Sara Hooker

TL;DR Sara Hooker is a leading AI researcher whose work spans model compression, interpretability, and the responsible development of large-scale machine learning systems.

Sara Hooker is an influential figure in modern artificial intelligence, known for her research on the efficiency, transparency, and societal impact of machine learning models. With a background that bridges cutting-edge technical research and global perspectives on AI access, she has played a significant role in shaping how the field approaches responsible, inclusive, and computationally efficient AI.

Sara Hooker began her career working at Google Brain, where she led groundbreaking research on model compression, pruning, and interpretability. Her work focused on making neural networks smaller, faster, and more efficient without sacrificing performance, helping expand access to powerful AI tools worldwide. She contributed to key studies showing how large models behave under stress, how they distribute information internally, and how they can be made more transparent.

She later joined Cohere as a research leader, bringing her expertise in deep learning systems to large-scale language model development. In addition to her technical achievements, she founded the nonprofit organization Delta Analytics, which provides data science training to teams working on social impact problems, demonstrating her long-standing commitment to equitable AI.

Her work continues to influence how researchers think about the costs, risks, and opportunities of state-of-the-art machine learning.

• Significant research contributions to model compression, pruning, and efficient neural networks

• Former research scientist at Google Brain, leading work on interpretability and robustness

• Research leader at Cohere, advancing large-scale language model development

• Founder of Delta Analytics, promoting global access to data science education

• Influential studies on model behavior, internal representations, and responsible AI

• Prominent advocate for equitable, transparent, and efficient machine learning