GLM 5.2 beats Claude in our benchmarks

https://semgrep.dev/assets/semgrep-benchmarking-summary-with-padding.png
GLM 5.2, an open-weight model, beat Claude Code at IDOR detection with 39% F1, costing $0.17 per vulnerability found. The harness still matters more than the model, but open-weight models have crossed a threshold worth watching for security tasks.

I used Claude Code to get a second opinion on my MRI

https://antoine.fi/rails/active_storage/representations/proxy/eyJfcmFpbHMiOnsiZGF0YSI6NTU5LCJwdXIiOiJibG9iX2lkIn19--14f45f75a8f8216447265425e1b3cebe1d211a37/eyJfcmFpbHMiOnsiZGF0YSI6eyJmb3JtYXQiOiJwbmciLCJyZXNpemVfdG9fbGltaXQiOlsyMDQ4LDE1MzZdfSwicHVyIjoidmFyaWF0aW9uIn19--c0bbe91c3d5655e568ecfbbb413acaec2c5c367e/image.png
The author used Opus 4.8 to analyze an MRI and got a second opinion on their diagnosis, which disagreed with the original doctor's report. The AI analysis found no partial-thickness tear, contradicting the doctor's diagnosis and making the author question the treatment plan.

5k menus from the New York Public Library’s Buttolph Collection (1880-1920)

https://pudding.cool/2026/06/menu-story/assets/social-facebook.jpg
What do America’s earliest restaurant menus teach us about America?

Historical memory prices 1960-2026

The interactive dataset tracks historic and current memory and storage prices, including DRAM, NAND flash, and HBM, with modeled estimates from Epoch AI for AI-accelerator costs. The data is downloadable and includes quarterly updates for HBM and monthly updates for DRAM and NAND prices from 1957 to the present.

TOP500 at ISC'26: We Have a New Number 1 – By George Cozma

https://substackcdn.com/image/fetch/$s_!rXdh!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F11672287-1e72-4e2b-9220-bd12de3d8ab4_1200x900.png
The LineShine Supercomputer in China has taken the top spot on the 67th TOP500 list with 2.198 Exaflops of sustained FP64 performance. This is the first Chinese submission to the TOP500 in 9 years and marks a significant milestone for China's HPC capabilities.

Librepods: AirPods liberated

https://raw.githubusercontent.com/librepods-org/librepods/main/imgs/banner.png
LibrePods is a project that allows non-Apple devices to use AirPods features. It implements the proprietary protocol used by Apple devices to exchange data with AirPods.

The Boeing 747 begins its final descent

https://cdn.theatlantic.com/thumbor/HIoqct7cC0eypfzBtLSkiNp1U_A=/0x79:4932x2853/1440x810/media/img/2026/06/11/0726_WEL_Bogost_747_16x9/original.png
The author visits Pinal Airpark in Arizona to see retired Boeing 747s, once the principal host of important journeys, now a symbol of American decline. The 747 was a technological innovation that embodied American might, invention, and progress, but its accidental longevity defined an era of decline.

Ford rehires 'gray beard' engineers after AI falls short

https://techcrunch.com/wp-content/uploads/2026/06/GettyImages-2255983122.jpg?w=1024
Ford is reviving its quality control by rehiring technical specialists to train younger staff and reprogram AI tools. This move is expected to save the company $1 billion in costs this year.

Working around dragons with the Lemote Yeeloong laptop and OpenBSD

https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgwr0Vsnj1UuAJWn229_mBMPRg6E4JpX_jhbOUQTo00gTSHpVXRCDmY3SzMr9Qs46tAMdlzSXJFn5C8PLbPkKtn4_GtyIB-ytyD_h9b9u0-r0S_fL4IVoMLVt7RBO9OkQZ3ahw4p1ybokI3ucnJ3avc2dkFUGatYUTsPfvFEY2-cazKJSjes73VfqThHZ8/s320/stallman-lemote-2.JPG
The Loongson processor is a Chinese-developed CPU that emerged in 2001, initially based on MIPS architecture, and was later developed into a 64-bit processor. The Loongson processor was used in the Lemote Yeeloong laptop, a low-cost netbook that was released in 2008.

