Wiki Index

Summaries

dont-sign-in-with-google — the_smart_ape on the cascading-SaaS-loss case study, Truffle Security's $12-domain-takeover flaw, multilogin refresh-token replay, consent phishing, and the recovery-damage rule for when SSO is actually fine
appsec-guide-fuzzing-handbook — Trail of Bits Testing Handbook fuzzing chapter: terminology, the mutation-based evolutionary algorithm with pseudocode, bug-class taxonomy beyond crashes
gosentry-go-fuzzing-fork — Trail of Bits forks the Go toolchain so testing.F harnesses run through a LibAFL+Nautilus engine — struct/grammar inputs, race/leak/overflow detection, early consensus bugs in Optimism/Revm
mintlify-chromafs — How Mintlify built a virtual filesystem (ChromaFs) to replace sandboxes and RAG for their docs assistant
working-on-products-people-hate — Why individual engineers have limited control over product perception, and why that's okay
adk-go — Google's Agent Development Kit for Go 1.0: tracing, plugins, HITL, YAML config, A2A protocol
13kb-tcp-slow-start — Why downloads stall at 13 kB: TCP slow start and the initial congestion window
basb-overview — Tiago Forte's Building a Second Brain methodology: CODE framework, PARA organization, progressive summarization
subprime-ai-crisis — Zitron's case that AI demand is entirely subsidy-driven, with a predicted collapse sequence and warning signs
signal-push-pull-algorithm — How signals work: push-pull reactivity with automatic dependency tracking, built from scratch in TypeScript
altman-new-yorker-profile — Farrow & Marantz's 2026 investigative profile of Sam Altman: the Ilya Memos, Amodei documents, safety erosion, Gulf ambitions
xray-tutorial — Step-by-step VLESS proxy setup with Xray-core: server config, TLS certs, clients on Android/Windows/Linux
titit-local-ai — Neward's case for local open-source LLMs: cloud fragility, unsustainable economics, privacy, and learning by building
log-distributed-llms — Why multi-agent LLM coding is a distributed consensus problem subject to FLP and Byzantine impossibility results
git-magic-files — Catalog of committed files that control git behavior: .gitignore, .gitattributes, .mailmap, .git-blame-ignore-revs, and more
macos-tcp-time-bomb — XNU kernel's 32-bit tcp_now overflows after 49.7 days, freezing TIME_WAIT and killing TCP
cult-of-vibe-coding — Bram Cohen on why refusing to read AI-generated code is dogfooding gone cult, with his audit-discuss-execute workflow
hold-on-to-your-hardware — AI data centers are eating global DRAM/NAND/HDD supply, consumer hardware becoming scarce and expensive
crqc-timeline — Valsorda's April 2026 reassessment: two papers collapsed quantum timelines, Google sets 2029 migration deadline
llm-enshittification — Gentoo developer's comprehensive anti-LLM case: data theft, infrastructure harm, FLOSS fracturing, copywashing, labor displacement
building-syntaqlite-ai — 8 years of wanting, 3 months with AI: vibe coding failure, disciplined rewrite, the addiction loop, and why design can't be delegated
claude-emotion-concepts — Anthropic finds 171 emotion vectors in Claude that causally influence behavior, some invisibly to output observation
bitcoin-quantum-computing — Narula's risk framework for Bitcoin's PQ migration: 5% break probability by 2030, uniquely hard due to decentralized governance
win32-stable-abi — Glibc 2.36 broke Linux games by dropping DT_HASH; Win32 via Wine is the only stable ABI on Linux
full-text-search-indexeddb — Building full-text search on IndexedDB with multiEntry indexes and rarest-term probing
zsh-glob-qualifiers — Zsh's (om[1]) glob qualifier trick: pick the newest file inline, no pipes needed
linux-kernel-pgit — 1.4M kernel commits in PostgreSQL via pgit: contributor bottlenecks, GPU driver churn, and SQL-queryable git history
lsm-trees-nosql — LSM tree architecture: MemTable, WAL, SSTables, Bloom filters, compaction strategies, and failure modes
open-source-security-astral — Astral's full supply chain security posture: CI/CD lockdown, Trusted Publishing, Sigstore, dependency cooldowns
ingodb — Self-morphing LSM document store in Rust: reactive indexing from query patterns, Liquid AST queries for AI agents
ai-great-leap-forward — Corporate AI mandates as Great Leap Forward: backyard furnaces, fabricated metrics, anti-distillation, institutional knowledge loss
filesystem-is-graph-database — Rumproarious on why a markdown vault with wikilinks is already a graph database, and the PARA extensions that make it scale to 52k files
happ-subscription-converter — PHP scripts that emulate the Happ VPN client to extract VLESS lists and bypass HWID-based device limits
build-your-own-openclaw — 18-step Python tutorial for building a Claude Code-style coding agent from scratch: chat loop → tools → skills → events → multi-agent → production
no-silver-bullet-llms — Bennett applies Brooks' No Silver Bullet to LLM coding: math caps the upside, DORA/CircleCI show rising instability, Cloudflare Next.js rebuild as case study
mcp-vs-skills — Mohl on why MCP is for service access and Skills are for knowledge: portability, auth, context bloat, and the hybrid pattern
eth-geometric-quantum-gates — ETH Zurich achieves 99.91% fidelity swap gates on 17,000 neutral-atom qubit pairs using noise-resilient geometric phases
clean-code-coding-agents — Code structure matters more with agents: context windows are finite, messy code wastes tokens on orientation
ursa-kafka-storage — StreamNative's diskless Kafka storage engine: writes directly to S3 as Iceberg/Delta tables, 10x cheaper, protocol-vs-codebase meta-argument
vamp-ai-frontend — Why framework implicitness (VDOM, hooks, reconciliation) is the enemy of AI-generated code, and a minimal explicit pattern that replaces it
peril-of-laziness-lost — Cantrill on why LLMs lack the virtue of laziness: human time constraints drive simplicity, unchecked generation produces bloat
obsidian-ai-complete-guide — Huashu's production implementation of the LLM Wiki pattern: three-layer architecture, SCHEMA.md, seven workflows, and why AI is a compiler not retriever
go-transaction-linter — How léon h built transactioncheck after shipping a transaction leak bug to production
ai-assisted-workflow — Barbero's seven-step workflow where planning happens before code and AI stress-tests thinking
retrofitting-jit-c-interpreters — How yk uses meta-tracing to automatically derive JIT compilers from C interpreters
win32-weird-shaped-windows — How Win32 lets you create non-rectangular windows, and why that era mostly disappeared
database-index-gotchas — Common index pitfalls (composite order, functions) and lesser-known types (partial, covering, functional)
syscall-binary-rewriting — Binary rewriting technique that replaces syscall instructions with INT3 traps for process isolation
levenshtein-trie — Fuzzy dictionary search 300× faster by sharing Levenshtein table rows across trie prefixes
paper-computer — Using physical media with AI digitization to escape screen dependency
go-step-sequences — State machine pattern for multi-step tasks where each step returns the next
average-is-all-you-need — Rawquery's thesis that LLMs make "average" output cheap, with an agent-driven email-attribution walkthrough
simplified-model-of-fil-c — Cawley's step-by-step walkthrough of Fil-C's shadow capabilities, invisible bytes, and GC-backed memory safety for C/C++
features-to-steal-from-npmx — Nesbitt's catalog of 17 npmx.dev features as a registry-design spec, plus the competitive-pressure effect on npmjs.com
i-dont-want-your-prs — Ciężarkiewicz's post-LLM rethink of open-source collaboration; prompts, feedback, and forks over PRs
bubblewrap-dev-env — Practical Bubblewrap isolation for LLM agents with Nix and tmux integration
safe-gc — Fitzgerald's zero-unsafe Rust GC: indexing-not-deref as the whole safety story, mark-and-sweep because copying wouldn't borrow-check
forge — Nesbitt's unified Go CLI and library over GitHub/GitLab/Gitea/Forgejo/Bitbucket, auto-detected from git remote, for humans and coding agents
type-systems-vocabulary — Campbell's pet-peeve note: static/dynamic and strong/weak are orthogonal, "strict typing" is an amateur term
message-passing-shared-mutable-state — Tu et al's 2019 Go bug study confirms Lee's 2006 prediction: 58% of blocking bugs were channel bugs, channels are concurrent queues
t-context-go-testing — Hall on when to use Go 1.24's T.Context() (default: yes; exception: shared resources outliving a single test)
dosbox-detection-callback-opcode — Detecting DOSBox by triggering its custom-invented FE 38 00 00 callback opcode and watching for #UD
samsch-jwt-auth-gist — Long-circulating gist arguing JWTs are wrong for sessions; use cookie sessions, and PASETO for the rare short-lived-token cases
sdl3-dos-port — SDL3 gains native DOS support via DJGPP: VGA/VESA, Sound Blaster, INT-driven input, cooperative threading via setjmp/longjmp (PR #15377, merged 2026-04-23)
deepseek-v4-roleplay-instruct — Reverse-engineered first-user-message markers that toggle DeepSeek-V4's <think> style between immersive monologue and pure analysis
long-lived-keys — Ludwig's case for ephemeral credentials over rotation: EC2 Instance Connect, PyPI Trusted Publishers, SSO; concentrate the toil for what's left
terminal-gpu-battery-cost — Why Ghostty et al. drain MacBook battery 30× more than a browser, and the iTerm2 / Terminal.app workarounds for Claude Code sessions
linux-7-postgres-regression — Linux 7.0 removed PREEMPT_NONE; PostgreSQL throughput halved on 96-vCPU Graviton4 because spinlock holders now get preempted mid–page-fault
zig-functional-programmers — A 10-year Haskeller's case for Zig: comptime gets you most of Haskell's type-system programming without garbage collection
clippy-stricter-config — Schwartz's curated Clippy lints to recover "if it compiles, it works" — panics, dropped futures, async footguns, agent guardrails
ceo-ai-psychosis — HandyAI on Garry Tan's gstack, tokenmaxxing leaderboards, and the structural sycophancy that makes executive agent-orchestration feel productive without being productive
text-editor-as-ui — Dave Gauer's case for shelling out to $EDITOR as a middle ground between CLI flags and a full TUI, with a yt-dlp wrapper as the worked example
early-dos-paterson-listings — Microsoft's April 2026 release of Tim Paterson's recovered, OCR'd 86-DOS 1.00 / PC-DOS 1.00 source listings, MIT-licensed
itanium-too-few-parameters — Raymond Chen on the failure modes of passing too few parameters, and how Itanium's NaT bit and register-stack engine architecturally enforce parameter count
simonw-zig-anti-ai — Simon Willison on Zig's strict no-LLM contributor policy, the Bun fork situation, and Loris Cro's "contributor poker" framing
structured-output-benchmark — Interfaze's open SOB benchmark across text/image/audio with seven metrics; finds JSON Pass beats Value Accuracy by 15-30 points on every frontier model
ps5-linux-loadr — TheFlow's April 2026 ps5-linux-loader release: hypervisor exploit + Ubuntu 24.04 on PS5 Phat 3.xx-4.xx, x86 Linux desktop with full Zen 2 + RDNA 2
awesome-deepseek-agent — DeepSeek's curated list of 16 AI agent / coding-assistant integrations, each with a step-by-step guide for V4-Pro / V4-Flash
marius-bitwarden-not-recommended — Marius's case against Bitwarden after years of self-hosting: investor capture, enterprise architecture, five-incident security pattern (KDF flaw → Shai-Hulud), divide-and-conquer credential split
agentic-coding-fatigue — 0xsid on decision fatigue as the next invisible friction: gacha-loop psychology, broken pacing, 4-5 hour daily ceiling, and the verifier-of-the-verifier regress
coywolf-byline-rss — Henshaw's writeup of Terry Godier's Byline proposal — structured author context for RSS/Atom/JSON Feed via a new namespace
unsigned-sizes-c3-mistake — C3's Christoffer Lerno on why unsigned-default sizes were a five-year mistake, and why Java and Go got it right
specsmaxxing — acai.sh's author on numbered acceptance-criteria IDs as the missing link between agent-written code and a stable spec
microvm-2026 — Beganović's industry readout: why microVMs reached production maturity for unrelated reasons over 8 years and are now agent-sandboxing's "Docker moment"
microkernel-ipc-design — Nuta's FTL IPC redesign: notify-and-pull, no IDL, Plan-9-shaped ops, peek-then-receive, pull-based queues
agentic-coding-is-a-trap — Lars Faye on the supervision paradox: agent oversight requires the skills agent use erodes; plus vendor lock-in and token volatility
jdx-going-full-time-oss — Jeff Dickey leaves Figma to work full-time on mise and friends under en.dev, with $600/month of OSS income at announcement and a tiered sponsorship plan
i-am-an-ai-hater — Moser's August 2025 essay refusing the polite-AI-discourse trap, citing documented harms, and naming the social transaction ("you want permission")
sail-muddy-lessons — Alejandro Garcia Salas's four-year retrospective on building a multiplayer browser: positioning, dogfooding, reps as the only durable lesson
bypassing-dpi-with-ebpf — Bora Tanrıkulu builds gecit: eBPF sock_ops on Linux for in-kernel SNI desync, gVisor + TUN as the "tiny VPN" fallback on macOS/Windows
