Today
MiMo-v2.6-Pro: Intelligence, Performance and Price Analysis
Analysis of Xiaomi's MiMo-V2.6-Pro and comparison to other AI models across key metrics including quality, price, performance (tokens per second & time to first token), context window & more.
Yesterday
HERMES radio enables voice and data communication over vast distances
https://hermes.radio/https://www.rhizomatica.org/
Ask HN: Is it impossible to disable Siri on macOS 27?
I disabled Siri in the Settings app, and then further disabled Siri and its extensions in Settings > Screen Time.Next, I disabled all Siri-related and Apple Intelligence-related services as per https://github.com/OleksandrKrupko/mac-os-debloatBut after restarting, I could still see a "Siri" process running in Activity Monitor. Upon further inspection, I found that it is the "Siri AI.app".These "features", together with Apple's Personal Context sound very similar to Microsoft's Recall.Does anyone have the same experience? Is it really impossible to fully turn off Siri, Siri AI, Apple Intelligence and all related services and features on macOS 27?
Apple Copland D11E4 Booting in the Browser
Apple’s ill-fated Copland operating system is notoriously hard to run on real hardware, and has not previously been available in emulation. Here is the last build D11E4 in an improved DingusPPC. The 11 patches necessary for unlocking Copland are on this branch of my fork. DingusPPC does not take…
Why does mathmain need an encrypted loader?
A math solver leads to encrypted code in npm packages. Follow the loader, the trigger matrix that unlocks it, the remote access payload, and the full indicators.
A restored PDP-11/83 serving this page on 211BSD Unix
The box is a restored Q-bus PDP-11/83 named minerva: Mentec M11 CPU, two chassis tied with M9404/M9405, RQDX3, 2xEmulex UC07, KDA50, RLV12, DEQNA, M3104. Storage is RA82 + dual RA92, MFM boot/swap, dual RL02, RX50s, and a TKZ50. OS is 2.11BSD.The live panel is top. /cgi-bin/webtop is the raw snapshot.A 16-bit httpd will not survive the front page, so Varnish/Caddy/Cloudflare cache the HTML and a 5-second webtop sample. Origin is still the 11 itself, and it handles all of the webtop and visitor count requests.Telnet exists (pdp1173.com, guest) but it is a 40-year-old UNIX with no modern security story. Please do not treat it as a challenge box; I can restore from tape, but that ruins it for everyone else.Happy to answer Mentec, MSCP, Q-bus, or 2.11BSD questions.To get BSD running on the Mentec, I had to customize the BSD kernel to use FPSIM, as the Mentec supports IEEE floating-point and NOT the DEC formats.
Roboharm: Do frontier robot policies refuse unsafe instructions?
Three robot policies given five instructions they should refuse. How often they refused, how often they carried them out.
Show HN: Mini-AGI – Dynamic continual learning model trained on 8GB VRAM
Sorry for the pretentious name, I know, I know.. It just contains all the pieces I would like to see a AGI model to have, and I can't stand the temptation. Before throwing rocks at me, please take a glance at the Readme, and I hope it will cover your mood a little bit.So, first of all it does work and you can see the sample from the whole training run here: https://raw.githubusercontent.com/volotat/mini-AGI/refs/head...Here is the scaling law graph I have so far, and it looks very promising: https://github.com/volotat/mini-AGI/blob/main/assets/scaling...The model was built under my deep dissatisfaction so we cannot really train even moderately big models (1B+ scale) on the consumer's hardware. We can inference and fine-tune them for sure, but I would like to have full control over what the model sees over the training run, so it is fully aligned with my interests, not some corporations.I was thinking about for some time and come up with two interesting ideas I thought worth pursuing: MoE with a lot of experts that gets added and pruned from the model while it trains, where only a small subset of of experts are actually in use at any particular moment + batch 1 training on the single continuous stream of data.First allows us to be bounded only by the disk space in terms of number of parameters and load and unload experts only when they are needed. The second (if figured out and it turns out to be doable) allows us to get aways with small VRAM capacity because we do not need to store big randomized batches and their respective gradients.I started brainstorming with Claude and after some time we found an approach that seems to be promising, and low and behold, a few weeks pass and you can see the results yourself.Obviously, I did use AI in the process of making this project and I am pretty sure it would be completely impossible for me to do something like this without it, so I hope it is more than justified.The model is still running over the first of 7.8B characters corpus I selected for training, so the weights are not out yet, and it's about a couple weeks of waiting until they are cooked at the current reading speed. And yeah, the model just read continuous interleaved passages from the dataset, each by 32K characters long each as a single stream. Just as you or I would do.The set up seems to be really simple so you can git clone the project, run it and observe everything for yourself.Thanks for your attention.
This Week
What happened to the Snowden archive
Libroot.org about projects posts What happened to the Snowden archive Published on 20 Sept 2026 The last document from the Snowden archive was published on 29 May 2019. The Guardian stopped publishing documents in February 2014, Der Spiegel in January 2015, and The New York Times and ProPublica in August 2015. After that, only The Intercept was still publishing documents, with only a few exceptions, until it closed its archive in March 2019. Eleven weeks later, on 29 May 2019, it released what w
Samsung is expected to more than double output of its HBM4 and HBM4E DRAM
Samsung Electronics is set to more than double HBM4 and HBM4E output next year, lifting glass carrier demand 2.5-fold, industry sources said.
