Despite ongoing debates about whether so called large language models /generative language (and other media) tools are ‘proper’ AI (I’m sticking with the shorthand), my own approach to trying to make sense of the ‘what’, ‘how’, ‘why’ and ‘to what end?’ is to use spare moments to read articles, listen to podcasts, watch videos, scroll through AI enthusiasts’ Twitter feeds and, above all, fiddle with various tools on my desktop or phone. When I find a tool or an approach that I think might be useful for colleagues with better things to do with their spare time I will jot notes in my sandpit, make a note like this blog post comparing different tools or record a video or podcast like those collected here or, if prodded hard enough, try to cohere my tumbling thoughts in writing. The two videos I recorded last week are an effort to help non-experts like me to think, with exemplification, about what different tools can and can’t do and how we might find benefit in amongst the uncertainty, ethical challenges, privacy questions and academic integrity anxieties.
The video summaries were generated using GPT4 based on the video transcripts:
Can I use generative AI tools to summarise web content?
In this video, Martin Compton explores the limitations and potential inaccuracies of ChatGPT, Google Bard, and Microsoft Bing chat, particularly when it comes to summarizing external texts or web content. By testing these AI tools on an article he co-authored with Dr Rebecca Lindner, the speaker demonstrates that while ChatGPT and Google Bard may produce seemingly authoritative but false summaries, Microsoft Bing chat, which integrates GPT-4 with search functionality, can provide a more accurate summary. The speaker emphasizes the importance of understanding the limitations of these tools and communicating these limitations to students. Experimentation and keeping up to date with the latest AI tools can help educators better integrate them into their teaching and assessment practices, while also supporting students in developing AI literacy. (Transcript available via Media Central)
Using a marking rubric and ChatGPT to generate extended boilerplate (and tailored) feedback
In this video, Martin Compton explores the potential of ChatGPT, a large language model, as a labour-saving tool in higher education, particularly for generating boilerplate feedback on student assessments. Using the paid GPT-4 Plus version, the speaker demonstrates how to use a marking rubric for take-home papers to create personalized feedback for students. By pasting the rubric into ChatGPT and providing specific instructions, the AI generates tailored feedback that educators can then refine and customize further. The speaker emphasizes the importance of using this technology with care and ensuring that feedback remains personalized and relevant to each student’s work. This approach is already being used by some educators and is expected to improve over time. (Transcript available via Media Central)