I want to feel this world the way a child does, and share my curiosity.
Filter by tag
-
The AI Inference Chip Spectrum — Seven Gradients from General GPU to Model-Etched Silicon
From NVIDIA GPUs to Taalas model-etched silicon, the 2026 AI inference chip landscape forms a seven-gradient spectrum from general to specialized. Each step right brings a 3-10× speedup at the cost of flexibility. The photonic compute path is blocked by the diffraction limit; the interconnect path has scaled.
-
CPU vs GPU Distributed Computing — Two Engineering Implementations of the Same BSP Theory
MapReduce's Reduce and NCCL's AllReduce both descend from MPI; Spark stage and PyTorch DDP step both descend from BSP. Same vocabulary, divergent engineering constraints — let's lay out fault tolerance, communication granularity, sync frequency, and programming model side by side.
-
A Decade of GPU Architecture Evolution and the Parallel Expansion of the CUDA Programming Model
From Pascal to Rubin, compute grew 2380× over a decade while CUDA core throughput grew only 10×; hardware complexity was absorbed by Tensor Cores, the programming model expanded from Thread to five layers, while torch.matmul did not change by a single line.
-
A Layer Map of the Modern Data Engineering Ecosystem
Organized by source / ingestion / processing / storage / query / serve into six layers, this note unpacks 16 mainstream open-source and commercial projects in the current data stack — their history, design, users, and typical usage.
-
Descartes · Hume · Nietzsche — Three Demolitions and Rebuildings
The three most important demolish-then-rebuild moves in three centuries — Descartes against scholasticism, Hume against Descartes, Nietzsche against morality. A relay, and the half each is misread on.
-
CoreWeave and Nebius — Two Divergent Paths for GPU Clouds
Side-by-side view of two listed Neocloud companies — CoreWeave, born of crypto mining, with Q1 2026 revenue of $2.1B and backlog of $99.4B; Nebius, restructured from Yandex, with ARR of $1.2B and a $17.4B Microsoft deal. Founding stories, teams, business, capital structure, and the essential differences from Hyperscalers.
-
Hong Kong Through Its Films and Songs — A Pilgrimage Map of Works, People, and Culture
Not going to Hong Kong to look for movies, but using movies to know Hong Kong again. From Chungking Express to Infernal Affairs, from Leslie Cheung to Beyond, from McDull to the Monster Building — a map of Hong Kong arranged by works, people, and culture.
-
The Mountains of Hong Kong — A Hiking Map of Trails, History, and City Memory
The 100-km MacLehose Trail, the Lion Rock that's been sung about for fifty years, the Spring Wind Pavilion at Pat Sin Leng that the fires can't burn down — 70% of Hong Kong is country parks, and every trail threads through institutional history, an old song, or a war. 10 routes × history × culture.
-
Li Bai, Daoism, and Nietzsche's Dionysian Spirit
Nietzsche's Dionysus, Zhuangzi's "I have lost myself," Li Bai's solitary drinking — three postures of "self-dissolution" meet at one rhyme and depart on another.
-
Where Ideas Were Born — A Pilgrimage Map of European Philosophy
The Academy in Athens, an army tent on the Danube, a tower in Bordeaux, an Ulm stove-room, the Sils-Maria boulder, a Black Forest hut — putting twelve philosophical classics back into the rooms, cities, and landscapes where they were actually written.
-
Wall Street's Rear-View Mirror
Institutions are triple-constrained "wait for earnings before deploying capital"; the AI capex relay race GPU → power → memory → networking → CPU lags 12–24 months — this is retail's window.
-
China's Hard-Tech Dividend Over the Next Five Years
Starting from systemic risks — demographics, property, local-government debt — projecting the next five years of policy dividends and sacrificed directions, and mapping them to A-shares / Hong Kong / hedges.
-
The Three-Way Rewrite of AI and Social Media
Generative AI is simultaneously rewriting the three layers of social media — consumption, production, and distribution. A sourced platform map showing each player's stance and three possible future scenarios.
-
Cracks in NVIDIA's Moat
NVDA moat contrarian view — architectural convergence lets CUDA be bypassed, training is fortress and inference is the soft spot, and the true inflection signal is non-GAAP gross margin, not market share.
-
Apple's Gatekeeper Distribution Power
Apple uses gatekeeper distribution power to run multi-vendor AI sourcing, netting +$18B/yr inbound; Mac becomes the household compute hub, and smart glasses 2027 stake out the next entry point.
-
Three Telescopings of the US Military Machine
2003 Iraq → 2025 Hammer → 2026 Epic Fury — same command chain, four focal lengths contracting across authorization, coalition, civil-military, and ground forces.
-
The World of the Volatility Trader
A complete guide to volatility trading — from RV vs. IV, VRP, skew, and the Greeks through the VIX complex, threading short-vol / long-vol / tail-risk strategies across eight chapters.
-
Foreign Exchange — A Primer
An FX primer — the \$7.5T/day OTC behemoth, from currency pairs and spreads to the four strategies (carry, trend, mean-reversion, event) at one go.
-
The Kelly Criterion and Position Sizing
Picking direction only gives you expected value; picking size gives you compounding — Kelly, vol targeting, mean-variance, risk parity, CPPI, and 12 sizing formulas.
-
The Eight Schools of Quant Trading
A field guide to the eight schools of quant — HFT, StatArb, CTA, macro, factor, ML, event-driven, crypto — 11 chapters and 40+ leading firms cross-referenced.
-
Methods of Company Valuation
A valuation primer — five families of methods (DCF, relative, DDM, asset-based, SOTP), plus WACC, terminal value, startup valuation, and 15 classic pitfalls.
-
US Earnings Report Reading Guide
A complete guide to reading US earnings — from the four financial statements to DAU / FCF / Rule of 40, 11 chapters and 80+ terms, each cross-walked to Reddit's actual numbers.
-
China's Industrial Hardware Goes Domestic
From DJI to Bambu to Hypershell — the systematic Chinese playbook of compressing industrial-grade equipment by 2-3 orders of magnitude, moving military and lab hardware into the living room.
-
LLM Inference Walkthrough — Tensor Shapes and Core Formulas Across the Whole Pipeline
Using Llama 3 8B as the reference, walk the entire inference path from token ID → embedding → Transformer → sampling, writing out every shape transition and every key formula.
No matching articles