Chat GPT, Midjourney, AI… there are a lot of terms flying around at the moment. But this also opens up an incredible number of new possibilities. At the Nerd Republic, we always involve AI for smaller jobs and as a creative sparring partner. So now we wondered, is it possible to create a self-learning course 100% using AI? We are going on an adventure and documenting our steps here. Will we be successful? Who knows…
What are we trying to achieve?
The aim is to have a self-learning course produced by an AI. We are aware that we will combine different AIs in order to achieve the best possible result. Simply pressing a button and three minutes later everything is there will not (yet) work.
We have chosen the upcoming course “Agile Estimation and Prioritization” as our self-study course. We are of the opinion that the topic can be grasped well, i.e. is challenging enough, while it should also be implemented well by AI. There are numerous sources and articles (the AI has been trained) that can be referred to. At the same time, the course can create good added value and can be easily differentiated.
Basically, we want to satisfy our curiosity a little. Will it work? How will this course feel? What will the feedback be like? Will it all be like this in the future? Where will boundaries arise? Everything is so exciting 😉
If you don’t just want to read the blog, but also want to experience the course, you’ll find it here:
Where do we see limits to course creation with AI?
We will see where the limits are at the moment. However, as we already have some experience with AI, we are currently assuming that it will probably not be 100% AI. To be honest, we had already thought about having this accompanying blog created entirely by the AI. Then again, it is probably important to have a “safe space”, and we ultimately want to document our steps as well as possible.
What we will probably do manually is the front end of the course. This is where we describe the content and the AI setting of the course. In addition, we bring in our knowledge of the subject and use it to create the prompts. The course will also have a few typical Republic chapters such as “Welcome”, “Change Log”, “Introduction Republic”. We will also manually pull in these elements from existing courses and adapt them accordingly.
What comes from the AI?
At the start of this project we assume the following setting:
- Course structure is created by the AI of Learnworlds
- eBook content is created by Chat GPT, Creaitor and others
- Images are created by Midjourney
- Learning videos are created by Synthesia
What questions are unanswered at the start of the AI project?
We have a few ideas that we would like to implement, but we need to delve deeper into the subject.
You probably know that there are different types of learners. Some learn better by reading, some by watching, and others like to listen or experience learning. We want to try to find the perfect solution. The whole thing should work via an initial assessment and on the basis of this, the course is put together for the learner in a personalized way, basically multilingualism would also be possible (mentally). We have ideas for this, but need to test it a little. Perhaps this is not the right way to go.
October 14th, 2023 – Online
The course is online. We have built the frontend and put the course on pre-sale. The frontend was largely created without AI. Too many elements are too special for this, especially with regard to the typical course structures of the Nerd Republic. Most of the images were created with Midjourney.
October 22nd, 2023 – The Course Structure
The time has come. We start with the structure of the course. An AI is connected to our learning platform. It is possible to enter parameters for the course structure via a UI. The AI then generates a structure from this.
We want to try to get it right at the first attempt. We lose some of the fun if we screw, turn and twist ten times. So we put in as much as possible and work with the structure we get. We keep potential deletions to ourselves and only include topics in a new iteration once our product has been produced in iteration 1.
Our parameters:
- Activities: Include videos, eBooks, audios, exams and a Certificate of Completion
- Build eight sections (we will add our classic introduction and conclusion manually later)
- Include six sections with activities per section.
- Use the following key topics: Challenge of Project Planning in VUCA, Basics of Agile Estimation, Story Point Estimation, Fibonacci Series, Planning Poker, Velocity, Prioritizing in Agile, Weightest Shortest Job First, Magic Estimation, Getting Things Done, MoSCoW Prioritization.
- Base the learning activities on the following paragraph:
- “Agile estimation plays a critical role in the effective planning and implementation of projects. Mike Cohn, a renowned expert in Agile methods, has made a significant contribution to this with his concepts. Using the Fibonacci sequence for estimation allows for a gradual and realistic estimation of effort as it better reflects the uncertainty and complexity of tasks. By breaking tasks down into smaller units, known as user stories, planning becomes more agile and manageable. This decomposition allows teams to set clear priorities and focus on those tasks that offer the highest added value. Another important tool in agile estimation is planning poker, which is based on the consensus of team members. This encourages collaboration and allows for a broader perspective when estimating. In summary, Agile Estimation emphasizes the importance of flexibility, collaboration and continuous adaptation to successfully meet the challenges of dynamic projects.” (Created by an AI)
- The teaching approach is to be implemented according to Merrill’s first principle of learning.
