H3-metal – Native MiniMax-H3 inference for Apple Silicon

273 swyx 2026-08-11 02:02:02 UTC github.com/antirez GoogleT5 Share 53

Native minimax-h3 inference for apple silicon. specialized 256-thread kernel gathers and quantizes each h3 row directly into the projection's row-major int8 buffer, eliminating the intervening full-width bf16 transpose without changing any output byte. cross-m5 measurements improved complete 512 forwards by about 0.2-0.8% avoiding a second device-memory read when

France to ban unsolicited telemarketing calls

117 aziaziazi 2026-08-11 09:02:02 UTC lemonde.fr GoogleT5 Share 72

If you are an authorized partner, a le monde subscriber, or would like to request permission d’accès to this content, please contact: licensing[@]groupelemonde.fr. please include your ip address and request id (rid) in your email - we'll try to resolve the problem as soon as possible. meanwhile, we have fixed the error page for you, which is now available on our website. we apologize for any

As AI eats the web, the internet’s collective memory is disappearing

299 awnird 2026-08-10 23:02:02 UTC thewalrus.ca Full Share 295

To Save C, We Must Save ABI

42 gurjeet 2026-08-11 06:02:02 UTC thephd.dev GoogleT5 Share 18

A new standard library for.c has been released, but it's still not ready for prime time... edward mccoy argues that this is the first step in addressing application binary interface (abi) stability issues he says if you compile against the wrong headers, you don't get warned that you have wrong definitions of things with the same name - but there is an easy way around it with n29

Chicken Scheme 6.0

193 eatonphil 2026-08-11 01:02:02 UTC code.call-cc.org GoogleT5 Share 21

- fix warnings about zero-sized memset in build system. fix'memory-statistics' by returning semi-space bytes like the documentation says. increase the "binary compatibility version to 9' a number of bugs have been fixed in the latest version of libchicken, including the following: r4rs no longer exports eval and srfi-

LFM2.5 2.6B model competitive with 4x larger models

77 nateb2022 2026-08-11 03:02:02 UTC huggingface.co GoogleT5 Share 18

Lfm2.5-2.6b is part of a family of hybrid models designed for on-device deployment. it builds on the lfm2 architecture, with an 128k context window and agentic post-training - the fastest model in its size class, reaching almost 15k output tokens per second at high concurrency, roughly 1.3b token per day on samsung h100s. the model is optimized for cpu inference

Show HN: Mcptoon – Token-efficient MCP CLI client

46 mcptokensaver 2026-08-11 06:02:02 UTC github.com/activeing123 GoogleT5 Share 35

Mcptoon is a cli client that connects to any mcp server (stdio or http) it outputs toon (token-optimized object notation) instead of json. works with every ai agent — claude code, codex, opencode, cursor, catpaw, anything that runs shell commands. configure the servers, tools, token savings once - every agent shares the

About Rx Kids

18 mooreds 2026-08-11 08:02:02 UTC rxkids.org GoogleT5 Share 2

Rx kids is the nation's first community-wide prenatal and infant cash prescription program. moms receive $1,500 during pregnancy, and babies receive $500 per month for 6 or 12 months after birth, based on funds raised. to qualify, you must be a resident of one of the participating locations. the program is administered by givedirectly, the global leader in safe and secure cash transfers. for ...

Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots

333 HenryNdubuaku 2026-08-10 20:02:02 UTC cactuscompute.com GoogleT5 Share 130

Cactus needle is a 2bit model that fits 45m parameters into 14mb. the model is trained on 115b tokens and post-trained on 38b engrams - jim snyder & joe wilson: the performance never lets us down. he says all numbers are measured end-to-end through the shipped c++ engine in its production configuration. "the tables answer one

Show HN: Scroll through all 43252003274489856000 Rubik's Cube states

190 Alen123 2026-08-11 00:02:02 UTC everycube.alen.is Full Share 62

Scroll through all 43,252,003,274,489,856,000 reachable Rubik's Cube permutations.

Mark Zuckerberg attacks 'closed' AI rivals as Meta returns to open models

506 root-parent 2026-08-10 15:02:02 UTC ft.com Full Share 469

The following information can help our support team to resolve this issue.

The “mechanical miracle” that ruined Mark Twain’s life

139 benbreen 2026-08-10 23:02:02 UTC resobscura.substack.com Full Share 72

Recycle – Floppydisks

63 calvinmorrison 2026-08-11 02:02:02 UTC floppydisk.com Full Share 27

We buy new disks in sealed undamaged packs. Quantities of 100 or more. If disks are outside their original packaging, we cannot treat them as new. For new disks, send us a picture of your disks, and call (800) 397-7890 for a quote.

