Matt Boulos

Matt Boulos serves as the Head of AI Safety & Policy and has made significant contributions to the field of artificial intelligence, particularly at the intersection of machine learning and systems. His work focuses on efficient training methods and managing long-range context in AI systems. Matt has explored hardware-friendly sparsity techniques and has been involved in maintaining quality while implementing sparsity in AI models. He has also investigated efficient inference from multiple perspectives including model design, software frameworks, and hardware optimization. Additionally, Matt has researched scalable alternatives to transformer models, particularly in terms of sequence length. He has co-authored an update on the ethical decision-making capabilities of large language models (LLMs) and contributed to a podcast episode discussing advanced topics such as FlashAttention, sparsity, quantization, and efficient inference. Matt is set to join Princeton University as an assistant professor, where he will continue his research and advocacy in AI safety and policy.