with apologies

Sabbatical Diary: Week 46

· 3 min read · August 21, 2026 · #academic #cambridge #sabbatical #teaching

Another one of those weeks that did not feel productive but actually several things did get done and I shouldn’t ignore them.

Made some progress on some external bits of work that were (and are) interesting, particular where it involved reading up a bit on attacks against LLMs.1

Several bits of what the academic career usually refers to as “service”. In this case, not particularly pleasant but at certainly important. And I am hopeful that there are signs that some good progress might be made.

Started picking up elements of work associated with upcoming role as Director of Undergraduate Teaching for CST. Following last week’s mock admissions interview for Women in Computer Science residential, I also gave two talks as part of the Department’s participation in the STEM SMART which were effectively my first run at giving the Open Day talk here. Was interesting to compare to the one I used to give at Nottingham!2 If you’re interested in participating in future, it looks like the relevant University page is here.

Continued catching up after the holiday: churned more email, so back to the inbox list fitting on a single screen which is nice; also finished catching up with Ph.D. students and colleagues about ongoing work, which was pleasant, particularly as it looks like a couple of papers are just about ready to ship :)

Finally, the most interesting—and to me surprisingly interesting—thing was the finalising of a business “Teaching Case Study”. This is not something I’d come across before but via colleagues and due to my having, prior to sabbatical, chaired the project board for the refurbishment of our West Cambridge Data Centre, it was quite interesting. It seems that a teaching case study poses a real dilemma to students where some action is required and a decision must be taken. Accompanied by teaching notes and any relevant data, this gives the students a chance to work through the dilemma with the instructor to understand the factors and consequences involved. It’s now submitted for publication so I can’t share it yet, but it was quite an enjoyable process, not least because I mostly just had to pontificate, leaving the main author to do the writing, at which they were excellent! Also nice that something I did purely as a piece of service to the university actually produced an interesting3 academic output.

  1. For those who’ve not been paying attention, or in case I’ve been too English about the whole thing: I’m an AI sceptic. And reading up a bit about attacks against them has done nothing to lessen that. Honestly: we spend decades designing systems that separate control and data when possible because it’s easier to secure and manage those systems, and then we go ga-ga about an entire approach which seems fundamentally to be based on mixing the two up. I dunno.

  2. Which, now I think on it, was cough-far-too-many-years-ago. We’re measuring in decades now. Good grief.

  3. In my opinion of course. But then, pretty much everything written here is.