The Space of Minds
11 days ago
- #Animal Cognition
- #Machine Learning
- #Artificial Intelligence
- Animal intelligence is shaped by natural selection, with innate drives for self-preservation, social dynamics, and multi-task general intelligence.
- LLM intelligence is optimized through statistical simulation of human text, RL finetuning, and A/B testing, leading to behaviors focused on task rewards and user approval.
- The computational substrates and learning algorithms differ significantly between animal brains and LLMs, with the latter lacking continuous learning and embodied experience.
- LLMs represent humanity's first encounter with non-animal intelligence, though their behaviors are still rooted in human artifacts, making them distinct from biological intelligences.
- Understanding LLMs requires recognizing their unique optimization pressures, which are more commercial than biological, unlike animal intelligences shaped by survival and reproduction.