Research Notes
Working notes on scientific practice, agents, research systems, world models, data, and embodied AI.
All views expressed here are my own.
Research Practice / Sustainable Performance
The Calm Control Plane: A Durable Edge in Competitive AI Research
As AI makes production abundant, calm judgment becomes the control plane for choosing questions, engineering reliable systems, navigating review, and remaining capable across many research cycles.
Read noteResearch Practice / Scientific Integrity
Beyond Acceptance: A Discipline for Durable AI Research
A research operating system for pursuing top-conference ambition through proportional rigor, portfolio design, epistemic ownership, and auditable human–agent research.
Read noteAgents / Software Engineering
Engineering with Coding Agents: Judgment, Skill, and Responsibility
Coding agents shift the bottleneck from code generation toward judgment, verification, responsibility, and the durable engineering skill that remains with the human.
Read noteAgents / Research Systems
Thoughts on AutoResearch
Research automation should tighten the loop between hypotheses and evidence without hiding the scientific judgment that makes results meaningful.
Read noteFoundation Models / World Models
Thoughts on World Models
A functional view of world models—and why LLMs may already be partial, implicit models of the worlds represented through language.
Read noteEmbodied AI / Foundation Models
Thoughts on Embodiment Foundation Models
How general models might connect multimodal understanding, action, embodiment, and reliable adaptation in the physical world.
Read noteEmbodied AI / Data
Ego Data: Today's Feasible Path, Not the Final Substrate
Why ego data is worth pursuing now, why recent scaling results matter, and why task-centric physical transition data may develop in parallel.
Read note