Ion Stoica Highlights Open-source Efficiency Across the AI Stack

Over a hundred students, researchers, and members of the community gathered in Davis Auditorium on March 28 to hear Professor Ion Stoica deliver the latest installment in Columbia Engineering’s Lecture Series in AI.

Stoica, a computer scientist at the University of California at Berkeley, director of Sky Computing Lab, and executive chairman at Databricks and Anyscale, delivered a talk titled “An AI Journey: From Scaling AI Workloads to LLM Evaluation.”

In his introduction, Vice Dean for Computing and AI Vishal Misra told the crowd that Stoica has been “both a very successful entrepreneur and a passionate supporter of open source.” Two companies that Stoica founded — Databricks and Anyscale — use open-source software to support large swaths of today’s AI infrastructure, Misra said. 

Image

Stoica’s talk addressed the multifaceted problem of scaling AI systems at every stage, from preprocessing through application and evaluation. He walked the audience through tools and concepts for more efficiently using high-value hardware and integrating AI workflows at the increasingly large scales at which AI systems operate. 

He discussed three projects — Ray, vLLM, and Chatbot Arena — that bring new approaches to workflows, allowing for more efficient use of computational resources. 

Image

Stoica closed his talk with a few words about open-source AI. After observing that the Internet was the result of cooperation among government, academia, and industry, Stoica advocated for an approach to open source that extends beyond open weights to include open data, code, and evaluation. 

“Unlike the internet, AI development is siloed across companies. If you believe that AI is so important and you want to accelerate progress in a responsible way, you have to have all of the best minds in AI collaborating,” he said. “They can’t do that without shared artifacts and a shared infrastructure that everyone can access.”

Lead Photo Caption: Ion Stoica during his Mar. 28 visit to Columbia Engineering as the guest speaker for the School’s Lecture Series in AI

Lead Photo Credit: Brandon Vallejo/Columbia Engineering