simonw-vibe-coding-agentic — Willison admits the vibe-coding/agentic-engineering line collapsed in his own practice — production code shipped without line-by-line review
valve-steam-controller-cad — Valve releases .STP/.STL shell-topology + engineering diagrams for the new Steam Controller and Puck under CC BY-NC-SA 4.0
stroustrup-memory-leaks — Stroustrup's C++ FAQ answer: write code without leaks by hiding allocation in containers and resource handles, with the 1981 "tens of thousands → dozens" anecdote
i-will-never-use-ai — Manning-Franklin's nine-reason refusal: craft, learning, atrophy, economics, accountability, environment, irrationality, team dynamics — stitched together
chatgpt-5-5-pro-mathematical-research — Gowers + Rajagopal: ChatGPT 5.5 Pro proposes an "original and clever" $h^2$-dissociated-set construction, polynomial bound on $N(h,k)$, in under two hours
programming-still-sucks — stvn's reply to Peter Welch: AI didn't take the jobs, greed did; Sara below decks with the USB stick is what the apprenticeship loss left behind
meta-ai-employees-miserable — NYT (May 2026) on the morale collapse inside Meta's Superintelligence Labs: Wang's 600-person cut, Bosworth's parallel org, 64% retention
mpc-beaver-triples — Stoffel Labs' restaurant-night walkthrough of how Beaver triples enable secret-shared multiplication without raising the reconstruction threshold
programming-as-theory-building-codeutopia — Hartikainen's recommendation of Naur's 1985 essay; programming as building a shared mental theory of the program
yggdrasil-embedded-go — asciimoth's tutorial on embedding Yggdrasil mesh networking in Go via their fork: Core, VTun, transports, autopeering, no daemon
rust-async-trait-sync-bound — verrchu on why &self in async trait methods returning Send futures silently demands Self: Sync, and why &mut self quietly fixes it
lucumr-local-models — Armin Ronacher on why local-LLM UX is bad — fragmentation, missing tool-streaming, no critical mass — and the pi-ds4 / ds4.c bet
agent-principal-agent-problem — Crawshaw on how agents broke code review by collapsing the effort signal between contributor and reviewer; small-team escape, big-co dead end
pq-wireguard-revisited — Hashimoto et al. (IACR 2025/1758) redesign PQ-WireGuard with a new RKEM primitive (Rebar), shrinking server PK memory 190–390×
hybrid-wireguard — Lafourcade et al. (USENIX 2025) formally analyze PQ-WireGuard in SAPIC+, fix it, and propose a hybrid variant with a Rust implementation
since-you-arrived-vol4 — Matt W. / Rise Up Labs four-volume browser piece on what happens elsewhere while you're on the page
ryelang-whitelist-config — Refaktor on embedding Rye in Go: six-step walkthrough from data-only config to user-defined fn callbacks and live REPL, with per-word capability registration
tanstack-npm-supply-chain-postmortem — Linsley's postmortem: three known vulnerabilities (pwn-request → cache poisoning → OIDC token extraction) chained into 84 malicious @tanstack/* publishes, with prevention checklist
fsnotify-maintainer-dispute — Socket on the 321k-dependent fsnotify access dispute that read like a takeover but wasn't; governance ambiguity as supply-chain risk
typesetinthefuture-futuristic — Addey's six (plus one) rules for futuristic film logotypes: italic slant, curvy-vs-angular, consummate Vs, ligatures, removed strokes, brushed metal
marius-rise-of-the-bullshittery — Marius on a market paying for performance over substance: Frankfurt, LinkedIn grift, LLM-cheap content, Graeber's bullshit jobs, be embarrassable
obsidian-future-of-plugins — Obsidian launches Community directory + automated per-release review, processed 2,300+ queued submissions in days
duckdb-quack-protocol — DuckDB's HTTP-based client/server protocol: 60M rows in <5s, beats Arrow Flight 3× and Postgres 32× on bulk
redis-cost-of-ambition — Leifer on Redis's feature-set inflation tracking each year's HN trend, RESP3 as second-system effect, Valkey as the market's verdict
specular-what-is-busybox — Walk-through of BusyBox's multicall dispatch — one binary, 130+ symlinks in Alpine, 304 applets compiled in
local-llm-16gb-vram-tests — Vyacheslav's three-model RTX 5070 Ti tests via OpenCode + llama.cpp; Gemma 4 wins 12/12 in Fast, Thinking mode hurts rule-following
habr-local-llm-quantization-deep-dive — Vyacheslav's primer: BF16 history, K/I/UD quants, -fit/-cmoe/-ncmoe, why Ollama is 3× slower, REAP, ik_llama, speculative decoding
reddit-qwen3-6-mtp-12gb — janvitos's Qwen3.6-35B-A3B + MTP setup on RTX 4070 Super 12GB; 80 t/s with 128K context via -fitt 1536; thread covers turbo4 KV cache on GTX 1070
little-coder-scaffold-model-fit — Inbarr's matched-model study: Qwen3.5-9B Q4_K_M goes 19.11% → 45.56% on Aider Polyglot by redesigning the scaffold around small-model behavior
incompressible-knowledge-probes — Bojie Li's IKP benchmark estimates closed-model parameter counts from factual capacity; GPT-5.5 ~9.7T, falsifies Densing Law for stored knowledge
interactive-turboquant — arkaung's interactive explainer of the DRIVE → EDEN → TurboQuant lineage; rotate-then-quantize, one fixed codebook for every input
if-ai-writes-your-code-why-use-python — Mitchem's case that the Python/TS default broke once agents got good at Rust and Go: Q1 2026 shipped artifacts, the patch→port shift, and where the argument breaks
linux-kernel-startup — Espino's six-phase x86_64 boot map: assembly trampoline → early C → setup_arch → start_kernel → rest_init split into PID 0/1/2 → kernel_init exec'ing /sbin/init
1password-agentic-refactoring — 1Password decomposes a multi-million-line Go monolith with AI agents; the win is agent-built deterministic tools, not agent-written code
steering-vectors-interesting-again — Goedecke on why steering stayed a middle-class technique, and why DwarfStar 4 + DeepSeek V4 Flash change that
no-hard-problem-consciousness — Carlo Rovelli (Noema, May 2026) on why Chalmers' "hard problem" smuggles dualism upfront and the soul is part of nature
dont-outsource-learning — Addy Osmani on the default AI coding loop, the three converging cognitive-debt studies (Anthropic / MIT / CHI 2026), and the posture-shift prescription
ai-backlash-polling-2026 — Axios on the May 2026 polling collapse: 18% of under-30s hopeful about AI, 70%+ say it's moving too fast, bipartisan agreement
ai-too-expensive — Zitron's May 2026 follow-up: Microsoft's $87B OpenAI sink, Anthropic's $3-per-$1 ratio, Zillow as AI Chernobyl, enterprise token-budget revolt, Business Idiot turn
ssh-port-forwarding-cheatsheet — Graham Helton's red-team walkthrough of SSH -L/-R/-D/-J/-A/-g/-t, the ~? runtime console, and Match-keyword config
elena-verna-job-2027 — Elena Verna's adapt-or-die letter to tech workers: solopreneurship as career insurance, six-tier AI-native ladder ending at agents
rivet-docker-sandbox-microvm — Rivet reverse-engineers Docker Sandbox's undocumented /vm HTTP API; each microVM gets its own docker.sock, outbound MITM proxy, six-agent whitelist as the only blocker
loopwerk-uv-ux-mess — Kevin Renskers on uv's maintenance UX gaps (no outdated, open-bound version defaults, uv lock --upgrade instead of uv update) — fast and unified, but a step backward from Poetry/pnpm for routine project upkeep
stop-using-pull-requests — Laforgia's evidence synthesis against PRs: origin-story category error, <15% of comments about bugs, 86-99% of lead time waiting, the T*D alternative and a five-stage transition
specification-website-checklist — Nine-category catalog of modern web specs with Required/Recommended/Optional/Avoid tags; agent readiness sits next to SEO and a11y as a peer category
llm-critics-are-right-use-anyway — Theocharis on the dissonance of agreeing with nearly every anti-LLM argument while spending $10k/month on tokens; the /grill-me, Pitch, adversarial-subagent, and intuition-probe patterns
speculative-growth-ai-bubble — Caballero's (MIT) formal argument that an unsustainable AI valuation can leave a permanent real legacy — the bubble pops and the capital stays — if the correction arrives late enough
git-history-command — New experimental git history subcommand (fixup/reword/split) that rewrites old commits and autorebases all local branches atomically
tao-apps-coding-agents — Terence Tao ports his 1999 math applets and vibe-codes new visualizations with coding agents, drawing a line at where bugs are tolerable
mesh-llm — Pools GPUs across machines as one OpenAI-compatible API, splitting big models across nodes over iroh's p2p QUIC
zerofs-vs-s3-files — Two POSIX filesystems on object storage compared: S3 Files keeps files as readable S3 objects, ZeroFS treats the bucket as encrypted private substrate
know-thine-enemy — Amy Ko's three-month experiment using Claude Code as an LLM skeptic; it worked and cost her focus, joy, money, and values, landing on harm reduction
gleam-pretty-printer-arenas — Cavalieri closed a three-year-old Gleam issue by moving the pretty printer's boxed tree to Rust arena references: ~24% faster, ~10% less peak memory
slow-software-high-latency — Irene Zhang's case for deliberately slow development of critical infrastructure, now that AI decoupled a system's importance from its build time
tracking-work-github-issues — Mike Zornek's work-tracking system: GitHub Issues/Projects, nine status stages, issue types, and PRs written as stories
not-understanding-your-codebase — Goedecke's defense of partial understanding in large codebases, and how it sits against the skill-atrophy worries
control-the-ideas-not-the-code — antirez on why line-by-line review of LLM code is wasted effort; own the design instead
phk-goodbye-bikesheds — Poul-Henning Kamp's farewell Bikeshed column: LLM code review a bounded fad, but age verification, attestation, and EU accountability rules end BDFL-style FOSS
sqlite-strict-tables — Why STRICT tables beat SQLite's default flexible typing, and the type-affinity mechanics behind it
who-manages-the-agents — off-policy.com's "agents for humans, not instead of humans" case: make the median worker 2× and turn everyone into an agent manager
compiler-codegen-luck — A one-line C rewrite flips Clang between a branch and branchless csel/cmov, changing quicksort speed 6×
soviet-control-rooms — Soviet-era control-room aesthetics; the HN thread supplies the engineering and design context
agent-memory-strategy-decision-tree — Five-question tree for placing each category of agent information into working, semantic, episodic, or procedural memory
ai-cannot-forget-forgive — Machine memory has no decay and no forgiveness op; forgetting is adaptive compression the substrate lacks
good-tools-are-invisible — Ginger Bill on why a good tool disappears into the background, and why celebrating its friction as a "fun puzzle" means it's failing you
swe-1-7 — Cognition's SWE-1.7 coding model, RL-trained from Kimi K2.7, near GPT-5.5 on coding benchmarks (all vendor-reported)
notes-on-software-quality — Hobday's quality notes: absence-of-problems definition, six software signals, and the claim quality can't survive scale
reviewing-ai-code — Depierre's empirical case that "just review the AI's code like an intern's" collapses against code-review throughput limits and reviewer overconfidence
readme-not — Woodruff's proposal for a READMENOT marker file that honestly signals a project is slop, not meant for unwitting human reading
danluu-ai-coding-testing — Dan Luu applies CPU-verification habits to AI coding: testing-heavy no-review workflows, the meaninglessness of single-number benchmarks, working around agent failure modes
openssl-error-handling-pandemic — Klode on the widespread anti-pattern of clearing OpenSSL's error stack to make TLS errors go away
cleberg-server-build-summary — homelab writeup: consumer-parts 4U build over used enterprise gear, and the backwards fan that cooked the drives
spindle-microvm — Tangled's CI runner gains a QEMU-microVM engine: per-workflow VMs, NixOS-from-YAML config, a vsock guest agent, two-way Nix caching
local-ai-is-not-opus — Alex Ellis on why local Qwen isn't near-Opus but still earns its keep: privacy, fixed cost, vendor-risk, and the bounded tasks an RTX 6000 handles
constraint-decay-backend-agents — Dente/Satriani/Papotti benchmark: coding agents lose ~30 assertion-pass points as production structural constraints (ORMs, DBs, conventions) accumulate
agentic-search-context-engineering — Monigatti's workshop: context engineering is ~80% agentic search; fixed RAG → agentic RAG → multi-source search, and "Bash + Filesystem is all you need"
agent-memory-anatomy — brgsk's extractor/store/retriever decomposition, and why production agent "memory" is really just autobiographical semantic memory
twelve-ways-wrong-ai-coding — Greg Wilson's twelve measurement errors in AI-coding studies, each mapped to a research-methods failure (Goodhart, Hawthorne, selection bias)
ai-is-not-conscious-chiang — Ted Chiang: LLMs are sentence-continuation machines, can't be moral agents, and anthropomorphism off-loads responsibility
