wanted to find a child’s voice saying the colours so I found a video and downloaded the audio. It took that audio and spliced it so I had the colours I wanted and created a new audio file to then put into my presentation. The last step was recording my screen whilst I timed the audio to each slide.
While making my screencast I kept in mind the principles of modality and redundancy. Both state that we learn best from visuals and narration, not visuals and text, so I made sure to not include any text and find audio narration. I also implemented the voice principle, which states people learn better with real voices, not robots and I thought that since my video is intended for small children, having a child speak the colours would be more inviting.
What I found challenging was the simplification of it all. It is very tempting to add more flair to these presentations, with more text and everything. However, I had to try hard to avoid adding distracting elements that would just become an additional load or extraneous load.
I imagine the audience for this screencast would be very young children who are still learning colours and their associations. It’s about getting the bare basics of these names of colours and what they might represent by hearing them over and over in association with the images on the screen. I would’ve made it longer but I think this gets the point across for us adults.
As an art history student much of my learning is presented to me in visuals next to text, so learning about dual coding theory was interesting to me. Many of these principles as I could tell, were just basic visual design principles, how to make something look good and understandable. It was interesting to hear those ideas shared in these videos like it wasn’t something already self-explanatory, like the rule of spatial continuity, for example, or signalling. A principle that I found interesting was the segmenting principle, which in one of the videos they compared to how a book has chapter breaks.
Many of the principles were pretty intuitive to me when I learned them, like signalling and spatial continuity but the redundancy principle I haven’t always followed too well. As an artistic person sometimes I like to take the simplicity out of things, so this is something I will try to focus on moving forward. I did however learn through this assignment about cognitive load, intrinsic difficulty, extraneous load and germane knowledge were new terms to me and helpful in categorizing these different principles that I will now try to remember and integrate into future assignments.
