Reading · GiTech Design
On the Origin of Digital Species
When Algorithms Meet Darwin · 硅基生命的进化论
GiTech Design · Written winter 2025
We look at today's noisy world of AI large models through the lens of biological evolution — the competition and iteration among algorithms turn out to rhyme with how species evolve in nature. Transformer is the common ancestor; each model is a species that branched off, each finding its niche in an ecosystem held up by three pillars: compute, data and algorithms. No solemn lecturing — just a little curiosity, told slowly.
A race is people chasing each other on a track; these algorithms are more like mushrooms after spring rain, popping up overnight from every corner.
A pace that would make Darwin reconsider
Biological
single-cell → multicellular, ~2.5 billion years
AI
Transformer → GPT-5, eight years
Speed-up
about 1 billion ×
Contents
- Prologue · Starting from a salted duck egg
- The origin of species in the digital world
- Survival of the fittest: the three-pillar game
- Mutation & recombination: the secret of capability leaps
- Niche differentiation: a place for each
- Co-evolution: everything connected
- Epilogue · Where does evolution end?
Download PDF ↗Free to read · our small, curious observation on how AI evolves.