Language model performance and development works on comparing the performance of multiple leading language models, including OpenAI’s GPT, Big Science’s BLOOM and models developed by the Center for Humanities Computing at Aarhus University (CHC). We investigate and document methods for understanding performance in both English and Danish, and also explore implications for traditional Natural Language Processing tasks and pipelines.
Human creativity and thinking explores the potential for language models to produce creative work as well as engaging in philosophical dialogues. We also collaborate with the project Fabula-NET, a project focused on the understanding of literary quality and preferences.
Cognitive Science and linguistic research works towards a general description of how large language models are challenging existing understandings of language and language acquisition. The purpose is to consider firstly what problems are posed to traditional linguistic research by state-of-the-art large language models; and secondly, to consider what challenges traditional linguistic research can pose to contemporary work in natural language processing.
Pedagogical application investigates how text generation can be used in pedagogical settings at primary, secondary and tertiary levels of education. The project isd involved with pilot projects on grammar assistants, automated translation and creative exploration of large language models.
Trust and authority concerns the popular interfaces between humans and large language models: conversational agents like ChatGPT. This topic considers users’ perceptions of a chatbot’s competence, trustworthiness, warmth, and authority are conditioned, and how these factors impact how much users trust and defer to information produced by chatbots – even when this information is unverified, false, or unethical.
