Hey {{ first_name | human }},

That Easter break is quickly coming up and so are my Master’s deadlines. Let’s dive right in.

🚨AI Oracy Assessment: The CEO of the Chartered College of Teaching has called on oracy to be assessed by AI to reduce the burden on teachers. But should we even consider this?

🧪AI Detectors Do Lie: In what should not be a surprise at all to anyone following this space, a new systematic evaluation of 13 AI detectors find that existing detection technologies are inadequate…

⚡️AI Tools update: Claude got an update to control your computer & ChatGPT interactive teaching widgets are live.

🚨 Call for AI Oracy Assessment > What it is: Dame Alison Peacock of the Chartered College of Teaching suggests that using AI to assess oracy skills could be beneficial for students without putting extra pressure on teachers. Speaking at The Speaking Summit in London, she highlighted the potential for AI in evaluating spoken language abilities.

Why it matters: The government has committed to publishing a new oracy framework for primary and secondary schools. Recent research has shown a decline in oracy opportunities in the early stages of secondary school, coinciding with a decrease in students' confidence in their speaking abilities. However, it is unclear to me at least, what an oracy assessment would entail, how it could be measured in a way that the results are both reliable and valid and whether AI is the right tool for this job. There is also the small matter of safeguards….

Safeguards: Voice recordings of children are biometric-adjacent data under UK GDPR. So schools could not upload student recordings to any third-party AI tool without explicit DPA coverage and parental consent. This problem would need to be solved on a national scale before it could be even implemented in schools.

🧪Systematic Evaluation of AI Detectors > What it is: A large-scale study evaluates 13 AI detection tools using over 280,000 authentic student artefacts, including coursework, theses, and engineering code, to test how well they can identify AI-generated content across real educational contexts.

Why it matters: The sector is increasingly relying on detection tools as a proxy for academic integrity, yet this study suggests that confidence may be misplaced. While detectors perform “reasonably” on long-form essays, they fail systematically on short-form tasks and coding—arguably the very formats most commonly used in assessment. More concerningly, STEM subjects appear disproportionately flagged due to their formulaic writing conventions, raising questions about bias and fairness. Perhaps most critically, simple editing strategies allow the majority, 88%, of AI-generated work to evade detection. This challenges the assumption that detection can underpin high-stakes decisions.

It should be pointed out that reasonable success on long-form essays is still a concern. If even one person can be accused of using AI when they have not purely based on these commercial detectors, then the systems are not fit for purpose. They could well provide a signal, but they should not be making high-stake judgements.

Why this DOES NOT work: GenAI tools are trained on human created text, which it then uses to create new text based on a user prompt. It is not copying and pasting from its vast data network, it is producing something novel (to the extent that any new written word is novel) based on human writing. The very notion that their is ‘AI writing’ is should be challenged.

⚡️ Claude Controls Your Computer > What it is: A new feature coming to certain paid tiers of Claude will allow it to access and control your computer and let it take actions based on instructions that you message it. The promise is this: Need to send your cover lesson to the department head because you are stuck in traffic? You could message Claude to find the resources needed in your files, open your email client to a new email and send them to your department lead just by asking it. It makes a great demo, but the proof will be in the pudding. Follow the link below to see it in action.

Last week I mentioned that ChatGPT was getting some interactive tools. Well it now seems that those are live. I asked it to explain the area of a Triangle and a very basic interactive widget appeared. You can control the base and the perpendicular height using the sliders. It could be useful for some live modelling where static images just do not quite do it. Unfortunately, I cannot add videos directly into these, but you can see an image of what this looks like below.

‘Till next week.

Mr A 🦾

Help a colleague save time by sharing this newsletter; distributing these ideas helps a friend get home on time and keeps our energy focused on what matters most: great teaching.

The tools and workflows mentioned are intended for professional productivity and educational enhancement. Users must ensure that any AI implementation remains compliant with their local data protection regulations and institutional safeguarding policies.