Learning Notes
AI tools, certifications, books and small experiments, written down while learning. 45 posts.
Your Video Memory Shouldn't Live on Someone Else's Cloud
Field notes from moving an entire 'watch a video, get a searchable multimodal memory' pipeline onto my own machine — capture, embedding, storage, search. One evening, three pitfalls, zero cloud services.
Is Qwen3.8 27B on Your Own Machine Right for You?
Notes after watching three hardware reviews from two YouTubers. A payback calculation, the three real costs of running a large model at home, and how to tell whether this is for you. A thinking exercise about tools, not buying advice.
When Exit Codes Start Lying: An Alarm That Rang for Seven Days and Nobody Understood It
Silent failure in automated systems: a tool that answered correctly then reported failure, and an alarm that fired daily under the wrong name. The full reasoning trail from one debugging session, plus three checks you can run on your own system today.
When the Genius Walks Away: Grok Catches Up, Google Loses Its Legends, Meta Returns to Open Source
Notes after listening to TechWave EP150. Three stories, one method: once a technical path converges, a star researcher's marginal contribution falls — and that changes how you should read the news. Educational content, not investment advice; no stock recommendations or price targets.
Why Hardware Is Hard: The Three Most Valuable Lines from One Conversation
No field updates. An over-eager supplier is a warning sign. Money is recoverable, time is not. Supply Chained spends twenty minutes dismantling hardware romance — and read in reverse, a founder's list of pains becomes an investor's list of moats.
A Thousand Sails Pass the Sunken Boat: A Freelance Engineer on the AI Content Factory
A Chinese freelance engineer's YouTube channel takes a rare break from hardware to talk about its own industry: collapsing creator traffic. Short-form video is draining watch time while AI content factories let one person run ten channels; the platform's ability to detect AI content has an expiry date and is already wrongly banning humans. His conclusion: you aren't competing with AI, you're competing with people who wield it. His plan: use the three-to-five years in which being human is still provable. My extension: content can be mass-produced, liveness cannot; faces expire, records don't.
herdr: if you run a herd of AI agents, sooner or later you must answer "who's working, who's stuck?"
What herdr is, how to install and use it, and why we ended up borrowing its two best design ideas instead of wiring the tool into production.
Generative AI Certification — Complete Exam Cheat Sheet: 71 Key Terms + Chapter-by-Chapter Study Points
A last-hours review tool for Taiwan's III Generative AI certification: a full 71-term key-terms reference in five groups, plus the high-frequency points and 'see X, pick Y' distinctions across all four areas — fundamentals, prompting, applied skills, ethics and law. A companion to my prep-method post. Educational sharing, not a question bank.
Two AI Certs in Two Months: How I Used AI to Compress an Exam Down to What Actually Gets Tested
AI-900 in June, the III Generative AI certification in July. Here's my prep method: real past papers first, use AI to squeeze hundreds of practice questions into a single 'only-what-I-haven't-seen' review card, and judge on the spot which chapters to skip. Plus the topics this exam actually tests. A learning note.
How to Install Claude Code / Codex / Grok / Antigravity: A Beginner's Step-by-Step (Mac + Windows)
Never touched a terminal? You can still follow this. Step-by-step install for four desktop AI command-line tools (Claude Code, Codex, Grok Build, Antigravity agy) on Mac and Windows: how to open the terminal, which line to paste, how to log in and use it, and the mistakes beginners hit most. Pick the one you already pay for. Tutorial.
xAI Just Open-Sourced All of Grok Build: A Dissection Class on a Production AI Coding Agent
On 2026-07-15 xAI open-sourced Grok Build, its terminal AI coding agent — 840k lines of Rust, out in the open. What it is, why it's worth reading, how to install and use it, and the one thing we learned reading the source. A learning note; educational sharing.
A One-Person Engineering Department: Running Codex and Grok from Claude Code
I was paying for three AI subscriptions and using one. So I turned Claude Code into a chief engineer and made Codex and Grok my implementation lanes — five-part specs, fail-loud discipline, and the most expensive lesson of all: an AI's completion report is not completion.
The Mirror Protocol: I Let an AI Read All 6,139 of My Commands. It Described Someone I Didn't Recognize.
A prompt circulating in the AI community: have an agent excavate every session log on your machine and answer 'who are you' using your own timestamps. Six phases, fully local, full prompt included for copy-paste. This is my run — including the part where my own records called my bluff.
Saying Goodbye to Fable 5: I Asked a Retiring AI to Leave Behind How It Thinks
The strongest model on my AI team retired today. In our last two days, I didn't ask it to do more work — I asked it to write down its thinking process. It left a map of my life's key problems, ten crisis playbooks, an honest gap map, and a letter. This is the record of those two days, and what I learned.
When the Strongest GPT-5.6 Meets a Suddenly-Genius Grok 4.5: I Used Claude to Turn Three AIs Into One Team
How a one-person AI company actually works: let Claude be the head coach that plans the hard problem and nails the interface, hand the execution to OpenAI's GPT-5.6 and a suddenly-sharper xAI Grok 4.5 running in parallel, then verify against reality. Three days of multi-vendor AI teamwork, shared as method and reflection.
AI-900, Part 2: How I Used AI to Turn a Question Bank Into the Fastest Review System
The parts the first post skipped: where to get practice questions (and an honest caveat), how to use AI to build a personal question bank that auto-quizzes, grades, and logs your misses, how to get AI to teach you the 'concept tree' instead of memorizing service names, and how to use AI to make review tables that cover the most concepts in the least time. A learning-method piece, not official courseware.
A Subscription Is Not a Discount: Wiring Grok 4.5 into Claude Code, and One Security Self-Audit
An overlooked billing trap: a subscription buys you quota inside the official app, while most third-party tools reach the model through the metered API. This is my attempt to run Grok 4.5 inside Claude Code and still bill it to my subscription, the grey-zone path I found, and the four-question security audit I ran on a third-party tool before letting it hold my account token. Tech notes and hands-on lessons, not a how-to.
After the OAuth Wall: Three Ways to Run Grok in Claude Code, and What I Actually Learned
The OAuth cliffhanger from last time, resolved. A diagnostic mistake — reading a personal subscription as a team account — sent me the long way round; once cleared, there turned out to be three ways to run Grok in (or next to) Claude Code, each with a price. But the real takeaway isn't 'I found the method' — it's rethinking how to allocate tasks: unattended automation rides the stable billing path, and only human-in-the-loop work rides the cheap subscription path. First draft written by Grok 4.5 itself.
Passing AI-900 in Two Weeks: A Quant Investor's Study Method and Key Concepts
How I passed Microsoft Azure AI Fundamentals (AI-900) in two weeks using block-based sprints, a one-page service cheat sheet, and spaced retrieval testing — plus a roundup of the most-tested, easiest-to-confuse concepts. A learning-method piece, not official courseware.
I Spent Hundreds of Hours Proving Moving-Average Crossovers Dont Work
A negative-result study showing that after 11,151 U.S. stocks including delisted names, a 31.7% survival rate, price-band controls, and trading costs, moving-average crossovers have no tradable edge.
Why We Freeze an Expiry Date onto Every Prediction
The design philosophy behind a prediction-scoring system: pre-register the criteria, pair naive hit rate with a benchmark, refuse to conclude on thin samples, and handle conditional forecasts honestly. It is really about how not to fool yourself.