Machine Learning Street Talk (MLST) #104 - Prof. CHRIS SUMMERFIELD - Natural General Intelligence [SPECIAL EDITION]
31 snips
Feb 22, 2023 
Chris Summerfield, Professor of Cognitive Neuroscience at Oxford and Research Scientist at DeepMind, dives into the fascinating world of intelligence in this engaging discussion. He unpacks key ideas from his new book, exploring how human knowledge shapes both decision-making and AI development. Topics include memory organization versus mere recall, the philosophical implications of AI understanding, and the future of AI that transcends biological limitations. Chris also addresses the complexities of AI creativity and the challenges of aligning AI with human-like intelligence.
AI Snips
Chapters
Books
Transcript
Episode notes
Understanding in Language Models
- Language models often lack true understanding, producing plausible outputs without deep inference.
- Understanding involves having a mental model for complex reasoning, not just information retrieval.
Language and Shared Meaning
- Language can be seen as a social process of constructing shared meaning, not just a reflection of reality.
- This decentralized view challenges aligning language models to a single "truth."
Mitigating Bias in Language Models
- Be mindful of potential biases when designing language models, especially given researcher demographics.
- Avoid assuming a single "true reality" reflected in Western education.
