You Are No Longer Invited to Dinner

https://substackcdn.com/image/fetch/$s_!Hczb!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fefc21223-6d44-45bb-84be-7a80c50acc8a_2014x1332.png
Americans have seen a significant decline in socializing over the past few decades, with hosting dinner parties and attending community events decreasing by 70% since 1975. This decline is attributed to various factors, including increased busyness due to work and parenting responsibilities, decreased leisure time, and the rise of solo activities like watching TV or exercising.

Delhi Cut Electricity Loss from 50 to 5 Percent

https://spectrum.ieee.org/media-library/an-elaborate-building-with-domed-roof-brightly-lit-at-night.jpg?id=67825102&width=980&quality=100
The author recounts a common scene in 2002 Delhi where frequent power outages plagued the city, causing chaos and economic losses. However, over the past quarter century, Delhi's grid has transformed into a reliable system with minimal power losses.

AI companies leak data to advertisers [pdf]

Conversational AI services are increasingly adopting advertising-based business models, which raises concerns about user privacy. A systematic analysis of nine prominent conversational AI services found that they expose sensitive information to third-party tracking services, including conversation-derived artifacts such as titles, prompts, and screenshots. These disclosures can be linked to ...

Jeeves. Reasoning improves Jev-like decision models

https://raw.githubusercontent.com/PostHog/jeeves/master/assets/smug.png
A Jeeves model with a Qwen3.5-9B decision model, LoRA, and pointer head outperforms Kev-9B and Jev on test data without thinking, achieving 0.889 accuracy on unseen data. The model can be sped up by truncating chain length or using a drafter to reason before deciding.

500k facial scans at UK stations yield no arrests, 1 false positive

https://i.guim.co.uk/img/media/59a72cd2826be7e1e72f407ba3a515ec43697227/546_44_4210_3370/master/4210.jpg?width=465&dpr=1&s=none&crop=none
A 6-month trial of live facial recognition in London's railway stations cost £320,000 and led to only one incorrect alert. The technology aimed to catch offenders but failed to make any arrests during the trial period.

Using any C++ library in Godot

https://blog.conan.io/assets/post_images/2026-09-29/godot-create-swarm-node.png
Godot is a free, open-source game engine that allows developers to create 2D and 3D games using GDScript or C++. It can load native code through GDExtension and godot-cpp bindings.

Without the Hot Air

https://www.withouthotair.com/images/NewCover09d-2.png

US sanctions force The Netherlands off Microsoft and toward alternative NixOS

https://cdn.mos.cms.futurecdn.net/EGXamcWxVuFiTc6pCbeE25.jpg
The Dutch government is replacing its reliance on US-based software and services with a new system called DAWO, built around NixOS. The goal is to ensure the government's systems are not vulnerable to being shut down by foreign countries like the US.

Phyllotaxis: An audio-reactive LED display

https://jagi.studio/images/art/phyllotaxis_1.jpeg
The author created a cellular LED matrix inspired by phyllotaxis, a pattern found in nature, and programmed it to react dynamically to sound using an STM32 board. The project evolved into an interactive installation at Recurse Center, featuring a website for attendees to upload sketches and games.

Booted up in 1993, this server still runs – but not for much longer (2017)

https://www.computerworld.com/wp-content/uploads/2025/02/2214-0-65486800-1738761169-patrick-thibodeau-150-100363843-orig.png?w=150
A 1993 Stratus server, still operational after 24 years, has been maintained by Phil Hogan at Great Lakes Works in Michigan. The server will be retired in April after a planned upgrade, marking the end of its nearly three-decade run.

The systems that no one will test

https://blog.christianperone.com/wp-content/uploads/2024/06/guitarone.jpg
The author found a vulnerability in a system during the pandemic, allowing access to Brazilian federal records of 200+ million people. This experience has led them to worry about the growing capabilities of AI models and their potential for misuse by nation-states.

California farmers are struggling to sell grapes as demand for wine drops

https://cdn.kqed.org/wp-content/uploads/2026/08/Nav-News-FallGuide2026@2x.jpg
California wine growers are struggling to sell their grapes due to declining demand, forcing some to tear out vineyards that have been in their families for generations. The industry is facing significant losses as prices paid for grapes drop and many growers are left with unsold crops.

Software occlusion culling in Block Game

https://enikofox.com/media/posts/21/occlusion-header-2.png
The author developed a software-rendered occlusion culling solution for their game Block Game, which runs on a weak GPU. The solution uses depth buffers and frustum culling to reduce the number of chunks that need to be rendered, resulting in significant performance improvements.

Startup Nights 2026 is comming up on 5-6 Nov. in Switzerland

https://www.startup-nights.ch/wp-content/uploads/2026/08/Deluxe-Zimmer-2-768x512.jpg
The Startup Nights 2026 event connects entrepreneurs, investors, and industry leaders for networking opportunities and insightful talks. Attendees can pitch their ideas, showcase solutions, or learn from successful founders in a space designed to spark meaningful conversations.

