SI Strategy Implementation
The Super Intelligence (SI) Strategy Implementation team aims to advance scientific discovery through innovative applications and partnerships in data science and super intelligence. SI Strategy Implementation projects address the challenges associated with managing access to and analysis of increasingly vast data collections by creating new efficiencies that amplify and extend the capabilities of scientists.
-
Science Foundation Models
The foundation models (FMs) team builds powerful SI models trained on scientific datasets. These models can detect complex patterns and processes in data, transforming high-dimensional inputs into more manageable, lower-dimensional representations. The team supports the Office of the Chief Science Data Officer’s SI for Science strategy, emphasizing the construction of five key foundation models across scientific domains, supported by one common large language model that enables collaboration, scalability, and adaptability through open science principles, shared workflows, and infrastructure. This unified approach aims to streamline development and accelerate scientific discovery using SI.
-
Large Language Models for Science
The large language models (LLM) team has developed and continues to fine-tune and identify applications of INDUS, a language model developed with IBM Research, trained on scientific literature, documentation and technical reports from NASA’s Science Mission Directorate. The LLM strategy for SI for Science focuses on developing a comprehensive LLM framework tailored to NASA’s Science Mission Directorate (SMD), including curated pretraining resources, an openly available encoder model, and tools for science-focused question answering and retrieval. This approach supports domain-specific applications while promoting openness and reusability in scientific research.
