Foundation Models: Strengths, Weaknesses, and Future Implications - Session 4

Explore the strengths, weaknesses, and societal implications of foundational models like GPT-3 in this 42-minute panel discussion from AI UK 22. Join experts from The Alan Turing Institute, University of Oxford, University of Bristol, and DeepMind as they delve into topics such as the limitations of these models, their knowledge of language, linear scaling, data choice, sustainability concerns, and the concept of "bullshit machines." Gain insights into the future of AI, the challenges of black box systems, and the importance of calibrated probabilities in this thought-provoking conversation about the latest developments in artificial intelligence.

Introduction
What are foundation models
Are foundation models limitless
Knowledge of language
Linear scaling
Bullshit machines
Data choice
How to make them more sustainable
Concerns
Realistic
Should you want school
Black boxes
calibrated probabilities
conclusion