This article has been machine-translated into English. Therefore, there may be nuances or errors in the content. The Danish version is always up-to-date and accurate. Read the original Danish article here.

Another week has passed, and here is a round-up of the news relating to artificial intelligence and education. The biggest story has been the launch of Google’s new language model, Gemma, and we take a closer look at how the model can be used in teaching. It has also emerged that there are challenges with Google’s Gemini, which, among other things, has problems generating historically accurate images and has been accused of being "woke". Meanwhile, Marie Bjerre (V) has rejected the idea of developing a Danish language model trained on Danish data. She believes the private sector can do this better than yet another large public IT project. At the same time, Læremidler.dk has published Learning Tech 14, focusing on AI and didactics in education.

Happy reading

Google Gemma

This week, Google launched a new language model called Gemma, which can run locally on your own computer without using the internet. This makes the model interesting for subjects such as digital technology understanding, communication and IT, informatics and programming, as students can run and test the model on their own computers. It also means that Google Gemma can be embedded in students’ own applications or used to test language models in teaching. That said, these offline language models are not yet on a par with what we see in ChatGPT or Google Gemini, and they struggle considerably with the Danish language.

Google Gemma comes in two different sizes, 2B and 7B. Gemma 7B is the largest model, capable of solving complex tasks such as mathematics and coding, while the smaller model, 2B, focuses more on accessibility on devices with limited computing power - e.g. laptops without a powerful graphics card.

Google Gemma is technically not an open-source language model, and it is not possible to further train the models. If you are up for it, you can test Google Gemma yourself via LM-Studio.

Read more about Gemma below:

Google’s New Gemma 2B and 7B Open-Source AI Models, But Do They Beat Meta Llama 2 7B and Mistral 7B?
Explore the new Gemma 2B and 7B open-source models released by Google and discover how they stack up against Meta Llama 2 7B and Mistral 7B.
DigiAlps LTDAleha Noor
Google Gemma LLMs small enough to run on your computer
Your own personal chatbot awaits
The RegisterKatyanna Quach

Development of a Danish language model ruled out by the minister

Minister for Digitalisation Marie Bjerre (V) will not support the development of a Danish language model, as there is a risk the model will not be "good enough". Although both Norway and Sweden are working on national language models, the minister believes the work is better placed with private companies.

In its digitalisation strategy, the government has allocated DKK 61 million to a "strategic initiative for artificial intelligence", and it is unclear which initiatives these funds will be used for. Marie Bjerre told Børsen that she cannot say precisely what the strategic funds will go towards, as they are to be determined together with the parties to the agreement during 2024.

Several have criticised the government’s new digitalisation strategy for lacking ambition in the area of artificial intelligence. The criticism is that the funds allocated are insufficient to create significant value, and they urge the government to make greater investments in concrete AI projects. By contrast, Minister for Digitalisation Marie Bjerre argues that the strategy establishes a foundation for future AI efforts and rejects the claim that the strategy is unambitious.

Critics line up: AI ambitions are missing in the new strategy
Industry organisations, a CBS professor and the IT spokesperson are calling for greater AI ambitions in the new digitalisation strategy, which Minister for Digitalisation Marie Bjerre…
Dagbladet BørsenMikkel Bødker Olesen
Behind a paywall
Minister lays Danish Chat GPT to rest
Minister for Digitalisation Marie Bjerre (V) was unsure, but nonetheless she would not rule out that the government…
Dagbladet BørsenAstrid Sturlason, Mikkel Bødker Olesen
Behind a paywall

Google Gemini’s image generator has been paused.

Google Gemini, Google’s new language model, designed to compete with ChatGPT4, includes, like ChatGPT, an integrated image generator comparable to ChatGPT's Dall-E 3. This image generator has now been paused after accusations of historically inaccurate images - some have called it extremely "woke".

@BasedTorba on X

For example, Gemini generated images of a pope depicted as a dark-skinned man or woman, and the Apollo 11 astronauts were portrayed as a Black man and a woman. Google has stated that they carefully adjusted the model during development to prevent the problems that have previously arisen with image generation technology, such as bias and inappropriate images, and to achieve more inclusive and diverse images. During this fine-tuning there has likely been an overcompensation, which has triggered these new challenges.

We have ourselves experienced Google Gemini’s attempts to be inclusive, in text as well.

Google’s image generator has been accused of being absurdly "woke", and it has now been paused
Gemini particularly struggled to create historically accurate images.
DRTobias Leth Klinge
Google explains Gemini’s “embarrassing” AI pictures of diverse Nazis
Gemini will undergo testing before it can start generating people again.
The Verge
Gemini image generation got it wrong. We’ll do better.
An explanation of how the issues with Gemini’s image generation of people happened, and what we’re doing to fix it.
GooglePrabhakar Raghavan

Learning Tech 14 published: focus on AI and didactics in education

Læremiddel.dk's latest special issue explores AI’s role in didactics and education. The articles range from the use of AI in language teaching to adaptive learning resources in mathematics, ethical considerations around the use of AI chatbots, and measuring reading skills using eye-tracking technology.

Read the full issue here:

Learning Tech 14: AI and didactics in education
The theme and title of Learning Tech 14 is AI and didactics in education. The issue was published in December 2023. Read it for free here.
Learning TechAI and didactics in education

New articles on Viden.AI

When we write articles on the blog, we mostly do so based on our own interests and our individual perspectives within the education sector. We write for the pleasure of it and because we want to inform and share knowledge about artificial intelligence across the entire education sector. If anyone would like to write for the blog, you are always welcome to email [email protected].

