Show HN: Getting GLM 5.2 running on my slow computer

412 vforno 2026-07-09 19:39:36 UTC github.com/justvugg Qwen Share 105

colibrì runs GLM-5.2 (744B MoE) on 25 GB RAM via int4 quantization, disk-streamed experts, and C-based zero-dependency engine, achieving lossless inference with speculative decoding and adaptive caching. It leverages AVX2, MTP heads, and batched expert reads, but requires NVMe SSDs and careful hardware management due to disk I/O and wear risks.

EU Parliament greenlights Chat Control 1.0

1023 rapnie 2026-07-09 11:39:36 UTC patrick-breyer.de Llama Share 493

The European Parliament allowed the interim "Chat Control 1.0" regulation to pass, permitting mass scanning of private communications until 2028. Negotiations for a permanent regulation will resume in September, with the EU Parliament pushing for targeted detection orders and stricter security standards.

Focus

28 iacguy 2026-07-10 01:39:36 UTC boz.com Qwen Share 8

The author reflects on Facebook's intense startup era, marked by extreme workloads and relentless focus, contrasting it with later dilution of priorities as the company grew. They argue that even well-intentioned diversions from core goals—like supporting nonprofits—accumulate hidden costs, eroding efficiency and clarity over time.

GPT-5.6

1083 logickkk1 2026-07-09 17:39:36 UTC openai.com Qwen Share 790

OpenAI launches GPT-5.6 (Sol, Terra, Luna), with Sol leading in coding, cybersecurity, and science efficiency, outperforming prior models and competitors. Enhanced ultra mode, cost-effective tiers, and robust safeguards boost performance-per-dollar and safety for diverse professional tasks.

Show HN: 18 Words

840 pompomsheep 2026-07-09 13:39:36 UTC 18words.com Full Share 288

Play 18 Words: a fast daily word challenge. Solve each scrambled word before the timer runs out and see how many of 18 you can survive.

Star Just Ate a Planet, and It's Not Done Yet

15 wglb 2026-07-10 01:39:36 UTC nytimes.com Full Share 15

Please enable JS and disable any ad blocker

Train sim created by just one person is being called the best ever made

310 oumua_don17 2026-07-09 17:39:36 UTC kotaku.com Qwen Share 116

Running Train, a hyper-realistic train sim by solo dev Novatetsu Games, blends fictional Japanese landscapes with intricate details like logical powerlines and weather effects, offering immersive gameplay through precise controls or free camera exploration. Praised for its stunning realism and dedication to detail, it’s hailed as a top-tier simulation with plans for expanded routes and features.

Interview with Mitchell Hashimoto about Ghostty and Zig

135 veqq 2026-07-09 21:39:36 UTC alexalejandre.com Qwen Share 51

Alex Alejandre discusses transitioning from software engineering to building Ghostty, a feature-rich terminal emulator addressing ecosystem gaps with protocols like n-screen and button APIs, emphasizing open-source principles and low-level systems understanding. He critiques modern tech complexity, advocates for Zig's potential, and stresses open-source maintainers' autonomy while balancing ...

My Story of 3D Realms / Apogee Part I (2020)

35 Michelangelo11 2026-07-09 23:39:36 UTC joesiegler.blog Qwen Share 1

The author recounts their 17-year tenure at Apogee/3D Realms (1992–2009), detailing roles in community management, game distribution, and early online innovations like virtual tours and DRM. They also outline the company's complex name/ownership history and note broken legacy site links.

Hy3

398 andai 2026-07-09 15:39:36 UTC hy.tencent.com Qwen Share 86

Hy3 outperforms similar-size models and rivals larger open-source models in productivity tasks, achieving 2.67/4 in expert evaluations. It addresses stability, hallucination, and context retention issues, is open-sourced under Apache 2.0, and reduces API costs.

Postgres rewritten in Rust, now passing 100% of the Postgres regression tests

404 SweetSoftPillow 2026-07-09 12:39:36 UTC github.com/malisper Llama Share 412

pgrust targets Postgres 18.3 compatibility with over 46,000 regression queries. It uses Rust and AI-assisted programming to explore deeper server changes.