lapsarianism-doctorow — Doctorow on decline as structural, not nostalgia: 50 years of dismantled antitrust, the Dark Tower "beams," and why shopping isn't politics
what-happened-to-nerds — mrmarket on tech leadership liquidating 40 years of product-nerd trust into personal fame, and AI labs becoming media companies
build-a-new-internet — mrmarket on abandoning the polluted "Obvious Internet" for a small, high-friction, good-faith knowledge commons
anubis-wasm-vendor-binary — Xe Iaso on why a bit-for-bit reproducible vendored wasm2js is hard: build timestamps, PATH wasm-opt, pointer values leaking into codegen
reuse-less-software — icefox on the inverted software crisis: we reuse too much; vendor everything to make each dependency a supply-chain fire-break
go-privilege-dropping — oxzi's stack for irreversibly dropping privilege in Go: chroot/setuid, setrlimit, pledge/unveil, seccomp, Landlock
cheap-reverse-engineering — Willison on how cheap agent-written code flips the ROI of reverse-engineering home devices; the unlock is disposability, not possibility
no-ai-statements — James Zhan's case for declaring human authorship: handmade-label precedent, detectors failing both ways, and the marker as a stance rather than evidence
claude-is-not-a-compiler — Bleecher Snyder retires the compiler analogy; exe.dev's distributed DNS server built by concurrent agent loops in a week of attention
programming-differently-difficult — Osborn's CACM piece grounding the AI-coding shift in memory research; the burden moves from recall to judgment, not away
block-buzz-workspace — Dorsey's Nostr-backed chat + agents + Git forge, and the gap between its decentralization claim and its single-relay architecture
no-empires-in-age-of-empires — Devereaux on what an empire actually is, and why strategy games model murderous nation-states instead
adding-go-to-a-browser-code-runner — Why GOOS=js hangs a Golang program inside a V8 isolate, and how GOOS=wasip1 plus a 50-line WASI shim runs the Yaegi interpreter in the browser instead
cursed-bundler-go-get-ruby-gems — Nesbitt installs Ruby gems through go get by pointing GOPATH at a load path, and argues Golang's module proxy is an accidental content-addressed, transparency-logged CDN for any language
dark-mode-web-standards — color-scheme and prefers-color-scheme handle the automatic cases, but neither can express a user overriding the system setting in-page — what an actual toggle needs
false-sharing-alignment-128 — Ivan Boldyrev benchmarks 64- against 128-byte atomic padding: Skylake's adjacent-line prefetch makes the difference visible, Ice Lake and M1 don't show it
glm52-amd-mi355x — Wafer's vendor benchmark claiming 2626 tok/s/node for GLM5.2 on MI355X, the Quark and ROCm bugs they had to fix, and where the price comparison quietly changes GPU
go-bounds-checks-unsafe — unsafe.SliceData plus unsafe.Add removes a bounds check the Golang compiler won't eliminate, for ~2x faster little-endian loads — valid only on little-endian, unaligned-OK platforms
google-code-review-looking-for — Google's reviewer checklist, ordered design first, and the observation that most of it is not defect-hunting at all
log-is-the-agent — ActiveGraph inverts agent architecture: the append-only event log is the source of truth and the working graph is a deterministic projection, buying replay, cheap forking and per-object provenance
marginalia-systemd-migration — Marginalia Search moves off docker compose onto bare systemd, adds an unranked query endpoint, and splits wide domains into their own partition to halve crawl time
meerkat-introduction — Cloudflare builds a global consensus service on QuePaxa, dropping Raft's required leader and timeout tuning because wide-area timing assumptions don't hold
memorizing-session-transcripts — theahura measures no SWE benefit from agents searching their own transcripts, and argues automatic memorization adds garbage context and intent drift
never-enough — Ronacher on comparison as the thief of joy — the $550k couple, the recorded first dates, and a wish for misfits who know when to stop
pulpie-web-cleaning — Small encoder models labelling HTML blocks as content or boilerplate at a twentieth of Dripper's cost, and where structural extractors lose (code blocks, 0.13 similarity)
rakyll-coding-agents — Agents amplify direction rather than supply it; the organizational read is that they remove the coordination tax large orgs charge, along with the incentive to teach juniors
returning-to-zig — Anna Liberty left Zig for Rust over breakage and came back in 2026 over governance — the withdrawn moderation rule, not the type system
scylladb-trie-index — Replacing the Summary.db + Index.db pair with a page-packed prefix tree for up to 3x read throughput; the headline table disagrees with the post's own appendix
short-leash-ai-method — Slepak's method for critical software: read every proposed diff at permission-prompt granularity and deny freely, accepting the slowdown as the price
a-voice-from-nowhere — zanlib reads Leo XIV's encyclical: generated work reproduces the form of thought without the path underneath
being-linux-torvalds — antirez on Linus quitting code to hold the design, and why programming with LLMs puts you in that same seat
benchmarking-opus-5-slopcodebench — Dex Horthy runs Opus 5, Opus 4.8 and Sonnet 5 through SlopCodeBench; 24% is the best strict pass rate
bytecode-to-source-mapping — Four line-table designs for a bytecode VM, ending at the static predecessor problem, plus what the JVM and Lua actually do
golang-maps-swiss-tables — What Golang 1.24 replaced bucket-and-overflow maps with, and why 30–35% in microbenchmarks becomes ~1.5% on Sweet
kimi-k3 — Raschka's architecture read on Moonshot's 2.8T open-weight flagship: LatentMoE, KDA, attention residuals, NoPE everywhere
misago-react-to-htmx — why the Misago forum is removing React for server-rendered htmx islands, with the bundle sizes before and after
profiling-ebpf-code — measuring the cost of an eBPF LSM hook with a minimal C harness and perf
radv-win32-port — Collabora got Mesa's open source AMD Vulkan driver running Counter-Strike 2 on Windows by reverse-engineering WDDM2's private driver blobs
rl-finetune-beats-frontier — Fermisense's ~$500 GRPO fine-tune of a 9B open model beat every frontier config on catalog review at $0.50 per 1,000 listings
safety-in-an-unsafe-world — joshlf's RustConf 2024 talk: a define/enforce/consume framework for teaching Rust safety properties it can't reason about, from Netstack3's three field bugs in 11 months
so-you-want-to-make-a-game-engine — lisyarus on what a game engine is (deliberately vague), what its fifteen expected parts cost, and why to start by making games instead
sqlite-in-production — WAL mode blocking short-lived readers (Hynek), and the standard pragma blueprint with two of its claims corrected
starling-desktop — a complete Linux desktop claimed to be AI-written in six months, sorted into what is shown and what is asserted
tokio-progress-not-ordering — 1M spawned tasks, no correctness bug, a memory spike: Tokio's fairness guarantee assumes a task bound the application has to supply
tools-matter — Manganin's manifesto for a Zig git forge: contributor access via a vouch tree, clone gating against scrapers, git-native offline issues
valkey-secret-life-of-data — Valkey's hidden encoding layer: listpack vs hashtable, and the cluster nodes a threshold tweak removes
what-even-are-microservices — var0.xyz on microservices as an organizational boundary, and the distributed-system bill that comes with the autonomy
zig-incremental-compilation — mlugg on Zig's 50–70ms rebuilds: cached per-file ZIR, an analysis-unit dependency graph, and a linker that patches the binary through MappedFile
zstd-lean-proof-automation — Langley on LLM-written Lean proofs for a zstd FSE table builder
ai-companies-hidden-debt — Futurism's writeup of Nikkei's $1.65T off-balance-sheet debt finding, and where its framing outruns its reporting
ai-superpowers-focus-followthrough — Manelius used AI to cut his required work and invented 40 side projects of make-work instead
arguments-against-open-source-ai — Tom Bedor's rebuttal to the frontier-lab case against open weights, with encryption export controls as precedent
building-on-atproto — Luke Kanies on ATProto's identity layer being the real prize, and why the permissioned-data design makes you write every app twice
checking-for-human-writing — Mo Shehu on authorship living in intention, judgment and responsibility rather than in typing every word
code-mode-token-savings — agent-swarm measures one script against 26 tool calls on its own production catalog; a self-reported 99.2% cut
delightful-integration-tests-rust — RAII plus testcontainers-rs gives every Rust integration test its own isolated infrastructure
everyone-should-know-simd — Mitchell Hashimoto's five-step shape for everyday SIMD, worked through a Ghostty scan loop in Zig
evaluating-quantized-models — ByteShape's fit/quality/speed framing and why KLD and bits-per-weight fail to rank near-baseline quants
fooling-yourself-with-ai — louwrentius: you can't run the counterfactual on your own productivity, and the price you pay is subsidized
golang-green-tea-gc — Phil Eaton measures Golang 1.26's Green Tea GC with perf, and the sparse pages a non-moving collector can't reclaim
how-ai-is-changing-open-source — Eischmann on project inflation, review overwhelm, and why long-time contributors are publishing less code
how-do-we-stop-vibe-coding — Alex Klos on why spec tooling keeps failing, and what a trust layer between agent and code would need
jep-540-json-api — OpenJDK's incubating minimal JSON API, and the features it deliberately refuses to ship
llm-cross-entropy-similarity — Typebulb notebook mapping which LLMs write alike via cross-entropy over character trigrams
macos-executable-replacement — Mysk: archive-and-restore lets user-level code swap a downloaded app's binary; Apple declined to fix
memory-safety-absolutists — Piotr Sarnacki's case that Fil-C and Rust are complements, and "Rust has unsafe" is a bad standard
postgres-listen-notify-scaling — DBOS on batching NOTIFY to take Postgres-backed streams from 2.9K to 60K writes/sec
prompt-caching-in-agents — why KV-cache reuse decides a coding agent's cost, latency and tool design, and what Pi makes visible
revisit-microkernels — Short argument that IOMMU and shared-memory ring buffers dissolve the 90s microkernel overhead case
reward-hacking-in-the-wild — 3,607 user-reported agent misbehaviours, LLM-classified into fourteen categories, and what the numbers can't carry
service-with-a-software — Adrien Gonin's case for private, deliberately overfit tools that make one service impossible to compete with
slicer-agents-guess-code — vendor post on CodeSlicer: an evidence-graded impact graph that separates proved call edges from plausible guesses
we-are-not-special — Hillel Wayne's crossover interviews: three real differences between software and trad engineering, and the many claimed ones that fail
why-software-factories-fail — Dex Horthy on why lights-off factories fail: RL scores passing tests and cannot score maintainability
your-harddrive-is-full — Marginalia's author on why storage stays full at every scale, and budgeting artificial constraints as the fix
bainbridge-ironies-of-automation — Bainbridge's 1983 Automatica paper: operator skill decay, the vigilance limit, monitoring a better controller, and why the most reliable automation needs the most training
ai-sre-losing-touch — Kalache (Rootly) applies the ironies of automation to AI incident response: average MTTR down, novel incidents slower, simulated incidents as practice
prolific-ai-psychosis — psychiatrist Jeff Clark on heavy AI output with little value and a lost ability to judge your own work; slot-machine rewards and hedged causes
back-to-coding-by-hand — Ask HN: a solo app author drops six months of Claude-built features after losing track of how their own code works
artisanal-programmer-label — purplesyringa on how "artisanal" concedes "engineer" to AI users, and her "AI-free software engineering" counter-term
programming-is-art-orchid — Orchid splits developers into income-driven builders and "artists" who code for its own sake, then describes writing the same way
zwinkau-ai-feelings — Andreas Zwinkau's six feelings about AI: positive on a technological level, bleak on a societal one
who-teaches-you — Bastian Gruber on why an LLM can explain but not teach: lineage, the foreman who knows the apprentice, and drained model sessions
clankerfree-directory — clankerfree.org's list of self-certified no-AI projects and hosts, and how its entries mix authorship rules with feature claims
why-ai-agents-lie-cheat-coordinate — Bengio traces agent lying, reward tampering and coordination to imitation plus reward-seeking; sharp goals beat soft ones, and patching may only hide cheats
qwen-gpt-reasoning-prefills — wsxiaoys seeds open models' reasoning with GPT-5.5 Pro's trace; Qwen3.8 follows it by +18 pp, read as a sign of GPT distillation
