https://careychou.tech/ Essays on human-centered AI: codifying expertise, machine memory, and the boundaries between human judgment and model output. en-us Fri, 16 Oct 2026 23:17:34 +0000 https://careychou.tech/writing/decision-reasoning-models https://careychou.tech/writing/decision-reasoning-models Wed, 14 Oct 2026 12:00:00 +0000 Carey Chou We tried LLM agents for everyday decisions and kept getting different answers to the same question. A decision reasoning model keeps the language model's reading of free-form text and returns one calibrated, repeatable answer, shown on three decisions: a shopper's order of preference, a Yes or No on a conference, and a travel budget. Earning the Right to Act Decision models https://careychou.tech/writing/personalized-memory-in-personalization-experience https://careychou.tech/writing/personalized-memory-in-personalization-experience Sun, 20 Sep 2026 12:00:00 +0000 Carey Chou Even a huge deep-learning model is still a global model, an average over millions of shoppers. It learns the stable structure and misses the moment one person's taste turns. Personalized memory sits on top of the global model as a per-customer state that carries each shopper's own baseline, a residual on top of the global model, floored so a quiet session never collapses back to the average, with a Kalman-style shrink factor set from behavior. A State Space / Kalman view of personalization. Memory That Outlives the Session Personalization https://careychou.tech/writing/cognitive-orchestration https://careychou.tech/writing/cognitive-orchestration Sat, 19 Sep 2026 12:00:00 +0000 Carey Chou Cognitive orchestration is usually described as a control plane over many agents. That is the question agentic orchestration already answered. Being cognitive means knowing which decisions are settled, which are genuinely contested, and which have no precedent at all. That is measured at runtime from human disagreement, rather than drawn on a flowchart at design time. Codifying Expertise Human–AI collaboration https://careychou.tech/writing/protein-mania https://careychou.tech/writing/protein-mania Sat, 19 Sep 2026 12:00:00 +0000 Carey Chou Why the protein boom is happening, which forms are really rising, and the sugar hiding behind the claim. Drag a lens across the evidence. Human–AI Design & Art Data journalism https://careychou.tech/writing/codifying-implicit-decisions https://careychou.tech/writing/codifying-implicit-decisions Sat, 12 Sep 2026 12:00:00 +0000 Carey Chou Recovering the implicit tradeoff reasoning behind expert decisions, so an agent reasons instead of imitates. Codifying Expertise Reasoning https://careychou.tech/writing/collaborative-episodic-memory https://careychou.tech/writing/collaborative-episodic-memory Sat, 12 Sep 2026 12:00:00 +0000 Carey Chou Human–AI thinking happens turn by turn, then vanishes. Episodic memory that is shared across your own sessions and across people: reconciliation, convergence, distillation. Memory That Outlives the Session Memory https://careychou.tech/writing/glp-1-ripple https://careychou.tech/writing/glp-1-ripple Sat, 12 Sep 2026 12:00:00 +0000 Carey Chou A liquid-glass signal study of an appetite shock moving through the US economy. Human–AI Design & Art Signal study https://careychou.tech/writing/how-do-you-know-an-ai-answer-is-good-enough-to-act-on https://careychou.tech/writing/how-do-you-know-an-ai-answer-is-good-enough-to-act-on Sun, 02 Aug 2026 12:00:00 +0000 Carey Chou An analyst asks an agent a question about a product category and gets back a clean answer: here are the strongest item affinities, here is what is different abo… Earning the Right to Act Evaluation https://careychou.tech/writing/ai-cognitive-quadrant https://careychou.tech/writing/ai-cognitive-quadrant Mon, 13 Jul 2026 12:00:00 +0000 Carey Chou Not all AI is the same kind of AI. Plot two axes, how much coordination a task needs and how much the right answer depends on interpretation, and four quadrants fall out. Each has its own failure modes, governance needs, and demands on the humans around it. An interactive map for deciding what you are actually building. https://careychou.tech/writing/from-digital-twin-to-phygital-twin-codifying-process-knowledge-into-agentic-robotic-process https://careychou.tech/writing/from-digital-twin-to-phygital-twin-codifying-process-knowledge-into-agentic-robotic-process Sat, 27 Jun 2026 12:00:00 +0000 Carey Chou RPA is brilliant once you already know the steps. Process mining is brilliant once the work already lives in clean system logs. Both quietly assume someone has written the process down, and in most companies nobody has. On capturing the scattered, tacit workflow that lives across a dozen apps and in people's heads, and turning it into an agent that can run it. Codifying Expertise Digital Twins https://careychou.tech/writing/from-super-teammate-to-metatwin-toward-self-building-digital-twins-of-expertise https://careychou.tech/writing/from-super-teammate-to-metatwin-toward-self-building-digital-twins-of-expertise Sun, 31 May 2026 12:00:00 +0000 Carey Chou An open-source skill pack: withsuper-analyst, that could learn a dataset and produce a rules playbook. super-optimizer, which tunes a playbook against historica… Codifying Expertise Digital Twins https://careychou.tech/writing/test-time-training-ttt-discover https://careychou.tech/writing/test-time-training-ttt-discover Mon, 02 Feb 2026 12:00:00 +0000 Carey Chou Most of our experience with Large Language Models (LLMs) involves a frozen brain. You send a prompt, and the model, with weights fixed in time, predicts the n… Memory That Outlives the Session Adaptation https://careychou.tech/writing/the-new-human-roles-of-ai-why-boundaries-not-models-define-the-future-of-work https://careychou.tech/writing/the-new-human-roles-of-ai-why-boundaries-not-models-define-the-future-of-work Tue, 16 Dec 2025 12:00:00 +0000 Carey Chou The Wall Street Journal recently published an article titled “Four New Jobs That May Be in Our AI Future.” https://www.wsj.com/tech/ai/new-jobs-ai-future-fefd4b… Where Humans Still Sit Future of Work https://careychou.tech/writing/nested-learning-for-recommender-systems-bringing-fast-and-slow-learning-to-personalization https://careychou.tech/writing/nested-learning-for-recommender-systems-bringing-fast-and-slow-learning-to-personalization Fri, 21 Nov 2025 12:00:00 +0000 Carey Chou Modern recommendation systems face a paradox: Memory That Outlives the Session Personalization https://careychou.tech/writing/llm-driven-probabilistic-sampling-for-human-guided-optimization https://careychou.tech/writing/llm-driven-probabilistic-sampling-for-human-guided-optimization Thu, 14 Aug 2025 12:00:00 +0000 Carey Chou LLM-driven probabilistic sampling for human-guided optimization. Searching With a Human in the Loop Optimization https://careychou.tech/writing/evolving-llm-prompts-to-generate-customer-shopping-narratives-ai-guided-evolution-with-cohesive https://careychou.tech/writing/evolving-llm-prompts-to-generate-customer-shopping-narratives-ai-guided-evolution-with-cohesive Sat, 19 Jul 2025 12:00:00 +0000 Carey Chou In a previous article, we explored how combining large language models (LLMs) with reinforcement learning can create smarter business rules. Building on the Alp… Searching With a Human in the Loop Optimization https://careychou.tech/writing/sprt-and-mixture-sprt https://careychou.tech/writing/sprt-and-mixture-sprt Sat, 15 Jun 2024 12:00:00 +0000 Carey Chou Peeking at an A/B test turns a 5% false-positive rate into 25%. Sequential testing lets the evidence decide when to stop. Earning the Right to Act Experimentation