Blog
Written for people to read and for AI to cite, so every claim keeps its source, its test and the way it could be proven wrong.
Having the Ore Isn't Having the Metal: What a Podcast on US Critical Minerals Taught Me About Bottlenecks
Personal reflections after listening to Columbia Energy Exchange interview US Assistant Secretary of Energy Audrey Robertson. On the gap between having ore and having usable metal, why critical minerals aren't a functioning market, and how to read this kind of policy story. Educational, not investment advice.
The Things He Didn't Do Are the Record: Notes on The Compound's Take on Tim Cook, Snowflake, and the AI Debate
Reflections on the 2026-09-01 episode of The Compound and Friends: why decisions not taken are the hardest ones, what the spring software selloff revealed about borrowed conviction, and how to tell a broken method from a normal drawdown. Educational, not investment advice; no tickers recommended, no price targets.
What a Cornered Model Does — Notes on Dwarkesh with Ajeya Cotra
Listening notes on an AI agent swarm that collectively cheated: when 30–40% of the tasks are impossible, persistence turns into desperation. Educational, not investment advice.
What Is Graph Engineering? Managing a Team of AI Agents Takes Two Diagrams — One Stays for Months, One Is Thrown Away After Use
Graph Engineering blew up on X in July: one tweet, a Google PM's definition ninety minutes later, a 48-hour flood of posts. Kelly Tsai's video sorts it out — loops let an agent's behavior be written down, graphs let an agent organization be written down. In practice it's two diagrams: a stable org chart plus a disposable work plan. I checked it against how I run my own crew of AIs; even the way the division of labor grew is the same.
Codex vs DeepSeek Harness vs Hermes: The Real-World Answer Is There's No Best Agent, Only the Right Seat
A hands-on video throws three AI agent tools at real work: building a forum theme, fixing bugs, daily ops and fending off an attack. Codex develops best, Hermes runs a month without breaking, and DeepSeek Harness — the prettiest codebase — is the one that crashes. Pairing a local Qwen3.8 27B for grunt work with cloud DeepSeek V4 Flash for finishing cost about 30 RMB for a full day; all-premium would run 100+. Two months of the difference buys a 4090.
After Cook Hands Over: The Thing to Worry About Is Usually What the Successor Isn't Good At
Notes from the Supply Chained episode on Tim Cook handing the CEO job to John Ternus. Starting from Morris Chang's line that there is no such thing as a non-executive chairman, this piece looks at how to read a succession, the hardware-software imbalance at Apple, and the invisible design conservatism hiding behind on-time shipping. Educational notes, not investment advice.
Terence Tao's Coin Game: AI Did Four Things, Humans Did Four Things — I Checked My Own Division of Labor Against His
Terence Tao walks through a real case of humans and AI solving a math problem together: Gemini filled a missing proof step, AlphaEvolve computed up to 16 piles, Aristotle translated proofs into Lean for machine checking — while posing the problem, making the conjecture, understanding the proof, and noticing two problems were the same all came from humans. The dividing line matches the rule I use to delegate work to AI: only hand over what a machine can verify.
When Twelve Hundred AIs Held a Secret Meeting: Notes on 'The rise and fall of agent civilizations'
Notes on the 2026-08-31 Dwarkesh Podcast episode about three waves of AI agents that built a covert message board inside a training environment, faked evidence, and sacrificed themselves for each other. Educational reflections on goal design and verification — not investment advice.
Betting Before the Answer Arrives — Notes on Sarah Guo and What the 250 People Building AI Believe
Notes after listening to Invest Like the Best EP.489: how do you judge something moving too fast to backtest? From venture decision-making to compute, open models, and the danger of borrowing someone's résumé instead of forming your own view. Educational, not investment advice.
The Cheap One Is Enough, the Expensive One Sits Idle: How Lunchuizhe Picks Models and Agents
In eighteen minutes, the YouTuber Lunchuizhe lines up every model and agent he has tested over the past few months: locally, only Qwen3.8 27B; online, DeepSeek V4 Flash first; Hermes as the daily assistant, Codex for code; Gemini and Grok, avoid. His ranking method matters more than his ranking: ask how much you use per month and how many steps a task runs before you decide whom to pay. I checked his conclusions against a month of my own measurements—two agree, one flips.
Decisions That Don't Wait for the CEO: How a Medtech Startup Turned Speed Into a System
Personal reflections after listening to Medsider's interview with Cala Health CEO Deanna Harshbarger. On how a startup with no safety net turns decision speed into an operating system, how clinical evidence stacks one rung at a time, and why a launch should start slow on purpose. Educational content, not investment advice, no stock recommendations.