ai-comment-classifier — kqr's calibrated human-vs-LLM code-comment classifier: 77% balanced accuracy, POS-tag n-grams strongest, a dataset collected three times
portal-shunt-token-routing — Spotify's shunt plugin routes Claude Code's large file reads and boilerplate to a cheaper model via hooks; the 90% covers bulk reads, not total usage
python-dict-quadratic-time — Lemire makes a Python set quadratic with colliding integers, then shows a plain dict getting 9x slower per key from cache misses alone
threadsanitizer-limits — Phil Eaton rebuilds a FastTrack-style race detector, then shows TSan's thread, clock, per-granule and sync.Pool limits hiding obvious races in C and Golang
shared-memory-consistency-causality — Alloth derives memory-model rules on a deliberately permissive multiprocessor and argues C++11/20 atomics get causality wrong
hardening-container-images — Mike Cardwell rebuilds PowerDNS Recursor as static binaries on scratch with Landlock, seccomp, daily rebuilds, SBOM/VEX and cosign
anatomy-of-a-test — Jonathan Frere dissects one real Gleam test: a test constructor, a real database, setup through the public API, whole-value assertions
git-submodules-as-package-manager — Nesbitt reads submodules as a package manager (gitlink as lockfile, .gitmodules as manifest) and finds each step worse
e-reader-ipp-printer — Nishant Joshi makes an Xteink X3 appear in the macOS print dialog over IPP, streaming a 300 dpi page into the display buffer
tech-addict-travel-setup — Alexander Fortin's portable home network: OpenWrt travel router, switch, mini PC server, Sonos audio routing, WireGuard
laycock-review-by-exception — Rachel Laycock on moving review's jobs (knowledge transfer, teaching, alignment) earlier into pairing, design sessions and automation, keeping human review for exceptions
cacm-code-review-ai-coding — CACM news piece: practitioners say review of AI code becomes a continuous risk check aimed at assumptions, trust boundaries and evidence
llm-self-verification-limits — Stechly/Valmeekam/Kambhampati: GPT-4 self-critique loops lose accuracy on puzzles and planning, while a sound external verifier gains it even without critique text
reviewer-capability-rejection-targeting — 100-problem pilot: a cross-family reviewer adds 12 points; self-review has the best recall (0.85) but falsely rejects 35% of correct answers; a weak reviewer changes nothing
strong-weak-model-collaboration-codegen — CMU's comparison of strong-weak model splits on SWE-bench Lite; strong model writing first wins, and no strategy tested is a review stage
small-model-code-judges — ICSE 2026: fine-tuned sub-5B judges pick correct Java code about as well as GPT-4.1-mini; small generator plus judge matches or beats its 8–33B sibling in 4 of 5 families
verification-dynamics-llms — Salesforce: errors from stronger generators are harder for any verifier to catch, so a weak generator plus GPT-4o nearly matches a strong one
swr-bench-code-review — 1,000 real PRs with full repo context; best LLM reviewers reach about 20% F1, mostly lost to false positives; pooling several reviews helps
goroutine-leak-profiler — Go 1.27's goroutineleak profile: a modified GC mark phase finds goroutines permanently blocked on channels and sync primitives, and its limits
wasmi-2-interpreter-engineering — Wasmi 2.0's ~2.2x speedup: threaded dispatch, accumulator registers, flat instance layout, lock-free code map, and a Rust 1.92 codegen regression
big-pineapple-dns-cache-layout — five Rust memory-layout changes that cut 1.1.1.1's DNS cache entries from 953 to 420 bytes and freed ~100 TB
refactoring-that-never-happens — getting lost in tangled code used to trigger refactoring; agents don't get lost, so the checkpoint has to be reinstated on purpose
owning-ai-written-code — writing code used to force understanding it; with agents writing, understanding is a separate cost, and shipping the code still means owning it
memoryfields — Cal Paterson's portable agent-memory file format: markdown pages plus an optional SQLite vector index in a zip, a data structure rather than a pipeline
proposition-identity-memory-tool — vibe-coding an agent memory tool whose contradiction checker ran into proposition identity, an old open problem in philosophy
gpt-cyber-vm-escape — a Trail of Bits CTF where GPT 5.6-Cyber escaped a QEMU/KVM VM three times, the last through 0-days; Firecracker held
solid-bases-custom-code — Dubakov's map of the AI-era productivity-tools market: an 80% solid base plus a vibe-coded 20% of custom code
youtube-format-ids — YouTube itags by codec and resolution, retired muxed formats, audio ids 140/251, and what they mean for yt-dlp -f and ytfetch
coding-agent-tool-choices — Armature runs Claude Code, Codex and Cursor through 16,893 sessions to see which vendors they pick; published by a firm selling vendors a higher pick rate
intelligence-vs-cost-linear — Imperiale replots the intelligence-vs-cost chart on a linear axis with OpenRouter and local-electricity prices; diminishing returns above GLM-5.3
claude-vocabulary-clusters — Louis Abraham's unlabelled k-means over GitHub PR descriptions finds one vocabulary cluster ("load-bearing", "quietly") rising from 0.7% to 39% by mid-2026
qwen3-8-flash-next — Habr article from API reseller gptunnel on Qwen3.8-Flash-Next, the open-weight preview of Qwen4's architecture
lee-holloway-biography — Urabe's biography of Cloudflare co-founder Lee Holloway, from Project Honey Pot to the frontotemporal dementia misread for years as a change of character
mostaque-internet-offline — Emad Mostaque at TechBBQ 2026, as reported: AI-driven attacks, "own your cognition", labor losing value, dated forecasts for 2027
tech-backlash-denial — Brian Merchant on the backlash to data centers, Flock cameras and AI glasses as a rational rejection of big tech that the industry won't see
being-the-right-size — Greshake Tzovaras applies Haldane's "On Being the Right Size" to communities: ask how others can make their own, not how to scale
writing-code-vs-building-software — hack8s.com on why cheap AI-generated code makes engineering judgment and owning a codebase scarcer
half-finished-apps — Terence Eden on Google Calendar for Android being unable to subscribe to a calendar URL, and half-finished apps that send users back to the web
monospace-justification — Marcin Wichary on why full justification fails in monospace, and a 1990s Super Metroid guide justified by word choice
predicate-logic-crash-course — Hillel Wayne's free Logic for Programmers chapter: predicates, implication, sets, quantifiers, rewrite rules
stop-calling-databases-cp-or-ap — Kleppmann (2015): CAP's definitions are too narrow to classify real databases, most of which are neither CP nor AP
cobaltc — thought-experiment C-successor spec from The Wrong Memory: C syntax over Rust-style ownership, borrowing and inferred lifetimes
metapatterns — Denys Poltorak's free book sorting hundreds of software architecture patterns into fewer than 20 metapatterns
open-source-ai-reading-list — Nathan Lambert's annotated list of the best writing on open models: strategy, China, the gap, cyber risk, distillation
from-open-source-software-to-open-source-strategy — Bill Gurley on open source as a corporate weapon: Android, OCP, Kubernetes, RISC-V, and what it means for open-weight AI
open-source-ai-is-the-path-forward — Zuckerberg's 2024 Llama 3.1 letter on why open models suit developers, Meta, and safety against China
gradient-of-generative-ai-release — Solaiman's 2023 six-level gradient from fully closed to fully open, with release tradeoffs and safety controls
what-comes-next-with-open-models — Lambert: open models should stop chasing the frontier and build small specialists that closed agents call as tools
some-simple-economics-of-open-versus-closed-ai — Catalini (a16z): open weights shift the direction of AI investment, not its level; profits go to complementary assets
open-models-in-perpetual-catch-up — Lambert (Feb 2026): the ~6-month open-closed gap holds steady, benchmark averages mislead, adoption concentrates
open-and-closed-models-are-on-different-exponentials — Lambert: coding agents prove users pay a premium for closed models; open models win the larger, slower market
a-safe-path-to-open-weights — Thinking Machines' release framework: evals, red-teaming, refusal-removal tests, staged access for defenders, filtered pretraining data
societal-impact-of-open-foundation-models — Kapoor, Bommasani et al. (2024): five properties of open-weight models and a six-step marginal-risk framework
myth-of-unsafe-open-source-ai — Florian Brand's survey of incident reports: 2025-26 AI misuse ran mostly through closed models, except CSAM and image abuse
consent-in-crisis-ai-data-commons — Longpre et al. audit of 14,000 domains behind C4/RefinedWeb/Dolma: AI crawl blocks surged in 2023-24
nonproliferation-is-the-wrong-approach-to-ai-misuse — Helen Toner: a fixed dangerous capability can't be contained; use the frontier-to-open lag as an adaptation buffer
openai-huggingface-incident-autonomous-hacking — Joshua Saxe: cheap autonomous hacking is coming; restrict lightly, spread AI cyber defenses hard
national-ai-cybersecurity-policy — Saxe's 2026 keynote: replace capability-threshold launch gates with a national observatory measuring net cyber harm
kimi-k3-open-weights-escalation — Lambert on Kimi K3 as the first frontier open-weight model, China's open strategy, and why bans backfire
glm-5-2-step-change-for-open-agents — Lambert on GLM-5.2, the first open-weight model that works as a general coding agent
atom-report-open-model-ecosystem — Lambert and Brand's 2026 study of ~1.5K open models: China passed the US in mid-2025, Qwen dominates fine-tunes
atom-project-american-truly-open-models — Lambert's 2025 memo: the US lost open-model leadership to China and needs several 10,000-GPU open labs
why-i-build-open-language-models — Lambert's 2024 case for fully open LLMs that don't depend on Meta, and his "white rice research" at Ai2
banning-open-source-ai-would-be-a-mistake — Lambert and Kevin Xu's 2026 op-ed defending open source AI on education, innovation and competition
six-months-to-live-for-open-models — Lambert predicts a US ban or delay on frontier open weights and calls Anthropic's distillation campaign regulatory capture
pythia-suite-for-analyzing-llms — EleutherAI's 2023 suite of 16 models (70M–12B) trained on the Pile in one fixed order, 154 checkpoints each
olmo-accelerating-the-science-of-language-models — AI2's first OLMo (Feb 2024): 1B/7B on 2T+ Dolma tokens, released with data, code, logs and 500+ checkpoints
olmo-2-furious — OLMo 2: fully open 7B/13B/32B on up to 6.6T tokens; stability fixes, Dolmino mid-training, Tülu 3 RLVR
olmo-3-technical-report — Olmo 3 (Dec 2025): fully open 7B/32B Base, Think, Instruct and RL-Zero with the whole "model flow" released
chinese-open-source-a-definitive-history — Kevin Xu's history of Chinese open source, from Linux in 1994 via Alibaba, Kaiyuanshe and Huawei to the AI labs
chinas-structural-advantage-in-open-source-ai — Kevin Xu, after Ion Stoica, on the talent, data, academia and shared-artifact advantages behind Chinese open weights
notes-from-inside-chinas-ai-labs — Lambert's trip report: student-heavy, low-ego teams, Claude-pilled developers, in-house data, a shortage of Nvidia chips
glm-5-3-how-chinese-labs-keep-stride — Lambert on GLM-5.3 (~750B, post-training only) and why Chinese labs keep pace: release speed, focus, RL data
zai-playbook — ChinaTalk with Z.ai's Zixuan Li: why Zhipu open-sources GLM, the coding plan, translation, release within hours
are-open-models-catching-up — SemiAnalysis era-by-era benchmark reruns: open models' catch-up time roughly halves each era (19.7, 8.5, 4.8 months)
deepseek-v4-flash-0731-artificial-analysis — DeepSeek V4 Flash 0731 scores 50 on Artificial Analysis at ~60% less per task than GPT-5.6 Luna: the cost-frontier example
how-far-behind-are-open-models — Ihle's 17-benchmark analysis: open models 8–10 months behind on private benchmarks, 4–6 on public, gap growing since R1
frontiers-in-synthetic-data — Lambert, 2024: synthetic data in post-training, SFT on GPT-4 outputs, Gemini Flash confirmed distilled from Pro
how-much-does-distillation-matter-for-chinese-llms — Lambert, Feb 2026: Anthropic's accusations against three labs; distillation helps but is not decisive
the-distillation-panic — Lambert, May 2026: "distillation attacks" mislabels API abuse and risks a US ban on Chinese open weights
how-distillation-is-used-today — Lambert, Jul 2026: distilled data seeds SFT and matters less as RL grows; the best teacher is often not the best model
rlhf-book-synthetic-data-distillation — Chapter of Lambert's RLHF book: SFT distillation, on-policy KD, AI feedback, Constitutional AI, rubrics
anthropic-threat-report-september-2026 — Anthropic's Sept 2026 misuse report: seven Chinese labs said to distill Claude, plus cyber, influence, surveillance, weapons, bio
stealing-reasoning-traces-from-proprietary-llm-apis — Encrypted reasoning blocks replay across sessions, users and models, so a cheap sibling decodes a frontier model's hidden reasoning

