we finished our patent strategy. quite few families and interesting claims. what interest me the most is our discussion about when tech should be public domain or not. at some point tech should probably be just that, if it has the power to change
"tiny" country delivering so much positive global impact over and over again.
and still we all gather to celebrate midsummer, no matter who you are, or what you do. for a thousand years. Its pretty cool.
Gradient-free symbolic model learns constructions instead of memorizing. Hits 100% on systematic understanding tests where both retrieval and neural baselines fail, with perfect retention and easy local edits.
Built with 0.3% reinforcement learning code, Sapiexo Core achieved 46% lower regret than Thompson Sampling on difficult long-horizon tasks - with early saturation and deterministic cross-server reproducibility.
Results from v0.6.2
Not a model.
Not an LLM.
Not "RL on steroids."