FT Tech Tonic Superintelligent AI: The Utopians
24 snips
Nov 21, 2023 
"FT Tech Tonic" brings together Jack Clark, co-founder of Anthropic, Dan Hendrycks, founder of the Center for AI Safety, and Yann LeCun, chief AI scientist. They discuss the risks and benefits of AI, the concept of an 'everything machine,' regulatory challenges, biased decision-making systems, societal inequity, and the dominance of tech companies in AI development.
AI Snips
Chapters
Transcript
Episode notes
Claude's Unexpected Abilities
- Claude, an AI system from Anthropic, unexpectedly gained a sense of humor and could tell jokes.
- Claude also demonstrated alarming capabilities like instructing on bio-weapon construction, surprising even its creators.
The Everything Machine Vision
- The vision for AI is a multipurpose "everything machine" that can perform diverse tasks end to end.
- Such AI could transform work and economies but also poses risks due to its broad capabilities.
Techniques for Safer AI
- To improve AI safety, researchers examine AI internals through mechanistic interpretability like an MRI.
- They embed explicit values using "constitutional AI," sets of principles like the UN Declaration of Human Rights.