This week at Viden.AI, we have published two new articles:

Who should take responsibility?

Malte has published an article that poses the big question of who should take responsibility for the implementation and governance of AI in schools. He points out that there needs to be clarity about the division of tasks and responsibilities throughout the chain of actors. At the same time, it is important to focus on the values we want in our education system in light of the new technological possibilities.

Who should take responsibility?
With the explosive development of systems that use generative artificial intelligence, a host of educational questions are arising. The question of who should take responsibility for the implementation and governance of AI in schools is becoming pressing. But do we need to act? Should we r…
Viden.AIMalte von Sehested

Analysis: How does AI affect teachers’ work?

To what extent will language models affect the education sector? Danmarks Statistik (Statistics Denmark) has published an analysis looking at how the Danish labour market is affected by large language models such as ChatGPT. Per has written a short piece about it with a focus on the education sector, but the full analysis from Danmarks Statistik is well worth exploring.

If you want to avoid artificial intelligence in your workplace, you’d better become a painter 🤔

Analysis: How does AI affect teachers’ work?
Danmarks Statistik, together with researchers from the Copenhagen Center for Social Data Science (SODAS) at Københavns Universitet (University of Copenhagen), has published an analysis examining how large language models (LLMs), such as ChatGPT, affect the Danish labour market. In this article, I will very briefly introduce…
Viden.AIPer Størup Lauridsen

News

Minister lays the Danish Chat GPT to rest
Minister for Digitalisation Marie Bjerre (V) was in doubt, but nevertheless she did not rule out that the government…
Dagbladet BørsenAstrid Sturlason, Mikkel Bødker Olesen
We need a Germanic language model | DataTech (PRO)
We are currently seeing a development where American language models dominate technological development. Most people know ChatGPT and Bard, which are language models in closed systems, primarily trained on major languages such as English and driven by commercial interests in the USA. But the large systems also have a w…
DataTechTorben Blach
China releases list of 184 AI education bases in schools
China’s Ministry of Education (MOE) recently announced a list of 184 primary and secondary schools selected as artificial intelligence (AI) education bases, with the aim of better promoting the development of AI education.
CGTNCGTN
It’s a problem that teachers take such different approaches to artificial intelligence
Students are concerned that access to artificial intelligence and the rules for its use should be the same at all schools.
Utdanningsnytt.noKari Oliv Vedvik
Do AI and the working environment have anything to do with each other?
3-8-Stine-Wolff-billede-scaled-aspect-ratio-348-234Stine Wolff
How should we use chatbots in the future?
Brian-Holm-Soerensen-Mariagerfjord-aspect-ratio-348-234Brian Holm Sørensen
Vice-dean: “We are integrating artificial intelligence into the education of the future”
AU (Aarhus University) now gives the green light for students to use generative artificial intelligence (GAI). It is a natural step if our programmes are to keep up with the times, says Lise Wogensen Bach, vice-dean for education at Health (the Faculty of Health at AU).
Logo of Aarhus Universitet Find English // Default alphabox with staff service to the utility box AU.alphabox.boxes.push(DEFAULT_util_staffservice_da); For students For PhD students For staff Local study portal Public Health Science Sports Science Clinical Dental Technician Master’s in Clinical Nursing Medicine Dentistry Optometry and Vision Science Health sciences Master’s degree programme Nursing Dental Hygienist Brightspace Contact Study Service Local PhD portal Health Shared pages for PhD students phd.au.dk Local staff portal Staff portal for Health Shared pages for staff at AU medarbejdere.au.dk150688au.dk

Brief news overview

In this section, we have selected AI news in bullet points:

Scientific articles

We have previously written about misinformation and disinformation, and it is something we will certainly continue to focus on for a long time.

In the article: "Comparing the willingness to share for human-generated vs. AI-generated fake news", researchers from the University of Lausanne and the Munich Center for Machine Learning investigate users’ reactions to fake news created by humans versus artificial intelligence. The study shows that people generally find fake news written by artificial intelligence slightly less credible than that produced by humans. However, this does not affect whether people will share the news; according to the study, it makes no difference who authored the fake news.

The study also found that conservative participants were more likely to accept fake news as true.

The Decoder has produced a rather good overview of the article, which can be read here:

Users can’t tell the difference between AI-generated and human fake news, study finds
A study has investigated how users react to fake news generated by humans and AI, and identified socio-economic factors that influence who falls for AI-generated fake news.
THE DECODERMatthias Bastian
Comparing the willingness to share for human-generated vs. AI-generated fake news
Generative artificial intelligence (AI) presents large risks for society when it is used to create fake news. A crucial factor for fake news to go viral on social media is that users share such content. Here, we aim to shed light on the sharing behavior of users across human-generated vs. AI-generat…
arXiv.orgAmirsiavosh Bashardoust

How generative artificial intelligence is transforming business and society

Oliver Wyman Forums has just published a report that provides an overview of how generative artificial intelligence will affect businesses, employees and consumers – including a bit about education.

It is a comprehensive report, and it can be downloaded here.

Report: How Generative AI Is Transforming Business And Society

Share this post

Written by

Per Størup Lauridsen
Per Størup Lauridsen
MSc in IT | Co-founder of Viden.AI | Head of IT and Communications at TietgenSkolen