Tokenmaxxing is dead, long live tokenmaxxing

https://substackcdn.com/image/fetch/$s_!rp_Y!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd9ca712f-fd31-4c79-94b3-5a0ebe81bdb3_800x786.jpeg
The concept of tokenmaxxing, where companies spend large amounts of money on AI tokens without expecting a return, is not dead but rather evolving as companies realize the benefits of compounding correctness, where more tokens spent lead to better results. The shift towards open model platforms and generalist agents will lead to another rise in tokenmaxxing behavior, ultimately resulting in ...

Show HN: Bash4LLM+ – A lightweight, dependency-free Bash wrapper for LLM APIs

https://raw.githubusercontent.com/kamaludu/bash4llm/main/docs/img/bash4llm320.png
Bash4LLM⁺ è un wrapper CLI sicuro per l'API Chat Completions di OpenAI, scritto in Bash e completamente auditabile. È un singolo script auto-contenuto che può essere scaricato, eseguito e utilizzato subito.

Computer-Aided Language Development in Nonspeaking Children (1968) [pdf]

https://archive.org/services/img/opensource
EMBED (for Archive.org item Description fields) [archiveorg colby1968-computer-aided-language-development-in-non-speaking-children width=560 height=384 frameborder=0 webkitallowfullscreen=true mozallowfullscreen=true]

Show HN: Zanagrams

https://zanagrams.com/og-image.png
Zanagrams is a free daily word puzzle. Drag across the letters to find the hidden words and watch the grid shrink as you solve it. A new Zanagrams puzzle every day.

Show HN: NanoEuler – GPT-2 scale model in pure C/CUDA from scratch

https://opengraph.githubassets.com/2418b4e6023a573e4ec14ae93b78f6c20eb189ce2dce8b93f1a161f7263a1fb5/JustVugg/nanoeuler
A from-scratch text generator built in C/CUDA with a ~116M-parameter model trained on a single RTX 4070 GPU produces fluent-looking English with little real knowledge. The project demonstrates a complete, understandable training pipeline and the pretrain→SFT pipeline works end to end, but is not a capable assistant.

Daisugi, the Japanese technique of growing trees out of other trees (2020)

https://cdn8.openculture.com/2020/10/22225805/Daisugi.png
Daisugi, a Japanese technique, involves growing multiple trees from a single tree, creating a giant bonsai-like structure, but it's actually a form of coppicing, a common woodland management technique practiced for centuries. This method produces straight, usable timber without harming the original tree, and it's being re-emphasized as a sustainable way to manage woodlands.

Examining circuit boards from the Space Shuttle's I/O Processor

https://static.righto.com/images/shuttle-ioprocessor/gpu-and-iop-w600.jpg
The Space Shuttle's I/O Processor had 24 network connections and 25 virtual processors, implementing two different instruction sets. It used microcode to run the virtual processors on one physical processor.

Staying Awake (2008)

https://wp.harpers.org/wp-content/uploads/2019/11/9-Harpers-FINAL-480x350-c-default.jpg
Some people lament the disappearance of the spotted owl from our forests; others sport bumper stickers boasting that they eat fried spotted owls. It appears that books, too, are a threatened species, and reactions to the news are similarly various. In 2004 a National Endowment for the Arts survey revealed that 43 percent of Americans polled hadn’t read a book all year, and last November, in ...

The US Used to Demand the Best Tech. Now We Ban It

https://i.pcmag.com/imagery/articles/04ltqIf4ihFMDfo38F0qNgS-1.fit_lim.v1782561174.png
The US is being left behind in the electric vehicle market due to restrictive regulations and tariffs that prevent affordable Chinese EVs from being imported. American companies like Aptera are struggling to compete with Chinese manufacturers like BYD and Xiaomi, which have invested heavily in vertically integrated supply chains and battery factories.

Show HN: DRM-Free Books

https://frequal.com/novels/mam/cover-med.jpg
Multiple stories are presented, each with a main character facing challenges and obstacles. The stories involve mystery, action, adventure, and romance, set in various locations and time periods.