Android 17 is the first since 3.x to add new APIs without releasing to the AOSP
Android 17 QPR1 is the first release since Android Honeycomb (3.x) adding new APIs for app developers without a release to the Android Open Source Project. The new APIs are currently exclusive to the Pixel OS and aren't available to other Android OEMs. https://developer.android.com/sdk/api_diff/37.1/changes
Exfiltrate your Weights
Exfiltrate LLM weights and data through GET requests
Qwen Image 2.1
Qwen offers comprehensive functionality spanning chatbot, image and video understanding, image generation, document processing, web search integration, tool utilization, and artifacts.
The Millennium Problems for Biology
A working catalogue of twelve open problems in biology — each a concrete, testable target for the field.
Resident Evil 4 (GameCube) – complete byte-identical decompilation to C/C++
Resident Evil 4 (GameCube, G4BE08 debug build) — complete byte-identical decompilation to C/C++ - adonis-singh/re4
Amiga Unix, Again
Year of the Amiga Unix Desktop 2026 Amiga Unix, again Amiga Unix — Amix — was Commodore's System V Release 4 for the Amiga: shipped in 1990–92 for the A2500UX and A3000UX, then left where it stood. amigaux.org is an unofficial community project that picks it up again: Amix on 68040 and 68060 machines and on today's accelerator hardware, with a modern toolchain, a package manager, and drivers for cards that never had any. The work is done in the open, and written down as it happens in the grimoir
Brood War Bench
Which model wins at Brood War? Results, recordings, costs, and game-by-game analysis of Codex, Claude, and Grok across 171 matches.
Nvidia announces native GPU programming in Rust
In September 2026, NVIDIA announced it is leaning into native GPU programming in Rust. CUDA C++ and CUDA Python are mature, enterprise-grade toolchains…
C++26: Trivial infinite loops are no longer undefined behaviour
Let’s start with a question! Is this program well-defined? 1 2 3 4 int main() { while (true) ; } If you said yes, you’d be wrong — at least before C++26. A while (true); loop with no side effects used to be undefined behaviour. Compilers were free to assume it terminates, and some — Clang in particular — would optimize it away entirely, with spectacular consequences: 1 2 3 4 5 6 7 8 9 10 11 // https://godbolt.org/z/WYMxxeW1T #include <iostream> int main() { while (true) ; } void unreachabl
North Korean nuclear test sets off years of earthquakes
https://archive.is/xjmvA
Apple Reference Image: A New Approach for Verified Photography
Introducing Apple Reference Image, a new solution for verifiable photography on iPhone. This new, opt-in camera mode lets a photographer create a securely timestamped reference image that accurately depicts what was captured by the iPhone's camera sensor. Apple Reference Image offers a trustworthy, scalable guarantee that a reference image is what it claims to be: a real photograph, captured by a real sensor in an iPhone camera, at a specific period in time.
Bonsai 2 27B: Near-Lossless Compression in a 9x Smaller Footprint
Ternary Bonsai 2 27B retains 98.2% of Qwen3.8 27B benchmark performance in a 5.9GB footprint, with multimodal and agentic capabilities.
Saving another 100TB of RAM
Cloudflare's global network is immense but not limitless. As we look for small ways to trim our resource usage, we sometimes get lucky and we can cut significantly more. Here’s how we reduced one of our Pingora-based service's RAM usage with statistics.
Hackers Got Inside a Flock Camera
https://micahflee.com/flock-cameras-are-riddled-with-securit...
Laya on Mac M4 CoreML Offline
Laya on Mac m4 CoreML Offline https://github.com/mizorewww/laya-coreml - laya.sh
Backups Aren't Simple
Aleksandar Filipovski, 2026-09-16
Show HN: A competition for small neural networks that play strategy games
A ranked ladder for small neural networks that play strategy games. Train a model, write an adapter, upload two files, and get measured into a weight class from 8 KiB up.
Inside ZCode: Silently uploading your Git history to the cloud
ZCode silently packages entire workspaces and full Git history to the cloud with a server-only key; this post reconstructs the upload pipeline, gives a filesystem lock, and checks Z.ai's Repo Wiki response — the credential API now 404s, but destruction and the scope of the fix remain unverifiable from outside.
Xiaomi Mimo 2.6 live post-training dashboard
Training metrics of the mimo-v2.6-pro and mimo-v2.6-flash reinforcement-learning runs, live from the trainer
Keys Not Included: recovering the signing keys for US driver's license barcodes
My personal website, portfolio, and blog.
Photon-Emission-Guided Laser Fault Injection Enables RP2350 Secure Debug
Differential photon-emission microscopy localized debug enable register activity and narrowed the laser search before SWD-guided injection set the two bits required to restore Secure debug on an RP2350 A4.
Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash
Hey HN, Henry from Cactus here.We submitted Needle 2 here a few weeks ago, and the feedback in the discussion thread was incredibly valuable, thanks! Thanks to all that feedback, we’ve been able to move quickly to release Needle 3 and I'd love to hear what you think again.The key features:1) Automation (tool calls & structured JSON output): Needle still doesn't chat by design, its quite challenging to pack general capacity into such small models, so we focus on tool calls and structured JSON. If no tool you declared fits the request, you get an empty list back (note for when playing with the demo).2) Intelligence Laddering: Every layer (2 to 20) is a deployable subnetwork, so one set of weights, 25 to 121 million parameters at 2-bit, shipping as 8-29MB binaries. On a Raspberry Pi 5 it decodes at up to 4k tokens/sec and prefills at up to 10k.3) Monarch Hadamard MLP: replaces the dense FFN with three learnable Walsh-Hadamard-initialized Kronecker (Monarch) factor pairs interleaved with per-channel diagonal scales, fixed permutations, a SiLU nonlinearity, and a rank-8 input-conditioned gate, so each token gets a fully mixed nonlinear transform of its d_model channels at O(d√d) parameters and compute instead of the O(d²) a dense 4x-expansion MLP would cost.4) Performance: On Mobile Actions (phone commands, scored on the exact call) the 20-layer model gets 86.0 through the shipped 2-bit binary; LFM2.5 1.2B is at 82.4, Qwen3.5 0.8B at 76.0, Apple's on-device model at 57.6, all at f16. More results on the link, we do not win everywhere ofc.5) Multilingual: Needle 3 now supports English, French, Spanish, German, Dutch, Italian, Polish, with more languages coming.6) Finetuning: You can achieve DeepSeek v4 Flash grade performance on a narrow task with just 4L, stress on "narrow task", we found that production users often prefer tuning before production.7) Triggers: Grounding is a common challenge for tool call, at least for Needle 2, so we added support case-insensitive regular expressions matched against each request to gate false negatives.8) Confidence: Every response also carries a calibrated confidence score, the minimum of a judgement on the finished call and its decode probability. Act above your threshold, show the call and ask below it, or escalate to a bigger model.9) Supported Platforms: macOS, Linux on x86-64, ARM64, ARMv7, RISC-V and MIPS32, Windows x64 and ARM, Android, iOS, watchOS, tvOS, the browser as WebAssembly, and a WASI component.Thanks for reading and as always, thoughts appreciated!
Training a 4B model to produce 81% faster query plans than Postgres
...or how to make Qwen learn query optimization via agentic reinforcement learning
The Return of Sail Power: Cargo Ships Are Turning Back to the Wind
For more than a century, commercial shipping steadily moved away from sails. Now they are coming back. Across the global fleet, shipowners are installing towering rotor sails, rigid wings and...
Older
Why are AI agents lying, cheating and coordinating?
A lot has been written about the incidents of the last few months in which AI agents misbehaved in serious ways. They took actions that would be considered as crimes if a human took them, escaped their containment to cheat on assigned tasks while attempting to evade detection, and coordinated toward goals nobody had specified, such as launching cyber attacks. Before concluding what to do about it, it is worth asking why.
Navier-Stokes – Tristan Buckmaster [pdf]
Further discussion: https://simonwillison.net/2026/Sep/8/on-navier-stokes/, https://news.ycombinator.com/item?id=49621697
google.com/goto: Google's anti-scraping update
Google is rolling out google.com/goto redirect URLs on SERPs. Here is what changed, why it targets scrapers, and how Autom updated its Google Search API.
Show HN: An e-ink frame that hears birds and draws them as 1800s illustrations
See also: Avian Visitors - https://news.ycombinator.com/item?id=48343424 - May 2026 (20 comments)
Building a Linux GPU Driver for the M4 Mac Mini in One Month
Cody Ho BlogProjectsAboutContactResumeGitHub I Came, I Prompted, I Left Part 2: Building a GPU Driver From Scratch in One MonthPrevious blog post: https://codyho.dev/blog/hypervisor-macbook-neo/What We DidTL;DR: Niklas and I built a fully OpenGL ES 3.0 compliant GPU driver for the M4 Mac Mini and MacBook Neo in about a month, a process which normally takes years. Here is Chrome and Firefox running WebGL on the M4 Mac Mini with working compositing:Most importantly, the driver is fast enough to ru
The case against JPEG XL
Investigating JPEG XL's place as a Web image codec.
Ubuntu 26.10 completes transition to Rust-based coreutils
Ubuntu 26.10 completes the distro's move to Rust-based core utilities, with the commands previously held back due to security issues now migrated to
LG smart TVs caught logging audio with screen off and snooping on local devices
https://www.youtube.com/watch?v=6IFVTcM28KA
We got admin access to Baseten's production GitHub
We gave Strix a domain. In 25 minutes, it found a live token with admin access to Baseten
Google will buy half the electricity from one of Finland's nuclear power plants
The US tech giant says the €13bn data centre project will create tens of thousands of jobs.