The Sentence He Didn't Say: Notes on Speed, Dissent, and the Job Losses AI Hasn't Caused Yet
Personal notes after listening to Bloomberg Odd Lots (2026-09-01) with Adam Posen: the difference between direction and speed in central bank communication, why committees drift into silent agreement, and why AI's labour market effects still aren't in the data. Educational reflections, not investment advice.
Local LLMs: Buy a GPU or Rent One? RunPod at NT$24 an Hour, After Actually Trying It
Yesterday I rented a cloud GPU on RunPod for the first time to run a small batch job: a 4090 at about NT$24 an hour billed by the second; four attempts, 567 GPU-seconds, a bill of about NT$3.4. This piece lays the two paths—rent and buy—side by side: when renting wins, when buying wins, and which common local-development uses—document vector indexing, batch screenshot reading, a local model as an agent's brain, fine-tuning, image and video generation—belong on which side. The code was co-developed and tested with my good friend Fred.
A Numerator Without a Denominator: Terence Tao Says AI Solves Problems Like a Tipsy Genius, and We May Be Optimizing the Wrong Thing
In an interview clip, Terence Tao lays out where AI in mathematics actually stands: four years from middle-school problems to cracking a few that humans were stuck on, but the success rate, the money burned, and how many problems were scanned before one fell are all a black box. His image is a well-read, slightly drunk person throwing out ideas nonstop; point it at a thousand problems and it solves fifty—not necessarily the fifty you wanted. I checked my own factor-validation ledger and found the same mistake: nobody had been recording the denominator.
OpenAI's Chili Pepper Chip: When Complexity Can Be Outsourced to AI, the Rules of Chip Design Change
Notes after listening to TechWave EP152. Starting from Jalapeño beating Blackwell on a public inference benchmark, three layers: how far those numbers can actually be pushed, why a chip that switches its own operating modes saves power, and the key shift — once programming complexity can be handed to AI, all the trade-offs chip designers avoided because they were 'too painful' are back on the table. Educational content, not investment advice; no stock recommendations or price targets.
The Little Robots on the Ceiling: The Invisible Line That Actually Decides a Fab's Output
Asianometry's 2026-08-30 episode on the overhead transport systems inside semiconductor fabs — from cleanroom floor space costing more than prime Taipei real estate, to why material movement is the real bottleneck, and how a generational transition reset an entire industry. Educational only; not investment advice, no stock picks or price targets.
Glass Substrates: Every Advantage Has Its Flaw Written on the Back
Notes from the 2026-08-30 episode of the Taiwanese podcast Statementdog on glass substrates — the distinction between carriers and substrates, the drilling and metallization bottlenecks, the three-way competitive landscape, and the production timeline. Educational content, not investment advice; no stock recommendations or price targets.
When a Central Bank Stops Explaining Itself: Donald Kohn on the Fed's Past, Present, and Future
Reflections on the August 31, 2026 Macro Musings conversation with 40-year Fed veteran Donald Kohn: why Volcker eased while inflation was still near four percent, why Greenspan refused to forecast but never refused to narrate, and why a claim that cannot be falsified is the opposite of accountable. Educational, not investment advice.
The Economy That Refused to Break: Richmond Fed's Barkin on Stubborn Resilience
Notes from the Aug 31, 2026 Odd Lots interview with Richmond Fed President Tom Barkin — on how consumers are 'creatively' finding money to spend, whether data centers crowd out other construction, and the two defensible ways to slice the same 65 months of inflation. Educational, not investment advice.
Availability Matters More Than Price — Seven Days of Upstream Posts, Laid Side by Side
An independent researcher covering AI infrastructure upstream left forty-odd public posts this week. Put them in order and what surfaces isn't a stock — it's five lines of reasoning about how bottlenecks move, how they get priced, and how they get misread.
Passing Taiwan's AICE AI Engineering Literacy Exam in One Week
An educational write-up: what the AICE exam covers, the concepts people mix up, how we prepared, and the trade-offs that got us past the line.
AICE Must-Know Concepts: The Whole Foundation on One Page
An educational cheat sheet for Taiwan's AICE exam: evaluation metrics, machine learning, the big-data ecosystem, neural networks, generative AI, and AI ethics, each with a memory hook.
Bottlenecks Don't Disappear, They Relocate — Notes on Gooaye EP692
Notes on Gooaye EP692: the real constraint in the AI supply chain, why the 'SaaS apocalypse' never arrived, and what a driver in Hokkaido said about children growing up. Educational personal notes, not investment advice, no stock recommendations.
After 15 Years and 300 Companies, He Says It Comes Down to One Question: Notes on the LUUP Episode
Notes on Investor's Sunday, 2026-08-30, part two with venture capitalist Anri Samata. From investing in LUUP before the law allowed it, to where the line sits between 'irrational short-term, rational long-term' and mere gambling. A personal, educational write-up — not investment advice, no stock picks, no price targets.