Chrome 148+ uses OS-specific math libraries for functions like Math.tanh, creating OS fingerprints. Scrapfly replicates exact OS math behaviors (e.g., libsystem_m on macOS) to spoof browsers indistinguishably, matching host libm, CSS trig, and Web Audio down to bit-level precision.

Cyberpunk Comics, Manga and Graphic Novels

https://shellzine.net/wp-content/uploads/2020/02/cyberpunkcomicfeat.jpg
A curated list of cyberpunk works (1975–2016) spans Western comics, manga, and graphic novels, featuring *Akira*, *Blade Runner*, and *Ghost in the Shell*, exploring dystopia, tech, and societal critique. Themes include virtual reality, AI, and post-apocalyptic settings, with entries like *The Long Tomorrow* and *Transmetropolitan* highlighting genre evolution and speculative futures.

Tiny Emulators

https://floooh.github.io/tiny8bit-preview/ext/visual6502remix.webp

So you want to learn physics (second edition, 2021)

A self-study physics guide created to help non-traditional learners has aided over 600,000 people, offering a structured curriculum from undergraduate to graduate levels. It emphasizes problem-solving, accessible resources, and addresses gaps in formal education, inspiring self-directed exploration of physics through comprehensive textbooks and practical advice.

Designing and assembling my first PCB

https://vilkeliskis.com/b/2026/i/pcb_breadboard_components.jpg
The user designed and assembled a custom BME280 sensor PCB using KiCad, successfully replicating its I2C interface and achieving plug-and-play functionality after soldering. This hands-on experience with PCB design, assembly, and testing boosted their confidence to tackle more complex hardware projects like integrating an ESP32, LCD, and sensors on a single board.

Ask HN: What Are You Working On? (July 2026)

Working on iOS app improvements and multiplatform C projects with X11/wasm. LLM coding diminished coding's joy, sparking thoughts of a new hobby.

Billion Dollar PDFs

https://billiondollarpdf.com/og.png
A Billion Dollar PDF is a document — a memo, a deck, a whitepaper, a thread — that crystallized a narrative at exactly the right moment and moved billions of dollars of capital around the field. The term is Jeremy Giffon’s. This is an open index of the ones people keep citing — browse it below, expand any entry for the story, and follow the link to read the original.

Migrating a production AI agent to GPT-5.6: 2.2x faster, 27% cheaper

https://storage.googleapis.com/ployai/d21bf4ad-2458-43ee-9561-54c28ab0e85f/user/98bfdea9-ai-generated-1783640379759.webp
Ploy upgraded to GPT-5.6 Sol, outperforming Claude Opus with 2.2× faster builds, 27% lower costs, and cleaner code, but required fixing tool-call handling, caching, and reasoning replay to address model-specific behaviors and ensure fair evaluation.

Claude Code sends 33k tokens before reading the prompt; OpenCode sends 7k

https://systima.ai/images/claude-code-vs-opencode-banner.png
Claude Code incurs ~33,000-token overhead per request (vs OpenCode’s ~7,000) due to extensive system prompts, tools, and mid-session rewrites, while OpenCode’s stable, cached prefixes minimize costs. Subagent workflows amplify Claude Code’s token usage 54x vs OpenCode’s lean design, with configuration files and MCP servers further inflating baseline costs.

Old and new apps, via modern coding agents

https://terrytao.wordpress.com/wp-content/uploads/2026/07/image.png?w=613
A mathematician successfully migrated 1999 Java applets to JavaScript via AI, reviving tools like honeycomb and Besicovitch set visualizers with minimal bugs. They also created new interactive visualizations for special relativity and the Gilbreath conjecture using AI, finding the process efficient and low-risk for supplementary educational tools.

How we can reduce traffic congestion

https://storage.googleapis.com/gweb-research2023-media/images/NetworkAwareRouting1_StudyOverview.width-1250.png
Google's study shows rerouting 2% of trips via navigation apps in 10 US cities improved traffic speeds by 2% and reduced emissions, demonstrating scalable, low-cost congestion relief. The experiment establishes a framework for network-level traffic optimization using connected systems.

Kode Dot Programmable pocket device for makers, pentesters and geeks

https://cdn.shopify.com/s/files/1/1033/7779/7462/files/kode-dot-front-back.png?v=1782403366&width=1152
Kode Dot is a handheld, all-in-one platform with dual ESP32-P4/C5 MCUs, AMOLED touchscreen, sensors, wireless, and 16 programmable pins for instant prototyping. It supports Wi-Fi, Bluetooth, NFC, and community-developed apps, born from a 16,000-backer Kickstarter success.

Architecture Description Languages [pdf]

The paper addresses the lack of consensus on architecture description languages (ADLs) by defining and classifying them. It compares existing ADLs, highlighting their strengths and deficiencies to guide future research.

A Peek Inside Jim Henson's Creature Shop, Where Whimsical Puppets Are Designed

https://th-thumbnailer.cdn-si-edu.com/sK34FknQ6mq3fP61GVedAc158y0=/1026x684/filters:no_upscale():focal(1000x567:1001x568)/https://tf-cmsv2-smithsonianmag-media.s3.amazonaws.com/filer_public/e4/9a/e49a498f-f8fd-4afd-8d3c-66831cf87af2/entrance_puppets.jpeg
Standing in front of a small crowd of Muppet enthusiasts, actress and Sesame Street veteran Jennifer Barnhart demonstrated how to perform with three different types of puppets. First came Tillie, a small hand-and-rod puppet with an orange mane and a passion for gardening, despite her seasonal allergies. A rod puppet shaped like a hot dog popped up next, lamenting the winter gear he’d packed ...

