Many of today’s grand scientific challenges expect teams to reason about models, data, and background knowledge across multiple disciplines. This presents a high barrier to entry, requiring large multidisciplinary teams that must organize their work around well-defined questions. Scientists are also having trouble keeping up with the exploding volume of scientific information, particularly with the growth of data-intensive and compute-intensive science.

Foundation models, which are large AI models trained on Internet-scale amounts of data using high-performance computing, can be deployed within interactive dialogue systems commonly known as chatbots. We believe that advances in multimodality, reasoning, trustworthiness, and more for foundation models are required to develop chatbots for scientists to use as research assistants. These assistants will help scientists answer questions about scientific literature, automate workflows, and construct scientific models and reason about ambiguous scientific data.

Stay tuned for updates on our research...