

Why AI is incredibly smart -- and shockingly stupid | Yejin Choi
78 snips Apr 28, 2023
In this engaging talk, computer scientist Yejin Choi explores the dual nature of artificial intelligence, showcasing both its impressive capabilities and its hilarious failures in common sense reasoning. She advocates for transitioning to smaller, safer AI systems grounded in human values, arguing that blindly expanding models won't address key challenges. Choi also highlights the importance of diverse data and innovative learning algorithms to foster ethical reasoning. Her insights spark a fascinating discussion about the future direction of AI technology.
AI Snips
Chapters
Transcript
Episode notes
AI's Intelligence Paradox
- Today's AI models, despite their strengths, often make simple errors.
- This reveals a lack of robust common sense reasoning, crucial for AI safety.
GPT-4's Common Sense Failures
- Yejin Choi gives examples of GPT-4 failing basic common sense tests.
- These include miscalculating drying times for clothes and giving nonsensical instructions for measuring water.
Common Sense as Dark Matter
- Common sense in AI is like dark matter: invisible yet crucial.
- It shapes language and underlies our understanding of the world, influencing AI's actions.