CADS - Composable and distributed systems group
Mon, 12 May 25
First share of our experimental call summaries: AI-generated digests that have demonstrated efficacy for weekly Yak Collective live study groups. Represents a step forward in our exploration of human-machine collaborative cognition and Yak Oracle and Yak Memory systems for enhancing collective intelligence capabilities.
### Discussion of AI and Brain Comparisons
- Article discussed: [https://www.quantamagazine.org/ai-is-nothing-like-a-brain-and-thats-ok-20250430/](https://www.quantamagazine.org/ai-is-nothing-like-a-brain-and-thats-ok-20250430/)
- Additional reference: [https://www.numenta.com/blog/2021/04/26/comparing-hinton-glom-model-to-numenta-thousand-brains-theory/](https://www.numenta.com/blog/2021/04/26/comparing-hinton-glom-model-to-numenta-thousand-brains-theory/)
- Key debate: Whether fundamental differences between AI and biological brains matter
- Brain characteristics:
- More hierarchical structure
- Fixed inputs and sensors
- Handles basic functions (organ operation, feeding) beyond language/math
- Contains evolutionary “pre-training” spanning billions of years
- Early cognitive development insights:
- Pre-verbal children and dogs can follow pointing gestures
- Unique to humans and domesticated animals, not found in chimps/monkeys
- Demonstrates early symbolic reasoning capabilities
### LLM Characteristics and Limitations
- Feels more like interacting with a process/environment than hierarchical system
- Uses significantly more training data compared to biological systems
- Lacks embodiment that even simple organisms (like maggots) possess
- Current limitations:
- “Lock-in syndrome” - limited sensory capabilities
- Restricted to language/symbolic processing
- Requires massive computational resources
- Questions about training efficiency
### Using LLMs as Research Tools
- Experience using Claude to understand complex articles:
- ELI5 to ELI16 range tested
- ELI8-10 proved most effective for this piece
- Asked for long-term valuable points vs temporarily relevant ones
- Distinction between deterministic vs non-deterministic uses:
- “Calculator vs Oracle” paradigm
- Reliability assessment challenges (Brandolini’s law)
- Trust verification overhead
### Technical Comparisons and Future Implications
- Birds vs. Planes analogy:
- Different mechanisms achieving similar outcomes
- Energy density and material constraints create different optimization paths
- Demonstrates how different approaches can solve same problems
- Key differences in implementation:
- Biological systems: evolved complexity, efficient but constrained
- AI systems: designed simplicity with massive computational power
- Future considerations:
- Quantum effects in brain function
- Sensor density and integration possibilities
- Scaling limitations in both biological and silicon systems