Emergent Concepts in Learning
Ishan discusses the fascinating emergence of object boundaries through self-supervised learning, highlighting how models can identify and group objects without explicit training. He suggests that fundamental concepts like object permanence and even counting could naturally arise from vast datasets, challenging our understanding of machine intelligence. The conversation explores the potential for these models to grasp deeper concepts, opening up new avenues for interpreting images.In this clip
From this podcast
Lex Fridman Podcast
Ishan Misra: Self-Supervised Deep Learning in Computer Vision | Lex Fridman Podcast #206
Related Questions