LARP – Revenue infrastructure for serious founders

LARP is a satirical platform mocking legal circular revenue deals where companies mutualize revenue without cash movement, critiquing financial reporting's focus on numbers over real economic value. It highlights how accounting rules allow such practices, blurring lines between legitimate growth and illusion.

Why write code in 2026

https://softwaredoug.com/assets/media/2026/cheat-at-search-social.png
Building a software factory requires agents for automation and humans to maintain architectural integrity through active code engagement. Writing code fosters ownership, precision, and system resilience, preventing fragility amplified by over-reliance on AI-generated solutions.

MacKenzie Scott's Giving, in Quality-Adjusted Life Years (QALYs)

https://maxghenis.com/_astro/og-mackenzie-qaly.CAkykvwl.png
Skeptical weights each effect by how well its study identifies causation. Credulous trusts every cited effect at face value. Central value (mode) of a 0.55–1.10 triangular draw for the share of the studied effect a marginal unrestricted grant delivers — the whole distribution is sampled, not this number alone. Real 2026 dollars. Default inflates each year's gifts ($26.39B nominal, ...

I Learned to Read Again

https://substackcdn.com/image/fetch/$s_!3AdV!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F23985988-01ea-4d7b-a4f5-446c2a4f0a0b_1600x900.jpeg
Sam Kahn traces his declining reading habit due to social media, work, and shifting cultural values, battling digital addiction and societal pressures. He revives reading through disciplined strategies, stressing its role in sustaining deep thought and cultural continuity amid modern distractions.

Profiling the "Abundance" housing bottleneck with real data

https://i.snap.as/wjoTv1yV.png
The book "Abundance" argues regulatory bottlenecks ("chosen scarcity") hinder progress in housing and energy, but the user's analysis of cities like Austin and San Francisco shows mixed results. Vienna's success with public housing reveals alternative solutions, emphasizing the need to verify claims with data and consider multiple approaches.

Circular Obstacle Pathfinding (2017)

Mar 2017 The A* pathfinding algorithm is a powerful method for quickly generating optimal paths. Typically, people demonstrate A* navigating grid-based maps, but A* isn’t just a grid algorithm! It can work on any graph. We can use A* to find a path through this world of round obstacles. How does the same algorithm solve both problems? Let’s start with a review of how A* works. The A* ...

Automation Without Understanding

https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png
The essay warns that AI's mathematical advancements paired with U.S. declining math education risk strategic failure by eroding critical infrastructure. It proposes treating math capacity as a strategic asset and requiring formal, auditable AI reasoning.

Mechanistic interpretability researchers applying causality theory to LLMs

Your action triggered a security block, possibly due to specific phrases, SQL commands, or malformed data. Contact the site owner with your activity details and the Cloudflare Ray ID for resolution.

Against Usefulness

https://substackcdn.com/image/fetch/$s_!PChk!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F568f9825-5d39-466d-919c-45f7811bf475_1536x2048.jpeg
The essay explores how foundational, often "useless" research fuels future innovations, exemplified by Folk Computer, an open-source physical computing system. It calls for independent thinkers and patrons to support paradigm-shifting work, bridging the gap between institutional legacy and grassroots creativity.

I love LLMs, I hate hype

The author is enthusiastic about AI progress, including LLMs and coding agents, but criticizes fear-mongering narratives of AI dominance. They argue that value creation, driven by Moore's law, isn't captured by closed labs, advocating for open-source solutions.

Deir El-Medina Strikes

https://upload.wikimedia.org/wikipedia/commons/thumb/2/21/Deir_el-Medina_2016-03-23c.jpg/250px-Deir_el-Medina_2016-03-23c.jpg
The Deir el-Medina strikes (c. 1158 BC) were Egypt's earliest recorded labor protests, driven by unpaid grain rations under Ramesses III, with artisans demanding supplies from the vizier. The skilled, literate workforce, organized into roles like scribes and deputies, leveraged their royal tomb-building role to negotiate, but ongoing grain shortages linked to Sea Peoples' invasions and ...

A Speed Limit for Computers

Ivan Illich argued that excessive energy use harms equity, a principle reflected in e-bikes' legal speed limits (25 km/h) that balance access and fairness. In contrast, computing's unchecked power growth, unlike e-bikes' restrained design, exacerbates inequality and autonomy loss.

The One-Step Trap (In AI Research)

The one-step trap assumes AI can accurately predict long-term outcomes by iterating one-step models, but errors compound and computational complexity grows exponentially, making it unreliable. Temporally abstract models like options and GVFs offer a better solution for long-term predictions.

Show HN: Agent Draw: An agent draws while you talk, built on TLDraw

https://techstackups.com/img/articles/tldraw-agent-draw/cover.jpg
The user built a tool called Agent Draw that uses an AI agent to draw on a shared canvas while a user presents, utilizing the tldraw infinite-canvas SDK for React and an official Agent starter kit. The tool captures audio, transcribes it, and uses the transcription to draw on the canvas, with the agent finishing the entire drawing in one call and the user able to queue up multiple captures.

Why Vanilla JavaScript

This blog is mostly expression of my ideas on different things in programming and IT culture

Flash-MSA: Accelerating Million-Token Training with Sparse Attention Kernels

https://nanduruganesh.github.io/flash-msa/plot.png
The author introduces the first open-source training kernels for Minimax Sparse Attention (MSA) in CuTeDSL, optimizing block-based selection with max-pooling, GQA groups, and efficient caching for Hopper/Blackwell GPUs. Key innovations include fused backward passes, reduced register usage, and scalable CP strategies, though challenges like low occupancy and tensor-pipe bottlenecks remain.