Machine Learning Street Talk (MLST)

#112 AVOIDING AGI APOCALYPSE - CONNOR LEAHY

31 snips
Apr 2, 2023
In this engaging conversation, Connor Leahy, CEO of Conjecture and AI safety advocate, shares his insights on the potential risks of artificial general intelligence (AGI). He stresses the crucial need for AI alignment and the importance of empathy in understanding these systems. The discussion dives into the complexities of AI training, the dangers of dehumanizing biases, and the challenges of balancing research with product development. Connor also reflects on personal growth, storytelling, and the pressing need to alleviate human suffering as we navigate the rapidly evolving AI landscape.
Ask episode
AI Snips
Chapters
Books
Transcript
Episode notes
ANECDOTE

Initial Skepticism and Shift in Perspective

  • Tim Scarfe was initially skeptical of GPT-3, seeing it as mere pattern matching.
  • Connor Leahy's examples and access to the API shifted Scarfe's perspective towards acknowledging GPT-3's reasoning abilities.
INSIGHT

Interface and Reliability over Capability

  • Connor Leahy emphasizes the importance of the interface in interacting with language models.
  • He argues that ChatGPT's success stems from its improved interface and reliability rather than fundamentally new capabilities.
INSIGHT

Language Models as Cognitive Extensions

  • Tim Scarfe acknowledges the revolutionary impact of language models as an extension of human cognition.
  • While acknowledging limitations, he emphasizes their utility across diverse applications, highlighting the potential of filling gaps and evolving usage.
Get the Snipd Podcast app to discover more snips from this episode
Get the app