How an AI-integrated anonymous chatroom boosted AI literacy and inclusion

Two hot topics in higher education right now – embedding AI literacy in course design and designing inclusive learning environments – are often treated as separate problems with separate fixes. But they don’t have to be. The right AI tool, used the right way, can help with both issues at once – as a side effect of how students use it, rather than an extra unit bolted on to the syllabus.

For the past two years, we’ve been developing YoChatGPT!, a platform that lets a class share a single live chatroom integrated with large language models, instead of each student opening a private chat on their own device. That one design choice – putting AI to use in public, in front of peers – turns out to do useful work on both fronts.

Making AI literacy intrinsic, not supplementary

Most AI literacy teaching is delivered as content: a slide on bias, a session on hallucinations, a policy on data. Students absorb it for an hour, then go back to using AI exactly as before. Intrinsic AI literacy means students learn these concepts because the course is structured so they encounter them, not because someone explained them.

A shared chatroom with LLMs does this directly, on three fronts.

Bias becomes visible rather than theoretical. When the model answers a prompt with a built-in assumption (about gender, geography or a default example), the whole group sees it at the same moment. The discussion of why that happened is anchored to something that just occurred, not a hypothetical case study.

Hallucination becomes a shared, memorable correction rather than a warning. A model inventing a citation in front of the whole class teaches the lesson that AI can be confidently wrong far more effectively than a slide with that title. Students start fact-checking the AI’s outputs, and each other’s prompts, as routine practice. This is the critical evaluation habit AI literacy frameworks are trying to instil.

Data stops being a one-off consent form and becomes a live decision: what to share with the room, what to upload, what to keep out. Students make that call every session, rather than once at the start of term.

Uploading your own course materials adds a further layer. Put the outline, schedule, lecture notes and rubrics into the room, and students can ask the AI about the course itself: when something is due, what a rubric criterion means, how this week builds on last week. 

• ‘AI literacy is everyone’s responsibility’
• Transition from expert to guide by embracing generative AI
• Three ways to use ChatGPT to enhance students’ critical thinking in the classroom

This works through retrieval-augmented generation: rather than answering from its general training, the model draws on your specific documents first. The result reflects what you actually said and assigned, not what a generic model assumes a course “probably” covers. For students, it’s a low-effort, always-on first point of contact. For AI literacy, it’s a concrete demonstration that an AI system is only as good as what it’s grounded in: the data question, made tangible.

Designing inclusive learning environments, online and off

A seasoned teacher knows that students who ask questions out loud aren’t a representative sample of the room. Students working in a second language, students who feel behind, students acutely aware of status differences with more confident peers and introverts are all systematically less likely to raise a hand, regardless of how good their question is. Most inclusive-design checklists address accessibility – captions, alt text, flexible deadlines – without touching this dynamic at all.

A shared chatroom with an anonymous posting option removes the specific barrier doing the damage: being identified as the person who didn’t already know. A student can ask the AI to re-explain a concept, challenge an idea or admit confusion without that question being attributed to them in front of classmates or a tutor. The status-signalling that shapes who speaks up in person, who has the confidence, who already sounds like they belong, drops away and the question gets answered on its merits.

This travels with the device, not the room, so it works identically online and in person. It costs nothing in curriculum time, and it doesn’t single out any student as needing accommodation: everyone gets the same anonymous option, which is the point.

Putting it into a course

In practice: set up a shared room for the module, load the outline, schedule and rubrics, turn on anonymous participation and resist explaining the AI literacy angle upfront. Let students notice the bias, the wrong answer, the data question for themselves, then debrief briefly when it happens.

The evidence base is still growing but the early signals are encouraging. YoChatGPT! has featured in published and presented research on inquiry-based mathematics learning, personalised concept teaching and collaborative problem-solving, with outputs from teams in Hong Kong and the Philippines, and has picked up recognition including a Silver award at the QS-Wharton Reimagine Education Awards. 

The broader point doesn’t depend on this one platform: if AI literacy and inclusion are to be intrinsic to a course rather than supplementary to it, look for design choices that make the right behaviour the easy, anonymous, default one, not the one students need reminding to take.

Fridolin Sze Thou Ting is senior lecturer in the Department of Mathematics and Information Technology at The Education University of Hong Kong.

If you would like advice and insight from academics and university staff delivered direct to your inbox each week, sign up for the Campus newsletter.