Learning AI from a Children's Book

Dr Ben Swift

Yesterday I took part in Ben Swift's "LLMs Unplugged: A tech-free guide to understanding (and teaching) AI" at the Australian National University's College of Systems & Society. The workshop is aimed at school teachers and earns the attendee professional development points. But this would be useful for anyone who wants to understand how Large Learning Models (LLMs) work. I have a technical computing background, so much of this is familiar to me, but this clear fun overview was still very useful.

Ben took us through several exercises using children's story books. The advantage of this is that the source material has a limited vocabulary & repetitive patterns, making analysis easy, and fun. But the exercise could be carried out using government policy documents, or military reports.

The next workshops are 16 September & 25 November.

Continuous Delivery of Assessment

Dr Ben Swift gave a very interesting talk to tutors at the Australian National University recently about the work he has been doing using Gitlab and Continuous Integration (CI) for grading. In courses such as Art & Interaction in New Media, the tool Gitlab is used to provide an online repository of the work of each student. Students are given a template for submitting their work. As soon as uploaded, the CI feature of GitLab is  activated to carry out a series of checks specified by the lecturer of what the student submitted. These can be basic checks, such as verifying the student has included the statement it is their own work, comparison checks looking for plagiarism, and performance checks to see the code submitted does what it is supposed to. This has the considerable potential to be applied more widely to provide less stressful, more realistic assessment.

This approach might be applied beyond just computer code, to large bodies of written work with a complex structure. As Ben suggested, it would help get students out of the mindset of waiting until the last minute to submit. I suggest this could be taken further to apply the computer concept of "Continuous Delivery". With this approach the student would be expected to build a body of work in the repository, to the required standard, throughout their course. All of this work would be time and date stamped, and the student would be assessed on improvement over time, as well as quality of the finished product. A student who only deposited work in the repository shortly before a deadline would receive a grade of zero, regardless of the quality of their work, as they did not show improvement over time.

A byproduct of this approach would be to make cheating much harder. A student would have to steal, or pay for, a whole consistent body of work to be produced, not just individual assignments. They would have to deposit this work in the repository over time, in credibly sized installments, from Internet addresses consistent with their location. Multi-factor, or biometric challenges during submission could be used to make cheating even harder. Part of the student's assessment could be questions automatically generated from their own work.