Location: Melati Ballroom (Level 4)
Abstract: Achieving the promise of AI “foundation” models for scientific discovery requires immense computational power and multidisciplinary collaboration. Although only a handful of organizations have the resources necessary to train these models at scale, new strategies have emerged using high-performance pretrained models, such as domain-specific fine-tuning, and agentic constructs. International collaboration is accelerating progress in key areas such as data preparation; evaluation for scientific reasoning, trustworthiness, and safety; and application-focused fine-tuning. This presentation highlights the progress of working groups convened by the international Trillion Parameter Consortium across these and other areas. Together, these efforts enable the scientific community to navigate a rapidly evolving AI landscape.
