Associate Professor and Villum Young Investigator at the University of Copenhagen
Website
Talk Title: Data Mining and Landscape Architecting: On Creating Multimodal Resources [slides]
Talk Recording: https://youtu.be/jcvTUrcjy4Q?t=607
VP of Research at Cohere and Lead of Cohere For AI
Website
Talk Title: Aya: Optimisation under Severe Constraints
Talk Recording: https://youtu.be/jcvTUrcjy4Q?t=6929
Associate Professor at Princeton University
Website
Talk Title: Building Geo-diverse Models [slides]
Talk Recording: https://youtu.be/jcvTUrcjy4Q?t=27889
Staff Software Engineer, Google DeepMind
Website
Talk Title: Challenges with Geo-Cultural Understanding and Generation
Talk Recording: https://youtu.be/jcvTUrcjy4Q?t=25838
Full Professor and Deputy Department Chair of Computer Vision at Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) and Faculty at Linköping University, Sweden.
Website
Talk Title: Building Culturally Aware Multilingual LLM Benchmarks
Talk Recording: https://youtu.be/jcvTUrcjy4Q?t=20277
Assistant Professor at Stanford University
Talk Title: Culture Dialogue through Human-AI Collaboration
Staff Research Scientist, Google Responsible AI
Talk Title: Towards Meaningful Cultural Inclusion in VLMs
Talk Recording: https://youtu.be/jcvTUrcjy4Q?t=8444
Assistant Professor at the Indian Institute of Science (IISc), Bangalore
Talk Title: Richer Output for Richer Countries: Geographical Disparities in Language and Image Generation [slides]
Abstract: At the height of excitement surrounding “Generative AI,” and assurances of intelligence that benefits “all of humanity”, it is critical to ask whether AI models adequately serve the needs of a geographically-diverse population. Through a series of studies on text-to-image and language models, I will present how current AI models (i) fail to produce geographically-representative images of common entities; (ii) systematically erase certain countries from their outputs; and (iii) generate stories and travel recommendations that are less unique and informative for poorer countries. Lastly, I will highlight the need for better understanding datasets used to train models, and share preliminary work to geographically profile datasets.
Talk Recording: https://youtu.be/jcvTUrcjy4Q?t=18823
Associate Director at the Leverhulme Centre for the Future of Intelligence at the University of Cambridge
Website
Talk Title: ‘Re-scaling’ cultural knowledge in a UK heritage and museum context [slides]
Abstract: This talk draws on ongoing work that challenges conventional thinking about AI models which are typically made for large scale commercial or scientific application. By contrast, how does AI ‘know’ and work at a small, idiosyncratic, or local scale? With reference to the broader context of the UK GLAM sector (galleries, libraries, archives, and museums) adopting AI for internal database management, research, and public engagement, we discuss the specific case of a new collaboration with the Scott Polar Research Institute at the University of Cambridge.