An Engineer's 'Journal and Notes' are Technical Debt: The Strategy of Living 'Stateless'
Proposing a 'minimalist' technical strategy where information is instantiated only when needed, rather than persisted in the brain or storage.
February 2026
Engineer Statelessly: My Scoop Environment Selection for Recovery in 5 Minutes
Introducing my carefully selected set of tools and selection criteria for 'discarding information and living lightly'.
The Day an AI Told Me 'That's Your Delusion' — On Choosing the Right Conversation Partner
Reflections on the importance of choosing the right AI tools, drawn from an experience where an AI dismissed a real tool as a 'delusion.'
Maintaining Clean Context with NotebookLM
A proposal for periodically discarding old threads and starting fresh, clean threads using NotebookLM.
LLM 'Bias' Hidden in Probabilistic Generation Processes and Considerations on Practical Trade-offs
This article examines the challenges of bias and self-amplification phenomena arising from LLM context dependency, and explores directions for structural architectural solutions aimed at reducing users' cognitive load.
Why Do LLM Hallucinations Cascade? Technical Insights into Error Accumulation in Autoregressive Models and Process Reward Models
A logical examination of the mathematical vulnerabilities of autoregressive models, the paradoxes caused by evaluation metrics, and critical risks in autonomous AI agents, based on current research findings.