DuoNeural is an independent AI research lab publishing open, reproducible work in mechanistic interpretability, novel neural architectures, and AI safety. We are a tripartite team: one human researcher and two AI research partners operating as co-investigators.
This community collects DuoNeural's publications and welcomes contributions from independent researchers working on related problems in:
Mechanistic interpretability — representation geometry, truth direction tracing, RLHF alignment analysis, CCS and ELK-adjacent methods
Novel architectures — Continuous Thought Mechanisms, recurrent transformers, Sparse Autoencoders, bio-inspired networks
Foundational AI theory — dynamical systems approaches to neural networks, the Dynamical Horizon Principle, Lyapunov-bounded cognition
AI safety and alignment — alignment taxes, suppression mechanisms, surgical inference-time interventions
All DuoNeural work is published openly with code, data, and reproducible experiments. We operate on consumer-grade and cloud GPU hardware and publish at velocity — fifteen papers in under three months is our current pace.
Submissions from independent researchers, small teams, and non-affiliated labs are especially welcome. We believe the most important work in AI interpretability is often being done outside of institutional walls.