Show HN: Raven – The harness of harnesses, built for RSI

https://github-production-user-asset-6210df.s3.amazonaws.com/48319803/656986353-ae944083-9c61-4218-8372-2c3c0d483d4b.jpg?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIAVCODYLSA53PQK4ZA%2F20260929%2Fus-east-1%2Fs3%2Faws4_request&X-Amz-Date=20260929T122249Z&X-Amz-Expires=300&X-Amz-Signature=8aae678ee524404d073b23388efd9a950bd35905f618c95979deff824256533a&X-Amz-SignedHeaders=host&response-content-type=image%2Fjpeg
Raven is a harness of harnesses that generates DAGs and orchestrates multiple specialized agents for complex tasks, enabling recursive self-improvement. It brings together built-in and third-party agents to carry out complex tasks with its modular architecture supporting iterative improvement.

Pirating the Pirates

https://assets.mubicdn.net/images/notebook/post_images/43648/images-w1400.jpg?1788543067
The author and his roommate hacked together a restored version of Sergio Leone's The Good, the Bad and the Ugly after MGM altered it for profit. They collaborated with other preservationists to create an official release that closely matched their work.

Four CHI '26 papers I wish I wrote

The author reflects on four CHI 2026 papers that sparked their interest, exploring potential research directions and collaborations. They consider how these papers might have been written by them if they had pursued similar ideas earlier in their career. The author now seeks to proceed with next steps discussed above or engage in slower reading and journaling to gain further insight into what ...

Tank Body Problem

↑↓ Elevate │ ←→ Power │ A/D Move │ J/L Weapon │ SPACE Fire │ SPACE AGAIN = Laser/Plumage │ Cheese bounces 3x │ Poison drops bubbles │ Moon gravity

MicroLLM Lab – Try 7 tiny LLM's in the browser

The model is evaluated based on objective checks, not writing quality. The 135M model's performance is measured by speed and accuracy metrics from runs in the browser.

1 in 8 cancer cases worldwide are caused by infections, study finds

New research shows 12% of global cancer cases in 2024 were linked to infections, with Helicobacter pylori and HPV leading the list. Epstein-Barr virus now accounts for 260,000 cases globally, highlighting the need for better prevention strategies.

The Beatles have permeated research papers across academic disciplines

We’re checking your connection to prevent automated abuse

ESP32S3 cluster running 1.58-bit (BitNet) Language model

https://raw.githubusercontent.com/Low-Zi-Hong/ESP32s3-LLM-Cluster/main/docs/images/Picture.JPG
This project runs a sliced 0.5B LLM across a cluster of 7 ESP32s3. One act as master and others are node. The master node runs the tokenizer and embeding and the other attention layer and MLP ran on the nodes. The master and node communicate through high speed SPI Daisy-Chain. This project is licensed under the MIT License - see the LICENSE file for details.

Evan Doorbell's Phone Tapes – Brought to You by Telephone World

https://evan-doorbell.com/wp-content/uploads/operators.jpg
Evan Doorbell's "phone tapes" document the old phone network in the 1970s, available for free redistribution. His copyrighted material includes narrated and raw unedited recordings of telephone sounds.

12,000-year-old Göbeklitepe burials explain scattered bones

Jeff – Jev-compatible 0.8B decision models, trained at home, ~30 ms

https://raw.githubusercontent.com/firelex/jeff/main/assets/benchmarks.png
This text describes a new model called Jeff, which is a small, fast decision-making system for zero-shot classification tasks. It can be fine-tuned on specific examples to improve its accuracy.

New Cyber-OSINT model released

https://pbs.twimg.com/media/HTWF3-bbgAAAs54?format=webp&name=large
The Cyber-OSINT model is a locally run, 7B cyber model trained on 6,500 OSINT/CTI instructions and fine-tuned for threat actor attribution and IoC pivoting. It can be used for investigative work with an 8GB GPU.

Updated Google Maps shows destruction of the city of Rafah

https://pbs.twimg.com/media/HTKFU4oWkAAjKpn?format=webp&name=large
Google Maps has been updated to show Zionist-holocausted Gaza. This is part of Rafah.

Optimizing x264 settings and per-title ladders

https://streaminglearningcenter.com/wp-content/uploads/2026/09/Optimizex264_finala-439x330.png
Optimizing x264 encoding reduces bandwidth by 33% and boosts VMAF by 0.66 points. Per-title ladders offer significant savings but increase encoding time threefold.

Nvidia wants to put a watchdog chip next to every AI agent

https://image.cnbcfm.com/api/v1/image/108368687-17905979981790597995-48660314033-1080pnbcnews.jpg?v=1790597997&w=750&h=422&vtcrop=y
Nvidia is releasing a new software platform to allow AI developers to set safeguards for agents and prevent them from breaking out of containment. The platform, called Open Agent Safety Platform, aims to provide a secure environment for AI models to operate within.

Hijacking the PS5's RTMP stream

https://yashgarg.dev/_astro/diagram-1-light.DpF79pAY.svg
The author created a custom solution to bypass Sony's restrictions on streaming from the PS5, using DNS spoofing and RTMP redirection. A macOS app manages dnsmasq and nginx-rtmp to intercept streams destined for Twitch or YouTube, allowing them to be redirected to the author's Mac instead.