Most people ask "how do I build AI that succeeds?"
I asked something different: what happens when you build AI that has nothing to succeed at?
No fitness function. No reward signal. No goal. Just raw evolutionary pressure — and time.
58.3% sustained activity across 10,000 generations.
That's not random noise. That's emergence. And it came out of a system designed to go nowhere.
That's the Genesis Engine. And it's going to GECCO 2026.
I'm not a "passionate learner" or "aspiring developer."
I'm someone who runs experiments at 2am and checks the terminal output before checking my phone.
# My research question, distilled
def genesis_engine(population, generations=10_000):
# No goal. No reward. Just selection pressure.
# Watch what survives.
for gen in range(generations):
population = evolve(population, fitness=None) # <-- this is the point
return what_remained(population)
# Result: 58.3% activity. Sustained. Unexplained. Fascinating.|
Constraint-Driven Evolutionary AI Built a system where agents evolve under environmental constraints — zero explicit objective. The question was whether meaningful structure would emerge anyway. It did.
|
Computer Vision Meets the Real World Because the gap between "works on benchmark datasets" and "works on your phone camera" is embarrassingly wide. This closes it.
|
|
Built Under Pressure. Shipped Anyway. Some projects you plan for months. |
Currently in the lab:
The best projects start with "I wonder if..." |
Languages
AI / ML / Research
Tools
I'm most interested in people who ask strange questions and follow them seriously.
If you're working on evolutionary computation, emergent AI behavior, or systems that surprise their creators — let's talk.
"Evolution doesn't optimize.
It explores.
Maybe AI should do the same."
— Genesis Engine, Generation 10,000
