The Loneliness of Responsibility
Competence teaches people to depend on you. Sometimes it also teaches them to stop asking whether you are all right.
Engineering
9 articles in this category.
The Things I Was Wrong About: Thirty Years of Updating My Mind
The beliefs that helped me become useful, senior, and trusted were often the hardest to outgrow because success had made them feel like identity.
Why I Mentor: The Most Durable Work I Have Ever Built
Software disappears and titles change. The judgment, confidence, and generosity we help other people develop can continue into work we will never see.
A Camera Can Estimate a Pulse. Clinical Readiness Is a Different System.
Camera-based pulse estimation can be technically real without being clinically ready. The difference is a complete system of intended use, validation, failure handling, and bounded claims.
AI Slop Is What Happens When AI Averages Ambiguity
Most AI slop is unresolved workflow ambiguity made fluent. Better results come from engineering the objective, context, tools, stages, and checks around the model.
The Disappearing Boundary Between Design and Engineering
As products become behavior-heavy systems, design and engineering can no longer operate as a clean handoff chain. They are becoming a co-owned operating loop.
Why Modern Software Can't Be Designed on Screens
Screenshots are useful snapshots, but modern software behavior unfolds across state transitions, policy layers, and temporal feedback loops. If you design only screens, you miss the system users actually experience.
Reinforcement Learning from Human Feedback (RLHF): Taming the Ghost in the Machine
The definitive guide to the engineering breakthrough that turned raw text predictors into helpful assistants. We dive deep into the math of PPO, the psychology of Reward Modeling, and why 'The Waluigi Effect' keeps alignment researchers awake at night.
Unveiling the Hidden Curriculum in AI Training: Steering Models Toward Meaningful Learning
AI models often learn unintended shortcuts rather than meaningful patterns. This article explores five powerful countermeasures to address the hidden curriculum problem in AI training, ensuring models develop true understanding instead of exploiting statistical correlations.