Brit is awesome- I totally relate to his quest to explore data compression- except for machine learning “knowledge.” Like- For an image recognized by a deep layered neural work- that image exists as a “constellation” as it were- of connected neurons. What are the critical connections and what are the trivial connections for a concept like “cat.” Or “dog.” Can the trivial ones be “shaved off” so to speak- what’s the minimal number of neurons needed for a given tool to recognize a concept? Think of human cave drawings that evolved into letters.
Brit is awesome- I totally relate to his quest to explore data compression- except for machine learning “knowledge.” Like- For an image recognized by a deep layered neural work- that image exists as a “constellation” as it were- of connected neurons. What are the critical connections and what are the trivial connections for a concept like “cat.” Or “dog.” Can the trivial ones be “shaved off” so to speak- what’s the minimal number of neurons needed for a given tool to recognize a concept?
Think of human cave drawings that evolved into letters.
Exciting episode! Keep up the great work on the podcast 💚
Finally a new episode! I think Vlad Ten will be happy. Please keep making the podcast!