Generative Textbooks | Democratizing Participation in AI for Education quick take and next steps

nrzmalik · 2025-08-20T05:41:50+00:00

I read David Wiley’s post about the *Generative Textbooks* prototype and wanted to share a short summary and reaction for the community. **What the project does** David and collaborators built a no-code system for authoring *generative textbooks*: authors create structured inputs (learning objectives, topic summaries, activities, and a book-level prompt) and the system assembles those pieces into prompts that drive an LLM to generate learning activities and study experiences. The prototype is open source and available to try. **Generative Textbooks** **Link:** **Blog:** **Key design decisions worth noting** * The tool currently favors openness in authoring while letting learners run the final prompt on the LLM of their choice. To avoid per-token hosting costs, the prototype copies the assembled prompt to the user’s clipboard and directs them to paste it into whatever model interface they prefer. This preserves free access for learners while enabling high-quality model use. * The prototype used open-weights models hosted via Groq, but the architecture is modular so different models (including locally hosted ones) can be swapped in. * Early conversations with instructors revealed a strong preference: educators care more that tools are free for students than that they are strictly open. That insight shaped the design tradeoffs. **Why this matters for instructional designers and eLearning developers** * It’s a practical example of combining OER-style authoring with generative AI while keeping authorship and learning-object alignment front and center. Good templates + good source content = better, safer AI outputs. * The clipboard workaround is an interesting pattern when budgets or API costs are a blocker; it’s a pragmatic compromise that still lets learners access top-tier models. **Open questions & opportunities for AI4ID.community** 1. How would you structure learning objectives and activity templates so an LLM reliably produces pedagogically sound outputs? 2. What analytics or privacy-safe telemetry would you accept if we wanted to measure efficacy without exposing student data? 3. Could we adapt the clipboard pattern into a smoother, institution-friendly workflow (SSO, institutional LLM access, or a secure proxy)? **Try it & share feedback** Play with the prototype at the link in David’s post and, if you try it, share one example prompt or learning activity you generated. The project repo and demo are linked from the post if you want to inspect the code or authoring flow. *** Would anyone like to run a small experiment together: pick one short chapter (one topic), author the objectives + summary + two activities, then test outputs across 2–3 LLMs and report differences? If yes, I’ll set up a quick template and a shared thread.

✨ AI in Instructional Design

💡Vibe Coding

Blog

Apps