Weekly Newsletter

Practical AI Strategies

Exams are not the answer (part two)

Hi everyone,

There's a narrative in the media and amongst a lot of people I speak with: AI makes coursework easy to fake, exams are hard to fake, so schools should do more exams. It sounds like common sense, and it's wrong on two counts. First, it's bad pedagogy, which teachers already know, and second, it directly contradicts the assessment authorities that certify every Year 12 student in this country, which fewer people realise.

I spent a while this fortnight going through the policy handbooks, assessment advice and syllabus documents of every senior secondary authority in Australia: QCAA, NESA, VCAA, SACE, SCSA, the ACT's board, and Tasmania's TASC. What I found is that no Australian certificate is built on exams, several systems cap how much of a school mark an exam can be, and at least one authority abolished its last closed-book coursework rule in the middle of the AI panic.

The exam-heavy response isn't the safe, conservative option people think it is: in most states, it would actually breach the rules.

Exams Are Not the Answer: Part Two

This is the follow-up to part one, where I looked at the media-led cheating panic driving the whole narrative. In part two, every quote is verbatim from an official document, and they consistently say the same thing: external assessment isn't privileged over school-based assessment, exams measure a narrower range of skills, and most of what we want to know about a student can't be examined.

The AI cheating crisis is genuine and is costing educators a lot of time and energy, but it's an authentication crisis, and the answer to it isn't to dismantle the system's design and put in more exams.

Building a Voice Assistant with Claude Code

And now for something completely different: over a few evenings I turned a forty dollar pocket computer into a hands-free voice assistant, directing the entire build from my phone via Claude Code. You hold a button, talk, and a Mac mini on a shelf does the transcription, thinking and speech, then answers back in a "warm British voice".

It was fun to build, but the reason I wrote it up is what I learned along the way: a two dollar chip can act as a voice interface for a capable AI, with all the hard work offloaded somewhere more powerful. That's the "AI everywhere" future I've written about before in miniature. Fair warning: this one's an advanced project, with hardware, networking, and an agent running with its guardrails deliberately loosened. It's kind of a security nightmare, a bit ridiculous, and not for the faint-hearted!

And a personal note...

And finally, some exciting personal news: I've officially graduated with a PhD in Generative Artificial Intelligence from Deakin University. I crossed the stage on Tuesday in my silly hat and cloak, and would like to thank everyone involved in the process for the past three years: my supervisors, my PhD research participants, and everyone who has read the blog and joined these conversations along the way.

Cheers,

Leon

PS: Term 3 of Practical AI for Curriculum Leaders is now full. If you'd still like a spot, reply to this email and I'll add you to the cancellation list in case something opens up.

You can see the cohort details here: https://leonfurze.com/curric-leaders/​

​​​