17 :: IN :: SELF-TAUGHT
Transistor-free computing on hydrogen-passivated silicon. Computation happens via resonant electron absorption in 5-atom dangling-bond clusters. Memory and compute share the same atoms - no cache hierarchy, no DRAM bus. On a 3 cm² die: 14.1 TB in-situ memory at 79 mW.
Three attention projections instead of four. Content routes through a rank-r state ⊙ action matrix grounded in RL navigation. 16× smaller KV-cache than standard attention at ≤1.6 ppl penalty. Trains across text, vision, audio, world states, and cancer genomics under one block class. Validated at small scale only.
Modular multi-agent cognitive architecture. 12 specialized domain experts collaborating through Web-of-Thought reasoning.
Fine-tuning that skips samples the model already knows. Compute is routed to hard samples; mastered ones are frozen. Up to 80% compute savings at scale.
> CS
-0F172A?style=for-the-badge&logo=c&logoColor=39FF14){kind=icon}
-0F172A?style=for-the-badge&logo=cplusplus&logoColor=4ADE80){kind=icon}
> NANOG1 :: ~200-300M param cancer foundation model, trained on multimodal oncology data (histopathology, radiology, TCGA) via Quatrix
> still training, still loading