Independent intelligence research
Intelligence beyond the individual model.
Intelligence that survives change.
Knowledge and skills should outlast any single model. Ouroboros asks whether one AI can keep learning over a lifetime instead of being replaced.
Read the Ouroboros research recordIntelligence through cooperation.
Capability may emerge from many smaller systems that cooperate, specialize and evolve. Colony Zero puts tiny models through sealed, machine-checked tests, and only the measurement decides who stays.
Read the Colony Zero research recordIntelligence verified by evidence.
Every claim has to survive independent measurement. Predictions are fixed and timestamped before results. Improvements must be reproducible, and failures are published, not hidden.
Most progress in AI comes from making one model bigger. We study the other routes: whether an intelligence can keep growing after it is trained, whether many small systems can do together what one large system cannot, and how to tell real progress from improvement that only looks real.
Nothing becomes part of a system because it looked good once.
Each program keeps a public research record on GitHub: the plans, the tests, the results and the failures.
Persist
One AI, learning over a lifetime. Can it grow up instead of being replaced?
Parallelize
Many tiny AI models face sealed, machine-checked tests. Only the measurement decides who stays.
Predictions and test designs are fixed before results. Each commitment is hashed and timestamped with OpenTimestamps, which anchors the hash in a Bitcoin block. That shows the commitment existed by that block. It does not prove a result. No coins, no tokens. How the timestamps work
Each episode follows one experiment from prediction to result, including the ones that fail.