AI Snips
Chapters
Transcript
Episode notes
Optimize For Novelty Not Just A Single Objective
- Novelty search and evolutionary algorithms optimize for interestingness rather than a single objective.
- Ken Stanley's work shows selecting for novelty creates stepping stones that recombine into surprising solutions.
Machine Creativity Differs From Human Intent
- Machines can meet psychological tests for creativity but follow different paths than humans.
- Alice and Sam argue ML produces novel, sometimes useful outputs yet often lacks intent and human-context grounding.
LLMs Simulate Understanding Without Grounding
- Large language models simulate understanding but do not truly understand the world.
- Alice warns relying on LLMs' apparent comprehension can mislead decisions because models learn statistical regularities, not grounded understanding.