Deep Learning Insights
Tim discusses the limitations of deep learning models like BERT, emphasizing their reliance on surface features rather than deeper understanding. He raises important questions about the empiricism versus nativism debate in AI, arguing that while engineering successes exist, they may exploit spurious statistics, leading to brittle solutions that fail to generalize effectively. The conversation highlights the need for a more nuanced approach in applying advanced techniques to complex problems.In this clip
From this podcast
Data Skeptic
BERT is Shallow
Related Questions