News of the week: Ezri - a new Danish AI chatbot

The past week has been an autumn break, so not much has been written about artificial intelligence in teaching. However, there has been an interesting article in Version 2 about a new language model named Ezri. On Viden.AI, we regularly talk to Gustav Grimberg about their project, and we find it very interesting how they have solved the GDPR problem. However, we still hope it will be possible to develop a Danish language model trained based on Danish or Nordic values.

The new chatbot, Ezri, focuses on data security.

Gustav Grimberg and Leon Mächler, in collaboration with the Center for ICT in Education (CIU), have worked to train a language model that can filter content before it is sent to OpenAI's European version of ChatGPT. Ezri distinguishes itself by being an open-source, privacy-first model that cleans prompts of sensitive personal data before sending it to the language model.

Gustav Grimberg explains that while other language models may have challenges filtering sensitive data in Danish, Ezri is specially trained in this. In addition to data protection, Ezri offers a "prompt library" of predefined questions for training.

Version2 has interviewed Gustav about the project, and you can read about it below (unfortunately, it is behind a login):

You can create an account on Ezri yourself and test the model here:

Home

Welcome to Ezri

New book release: AI in high school

Praxis has just published a webbook focusing on artificial intelligence in high school.

The author, Sine Zambach, writes in the introduction to the book: The book is intended as a platform where, in addition to delving into the history and technical details of artificial intelligence, I also discuss philosophical and societal impacts to give the" most interesting ones a holistic picture of what artificial intelligence can and cannot do.

AI in high school includes chapters:

History of AI
Introduction to AI
AI in Danish (by Mischa Sloth Carlsen), English (by Bodil Aase Frandsen Schmidt), Mathematics (by Carsten Vium), Philosophy (by Mads Rangvid) etc.
What about the deep learning?
With other glasses
Didactic toolbox

The WebBook can be read here:

News of the week

Scientific articles

In our studies, we often come across new scientific articles, and we would like to highlight the most interesting ones.

Beyond Memorization: Violating Privacy Via Inference with Large Language Models

Current privacy research on large language models (LLMs) primarily focuses on the issue of extracting memorized training data. At the same time, models’ inference capabilities have increased drastically. This raises the key question of whether current LLMs could violate individuals’ privacy by infer…

arXiv.orgRobin Staab

Recommendations of the week

IT-Vest will publish a series of webinars on various topics within generative AI, and on Thursday, 26 October 2023, from 16-17, Per and Claus will talk about the upcoming AI regulation.