AI's Language Leap: MIT CSAIL Associate Professor Jacob Andreas Explores NLP and LLMs

For a full, uncut video version of this show (shot in Professor Andreas' MIT CSAIL office!), watch here: https://www.youtube.com/watch?v=QBV5Bj_NnbA MIT CSAIL Associate Professor Jacob Andreas walks listeners through how LLMs like ChatGPT evolved from academic curiosities to industry-disrupting technologies. Reflecting on the transformations he has observed in AI research, Professor Andreas discusses how some ideas—like linking NLP research to cognitive science—have taken a backseat, while others—like the importance of large-scale training data—remain central. He offers insights on AI use cases, the emergence of models like DeepSeek, and the critical role of human oversight in AI deployment. Acknowledging both the excitement and concerns surrounding AI, Professor Andreas believes that educators and industry alike face a challenge in keeping pace with the ‘moving target’ of AI progress. Plus: learn how AI can be used to decode the language of sperm whales! About the guest: Professor Andreas is an MIT EECS Associate Professor and CSAIL PI. Before joining MIT, he earned his PhD at Berkeley. His group's research focuses on three themes: building machine learning models that learn language in human-like ways, enabling the use of language as a general-purpose tool for building intelligent systems, and developing techniques for understanding machine learning models using language. His group envisions a future built around training, explaining, and interacting with intelligent systems for users of all the world's languages. Learn more about Professor Jacob Andreas and his research here: https://www.mit.edu/~jda/ Want to know more about Professor Andreas' work on the language of sperm whales? Read here: https://news.mit.edu/2024/csail-ceti-explores-sperm-whale-alphabet-0507 MIT's CSAIL Alliances Podcast is your guide to how AI & computer science research impact how we live, work, play, and learn. Listener discounts, meet the host, and more: csail.mit.edu/podcast