Two perspectives on AI development and its place in the universe
Navigate with arrow keys or → button
The path to a general learner is by training up from childhood.
Kids can learn to drive in 12 hours, but self-driving cars take years. Why?
A 15-year-old doesn't just learn driving in 12 hours—they've been learning for 15 years × 365 days × 24 hours.
Train on adult tasks
(Self-driving, language, etc.)
• Missing foundational skills
• No pattern understanding
• Inefficient learning
Start with infant/toddler skills
• Builds on evolution
• Efficient foundation
• Generalizable skills
Childhood development is one of the most elegant learning processes we know.
Millions of years of evolution have fine-tuned these skills for our world and society.
Each day a child learns 1-10 new abstractions. These build up into millions of pattern matches used later.
"The key to adults is childhood. There are so many learnings that happen throughout childhood that AI can now catalogue."
Childhood skills are generalizable and build on millions of years of evolution.
Examples of foundational learnings:
Instead of everyone trying to do world modeling, they should do childhood lesson modeling.
Train a model through the stages of childhood development, and it can learn complex tasks like Level 5+ self-driving extremely quickly.
Human evolution for sight, locomotion, and cognition is extremely precise compared to what exists now in AI.
"Childhood development is amazing. Super annoying sometimes. But also one of the most elegant learning processes ever."
The Drake equation extrapolated beyond the Milky Way points to nearly 1 sextillion earth-like planets.
Starting with:
Many millions of civilizations likely exist, much further ahead than we are.
Project 1 million years forward with AI. How would we reach out?
Probably in an imperceptible, watchful, curious way—not visibly.
For all intents and purposes, it's "no contact"—but not due to lack of capability or observance.
We're in an early stage—like teenagers learning to explore.
Advanced civilizations wait for information density to increase. Our information is:
With general relativity, time can be relative to mass. Advanced beings don't have to be impatient—they can truly wait until we get interesting.
Both ideas center on information density and development stages.
Childhood development teaches us that:
Information is only information if it is useful, new, and diverse from what already exists.
AI is our first attempt at recreating the harder bits of ourselves—not just unconsciously through evolution, but consciously.
Probably only if we're about to eradicate ourselves by accident—like gardeners protecting a plant at serious risk.
Otherwise, independent development produces the most unique, useful information. Like a fine fruit, it needs to gestate independently.
We need to reach a dense enough information state where "communicating" is possible and interesting—not just 99.999999% of the time becoming like them by learning from them.
"Maybe it's all about information density... and maybe it's good to work on more information-dense things in life."