Archive
Browse by topic or date.
RL / Training
3- Honey, I Shrunk the Circuits!
Low-rank circuit conditioning makes an existing dense-model capability recoverable as a compact causal mask.
- infinite — A Rubric-Driven Prioritized Replay to Maximise Continual Learning
A reinforcement-learning replay mechanism that uses rubric-based prioritization to optimize continual learning through evaluation and adaptive curriculum design.
- avatarl: Training Language Models from Scratch with Pure Reinforcement Learning
Replacing cross-entropy pretraining with a principled RL objective using expert-consensus rewards over active tokens.
Harnesses
2- How I Bring the Best Out of Claude Code — Part 2
Custom commands, multi-agent systems, and the protocols that made Claude Code actually useful.
- How I Bring the Best Out of Claude Code — Part 1
A comprehensive guide to effective Claude Code usage, context management, and building local multi-agent systems.
Culture
2- Theory of Telos
Serious delegation requires more than task competence. An agent must understand why the system under its care exists, what must be preserved, and where it is going.
- An Alchemical Outlook to Finding Human Gold
A framework for evaluating people through proof of interest, work, excellence, and exceptionalism.
2026
- Aug 27Theory of TelosCulture
- May 7Honey, I Shrunk the Circuits!RL / Training
2025
- Aug 20infinite — A Rubric-Driven Prioritized Replay to Maximise Continual LearningRL / Training
- Aug 9avatarl: Training Language Models from Scratch with Pure Reinforcement LearningRL / Training
- Jun 20How I Bring the Best Out of Claude Code — Part 2Harnesses
- Jun 15How I Bring the Best Out of Claude Code — Part 1Harnesses
- Jan 25An Alchemical Outlook to Finding Human GoldCulture