News of the Week: Exploring Bias in Language Models

One of the significant challenges in language models is that they contain Bias, which can be difficult to do anything about. Therefore, it is pretty interesting that a language model has been developed that can be used to explore the Bias built into the training set.

Using Bias in Language Models to Explore Bias

We have written several articles on Bias in language models, which we see as a significant challenge in generative AI. A team of researchers from Humboldt University in Berlin has developed a language model with built-in Bias, which they have called OpinionGPT - A very biased GPT model.

The researchers have trained their language model on content from subgroups on the social media site Reddit. They selected content with an apparent demographic reference and examined specific social dimensions such as politics, geography, gender, and age.

It is possible to access and use the model in teaching to discuss Bias in language models. However, the researchers point out that even if we have succeeded in creating a language model that represents nuanced biases, it will still only be representative of a Reddit variant of demographic data, not necessarily of society as a whole.

You can test OpinionGPT on their website and read the scientific article OpinionGPT: Modelling Explicit Biases in Instruction-Tuned LLMs on arXiv.org.

New articles on Viden.AI

In the past week, we have been around the country to give presentations, which has meant that we have unfortunately not written new articles for the website. But on Tuesday, an article from two new authors on Viden.AI will be published, and Claus is working on an article that goes into more depth with Bias.

News

Interesting scientific articles

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

Exploring User Perspectives on ChatGPT: Applications, Perceptions, and Implications for AI-Integrated Education

Understanding user perspectives on Artificial Intelligence (AI) in education is essential for creating pedagogically effective and ethically responsible AI-integrated learning environments. In this paper, we conduct an extensive qualitative content analysis of four major social media platforms (Twit…

arXiv.orgReza Hadi Mogavi

Detecting ChatGPT: A Survey of the State of Detecting ChatGPT-Generated Text

While recent advancements in the capabilities and widespread accessibility of generative language models, such as ChatGPT (OpenAI, 2022), have brought about various benefits by generating fluent human-like text, the task of distinguishing between human- and large language model (LLM) generated text…

arXiv.orgMahdi Dhaini

This Week's Recommendation

We have long followed Inga Strümke, an associate professor and researcher in artificial intelligence at the Norwegian Open AI Lab at NTNU. Inga is very involved in the Norwegian debate on artificial intelligence, and her message is that it's essential to participate in AI development while setting high standards for understanding and responsible use - always with a human-centric approach.

Today, her book, Machines That Think, is finally published in Danish. We have received it but have not yet had the opportunity to finish reading it, so it would be unfair to review it. However, what we have read so far gives a good insight into artificial intelligence and the development that has taken place, as well as the movement toward machines that are smarter than humans. We jumped a bit in the content to take a closer look at the section on artificial intelligence in education, and it's not very comprehensive. Therefore, you should probably primarily read the book if you are curious about what lies behind artificial intelligence and where the development is headed.