Independent Researcher · Systems Architect

Haelio Tang

I study the dynamical structure of intelligence — from the hidden-state trajectories of large language models to the thermodynamic limits of distributed coordination.

Distributed Coordination
Coordination as inference under partial observability in multi-agent systems
Active Inference
Free energy minimization and belief-driven behavior in distributed settings
Language Model Dynamics
Trajectory-level coherence and recurrence structure in neural language model hidden states
Thermodynamic Computing
Physical bounds on information processing, Landauer limits, and dissipation
Systems Architecture
Scalable infrastructure for real-time multi-agent coordination
The instruments we use to evaluate language models are structurally blind to the dynamics that matter most for long-horizon coherence. The bottleneck isn't control. It's observability. — from Trajectory Observatory, RN-001 (2026)
2026 · Active
Trajectory Observatory
A live research system tracking the hidden-state dynamics and spatiotemporal geometry of large language models. Research notes document experiments on observability gaps, scale-selective recurrence diagnostics, and causal trajectory steering — published as they emerge.
LLM Dynamics Trajectory Physics STR Causal Intervention

Papers currently in preparation. Preprints and research notes are available through the Trajectory Observatory.