Stowaway – Take the window seat on any plane or satellite overhead

269 thunderbong 2026-08-10 19:02:02 UTC stowaway.live GoogleT5 Share 32

The sky here is the real one outside — your location, your weather, this minute's light. click one and the camera follows it; stow away and it takes a seat on board, over real terrain. all of that is drawn in real time, which needs javascript and webgl 2. switching javascript on for this site should do it. if you want to use the site in mobile, go to http://www.satellite

The UK's war on anonymity has come to America

502 slowin 2026-08-11 00:02:02 UTC effort.news GoogleT5 Share 423

Effort investigation identifies coordinated operation to influence american lawmakers. five foreign non-government organizations and their us affiliates are involved, authors say syria khan and edward mccaffery report reveals. they are attempting to pass a patchwork of digital id laws in 21 states and the us congress - they say. cnn's john defterios and david

Sonic Pi v5

376 samaaron 2026-08-10 16:02:02 UTC patreon.com Full Share 89

Programming the Gigatron

3 shakna 2026-08-11 09:02:02 UTC iwriteiam.nl Full Share 0

Faster floating point math with Rust's new API

39 subset 2026-08-11 04:02:02 UTC pythonspeed.com GoogleT5 Share 2

Floating point math is often slower than integer math because the compiler is being conservative. starting in version 1.98, rust will allow telling compiler it can optimize your code further, but with extra control so that you can still write numeric algorithms with minimal rounding errors. if you want to sum an a large number of floating point numbers, add numbers with algebraic operators ...

Rust SIMD on the GPU

185 sagacity 2026-08-10 19:02:02 UTC vectorware.com GoogleT5 Share 92

Vectorware is building the first gpu-native software company. we are mapping rust's portable simd onto the gpu's native execution model. the ir itself is architecture-agnostic, so code and libraries can be used on gpus without rewrite. if you are interested in learning more, contact us at [email protected] - e-mail: adam@v

How Claude marks AI-generated content

179 mfiguiere 2026-08-10 22:02:02 UTC support.claude.com GoogleT5 Share 136

Claude models launched in the eu on or after august 2, 2026 will support machine-readable marking at launch. embedded watermarks and digitally signed provenance metadata will apply to all generated text - and files if supported by the platform ad hoc or the underlying file type. lack of detecting mark doesn’t mean the content was not ai-generated or processed – it just indicates that it may ...

What's the best programming language for coding agents?

180 chaychoong 2026-08-10 17:02:02 UTC danluu.com GoogleT5 Share 118

A post suggests that languages that represent things more concisely are more token efficient. the median time to solve the tasks is in the hundreds of seconds - sam taylor-mcgraw-smith et al. he says this is not always obvious how much the issue generalizes across tasks or across setups. for these kinds of data-y projects, llms massively reduce the amount of effort it takes to get

Confessions of a Long-Distance Sailor

117 AntiRush 2026-08-10 21:02:02 UTC arachnoid.com GoogleT5 Share 32

In 1988, i sailed away from the west coast of the united states in a 31-foot sailboat. this is my story — my account of... around the world solo sail - for 3 1/2 years, from fearing shore, to wanting never to touch land again. read my book, sail with me! " ''i'm not sure if it's possible,' he says. but it is

Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

1116 riordan 2026-08-10 11:02:02 UTC research.meta.ai Llama Share 608

Meta Superintelligence Labs introduces Muse Glimmer, a 30-billion-parameter model optimized for local agent workflows, enabling use anywhere, anytime. Muse Glimmer delivers strong performance on key agentic use cases and benchmarks, with optimized integrations and developer documentation available for building and running agents.

Publishing Schematics Before “Open Source” Was a Word

90 extralongdivisi 2026-08-10 20:02:02 UTC fabscene.medium.com GoogleT5 Share 19

This website is using a security service to protect itself from online attacks. the action you just performed triggered the security solution. you can email the site owner to let them know you were blocked. email them with the cloudflare ray id found at the bottom of this page. if you have any questions, please contact [email protected] - we'll be happy to help. click here for ...

Squeak 6.1

267 fniephaus 2026-08-10 13:02:02 UTC squeak.org Llama Share 129

Squeak 6.1 release notes highlight major changes, including a new tree browser, improved kernel infrastructure, and enhanced toolset and user interface.

Learning more about Claude's mathematical capabilities

206 tosh 2026-08-10 18:02:02 UTC anthropic.com GoogleT5 Share 138

Claude has improved on a longstanding lower bound for the fraction of zeros that satisfy the riemann hypothesis. the result draws heavily on techniques that allow the hypothesis to be true without it, john sutter says - and he says it's not surprising that it did! if you're interested, you can find the paper here, or read the full explanation here. d.j. suriajaya: the results

The Water Footprint of AI

13 MASNeo 2026-08-11 07:02:02 UTC doi.org Full Share 2

Tail-call optimization in C is relatively recent (2025)

150 prakashqwerty 2026-08-10 12:02:02 UTC lwn.net Llama Share 135

The C calling convention has historically not supported tail calls, but GCC now supports them with a separate calling convention. This allows for improved performance in interpreters like Gforth and a toy Forth interpreter.

Show HN: Ante, a coding agent in a single binary that runs offline

134 ubermon 2026-08-10 16:02:02 UTC github.com/antigmalabs GoogleT5 Share 79

Ante is a self-contained coding agent that lives in your terminal. it can also be the optimized core for building your own harness and high-performing assistants. ante is fast, lightweight, and the only terminal agent with native local inference built in. read more in our philosophy and agent organization patterns. to get started, download the alpha version of antigma.ai/eval.xml - it's available now

Exploiting System Management Mode with a very long interrupt

162 WhiteDawn 2026-08-10 17:02:02 UTC github.com/xoreaxeaxeax GoogleT5 Share 58

Exploiting system management mode with a very very long interrupt. to break this, all we need is someone too busy to notice they're supposed to join smm - at this point, core 1 can attack core 0 while the other is out of it. there are 100+ smm toctou cves out there: an mmap handler checks and then uses it; all you need for exploitation is to rewrite