- Published on
LLM360: From 360° Open Source to 360° Collaboration in AI
- Authors
- Name
- MLSys @UCSD
- Name
- Host: Hao Zhang
- Name
- Host: Daniel Zhao
This week, our MLSys seminar is pleased to present a talk by Dr. Zhengzhong (Hector) Liu scheduled on Thursday, February 20 @ 6:30 PM (PST). We welcome all interested students and faculty to attend the talk on Zoom: https://ucsd.zoom.us/j/91813834040 (Zoom-only)
Talk title: LLM360: From 360° Open Source to 360° Collaboration in AI
Talk Abstract: The LLM360 project is dedicated to advancing AI by open-sourcing large-scale foundation models and high-quality datasets. This talk will explore key initiatives within LLM360, including the K2 project, the most capable fully open-source language model, and TxT360, a curated dataset designed to support transparent and reproducible model training.
As open-source AI continues to gain momentum, it presents both significant challenges and opportunities. This talk will examine the true meaning of open source, emphasizing the principles of transparency, reproducibility, and accessibility. While many projects claim to be open, there is a growing need to distinguish between partial and fully open-source efforts. By reaffirming the spirit of full openness—what we call 360° open—we propose a new approach to academic and industry collaboration in open-source AI. Using K2 and TxT360 as case studies, I will illustrate how a truly open ecosystem can drive meaningful progress and shape the future of AI development.

Bio: Zhengzhong (Hector) Liu is the Director of the MBZUAI Institute of Foundation Models (IFM) Silicon Valley Lab, where he leads the university's large language model (LLM) training efforts. He drove the development of open-source foundation models through the LLM360 project and led the development of K2, the best fully open-source language model, and played a key role in Jais, the leading Arabic language model series.
Previously, Hector was the Head of Engineering at Petuum, where he led initiatives in Natural Language Processing pipelines, MLOps, and LLM development. He holds a Ph.D. from Carnegie Mellon University (CMU), where he worked with Professors Teruko Mitamura and Eduard Hovy and co-founded the CASL project with Professor Eric Xing.