What does GPT-4 know about people like me? This interactive workshop peeks under the hood of large language models to understand what goes into the datasets that shape them. We will discuss how these giant text corpora represent us in ways that feel hollow, harmful, and incomplete. I will share my analysis of existing datasets in an artistic research context, plus collaborative research that proposes building alternative models, methods, and datasets featuring texts that orient toward life and liberation. These texts will include, but not be limited to, works that center decolonial perspectives, queer love, ethics of care, and practices of commoning and mutual aid--created with transparency, community input, and contributor consent. Our collective Unsupervised Pleasures is developing more intersectional machine learning methodologies, and in our discussion we will unpack the issues with existing tools and imagine different tools that can support the communities they represent instead. Are these the automated systems we want? Whose voices, visions, and stories are captured in them and whose are excluded, harmed, or undermined? Let’s start building the language tools we want to see in the world.
Open to any level of technical knowledge, from zero to expert. All backgrounds welcome!
This Talk was translated into multiple languages. The files available for download contain all languages as separate audio-tracks. Most desktop video players allow you to choose between them.
Please look for "audio tracks" in your desktop video player.