New Learning’s Updates

Generative AI Breakthrough in the College of Education at Illinois

The CGMap application

February 7, 2024

We are pleased to announce a breakthrough in the use of Generative AI, calibrated to align closely with the disciplinary context of students and able to offer rich, immediately accessible formative feedback.

This has been achieved in a new version of our Generative AI application, CGMap, launched in our classes today. CGMap offers detailed, closely targeted feedback to learners in ways that are more relevant and specific than the unstructured and unfiltered use of AI in publicly accessible Generative AI interfaces such as OpenAI or Gemini.

One year since our first launch of CGMap, 351 graduate students in 15 masters and doctoral courses at the University of Illinois College of Education have demonstrably benefited from engaging with AI reviews in their required tasks. This has been achieved by combining AI and peer review. All AI review feedback is critically moderated with human review. This is a technology innovation grounded in sound educational values and cyber-social collaboration.

Here are the main features of the new release (note: tech trigger alert!):

  • This release builds on OpenAI’s latest GPT4 model, but in a stand-alone app that will be able to connect to any LLM (Large Language Model), including open source and locally hosted LLMs as they reach comparable richness and depth.
  • For AI reviews, CGMap uses Retrieval-Augmented Generation (RAG) with a vector store containing over 35 million tokens. Translated into non-tech talk, every student project under review is analyzed from a knowledge store of approximately 35 million words, consisting of Learning Design and Leadership (LDL) faculty research publications and our LDL masters and doctoral student projects since 2019. We're putting collaborative intelligence to work!
  • Unlike earlier versions of CGMap application, each student’s full submission (approximately 3,000 - 5,000 words) is included in the prompt producing richer and more relevant feedback.
  • The software runs student work through the GPT in a separate pass (prompt) for each of ten rubric criteria, for instance: theoretical framing, empirical evidence, critical analysis, academic language—the meaning of which is spelt out in explicit detail for both the students for peer and self-review, and the GPT.
  • GPT prompts are fully transparent and always moderated with human (peer and instructor) review against the same review criteria. This makes formative assessment an integral part of learning, and eases the burden of summative assessment through granular, big data progress visualizations.

We're watching Generative AI developments closely by the day, but we believe these are game-changing firsts: this use of a domain-specific knowledge bank customized for a particular curriculum, and multi-pass, rubric-based formative assessment.

CGMap is the latest app to be added into the experimental CGScholar e-learning platform. CGScholar can be embedded into all major Learning Management Systems via the LTI (Learning Tools Interoperability) API (Application Programming Interface).

Read More:

The Learning Design and Leadership (LDL) Team:

  • Bill Cope, Professor, Program Leader <>
  • Mary Kalantzis, Professor
  • Vania Castro, Teaching Assistant Professor
  • Duane Searsmith, Software Developer, AI Specialist
  • Olnancy Tzirides, Research Co-ordinator
  • Kara Francis, Academic and Program Advisor
  • Akash Saini, Teaching Assistant
  • Stuart Mattingly, Teaching Assistant
  • And the other members of the LDL Cyber-social Learning Lab