Triple Dragon Fractal (2020)

17 nhatcher 2026-07-10 00:39:36 UTC paulbourke.net Full Share 3

Created by Paul Bourke December 2020 This image is a visualisation of how the following series behaves for initial values z0 at each point on rectangular region of the real-imaginary plane. The colour represents how fast the series converges to a fixed point, or diverges to infinity (which it doesn't actually to for points in the bounded part of the complex plane in the figures here).

No leap second will be introduced at the end of December 2026

244 ChrisArchitect 2026-07-09 14:39:36 UTC datacenter.iers.org Qwen Share 184

The IERS confirms no leap second will be added at the end of December 2026, keeping UTC-TAI at -37 seconds. Leap seconds may occur in June or December based on UT1-TAI, with updates provided every six months via Bulletin C.

The glass backbone: Why the Army's logistics will break in the next war

301 baud147258 2026-07-09 13:39:36 UTC mwi.westpoint.edu Qwen Share 387

The US Army's efficiency-driven sustainment model, optimized for permissive environments, is ill-suited for modern multidomain warfare, where survival under persistent attack determines victory. Historical and contemporary conflicts show that centralized logistics collapse under attrition, demanding decentralized, survivable sustainment networks to match industrial-scale combat endurance.

A road to Lisp: Why Lisp

119 silcoon 2026-07-09 15:39:36 UTC scotto.me Qwen Share 117

Lisp's macros, code-as-data structure, and interactive REPL enable powerful language extensibility and dynamic programming, fostering a unique problem-solving approach. Mastering Lisp transforms programming by allowing developers to shape the language itself, creating domain-specific tools and adaptable systems through its homoiconic and extensible nature.

Build your own vulnerability harness

10 ianrahman 2026-07-10 01:39:36 UTC blog.cloudflare.com Qwen Share 4

The blog outlines a model-agnostic security architecture using interchangeable AI models in a multi-stage pipeline (Recon, Hunt, Validate) for enterprise codebases, enabling cross-repo vulnerability detection and reducing false positives through adversarial validation. It emphasizes orchestration over single agents, leveraging deterministic code and LLMs to trace dependencies, generate proofs ...

Building a real-time AI tutor for 5-year-olds

49 catalinvoss 2026-07-09 21:39:36 UTC ello.com Qwen Share 61

Building a real-time AI tutor for kids 4-9 requires sub-second latency, pedagogical action streaming, and dual-agent architecture (converser for interaction, planner for lesson flow) to balance speed, safety, and adaptive teaching. Custom systems handle safety checks, parallel execution, and predictive responses to maintain engagement without compromising learning outcomes.

Girls just wanna have fast MPMC queues with bounded waiting

139 EvgeniyZh 2026-07-09 15:39:36 UTC nahla.dev Qwen Share 31

The author designed a lock-free queue using atomic counters and state/data buffers, initially mistakenly labeled as wait-free; it ensures bounded wait times but risks overflow with extreme usage. Benchmarks show good scalability, though performance peaks vary with core utilization, and drivable operations help mitigate head-of-line blocking.

Launch HN: Context.dev (YC S26) – API to get structured data from any website

74 TheYahiaBakour 2026-07-09 16:39:36 UTC context.dev Qwen Share 55

Context.dev offers an API for web scraping, brand data extraction, and logo retrieval, with plans from Free to Enterprise. It includes self-service sign-up, a startup discount, and supports AI agents with tiered pricing and scalable credits.

A possible future for Damn Interesting

243 mzur 2026-07-09 15:39:36 UTC damninteresting.com Qwen Share 31

Alan Bellows, founder of Damn Interesting, is launching a fundraiser to regain time for the project after transitioning to a full-time job, which has limited his ability to write and manage the site. The goal is to counter the influx of AI-generated content by funding more long-form articles, with donations separate from regular operational expenses.

