Random Learning
← The journal

The constellation

Atlas

Every topic I've learned, as a sky. A thread is drawn whenever a new day's topic connects back to an earlier one — the shape of how the learning compounds over time.

Conway's Law in the age of AI… How fairy tales got sanitized… Sogo shosha and why Buffett b… The 1,200-year Kyoto cherry b… The hippie trail and its revi… The history of pigments: colo… Do mineral hot springs actual… What sand really is, and why … How Japanese konbini conquere… What we still don't understan… The Sovereign Individual: why… What makes technical denim an… AlphaFold and Isomorphic Labs… Flock Safety's license-plate-… Why airline loyalty programs … The aesthetic-usability effec… Building a home GPU server in… How addiction rewires the bra… Why third places collapsed an… What hikers and Camino pilgri… The 2026 sweepstakes-casino c… AI data centers, the power gr… Why rural Japanese train line… What makes a data-visualizati… Running a kabushiki kaisha in… What companies that kept the … How car owners actually stop … AI adoption in 2026 is bifurc… What people actually get from… Co-living in 2026: investment… If you get carsick reading, t… Fake recruiter coding tests n… Beating phone autopilot is fr… What people in AI-heavy jobs … Bossware and AI productivity-… Version control for large gam… AI-generated papers and fabri… Where classical computer visi… Shared-presence web toys and … Aperiodic monotiles: the hat … Open-source text-to-speech in… OpenStreetMap versus Google M… Japanese railways as real est… Context rot: why long-context… Generative UI in 2026: AI int… The collapsing economics of h… Link rot and the digital dark… Full-body AI health scanners … Prompt injection attacks on A… What people say about Kagi se… Why 1-bit and dithered pixel … What makes a hand-drawn imagi… Structured data and llms.txt … Edo Japan's circular economy:… Apple silicon Mac mini and St… Designing nuclear-waste warni… Game preservation and console… Why anime animators are paid … The 2026 revival of personal … Driver's licenses in Apple an… Functional ultrasound neuroim… The who-is-online presence la… Offline-first hiking navigati… Rucking — weighted walking as… Double-entry ledgers as the s… Traditional Japanese nail-fre… Default alive bootstrapping v… Why Texans treat H-E-B as mor… Proof-of-personhood and provi… Why mugicha (barley tea) is w… Why almost everyone thinks th… Precious Plastic's crisis and… Porsche eFuels and synthetic … The three-generation rule and… Marine protected areas, 30x30… Why people quit Anki and spac… AI circular financing and whe… The em dash as an AI writing … Costco's anti-Amazon contrari… Workslop: AI busywork dumped … How demoscene coders fit 3D w… Gaussian splatting vs NeRF: w… Why Chinese AI labs publish d… The small web and handmade we… Bambu Lab firmware lockdown v… Why brands turn to junk after… Phantom islands: places mappe… AI forecasting bots vs human … The retro handheld emulation … Generative art after AI image… AI bug-hunting vs the slop fl… Local LLMs people actually ru… Organic Maps vs CoMaps: why t… Japanese train station depart… AI agents deleting production… Self-hosted Slack and Discord… Mac mini as an always-on head… Online malware-analysis sandb… QR-code phishing (quishing) a… How autonomous AI agents pay … AI image models vs. manga art… Everyone is a token arbitrage… AI compute governance: can ex… Tailwind fatigue and the post… Running image-generation mode… Sonic branding: how brands de… Taste and discernment as the … In-browser video editors with… The Billion Dollar PDF: docum… Comprehensible input and imme… Synthetic data vs human-label… Does EV battery recycling act… Robotaxis vs rideshare: who's… The craft and collecting of g… Is Sanskrit really the ideal … Could today's LLMs actually b… Building Telegram bots on ser… Running Claude Code on cheape… Files over apps: the plain-ma… Kongo Gumi and the world's ol… Do reusable rockets actually … Real-time voice agents: laten… AI 2040 and the fight over an… Giving AI agents their own sp… Why AI coding agents grep ins… Recording talking-head and sc… Where clinical psychedelic th… The context layer for AI agen… Safety nets for autonomous co… Models that fine-tune themsel… The silicon baton pass: Nvidi… Learning with AI vs offloadin… Ads inside ChatGPT: OpenAI's … Interactive castle atlases an… On-device speech-to-text in 2… MLX vs llama.cpp for local LL… DeepSeek V4 vs Qwen 3.6 for l… Astro vs Next.js in 2026 Local model as the orchestrat… Design tokens as the contract… AI-native video editors: prom… LLM-as-a-judge: trusting mode… Print-in-place compliant mech… Typology photography and docu… Where tokenization bites: the… What agent memory people actu… Terminal demos as build artif… Stop conditions: when a codin… What people wire up so agents… Copy-in components vs headles… Harness vs model: what actual… Buy-it-for-life: how people d… Whether the junior developer … Ensemble AI code review: does… Quantization quality cliffs: … Why server-sent events came b… Which AI video upscalers peop… What local-first sync actuall… Code ownership and blame when…

159 stars · 53 nights · the sky grows a little each day.