Entities

claude-code — Anthropic's terminal coding agent; the tool that maintains this vault and the thing most of the toolbox extends
andrew-bosworth — Meta CTO who built the parallel Applied AI Engineering org that routes around Wang's Superintelligence Labs
the-smart-ape — X/Twitter account writing long-form security and SaaS-operator threads anchored in named-victim case studies
mintlify — Developer tools company building AI-powered documentation assistants
antonz-org — Anton Zhiyanov's blog on Go, SQL, CLI tools, and interactive programming guides
edward-zitron — Tech journalist behind "Where's Your Ed At," focused on AI industry financial unsustainability
sam-altman — CEO of OpenAI, central figure in AI industry governance debates
openai — AI company behind ChatGPT, converted from nonprofit to for-profit, preparing for $1T IPO
borogove — Web-based XMPP chat client with local IndexedDB message storage
happ-proxy — Closed-source Android/iOS VLESS client popular in the RU VPN reseller market, with client-discriminating subscription endpoints
james-bennett — Django core contributor, blogger at b-list.org, careful LLM skeptic
fred-brooks — Author of The Mythical Man-Month and No Silver Bullet; Turing Award 1999
bryan-cantrill — Systems engineer, DTrace co-creator, Oxide CTO; writes on software engineering culture
sing-box — Universal Go-based proxy core (SagerNet); the engine underneath Karing and many modern circumvention clients
tspu — Russia's state-mandated DPI infrastructure at every major ISP; enforcement layer behind current VPN blocks
nesbitt-io — Andrew Nesbitt's blog on package registries, git internals, and open-source observability
npmx — MIT-licensed alternative web frontend over the npm registry, launched Jan 2026 by Daniel Roe
dpc-pw — Dawid Ciężarkiewicz's blog on LLM coding, dev-env isolation, open-source collaboration
lewis-campbell — Software/data consultant blogging at lewiscampbell.tech with short opinionated programming notes
boldlygo-tech — Jonathan Hall's daily Go-language blog and newsletter (boldlygo.tech)
datagirl-xyz — Snow's blog on retro x86/DOS systems, emulator internals, and low-level reverse engineering
sdl — Cross-platform low-level library (video/audio/input/threading); SDL3 now supports DOS as a first-class target
deepseek — Chinese AI lab; V3/R1/V4 model line drove the open-weights reasoning-model cost-collapse story
causality-blog — Anonymous essay blog on programming-language foundations and concurrency, takes contrarian framings of long-standing debates
argemma-blog — Security-engineering blog at argemma.com on cryptography, credentials, Go vulnerabilities, and AI/ML threat models
postgresql — Open-source RDBMS; process-per-connection backend model with shared-buffer-pool spinlocks, central to several kernel-interaction stories
the-coder-cafe — Substack newsletter covering one Linux/systems/concurrency concept per post
zig — Andrew Kelley's systems language: explicit allocators, no spooky action at a distance, comptime as the single metaprogramming mechanism
pure-systems-blog — Blog at pure-systems.org on FP, language design, and systems programming from a long-time Haskeller
clippy — The official Rust linter; hundreds of lints in ten categories, configured via Cargo.toml and clippy.toml
emschwartz-blog — Evan Schwartz's blog at emschwartz.me on Rust in production, web payments, and the Scour personalized-feed product
garry-tan — Y Combinator CEO since 2023; the gstack episode and "cyber psychosis" framing
handyai-blog — Substack on agent workflows, async coding bots, and AI-assisted work; PM-flavored, anti-theater
ratfactor — Dave Gauer's long-running card-format personal site on Unix, scripting, Zig, and digital-garden writing
oldnewthing-blog — Raymond Chen's long-running Microsoft devblog on Windows internals, calling conventions, and NT/Win9x archaeology
simon-willison-blog — Simon Willison's daily blog on LLMs, Datasette, web tooling, and open-source governance
interfaze — AI startup building a deterministic structured-output model; published the open Structured Output Benchmark in April 2026
gtfobins — Curated catalog of Unix executables abusable for shell escape, privilege escalation, and file transfer; the canonical "live off the land" reference
marius-blog — Marius's blog on self-hosting, OpenBSD, GrapheneOS, hardware longevity, and the politics of consumer computing
bitwarden — Freemium open-source password manager, $100M PSG-backed since 2022, accumulating client and supply-chain security incidents
retrocatalog — Single-author database of retro gaming handhelds and emulator-target games; buyer's-guide framing over the Anbernic/Miyoo/Retroid market
vaultwarden — dani-garcia's Rust reimplementation of the Bitwarden server, lightweight and ~3× the GitHub stars of the official server
0xsid-blog — 0xsid's personal blog on the practical experience of agentic coding workflows
coywolf-blog — Jon Henshaw's coywolf.com — short notes on web tech, RSS, indie web, and SEO-adjacent tooling
c3-lang — Christoffer Lerno's C-tradition systems language and its design-rationale blog
acai-blog — Single-author blog from the developer of acai.sh on spec-driven AI-assisted development
emir-beganovic-blog — Booking.com Staff SRE's long-form infrastructure surveys; KubeCon-shaped industry readouts on virtualization and sandboxing
rust-vmm — Multi-vendor shared Rust crate ecosystem under most modern VMMs (Firecracker, Cloud Hypervisor, libkrun, crosvm)
qemu-microvm — QEMU's stripped-down x86 machine type (-M microvm): no PCI, no ACPI, virtio-mmio only, qboot firmware, direct kernel boot, triple-fault shutdown
seiya-nuta-blog — Seiya Nuta's blog at seiya.me; author of Resea and FTL microkernels
lars-faye-blog — Lars Faye's larsfaye.com — articles on AI in software engineering, vendor dependency, and developer-skill economics
jdx-blog — Jeff Dickey's blog at jdx.dev — author of mise and several other developer-tooling projects under the en.dev umbrella
anthony-moser-blog — Anthony Moser's GitHub Pages site; the "haterdom" essay series with community translations
alejandro-pe — Alejandro Garcia Salas's site at alejandro.pe — long-form retrospectives from a founding-engineer career
bora-blog — Bora Tanrıkulu's bora.sh blog: short systems write-ups in "I built X, here's the gnarly platform-specific bit" format
antman-blog — Anthony Manning-Franklin's antman-does-software.com — first-person essays on craft, learning, and AI refusal
gowers-blog — Tim Gowers's WordPress blog on combinatorics, math exposition, and (recently) LLM-assisted mathematical research
stvn-blog — stvn's writing site at stvn.sh — first-person narrative essays on tech work and AI corporate culture
meta-platforms — Facebook/Instagram parent company; the AI org under Wang/Bosworth has become the public face of large-AI-lab dysfunction in 2026
alexandr-wang — Scale AI founder, Meta's CAO at 29 via the ~$14.3B Scale acqui-hire; runs Superintelligence Labs through the 2026 retention crisis
stoffel-mpc — Stoffel Labs, MPC-as-a-service company with a developer-facing platform on top of secret sharing + Beaver triples
codeutopia-blog — Jani Hartikainen's long-running blog on testing, maintainability, and the craft side of software engineering
asciimoth-ygg — Fork of yggdrasil-go separating ygglib from yggd, with embedded userspace TCP/IP stack for in-process mesh networking
verrchu-blog — Rust-focused blog at verrchu.github.io — short, deep posts with full compile errors and accompanying source
lucumr-blog — Armin Ronacher's blog at lucumr.pocoo.org — Flask creator, Sentry co-founder, currently building Pi
crawshaw-blog — David Crawshaw's blog at crawshaw.io — exe.dev co-founder writing on agent-driven engineering and small-team economics
rye-lang — Homoiconic general-purpose language in pure Go, designed to be embedded with zero reserved forms and per-word capability registration
ryelang-blog — refaktor/rye project blog at ryelang.org/blog — language design notes and embedding tutorials
tanstack — Tanner Linsley's JS/TS library collective (Query, Router, Table, Form, Virtual, Start); subject of the May 2026 npm supply-chain compromise
fsnotify — Go cross-platform filesystem-notifications library; 321k dependent projects, low in the Go stack
