Thought Leadership
Chat GPT & Large Language Models in Education
How customized GPT agents are redefining modern curricula, student tutoring, and STEM lab education frameworks.
Language models are genuinely useful in a classroom setting, and genuinely risky in ways that a procurement checklist rarely captures. Drawn from our work building STEM and tinkering lab programmes, here is where the line sits.
Where the technology helps
The strongest use is patient, unlimited explanation. A student working through a robotics problem at ten at night can ask the same question five different ways without embarrassment, which is something no timetable can provide.
It is also effective at the teacher's side of the work: generating practice variations, drafting differentiated versions of a worksheet, and producing first-pass feedback that a teacher then edits. That reclaims preparation hours without putting the model between the teacher and the student.
Where it does not
Assessment is the clearest boundary. A model's confident tone does not correlate with correctness, and a student cannot reliably distinguish the two. Anything that determines a grade needs a human decision.
It is also weak at knowing what a specific student already understands. Without that context it will explain at the wrong level — too advanced for the struggling student, too slow for the one who is ahead. The teacher supplies what the model cannot.
Designing for a school environment
School deployments have constraints that consumer products do not. Student data protection is a legal obligation, not a preference, and the network in a lab is frequently slower than anything the vendor tested against.
- Restrict the assistant to the curriculum scope rather than exposing a general-purpose chat interface
- Keep student identifiers out of prompts entirely
- Give teachers visibility of how the tool is being used in their class
- Ensure the lesson still works when connectivity fails, because it will
Teach the tool, not around it
Students will use these systems regardless of policy. The more valuable curriculum response is to teach verification as a skill: how to check a claim, how to spot a confident error, and when the answer needs a source. That is a durable competency, and it happens to be the exact skill that makes the technology safe to use.
Working through this in your own organisation?
Tell us where you are stuck and we'll respond with a practical next step — no pitch decks.
Start a conversation