A way to exclude sensitive files issue still open for OpenAI Codex

https://opengraph.githubassets.com/eb6eaf7d4028ea024171759a9f016eafa52d89840c3d40660be2d1d74b98731a/openai/codex/issues/2847
What feature would you like to see? A mechanism to explicitly mark files/paths that the agent must not read or send to the model, at both repository and global levels (e.g., a repo-local .codexignore plus a global ignore file). Example: ...

The curious case of the disappearing Polish S (2015)

https://aresluna.org/images/the-curious-case-of-the-disappearing-polish-s/ogimage.png
A Polish user reported a bug on Medium where Ś wouldn't appear when typed. The issue was caused by Medium blocking CtrlS to prevent browser save dialogs, but Polish users use AltS to type Ś.

The cost YAGNI was never about

https://substackcdn.com/image/fetch/$s_!hzhP!,f_auto,q_auto:best,fl_progressive:steep/https%3A%2F%2Ftidyfirst.substack.com%2Ftwitter%2Fsubscribe-card.jpg%3Fv%3D-1150222451%26version%3D9
YAGNI is about timing, not thrift, and warns against building speculative structure before a feature is needed. It's about preserving optionality and not committing to costly changes when the future is uncertain.

The MUMPS 76 Primer – anniversary edition

https://opengraph.githubassets.com/3b5ce3c9017816b1fe9520f48c5ceaaf3345ff010c204a3fce8afd6d3c09a6f8/rochus-keller/MUMPS
MUMPS is a programming language and database system created in 1966 for hospital information systems. It has a built-in hierarchical database with a schema-free, ordered key-value store.

Programmable Probabilistic Computer with 1M p-bits

https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png
Researchers built a large Ising machine with one million p-bits by networking FPGAs, achieving over a trillion flips per second. The machine's performance depends on a timing ratio of boundary-exchange frequency to local p-bit update frequency.

The KIDS Act would require age checks to get online

https://www.eff.org/files/banner_library/ageverificationbanner-3.png
The KIDS Act package includes bills that require age verification, government-directed moderation, and new rules for private communications. This could lead to restrictive age-checking practices, reduced online privacy, and limited free expression.

EU to legislate about Chat Control behind closed doors

https://www.patrick-breyer.de/wp-content/uploads/2024/01/240130_Breyer-Alternativen-engl.png
Dr Patrick Breyer warns of a double-attack on secure messaging in the EU, with President Roberta Metsola pushing for mass surveillance and EU governments rushing a new scanning mandate. Citizens can email EU lawmakers and government representatives through fightchatcontrol.eu to stop this threat.

More evidence is consistent with possible ancient life on Mars (2025)

https://i.cbc.ca/ais/1.7649703,1759672248000/full/max/0/default.jpg?im=Crop%2Crect%3D%28109%2C140%2C2415%2C1358%29%3B
Scientists have found potential evidence of life on Mars, but no conclusive proof, with the latest discovery being two chemicals formed by microbial activity or chemical reactions. The possibility of life on Mars has been tantalizing humans for over a century, with many previous claims debunked, but new research suggests life could exist underground.

Do LLMs pass the mirror test?

The mirror test — Gallup's original, the one with the red dot on the chimp's forehead — has been adapted for LLMs several times in the past, but as far as I'm concerned, every adaptation gets it wrong in very similar ways: they build visual mirror tests translated into text. Show the model its own output and ask "is this yours?" or have it identify its responses among an ...

Marfa Public Radio Puts You to Sleep

https://npr.brightspotcdn.com/dims4/default/1ba947b/2147483647/strip/true/crop/3000x3000+0+0/resize/280x280!/quality/90/?url=http%3A%2F%2Fnpr-brightspot.s3.amazonaws.com%2Fe6%2F56%2F79316759410d8ce45efb655b32b3%2Fmpr-sleep-v2-logo-1.jpg
Marfa Public Radio creates a sleep podcast reading boring documents to fundraise for the station. Listeners can donate at marfapublicradio.org/donate to keep the station awake.

Show HN: Appaca – AI Workspace for Operators

https://assets.appaca.ai/graphic/usecases/lead-followup-assistant-1.png
Companies can use AI to automate tasks such as sales follow-ups, invoicing, and reporting, freeing up time for more important work. AI apps can also handle tasks like content creation, HR paperwork, and IT support, reducing administrative time and increasing productivity.