A New and Safe GenAI Tool for the Erasmus University Rotterdam — GenAI Monthly Meetup #4
On October 17th we had the fourth GenAI Monthly Meetup. The aim of these meetups is to exchange ideas and experiment with developments in AI to find new ways it can be applied to teaching and learning. Each month we have a guest speaker and a diverse group of participants from the EUR community.
Here we will recap the discussion from the fourth monthly meetup. If you are interested, you can check out the recaps of our previous meetups here.
Fourth Monthly Meetup
Our guest for this meetup was Bas Smit, Strategic Information Manager on Education at the Erasmus Digitization & Information Services (EDIS). Bas talked about Erudite 2.0, a new Generative AI tool that offers the university community access to a range of LLM functionalities in a privacy-safe environment. He discussed the properties and strengths of Erudite and gave the participants a preview, after which we discussed the possibilities and limitations of the tool.
Erudite 2.0 and its Strengths
Erudite 2.0 is a powerful tool that has all the functionalities of others LLMs that students and teachers might be familiar with while removing the risks associated with sensitive data. Its development was financed by NPULS, a national initiative to enhance education through technology, currently working on developing a new EduGenAI platform.
While the current proof of concept of Erudite 2.0 is developed in collaboration with Microsoft, the goal is to make the final version supplier-independent. To guarantee the privacy of its users, the tool has its own front end where they can safely upload data and documents. These inputs then go through a translation model, which can “talk” to LLMs like ChatGPT and LLAMA to generate outputs, but information doesn’t go to any US servers. Instead, all information is stored securely in the Netherlands by SURF. Moreover, as part of its agreement with SURF, Microsoft cannot access or use the data that users put into Erudite or the answers it generates. In fact, all GenAI models offered within Erudite 2.0 run natively on these same servers in the Netherlands, which also removes the necessity of having to do API-calls to US servers.
As with other platforms, each user has access to their own chat history. The platform also has speech-to-text and text-to-speech functions, as well as image generating and interpreting. But what it also offers to users is the ability to choose between different GenAI models like ChatGPT, LLAMA, and Gemini, with more models expected to be offered in the future.
Another feature of Erudite 2.0 is the ability to create and customise personas for specific purposes and adding your own documents in a privacy-safe environment. These can be shared with colleagues and students, making them perfect for implementing GenAI in curricula. Erudite 2.0 also supports Retrieval Augmented Generation, or RAG, allowing users to upload their own documents from which the models can draw information to respond more informed (and focused) to certain prompts. Again, none of these documents will be uploaded to US servers but will instead be stored on Erudite’s servers.
Concerns
A number of participants brought up concerns regarding the use of AI in education, some of which go beyond Erudite 2.0:
- Teachers in the social sciences are especially concerned about grading papers and essays written with the help of AI. They are forced to reconsider how to evaluate student understanding.
- Some wonder whether the use of AI should be mandatory for certain assignments, and whether teachers can and should compel all students to use AI.
- A big concern is that Erudite 2.0 will encourage students to use AI more, which could increase its (negative) environmental impact. It is important that users are aware of how much energy their use of GenAI (and thus Erudite 2.0) requires in order to make better informed decisions.
Overall, these concerns point to a need for a cohesive policy on AI within the university, as well as support for the community in implementing it. Some participants note that introducing Erudite 2.0 could be a first step toward getting everyone on the same page.
Possibilities/Opportunities of Erudite 2.0
Despite the concerns, participants saw a lot of possibilities for using Erudite 2.0. Privacy protection allows all members of the university community to use LLM functionalities safely. Here are some of the possible use cases that the participants brought up:
1. Teaching and Learning Support
1.1 Curriculum Development: Saving teachers time in setting up the curricula for courses by, for example, allowing them to quickly research syllabi at other institutions.
1.2 Supplementary Learning Resources:
- Providing summaries of additional articles for students to enhance their understanding beyond mandatory readings.
- Encouraging students to draw on resources beyond what is included in the syllabus.
1.3 Pre-training and Onboarding: Facilitating pre-training or onboarding for complex topics such as scientific programming, neuroscience systems, or dynamical systems to ensure all students are prepared for courses.
1.4 Case Teaching and Experiential Learning: Setting up personas for case teaching and experiential learning; For example, creating personas for different characters in the case, allowing students to engage with them and apply skills they are learning in the case.
2. Student Engagement and Skill-Building
2.1 Experimentation with AI: Creating a safe environment for students to experiment with AI tools, guided by teachers, to build the necessary skills for effective use.
2.2 Democratizing GenAI: Providing all students with the most advanced GenAI models available, closing the current gap between students that pay for advanced GenAI models and those that don’t and/or can’t.
3. Administrative Efficiency
3.1 Support Staff Productivity: Quickly addressing FAQs from students, such as questions about applications or deadlines.
3.2 Internal Communication and Strategy:
- Assisting in drafting clear communication to students.
- Preparing presentation materials and internal reports.
4. Research Efficiency
4.1 Analyzing datasets: Offering researchers the ability to analyze complex datasets with GenAI in a privacy-safe environment (currently only possible for researchers that know how to run LLMs locally on their own computers, requiring advanced knowledge of how to set such systems up and how to train models).
Takeaways
The discussions highlighted a shared challenge: both students and teachers are still unprepared to fully integrate Generative AI into the classroom. While Erudite 2.0 offers a range of valuable features beyond text generation, many educators lack the training and confidence to effectively leverage these tools. Furthermore, students need to develop the skills necessary to critically assess the outputs they receive.
Erudite 2.0’s key strengths are its security features and its ability to provide access to multiple LLMs, offering a safe and more versatile alternative to publicly available models. However, there remains a significant need to raise awareness among students and staff about the advantages of using such a tool and to educate them on its responsible and effective use.
In short: The tools to incorporate Generative AI effectively and safely within education are finally around the corner. However, the educational landscape demands much more than just access to these tools. Universities must prioritize essential elements like support, guidance, upskilling, and beyond to truly embrace and implement Generative AI within their systems.
Join Us for the Next GenAI Meetup! 😁
To participate in the meetups and receive follow-ups, we invite you to become a member of our community! Every month we have an exciting new meetup to look forward to, and it’s all up to you whether you join or not (there are no obligations).
👉 You can join us here (EUR members only)!
