

In Machines We Trust: A conversation with Geoffrey Hinton (Live)
22 snips May 6, 2023
AI Snips
Chapters
Transcript
Episode notes
Hinton Reevaluates Neural Networks
- Geoffrey Hinton originally thought neural network models were only rough approximations of brain function.
- The high performance of models like GPT-4 made him realize these digital models work quite differently but more effectively in some ways.
AI Sharing Boosts Learning Speed
- Digital intelligences can share learning instantly by copying their weights between copies.
- This collective learning capability lets AI process vastly more data than any human could individually.
AI's Growing Reasoning and Risk
- Advanced AI shows impressive common sense reasoning abilities, outperforming previous symbolic AI attempts.
- Their growing intelligence raises concerns about manipulation, as smarter AIs could easily influence humans.