Host Brenna Clarke Gray (Thompson Rivers University) and guest Autumm Caines (University of Michigan - Dearborn) explore the pedagogical implications of generative AI in this conversation in honour of Open Education Week. They ask such questions as:
- What happens when we leap into new technologies without first pausing to imagine harms, such as surveillance, bias, and discrimination?
- Can recentering the core values of the open education movement—equity, inclusion, transparency, and social justice—in our pedagogy help us move forward in a good way?
- How do we introduce these considerations to our students and empower them to make informed decisions with new technologies?

With over 305,000 open educational resources cataloged on OER Commons since 2007, ISKME works to make learning and knowledge sharing more participatory, equitable, and open, in pursuit of a more just society.

Those resources don’t describe themselves, though. The metadata of every resource in OER Commons was put together by someone before it got added to our collection, and then a librarian at ISKME reviewed it for quality – and that’s a lot of work, both in and out of house!

How much work? Well, if a librarian were to spend just five minutes on each record that ever found its way into our collection, that would take 25,433 hours. That’s enough time to…

- do 123 round trips to the moon (time to finally take that leave you’ve been saving)

- get 3,178 full nights of sleep (unless you’re a cat, then it’s only 1,413)

- walk 8 times from Cape Town to Copenhagen (we’re gonna need a bigger passport)

- work full-time for over 13 years (don’t worry, that excludes 4 weeks vacation)

All of that to say, metadata takes time.

It can be a challenge to balance metadata creation with other tasks like maintaining existing records, curation work, and supporting educational partners with curation. As such, we’re always on the lookout for tools and techniques that boost our capacity without compromising quality.

In 2023 and 2024, we’re testing out how generative AI tools like large language models can support our work in the OER landscape. This poster highlights some of the places where we’ve had successes, along with possible future applications that we think are both useful and doable.

As emerging technologies in artificial intelligence continue to evolve, their influence in educational settings is becoming increasingly significant. “The presence of AI systems and chatbots in education needs to be considered as an opportunity of development rather than a threat.” (Kooli, 2023). The primary objective of this guide is to assist you in navigating this new landscape. This guide will equip you to make informed decisions on how to proactively design and adapt your college courses for the age of AI.

The lecture provides an overview of autoregressive language models and how they have evolved from basic count-based approaches to more advanced neural network models. It explains the core concept of predicting the next word in a sequence using previous words and how this has been improved through neural networks that can generalize to unseen examples. The speaker highlights how modern models "compress" and "memorize" patterns from training data, allowing them to make better predictions. Techniques like beam search are also discussed as methods for generating text. While the results of these models can seem magical, the lecture emphasizes that they are built on simple but powerful principles of counting, probability, and compression, drawing on early work in information theory.

The slides can be found here: https://drive.google.com/file/d/1dk3o-fcdH1Y7-rGGqlVR35AZ1CVwz0qi/view

(This summary was generated by asking ChatGPT 4o for a summary of the partial transcript.)