In a groundbreaking initiative, Northeastern University has secured a $9 million grant from the National Science Foundation to investigate the inner workings of large language models, the powerful artificial intelligence (AI) systems behind tools like ChatGPT and Google’s Gemini. This project, dubbed the National Deep Inference Fabric (NDIF), aims to provide an unprecedented level of access and understanding into how these advanced AI models operate.
Despite the rapid rise and widespread adoption of large language models, their underlying mechanisms remain largely opaque, often referred to as “black boxes” within the scientific community. Even the brightest minds in AI are still grappling with how these neural network-based systems reason and make decisions, according to David Bau, a computer science professor at Northeastern and the lead principal investigator for NDIF.
“We fundamentally don’t understand how these systems work, what they learned from the data, what their internal algorithms are,” Bau explains. “I consider it one of the greatest mysteries facing scientists today — what is the basis for synthetic cognition?”
The NDIF project seeks to address this knowledge gap by creating a computational infrastructure that will equip researchers with deep inferencing tools to delve into the inner workings of large language models. This capability does not currently exist, limiting both researchers and companies across multiple sectors in leveraging the full potential of large-scale AI.
“With NDIF, U.S. researchers will be able to peer inside the ‘black box’ of large language models, gaining new insights into how they operate and greater awareness of their potential impacts on society,” says Sethuraman Panchanathan, director of the National Science Foundation.
In addition to establishing this groundbreaking infrastructure, NDIF aims to democratize access to large language models. Northeastern will build an open software library of neural network tools, enabling researchers to conduct experiments without needing their own resources. Educational materials and workshops will also be developed to train scientists and students in using the NDIF platform.
The implications of NDIF’s research extend far beyond the scientific community. By demystifying the underlying mechanisms of large language models, the project could inform policymakers, creatives, and others on the capabilities, limitations, biases, and potential safety issues associated with these powerful AI systems.
“The goal of understanding how these systems work is to equip humanity with a better understanding for how we could effectively use these systems,” Bau says.
NDIF will leverage significant computational power and resources from partnerships with institutions like the University of Illinois Urbana-Champaign’s National Center for Supercomputing Applications and New America’s Public Interest Technology University Network. The project will also prioritize the ethical use of AI, with a focus on social responsibility and transparency.
As the world increasingly relies on AI-powered tools, Northeastern University’s NDIF project represents a crucial step towards unraveling the mysteries of these advanced systems, paving the way for more informed and responsible development and deployment of AI technologies.
