4 ms·
My mental model is a bit different: Context -> Attention Span Model weights/Inference -> System 1 thinking (intuition) Computer memory (files) -> Long term m
by Difwif 1y ago
My mental model is a bit different:
Context -> Attention Span
Model weights/Inference -> System 1 thinking (intuition)
Computer memory (files) -> Long term memory
Chain of thought/Reasoning -> System 2 thinking
Prompts/Tool Output -> Sensing
Tool Use -> Actuation
The system 2 thinking performance is heavily dependent on the system 1 having the right intuitive models for effective problem solving via tool use. Tools are also what load long term memories into attention.
- yellow_postit 1y agoI like this mental model. Orchestration / Agents and using smaller models to determine the ideal tool input and check the output starts to look like delegation.
- mark_l_watson 1y agoVery cool, good way to think about it. I wouldn’t be surprised if non-AGI LLMs help write the code to augment themselves into AGI. The unreasonable effectiveness of deep learning was a surprise. We don’t know what the future surprises will be.