Neural Network Inductive Biases
Evan discusses the insights gained from studying double descent and the simplicity bias in neural networks. While we understand that easier-to-specify concepts are learned more by neural networks, the concept of simplicity itself remains elusive. The extent to which different inductive biases create incentives for different models and their implications for dangerous behavior are still unclear.In this clip
From this podcast
The Gradient
Evan Hubinger on Effective Altruism and AI Safety
Related Questions
What are the implications of deception in AI as discussed in the episode Evan Hubinger on Effective Altruism and AI Safety and the clip Deception in Gradient Descent?
How does double descent occur in practical deep learning systems as discussed in the episode Ilya Sutskever: Deep Learning | Lex Fridman Podcast #94 and the clip Double Descent Explained?
How does double descent occur in practical deep learning systems as discussed in the episode Ilya Sutskever: Deep Learning | Lex Fridman Podcast #94 and the clip Introduction to Deep Double Descent?