- Course level: Beginners
- Course participants: People with initial contact with agile frameworks. Here especially in the implementation of first methods.
- Tonality: Friendly
The AI has used these parameters to create a course structure. You can find it on the course page and in the course. At this point, we recommend that you get the course… then it’s even more fun to follow the whole thing.
October 26th, 2023 – Create a book list for recommendation
In almost all our courses, we recommend books from the relevant subject area at the end of the course. These are given a short summary to make them easier to categorize and are linked to the Buch7 page (you should definitely take a look – the sustainable book trade).
This task smells very much like a task for the AI. We choose the now well-known chat GTP for this. We also use the KISS approach (Keep it Simple, Stupid) and don’t mess around with the prompt too much:
“You are to create a book recommendation list for a group of agile coaches. Create a list with 10 books related to the topic of agile estimation and prioritization.”
We receive a list of works we know. A few seem new after all. But that’s great, you can’t know everything. We also get a short one-liner for each of the works. That’s not enough for our book description. So we ask the AI to add more here. The second round is also a bit short. When we specify that we need one paragraph per work, it goes like butter.
We are basically happy with the result and start transferring it to the course, picking out cover images and linking them to Buch7. As we transfer the texts, we realize the disadvantages of our prompt. Basically, every book explanation has a subordinate clause along the lines of: “Agile coaches will like this book because…”. This is a bit monotonous and also has an exclusionary effect. The AI has simply stuck to the single person in the prompt and thus not only refined the content selection, but also targeted the person linguistically. This is not tragic, but it could have a bit more pizzazz. In line with our rule of deferring changes to the next iteration, we put this in the course backlog. Certainly, another text should be generated for each work.
But we also noticed: three of the ten works could not be found at all.
We couldn’t find the works on Buch7. A cross-check on Amazon did not return any results either. The title and author on Google didn’t give us anything either. But no problem. After all, we have an AI at the start.
We take a sample with: “Agile Estimating and Planning Your Project” by Mike McLaughlin.
This was advertised to us as:
“In ‘Agile Estimating and Planning Your Project,’ Mike McLaughlin provides Agile coaches with a practical guide full of actionable tips. The book goes beyond theoretical discussions and provides coaches with tangible strategies for effective agile estimating and planning. Using real-world examples, McLaughlin ensures that coaches not only understand the concepts, but can apply them in the dynamic environment of agile projects.”
Sounds good. But can’t be found. So we ask:
“The work ‘Agile Estimating and Planning Your Project’ by Mike McLaughlin cannot be found. Can you provide further information that would allow us to find the work.”
Here comes the sobering answer:
“It appears that there is a mix-up or error regarding the book title “Agile Estimating and Planning Your Project” by Mike McLaughlin. To the best of my knowledge, which extends to January 2022, there is no book with this exact title by Mike McLaughlin.
It could be that it is a newer publication, published after my last knowledge update in January 2022. To find the book, I recommend searching the latest book directories, online bookstores or libraries. Alternatively, the exact title or author may be slightly different.
Please check the exact details or give me more information so that I can help you better.”
Small downer. We assume that the other two plants have the same problem. Check. AI task solved rather averagely. Once again, without curation, it’s not perfect.
November 15th, 2023 – We feed the AI for our articles
Once we have set up the structure, we start feeding the AI for the eBook articles. Here we rely in particular on Creator.ai. The tool has proven to be quite easy to use and creates good articles. We use the well-known Chat GTP (3.5) from time to time. Robinize also gives us a good idea of how other articles are structured. This helps with the structure, but is only ever a guide. We do not want to provide “trending” articles but rather informative added value.
This phase will now take a while. The progress can always be viewed within the course as well as in the front end of the course. There you can always see an insight into the course chapters and how much “reading time” the system is calculating.