The Art of Leaderboarding in the Era of Extreme-Scale Neural Models

Explore a keynote address from the 2022 Research Showcase that challenges the notion of scale as the ultimate factor in AI model performance. Delve into Yejin Choi's presentation on how smaller academic models can compete with larger industry-scale neural networks through the use of knowledge and reasoning algorithms. Learn about "symbolic knowledge distillation" and its role in creating machine-authored commonsense knowledge bases that outperform human-authored ones. Discover how unsupervised, inference-time reasoning algorithms can enhance complex language generation tasks. Gain insights from a MacArthur Fellow and award-winning researcher in natural language processing and artificial intelligence on the future of AI development and the importance of balancing scale with knowledge and reasoning capabilities.