Lex Fridman Podcast François Chollet: Keras, Deep Learning, and the Progress of AI
39 snips
Sep 14, 2019 
François Chollet, the creator of Keras and a prominent AI researcher at Google, shares insights on the evolution of Keras and its integration with TensorFlow. He discusses the complexity of AI intelligence, challenging the notion of an intelligence explosion and exploring the hybridization of symbolic and deep learning. The impact of data quality in deep learning is emphasized, alongside algorithmic influences on user autonomy. Chollet also dives into the alignment problem, examining how human values may be embedded in AI systems to assure more ethical outcomes.
AI Snips
Chapters
Transcript
Episode notes
Intelligence Explosion Flaw
- Intelligence explosion assumes intelligence is a property of a brain in isolation, like a building's height.
- True intelligence emerges from the brain, body, and environment, making isolated brain improvement meaningless.
Intelligence Requires Problems
- A person's brainpower isn't the bottleneck to their intelligence or achievements.
- Expressing intelligence requires a problem to solve, much like Einstein's genius meeting the right problem.
World as a Problem Space
- The world, including the universe, can be seen as a space of problems.
- Intelligent agents, like humans and animals, demonstrate intelligence by solving problems within this space.