Tussen wiskunde en woorden: taalmodellen verklaard

How do language models work behind ChatGPT, Co-pilot, Perplexity and similar tools? This lecture gives an overview of how language models work: What's their architecture like, how are these models trained? What happens when they're used? How can we further analyse how they work? Besides mathematical aspects, the conceptual ideas are discussed. Because why can these models deliver such impressive results? And why still fail so spectacularly?

Antske Breedings is professor of Computational Linguist Methods. The research in her chair focuses on the question of how language works and how to model this computationally. This question is often answered in interdisciplinary context, by looking at how language technology can be used in different research fields and applications. This context fits perfectly with the main subject of the chair: given how language and our models work, what can we expect if we use them in practice? What are options and risk derivatives? But also: what do we learn about the boundaries of our models?