The Theory of Anything Episode 108: AI and Obedience (with Dan Gish)
May 27, 2025
This conversation features Dan Gish, a software engineer and former educational software company owner, who offers a positive perspective on AI. He discusses the potential of large language models and advanced algorithms in fostering human-level creativity. Gish critically examines skepticism surrounding artificial general intelligence (AGI) and the implications of AI's obedience versus human disobedience. He also explores how innovations like the DeepSeq model redefine creative processes, suggesting that AI advancements are essential for the evolution of truly creative systems.
AI Snips
Chapters
Books
Transcript
Episode notes
Obedience vs Disobedience Misconception
- Creativity involves the ability to choose goals, not just disobedience in behavior.
- Humans are obedient to their own program, so obedience can also be creative.
LLMs Capture Interestingness
- Large language models (LLMs) have a sense of interestingness emerging from training on vast internet data.
- This ability to assess interestingness is key to open-ended creativity algorithms.
LLM Solves Tough Epistemology Problem
- Bruce tested an LLM that independently solved an epistemology problem he struggled with for years.
- He views that feat as an act of considerable creativity by the LLM.