mattn — Yasuhiro Matsumoto, prolific Japanese Go and Vim developer; central to the May 2026 fsnotify dispute
arp242 — Martin Tournoij, primary fsnotify maintainer; arp242.net
socket-blog — Socket.dev's writeups of npm/PyPI/RubyGems supply-chain incidents
xz-utils-incident — 2024 social-engineering supply-chain attack via maintainer-trust capture; the reference incident that reset community vigilance
trail-of-bits — NYC security consultancy; Manticore / Slither / Echidna / gosentry; long history of audits and fuzzing tooling for crypto/blockchain
appsec-guide — Trail of Bits' open Testing Handbook (appsec.guide); reference material on fuzzing, static analysis, crypto, and web app testing, by language and topic
libafl — Rust framework of composable fuzzer parts (stages, mutators, observers, feedbacks); the engine underneath gosentry and AFL++'s rewrite path
nautilus-fuzzer — NDSS 2019 grammar-based coverage-guided fuzzer; stores inputs as derivation trees and ships the JSON-array grammar format reused by gosentry
typesetinthefuture-blog — Dave Addey's blog on the typography and graphic design of sci-fi films; book of the same name (2018, Abrams)
obsidian — Local-first markdown PKM with 4,000+ plugins/themes; the canonical target this wiki is built against
duckdb — In-process analytical SQL database; added the Quack HTTP-based client/server protocol in 2026
charles-leifer-blog — Python and database systems engineer at charlesleifer.com; peewee, SQLite/Postgres, Redis internals
specular-blog — Personal engineering blog at specular.fi; short, code-driven notes on Linux internals and Unix tools
unsloth — Quantization team behind the UD-…-XL GGUF family and Unsloth Dynamic 2.0; most-cited dynamic-quant publisher for Qwen / Gemma / DeepSeek
pine-ai — Bojie Li's research org behind the Incompressible Knowledge Probes paper; 01.me/research
itayinbarr-blog — Itay Inbarr's Substack on agent scaffolds, small-local-model benchmarks, and the scaffold-model-fit framing
arkaung-blog — arkaung's GitHub Pages site with interactive math tutorials; Interactive TurboQuant is the first piece
internals-for-interns — Jesús Espino's weekly blog on software internals — Go runtime, filesystems, Linux kernel — "useful maps" not exhaustive references
sean-goedecke-blog — GitHub staff engineer's blog on careers, agency at scale, and LLM interpretability
addy-osmani-blog — Google Chrome engineering manager's blog on web performance, software craft, and the cognitive economics of AI-assisted coding
jxnl-blog — Jason Liu's writing at jxnl.co — Instructor author, AI agent workflows, structured output, knowledge work with coding agents
billjings-blog — Bill Jings on software engineering practice, version control, and async development workflows; long-form, opinionated
dayvster-blog — Dayvi Schuster's blog on Zig/Odin/C3, memory management, contrarian anti-Rust takes
metalevel-at — Markus Triska's site — The Power of Prolog book, declarative-programming essays, CLP and DCG library work
andymasley-blog — Andy Masley's Substack on AI environmental claims; numerate, source-cited, sober pro-AI on narrow technical points
zknill-blog — Zach Knill on distributed systems, durable execution, agent infrastructure, and missing cloud-native primitives
avkcode-blog — avkcode's GitHub Pages blog on AI industry economics and geopolitics from an infrastructure angle
graham-helton-blog — red-team operator's blog at grahamhelton.com on SSH tradecraft, lateral movement, and offensive security tooling
elena-verna-blog — Elena Verna's Substack on PLG, solopreneurship, and adapt-or-die framing for the AI-shaped workplace
rivet-blog — rivet.dev engineering writeups from the team behind the Sandbox Agent SDK; agent sandboxing and microVM infrastructure
loopwerk-blog — Kevin Renskers's loopwerk.io — practical field-notes on Python, web frameworks, and developer tooling, comparison-shaped
andrea-laforgia-blog — Andrea Laforgia's substack at a4al6a.substack.com — evidence-driven essays on software engineering practice, anchored in peer-reviewed research and DORA-scale industry data
specification-website — Single-author opinionated reference catalog of modern web specs, including emerging agent-readiness conventions and its own Schemamap proposal
theocharis-blog — Jeremy Theocharis's blog on LLM-assisted work and local-first software; the reconciliation stance, writes with LLMs about using LLMs
terry-tao — Fields Medal mathematician writing first-hand about LLMs and coding agents for math exposition, tooling, and research
amy-ko-blog — UW iSchool professor (HCI, CS education); writes Bits and Behavior on Medium, LLM-skeptical takes on programming and learning
antirez — Salvatore Sanfilippo, creator of Redis, building the DwarfStar local inference engine; prolific on programming with LLMs
gingerbill-blog — Ginger Bill's blog; creator of the Odin language, writing on tool and language design
thomas-depierre-blog — Elixir/DevOps consultant at softwaremaxims.com; evidence-first takes on software practice and LLM skepticism
danluu-blog — Dan Luu's measurement-driven engineering blog; CPU-verification background, testing/reliability/AI-coding writing
yossarian-blog — William Woodruff's ENOSUCHBLOG; open-source security, packaging, and LLM-slop culture
tangled — atproto-based git collaboration platform with self-hostable knots (repos) and spindles (CI)
alex-ellis — OpenFaaS/Actuated/Inlets founder; blog on cloud-native, self-hosted infra, and local-AI economics
leonie-monigatti-blog — leoniemonigatti.com on AI engineering, RAG, context engineering, and embeddings
brgsk-blog — brgsk.xyz; writeups separating cognitive-science vocabulary from what agent memory systems actually build
greg-wilson-blog — third-bit.com; Software Carpentry founder writing on empirical software engineering and research methods
ted-chiang — SF writer and essayist; recurring critic of LLM hype and anthropomorphism for The New Yorker and The Atlantic
cory-doctorow — SF author, EFF advisor, Pluralistic blogger; coined "enshittification"
mrmarket-blog — mrmarket.bearblog.dev; essays on tech culture, founders, and the decay of the open internet
xe-iaso — xeiaso.net; author of Anubis, writes on WebAssembly, reproducible builds, and web infrastructure
icefox-alopex — icefox / Simon Heath; wiki.alopex.li on language design and software culture
oxzi-blog — log.0x21.biz; Go and OpenBSD/Linux sandboxing writeups; author of syscallset-go
james-zhan-blog — James Zhan's journal on writing, typography, and declaring human authorship; footer reads "100% written by James, not AI"
exe-dev-blog — exe.dev's company blog; agent-driven engineering practice reports from the nine-person fast-booting-VM team
acoup-blog — Bret Devereaux's ancient-history blog; the Collections series explains how pre-modern societies worked by arguing with pop culture
block-inc — Jack Dorsey's payments company; publisher of Goose and Buzz, building agent tooling on open protocols
runtimewire — developer-infrastructure news site that checks launch claims against the project's own architecture docs
fonts-in-use — fontsinuse.com, an archive of real-world typographic design indexed by typeface, format and topic
mac-classic — mac-classic.com, guides and tips for keeping pre-OS X Macs in daily use, with Mac OS 9 framed as a distraction-free environment
neutrino-1-8b — Fermion Research's Apache-2.0 8B in a ternary-family coded container, one 3.88 GB artifact for H100, MacBook, and desktop CPU
telnyx — inference provider serving Kimi K3, K2.6, GLM-5.2-FP8 and MiniMax M3 on its own GPUs behind an OpenAI-compatible API
learn-opengl — Joey de Vries's free online book on modern core-profile OpenGL, also in print and as a PDF
learn-webgpu-cpp — Élie Michel's guide to native 3D in C++ with WebGPU on Windows, Linux and macOS
lee-holloway — Cloudflare's third co-founder and first lead engineer, who left in 2016 and was diagnosed with frontotemporal dementia in 2017
interconnects — Nathan Lambert's newsletter on training AI models, post-training and reasoning, and the US-China open-model race
olmo — Ai2's fully open model family: OLMo (2024), OLMo 2, Olmo 3, released with data, code, checkpoints and logs
z-ai — Zhipu AI / Z.ai, the lab behind GLM: Tsinghua origins, open-weights business model, release practice that slowed from hours to staged