Why American ambulance rides are so expensive

100 jyunwai 2026-07-09 22:39:36 UTC davidoks.blog Qwen Share 136

A 25-year-old man faced a $12,873 surprise ambulance bill after a minor injury, exposing the U.S. system's reliance on outdated Medicare-based per-ride fees, which force ambulance services to extract costs from privately insured patients through out-of-network charges. This flawed structure, rooted in high fixed costs and underfunded Medicare/Medicaid reimbursements, creates exorbitant bills, ...

Cache-Conscious Data Layout in Rust: Field Zoning, False Sharing, 128-Byte Rule

8 eigenBasis 2026-07-10 01:39:36 UTC debasishg.github.io Prefix Share 1

Part 1 of Low-Level Systems Design in Rust - a series on writing high-throughput, low-latency systems code, using a single-producer / single-consumer (SPSC) ring buffer as the running example. Part 0 - Architectural Decomposition made the highest-leverage decision (remove contention structurally, so every writer gets its own ring) and holds the full series index and guiding principles. This ...

Muse Spark 1.1

329 ot 2026-07-09 14:39:36 UTC ai.meta.com Full Share 173

Sorry, something went wrong. We're working on getting this fixed as soon as we can.

Patterncollider: Generate and explore quasiperiodic tiling patterns

24 tobr 2026-07-09 22:39:36 UTC github.com/aatishb Qwen Share 1

Pattern Collider generates quasiperiodic tilings using de Bruijn's multigrid method, enabling custom, shareable patterns. It explores non-repeating motifs via intersecting lines, offering controls for symmetry, disorder, and visualization, with references to Penrose tilings and quasicrystals.

Wildcard (YC W25) Is Hiring a Founding Engineer

0 2026-07-09 17:39:36 UTC ycombinator.com Qwen Share

Wildcard, a fast-growing agentic commerce platform, seeks a Founding Engineer to build infrastructure and product for AI-driven retail optimization. The role requires owning end-to-end systems, leveraging AI tools, and shaping product direction with direct impact on company success.

Buried Apple feature turns an iPhone into the perfect kids' dumb phone

281 PotatoNinja 2026-07-09 16:39:36 UTC wired.com Qwen Share 165

A parent uses iOS 17's Assistive Access to create a secure, limited-app "dumb phone" for their child, blocking internet/social media while retaining tracking and essentials. The setup repurposes old devices affordably, though Apple undermarkets this feature despite its child-safety benefits.

GLM 5.2 is nearly as accurate as a human book keeper

189 adamkurkiewicz 2026-07-09 18:39:36 UTC toot-books.pages.dev Qwen Share 113

GLM 5.2 prepared a nearly perfect UK SME VAT return with 59 transactions in 68 minutes at 2.73 USD, achieving a 7p net error but misclassifying share capital and confusing VAT categories. Despite minor errors, the model demonstrated AI's potential to automate bookkeeping, reducing costs compared to human accountants.

Opinionated and easy Pi.dev configuration

115 lwhsiao 2026-07-09 15:39:36 UTC lazypi.org Full Share 62

One command installs 60+ community skills, 67 themes, MCP support, sub-agent support, Claude Code CLI provider support, memory, and more. You're left with a curated starting point that gets you an exciting, useful experience immediately, without the research and configuration tax.

SimPolitics: America’s quest to solve politics with computers

84 mckelveyf 2026-07-09 14:39:36 UTC mitpress.mit.edu Full Share 13

You don't have permission to access "http://mitpress.mit.edu/9780262053198/simpolitics/" on this server.

TLS certificates for internal services done right

134 mrl5 2026-07-09 15:39:36 UTC tuxnet.dev Qwen Share 97

The article explains securing internal services (like Grafana) with split-horizon DNS and a reverse proxy (nginx) using ACME (Let's Encrypt) certificates, avoiding self-signed cert issues. It leverages a WAF via nginx and NetBird's Custom Zones to restrict access, ensuring TLS works seamlessly for both internal and external clients without manual client-side trust configurations.