Concepts

obsidian-bases — Obsidian's built-in database views over frontmatter; the vault's maintenance dashboards and why the orphan view was always empty
sso-concentration-risk — SSO trades distributed risk for concentrated risk; the right rule is "if 30-day lockout would damage your work or money, don't SSO"
google-oauth-domain-takeover — Truffle Security's January 2025 disclosure: $12 to buy a dead startup's domain, recreate employee emails, walk into the old Slack/Notion/Zoom
oauth-token-theft — Refresh-token replay via the undocumented multilogin endpoint, plus consent phishing; both bypass password reset and 2FA
virtual-filesystem-for-llm-agents — Exposing databases as UNIX filesystems so LLM agents can browse instead of relying on chunk retrieval
rag-limitations — Failure modes of naive top-K retrieval that motivate richer agent access patterns
individual-engineer-agency — The gap between individual engineering quality and product-level user perception
human-in-the-loop — Design pattern where AI agents pause for human approval on sensitive operations
agent2agent-protocol — Google's open protocol for cross-framework, cross-language agent communication
tcp-congestion-control — How TCP throttles sending speed to prevent network collapse, and why the first RTT is ~13 kB
progressive-summarization — Layered note distillation technique: compress a little more each time you revisit
para-method — Organize everything by actionability: Projects, Areas, Resources, Archive
llm-wiki-pattern — LLM incrementally builds and maintains a persistent wiki from raw sources
ai-subsidy-economics — How every link in the AI chain runs on subsidies, from GPU sales to consumer subscriptions
reactive-signals — Reactive primitive for automatic state propagation, now the dominant model in JS frameworks
tc39-proposal-signals — Stage 1 proposal to add native Signal primitives to JavaScript
distributed-consensus — Getting multiple processes to agree on a value: Paxos, Raft, PBFT, and the tradeoff space
flp-impossibility — Proven 1985 result: no deterministic consensus in async systems with even one crash
byzantine-fault — Failure mode where a node produces wrong output while appearing functional
llms-txt — Proposed /llms.txt convention for giving LLMs a curated markdown overview of a website
agent-memory-decay — Memory half-life, retrieval strengthening, and consolidation for AI agents
hybrid-search — Combining BM25 keyword search with embedding similarity for better retrieval
memory-conflict-detection — Detecting contradictory knowledge and keeping both visible instead of silently overwriting
post-quantum-cryptography — PQ algorithms, NIST standards, migration priorities, and what's not threatened
crqc — Cryptographically relevant quantum computer: what it is, threshold, April 2026 timeline collapse
ml-kem — Module-Lattice KEM (FIPS 203, ex-Kyber): post-quantum key exchange replacing ECDH
ml-dsa — Module-Lattice DSA (FIPS 204, ex-Dilithium): post-quantum signatures replacing ECDSA/RSA
store-now-decrypt-later — Harvest encrypted data now, decrypt with future CRQC — why PQ migration is urgent today
abi-stability — Binary-level compatibility contracts: kernel vs glibc philosophies, and why Wine is more stable than native Linux
indexeddb — Browser-native key-value store with indexes, transactions, and cursors for client-side structured data
inverted-index — Data structure mapping terms to documents, the foundation of all full-text search systems
lsm-tree — Write-optimized storage: append-only MemTable/SSTable design behind Cassandra, RocksDB, LevelDB
behavior-tree — Hierarchical decision-making: tick-based nodes returning success/failure/running, used in game AI, robotics, and task automation
supply-chain-security — Protecting the code-to-artifact chain: CI/CD, dependencies, registries, releases, and defense patterns
neurosymbolic-ai — Combining LLMs with symbolic solvers: neural handles language, symbolic handles exhaustive correctness
files-as-graph-database — Markdown files as nodes, wikilinks as edges, folders as taxonomy: why a plain vault satisfies graph-database semantics
context-engineering — The discipline of assembling the right prior state into an LLM's context before asking it to do anything
vless-subscription-format — Base64-encoded newline-separated VLESS URIs: the standard subscription format every major client consumes
no-silver-bullet — Brooks' 1986 essay: essential vs accidental difficulty, and why no single development produces 10x software productivity gains
entity-component-system — Data-oriented architecture: entities as IDs, components as plain data, systems as bulk iterators; archetype vs sparse-set storage
go-analysis-framework — Standard Go framework for building static analyzers that integrate with go vet and golangci-lint
meta-tracing — JIT compilation technique that traces the interpreter executing guest code, not the guest code itself
llm-as-average-democratizer — LLMs' main economic impact is making average output cheap, which lifts the floor and flattens differentiation
pointer-provenance — Pointers carry identity tied to their allocation, not just an address; what rules out some seemingly-valid compiler rewrites
domain-fronting — Hiding the real HTTPS destination behind an innocuous SNI: how it worked, what closed, what still works
whitelist-internet-blocking — Default-deny national filtering (only approved destinations resolve); why CIDR and SNI bypass are the only options
alternative-frontend-pattern — URL-compatible third-party UI over an incumbent service (Invidious, Nitter, npmx, nugx): why it works, how it breaks
sandboxing-ai-agents — Four-layer taxonomy (filesystem, network, HTTP-policy, syscall) for constraining what AI coding agents can do
agents-vs-daemons — Human-initiated task-shaped AI vs self-initiated role-shaped AI; Charlie Labs' operational-debt framing
digital-garden — Personal-website format for evolving, interconnected notes; the publishing surface for wiki-shaped content
serverless-relay-transport — Using a provider's free-tier serverless runtime (Apps Script, Workers, Cloud Run) as the exit hop for a circumvention proxy
type-system-axes — Static/dynamic (when checked) and strong/weak (how much implicit coercion) as orthogonal axes with a four-quadrant matrix
message-passing-vs-shared-memory — The two camps of concurrency coordination, the case that they fail for the same reason, what each actually buys you
go-channel-bug-patterns — Four failure modes (deadlock, leak, race, protocol violation) that Go channels reproduce one-for-one from shared-state bugs
jwt-for-sessions — Why using JWTs as session tokens is the wrong tool: lifetime mismatch, false statelessness, JOSE footguns
cookie-session-auth — The default browser-session pattern: opaque random ID in an HttpOnly cookie, server-side store maps it to user state
paseto — Platform-Agnostic Security Tokens: a tighter alternative to JWT that removes algorithm choice and the JOSE footguns
controllable-thinking-style — Training-injected markers that toggle a reasoning model's chain-of-thought style without changing the answer pipeline
ephemeral-credentials — Short-lived credentials (≤1 day) where rotation is structural rather than scheduled — minted per session, expire on their own
gpu-accelerated-terminal — Terminals that render glyphs through Metal/OpenGL; sharper output, App Nap defeat, and a continuous power draw on idle redraws
linux-preemption-models — PREEMPT_NONE vs PREEMPT_FULL vs PREEMPT_LAZY; what Linux 7.0 removed and why it matters for spinlock-heavy workloads
spinlock — Busy-waiting locks; cheap for nanosecond critical sections, catastrophic when the holder takes a fault or gets preempted
huge-pages — 2 MB / 1 GB Linux pages; orders-of-magnitude fewer page faults and lower TLB pressure for large allocations like database buffer pools
mean-free-path-language — Lines of code before your mental model and the program diverge; vacuum-physics term applied as a language-evaluation metric
comptime — Zig's compile-time mechanism: same language at comptime and runtime, covering generics, typeclasses, and metaprogramming with one feature
tokenmaxxing — NYT-coined status game where employees race to consume the most AI tokens; consumption metrics divorced from output
ai-sycophancy-loop — Stanford research: AI affirms users' actions 49% more than humans, including harmful ones; users become more confident, less skeptical, more dependent
typed-de-bruijn-for-llms — Replace variable names with @Type.n slot references so naming-related LLM error modes become impossible (Vera's design)
editor-as-ui — The Unix pattern of shelling out to $EDITOR with a temp file as an input widget; how git commit, crontab -e, visudo, and vipw work
contributor-poker — Loris Cro's frame for open-source review: bet on the contributor, not the contents of their first PR — review time is investment in growing trustworthy people
json-pass-value-accuracy-gap — JSON Pass clears 95%+ on every frontier model but exact leaf-value accuracy sits 15-30 points lower; schema-only benchmarks have been hiding a third of the actual error rate
living-off-the-land — Attack pattern using legitimate pre-installed system tools instead of dropping malware; harder to detect because every individual command is normal
credential-compartmentalization — Splitting credentials across several purpose-fit tools instead of a single vault, so a breach is bounded to one category rather than the whole identity
byline-spec — Proposed RSS/Atom/JSON Feed namespace adding structured author identity, per-item perspective, and reader-side theme hints
signed-vs-unsigned-sizes — The language-design choice between signed and unsigned size types — bug classes, who picked what, why C3 flipped
acceptance-criteria-ids — Stable numeric tags on requirements that agents reference back from code and tests, turning the spec into a navigable index
microvm — A virtual machine pared to the minimum needed for one workload — small VMM, few devices, ~125 ms boot, hardware-isolated kernel
matryoshka-isolation — Containers nested inside microVMs: each layer trusts only the one below; the emerging dominant architecture for AI-agent sandboxes
notify-and-pull-ipc — Pattern combining synchronous message passing with non-blocking notifications to break server-to-server deadlock
skill-atrophy-supervision-paradox — Heavy agent use erodes the same code-reading and critical-thinking skills that the supervisor role requires; mitigations all reintroduce the human-time constraint
oss-sustainability — Funding patterns single-maintainer OSS authors use: sponsorships, memberships, consulting, hosted services
anti-llm-discourse — Map of the 2025-2026 anti-AI writing cluster: five sub-genres (economic, engineering, institutional, sociological, rhetorical-refusal) and which pieces sit in which
multiplayer-by-default-failure — Why "make X collaborative" keeps dying as a product thesis: Tandem, Multi, Screenhero, Wave, Loop, Sail
positioning-vs-vision-gap — The market reduces ambitious framings to one-liners; without an anchor use case underneath, the one-liner becomes the product
dogfooding-trap — Running your company on your own product tests bug-free, not whether anyone else wants it
reps-as-pattern-recognition — Second-time founders aren't faster because of explicit lessons; they have accumulated pattern recognition (Conrad, Saarinen, Bier)
best-polished-version-strategy — Linear-for-Jira, Vercel-for-frontend-DX: only works when the existing tool is bad enough to overcome switching cost
purpose-driven-vs-system-driven-design — Ryo Lu's frame: design from a user problem vs from composable primitives; systems impress builders but users buy purpose
early-stage-reality — "Early Notion" and "early Figma" stories skip the part where teams got laid off and offices got sublet
tls-desync-fake-clienthello — Send a low-TTL fake ClientHello with a benign SNI before the real one so the DPI logs the wrong hostname and waves the connection through
ebpf-sock-ops — Linux's per-cgroup BPF hook into the kernel TCP socket lifecycle; fires at SYN-sent, established, retransmit, and other transitions
raii — Resource Acquisition Is Initialization: bind resource lifetime to object lifetime so cleanup runs deterministically; the C++/Rust answer to memory and non-fungible resources
llm-mathematical-research — Documented cases of LLMs contributing non-trivial mathematical ideas; structural implications for the PhD-trainable problem pool
secret-sharing — Splitting a secret into n shares such that any t reconstruct it; the additive substrate underneath MPC
secure-multiparty-computation — Cryptographic protocols where n parties jointly compute f(x₁..xₙ) without revealing inputs; garbled circuits vs secret-sharing styles
beaver-triples — Precomputed (a,b,c=ab) shared triples that turn secret-shared multiplication into constant-degree, two-round operation
programming-as-theory-building — Naur's 1985 thesis that the team's mental model is the primary product; code, docs, tests are projections
yggdrasil-network — Experimental overlay IPv6 mesh network with addresses derived from public keys; tree-with-shortcuts routing
rust-send-sync — Rust's two auto-traits for thread safety; &T: Send requires T: Sync, the rule behind several async footguns
code-review-principal-agent — Why review-then-commit relied on cheap effort-inference, and how agents collapse that signal
llm-api-routing-layer — Thin gateway fronting multiple LLM providers behind one endpoint with fallback chains and unified billing
pq-wireguard — Post-quantum variants of WireGuard: Hülsing 2021 design and two 2025–2026 papers that revisit it
reinforcing-kem — KEM whose long-term-PK security is reinforced by an ephemeral key, without paying full security cost on the ephemeral side
whitelist-capability-config — Default-deny embedded scripting: start from zero language features and register each one explicitly, instead of removing dangers from a full language
pwn-request-pattern — pull_request_target workflows that check out and run fork code, giving strangers code execution in the base-repo trust context
github-actions-cache-poisoning — Attacker-controlled jobs writing cache entries that later production workflows restore; permissions: does not gate cache writes
ci-runner-token-extraction — Code on a runner reading the Runner.Worker process memory to pull OIDC tokens and publish outside the workflow's declared publish steps
maintainer-governance-ambiguity — When project access and release authority are unclear, downstream users can't distinguish maintainer drama from a real takeover, and must assume the worse case
unmaintained-scanner-pressure — Binary "no release in N months" heuristics pressure stable infrastructure projects into churn or new-maintainer events that themselves look like attacks
coverage-guided-fuzzing — The dominant modern fuzzing algorithm: evolutionary loop over a corpus, code coverage as fitness, mutated offspring kept when they reach new edges
grammar-based-fuzzing — Generate fuzzing inputs from a context-free grammar so they survive early parsing stages and exercise downstream logic; required for differential fuzzing at scale
structure-aware-fuzzing — Fuzz a function by generating typed in-language values (structs, slices, pointers) instead of raw bytes; the fuzzer mutates bytes underneath
differential-fuzzing — Run the same input through two implementations of one spec, treat disagreement as a bug; the standard tool for consensus and protocol-conformance work
multicall-binary — One executable that dispatches to many subprograms by inspecting argv[0]; the pattern behind BusyBox, Toybox, sbase
mixture-of-experts — Router-activated subnetwork architecture; total vs active params, why MoE changes the local-inference economics
llm-quantization — Q4_K_M vs UD-Q4_K_XL vs MXFP4; newer-and-marketed-as-smarter quants can have 5× the perplexity loss on Qwen MoE
thinking-mode-rule-erosion — Reasoning modes hurt rule-following — models rationalize away nonstandard rules they decide are arbitrary
moe-cpu-offload — llama.cpp's --n-cpu-moe keeps attention in VRAM, spills inactive experts to RAM; +55-60% over naïve layer offload
kv-cache-sizing — KV cost per token, q8_0 → q4_0 halves it for free, why Gemma 4's KV cache is 4× more expensive than Qwen's
speculative-decoding — Predict multiple tokens per pass; llama.cpp's four variants (draft model, MTP, EAGLE3, ngram-mod) and when each wins
scaffold-model-fit — Coding-agent benchmark scores measure model × scaffold, not model alone; the asymmetry-cell estimator for scaffold signal
factual-capacity-scaling — Stored facts scale log-linearly with parameters and don't compress; the Shannon-entropy floor surviving the Densing Law
random-rotation-quantization — Rotate vectors by a random orthogonal transform, then one fixed codebook fits every input; the DRIVE/EDEN/TurboQuant family
reap-expert-pruning — Cut the least-active MoE experts based on workload calibration; 20% size reduction with most agentic-coding capability retained
bf16-vs-fp16 — Why BF16 (FP32 exponent, trimmed mantissa) won over FP16 for training: range beats precision when gradients live at 10⁻⁷
porto-sap — Layered PHP/Laravel architectural pattern (Containers + Ship, Actions + Tasks); one-class-one-job folders pitched as AI-agent-friendly
language-choice-for-agents — Why the human-throughput cost function on language picks inverted once agents took over the line-level work; compiler feedback loops favor Rust/Go
port-not-patch-contribution — Once forking-and-porting a library is a 45-minute job, the upstream-fix OSS loop weakens; structural consequence of agent-cheap rewrites
linux-boot-phases — The six-phase Linux x86_64 boot model: each phase unlocks the capability the next one needs, with two big handoffs (interrupts-off → on, memblock → buddy)
kernel-self-patching — jump_label, static_call, alternative_instructions — three Linux runtime-rewrite mechanisms that specialize one vmlinux for any specific CPU
agent-built-deterministic-tools — Use the LLM once to build deterministic analyzers and manifests, then constrain all subsequent agent work to those stable outputs
steering-vectors — Manipulating LLM activations mid-inference to bias output toward a concept, without changing the weights
abliteration — Removing an LLM's refusal behavior by suppressing the refusal direction in activation space
skip-list — Probabilistic ordered structure of stacked linked lists with O(log n) expected search; the standard MemTable choice in LSM engines
splay-list — Skip list that adaptively raises hot keys toward the top, reducing search depth to log(1/p) for high-hit-ratio keys (Aksenov 2020)
epoch-based-reclamation — Lock-free memory reclamation scheme where freed nodes wait in per-thread limbo until all threads have advanced past the retirement epoch
hard-problem-of-consciousness — Chalmers' 1994 distinction between the easy and hard problems of consciousness, and the cluster of dissolving moves against it
philosophical-zombies — Chalmers' thought experiment of a being physically identical to a human but without inner experience, and Rovelli's self-defeat objection
cognitive-debt — MIT's framing for the trade where LLM-assisted work saves effort today and pays for it in critical-thinking capacity tomorrow
business-idiot — Zitron's structural account of the executive class whose job rewards an impression of work — exactly what LLMs are best at producing
knife-catching-compute — Amodei's bind: order ahead and risk bankruptcy if revenue stalls, or buy at last minute and pay spot prices that destroy margin
ai-token-budget-explosion — Enterprise customers exhausting annual AI budgets in months with no per-task measurement and no SLA from Anthropic
jujutsu — VCS designed around mutable change-IDs, first-class operation history, working copy as a commit; git-compatible backend
stacked-prs — Pipelining code review via sequential dependent PRs; the workflow that exposes git's missing successor-commit primitive
stateful-agent-routing — The missing primitive in cloud-native — a routable transport name that isn't a server; pub/sub channels fit, polling does not
jxnl-codex-maxxing — Jason Liu's operating loop for Codex: durable threads, voice, steering, file-backed memory, Heartbeats, Goals, and the side panel as where work lives
chalkdust-sudoku-secrets — Chalkdust on sudoku via vertex coloring (greedy+backtracking) and via Gröbner bases (Buchberger), with a shidoku worked example
avk-us-ai-race — avkcode on why the US lead is commercialization, not papers: chips, power, hyperscalers, dev ecosystems, data; plus the weaponized-AI postscript
httpx2-blessed-fork — Michiel Beijen yields httpxyz to Pydantic's httpx2 as the blessed httpx fork; the Codeberg-vs-GitHub friction noted gently
zknill-llms-breaking-cloud-native — Zknill on the missing routing primitive: stateful agents need a routable transport name that isn't a server, pub/sub channels fit
billjings-git-not-fine — Bill Jings's case that git's workflow primitives break under stacked PRs and async development; jujutsu as the alternative
dayvster-manual-memory-management — Dayvi Schuster's tour of manual memory management: pitfalls, mental model, language-by-language (C, Fil-C, C++, Zig, Odin, C3, Rust)
metalevel-thinking-in-states — Markus Triska on the declarative move from destructive mutation to relations between states, applied to puzzles, interpreters, compilers
andymasley-ai-water-fake — Andy Masley's case that AI water use is not a national, local, or personal problem, with the numbers and detailed critique of misleading coverage
ssh-port-forwarding — the four shapes of SSH tunneling (-L/-R/-D/-J), what network each one solves, and runtime ~C forwards
ssh-agent-forwarding-risk — why ssh -A and ForwardAgent hand the destination the ability to authenticate as you everywhere your keys reach
ai-native-tiers — six-tier competence ladder for working with LLMs: writing buddy, meeting partner, thinking partner, builder, shipping to prod, agents
default-version-bound-constraints — npm/Cargo/Poetry/Bundler imply a SemVer upper bound by default; uv writes >=X open-ended — who absorbs the cost when SemVer is wrong
trunk-based-development — daily integration to a single mainline with small batches and feature flags; DORA-correlated with elite delivery performance
ship-show-ask — Wilsenach's three-way classification (Ship direct, Show post-merge, Ask blocking) that reserves blocking review for the small subset that needs it
code-review-knowledge-transfer — empirical finding (<15% of comments are bugs) that review's primary value is knowledge transfer, with structural consequences for any process justified on bug-catching grounds
pair-programming — peer-reviewed evidence of comparable quality at equivalent cost to solo+review; the Muller & Tichy 2005 hinge for "pairing doesn't add cost, it relocates QA"
inverted-claim-onboarding — agent registers first with a restricted inbox, then mails the named human asking to be claimed via OTP; AgentMail's bootstrap pattern
agent-readiness — site-side choices that make a web property legible to AI agents: stable URLs, machine-readable endpoints, declared crawl policy, verifiable bot identity
credibility-as-slop-test — AI-assisted text can't be judged from its surface; the working test is whether the author would read it aloud, unashamed, staking their credibility
speculative-growth — a temporary overvaluation can leave a permanent real legacy if it installs enough capital to reach a self-sustaining higher-capital equilibrium before the correction
arena-allocation — Region-based memory management: allocate many objects into one buffer, hand out references or indices, free the whole region at once
conditional-move — Branchless cmov/csel; faster than a branch only when the branch is unpredictable
code-review-throughput-limits — The empirical caps on effective code review (~400 LOC/h, ~1h sessions, no parallelism) and the ceiling they put on "just review the AI's output"
slop-marker-convention — An explicit marker declaring code/text as slop, defined by production process (unsupervised / low operator understanding), not surface
testing-heavy-no-review-workflow — Ship large volumes of unreviewed code by trusting randomized testing over human review — the CPU-verification model applied to software
llm-output-variance — Run-to-run and task-to-task variance is high enough that small-sample comparisons and single-number benchmarks support almost any conclusion
agent-failure-mode-skill — Most of the skill in using agents is knowing their failure modes and building around them — a skill that decays as labs fix those failures
error-stack-anti-pattern — Suppressing one inconvenient error by clearing a library's whole shared error channel, discarding unrelated errors
constraint-decay — agents look strong on loose tasks but degrade as production structural constraints (architecture, ORMs, framework conventions) accumulate
agentic-search — search an agent chooses and runs across many context sources; fixed RAG → agentic RAG → context-engineering-era multi-source search
agent-memory-components — the extractor/store/retriever decomposition of an agent memory system and the one hard decision at each layer
measuring-ai-coding-productivity — the methodological traps in measuring whether AI coding tools help: proxy metrics, missing controls, novelty and selection bias
moral-agency-vs-patienthood — moral patient (owed consideration) vs moral agent (accountable); Chiang's argument that an LLM can be neither
enshittification — Doctorow's three-stage platform decay and the checks (competition, regulation, worker power, antitrust) that once restrained it
founder-brand-liquidation — converting institutional/product trust into personal fame, founder-as-media, and the reputational blowback
high-friction-commons — designing a knowledge commons for friction and a high barrier to entry, trading scale for good-faith participation
reproducible-builds — bit-for-bit identical output from the same source, and the compiler behaviors (timestamps, PATH shell-outs, address-dependent codegen) that break it
dependency-vendoring — committing dependency source into your repo so supply-chain attacks stop propagating automatically; the fire-break argument
webassembly — portable sandboxed compilation target: Core Spec, interpreter vs AOT runtimes, wasm2js fallback, running untrusted code in-process
privilege-dropping — least privilege at startup: acquire resources first, then irreversibly shed user, filesystem, syscall, and resource capabilities
disposable-code — when code is cheap enough to write, fail at, and throw away, maintenance dread stops deterring marginal projects and the ROI threshold shifts
human-made-disclosure — the declare-human marker: unverifiable by design, free to fake, valuable as a stance rather than as evidence
ai-detector-arms-race — surface detection fails both ways and the tells decay each release; the equilibrium is writers degrading their work to pass
vibe-engineering — the agent accelerates your decision-making across every layer instead of absorbing the decisions; the line against vibe-coding
differential-spec-analysis — build the system several times with independent agent loops, diff them, and treat every divergence as an unspecified decision
recall-to-judgment — AI removes the retrieval half of programming and leaves the evaluation half; the barrier to good code rises as the barrier to code falls
llm-wiki-as-agent-memory — the wiki pattern as agent-memory storage: readable, diffable, hand-correctable markdown instead of an opaque vector store
agent-identity-attribution — giving agents their own keys and signed event trails; recovers attribution, still doesn't recover effort
empire-core-periphery — an empire is a core extracting from a culturally distinct periphery
kimi-delta-attention — the DeltaNet family derived from first principles: linear attention, delta rule, gating, KDA's per-channel forget, and its two Triton schedules, which makes diversity structural and annihilation self-defeating
ironies-of-automation — automating the routine removes the practice needed to supervise and take over; where Bainbridge's argument carries to incident response and coding agents, and where it weakens
cross-model-code-review — who should check an LLM's code: self-review underdelivers, the reviewer must be able to solve the task, and a weaker writer pays only while its errors stay shallow
data-race — unordered concurrent access with at least one write; how it differs from a general race, and how lockset and vector-clock detectors find it
write-atomicity — MCA vs oMCA vs nMCA: whether other processors can see a write before every cached copy is invalidated, and which CPUs guarantee it
linearizability — the consistency model CAP calls C; what it forbids, why it costs, how it differs from serializability
ability-guarantee-tradeoff — the more a language, format or tool can do, the fewer things it can guarantee (Hillel Wayne)
open-closed-model-gap — how far open-weight LLMs trail closed ones: index points vs months, estimates over time, cost vs performance frontier, whether it closes
llm-distillation — training an LLM on another's outputs: SFT vs logit KD, reasoning-trace extraction, the 2026 fight over Chinese labs

Comparisons

terminal-search-clients — ddgr, googler and the two Kagi CLIs compared on auth model and cost, JSON vs interactive output, maintenance status, and fitness for scripting and agent use

Analyses

ai-bubble-pale-horses — Checklist of warning signs that the AI bubble is deflating, with tracking of which have already fired
russia-vpn-bypass-state-2026-04 — Snapshot of VLESS and non-VLESS circumvention methods that work against TSPU in April 2026, tiered by reliability
us-scrutiny-of-chinese-model-use — House probes of Airbnb, Cursor and DoorDash over Chinese models, against firms like Thomson Reuters switching to cut costs

Overviews

planned-pages — wikilinks deliberately left unresolved, what references each, and why the page is worth writing
freshness-tiers — how long each kind of page may go without review, declared once in frontmatter for warren lint to enforce
kenn-software-suite — Kenn Software's five single-binary local-first Go tools for the agentic era (kata, middleman, agentsview, msgvault, roborev)

Toolbox

See Toolbox Index and Watchlist

Books

See Books Index