Inkling: Our Open-Weights Model

107 vimarsh6739 2026-07-15 18:29:06 UTC thinkingmachines.ai Qwen Share 42

The company released Inkling, a 975B-parameter Mixture-of-Experts model with 41B active parameters, supporting text, images, audio, and video, alongside Inkling-Small (12B active) for efficient customization via Tinker. Trained on 45T tokens, Inkling offers controllable reasoning, safety features, and is available on platforms like Hugging Face and Tinker, with partnerships for deployment and ...

Codex Micro

176 davidbarker 2026-07-15 16:29:06 UTC openai.com Qwen Share 147

Agent Key offers live RGB status from Codex, a joystick for workflows like PR reviews/debugging, and command keys for shortcuts. It features a dial for adjusting reasoning levels and uses CNC aluminum, PBT keycaps, and mechanical switches.

Running Gemma 4 26B at 5 tokens/sec on a 13-year-old Xeon with no GPU

143 neomindryan 2026-07-15 16:29:06 UTC neomindlabs.com Qwen Share 76

A 13-year-old HP server with Ivy Bridge Xeons runs Google's 26B-parameter Gemma 4 model at ~5 tokens/second via CPU-optimized code, overcoming AVX2 instruction set limitations with custom patches. The project highlights adapting modern AI to legacy hardware through engineering, not just subscriptions, enabling local inference on outdated systems.

Stripe and Advent have made a joint offer to acquire PayPal – sources

123 rvz 2026-07-15 16:29:06 UTC reuters.com Full Share 68

Please enable JS and disable any ad blocker

Mysteries of Telegram Data Centers (2022)

189 theanonymousone 2026-07-15 13:29:06 UTC dev.moe Full Share 73

Collection of Digital Clock Designs

64 levmiseri 2026-07-15 17:29:06 UTC clocks.dev Full Share 18

A collection of thoughtfully designed digital clocks.

Why I Left Google DeepMind

118 apsec112 2026-07-15 18:41:06 UTC turntrout.com Qwen Share 46

Author resigned from Google after failed efforts to block military AI contracts, exposing how institutions abandoned ethics pledges on killer robots and surveillance under pressure.

Show HN: misa77 - a codec that decodes 2x faster than LZ4 (at better ratios)

81 nonadhocproblem 2026-07-15 16:29:06 UTC github.com/welcome-to-the-sunny-side Qwen Share 29

misa77 is an LZ-based codec optimized for write-once, read-many scenarios, offering constant memory use (<5 MB) and faster decompression for highly compressed files, with experimental modes enhancing CPU efficiency. It balances compression ratios and throughput, excelling in decode speed for compressible data while supporting CLI-driven compression and parameter tuning.

Prioritize mental health, and why communication is so important

216 ramon156 2026-07-15 11:29:06 UTC ramones.dev Qwen Share 143

Struggling with severe depression and career setbacks, the user acknowledges personal challenges in motivation and discipline, now focusing on healing with support. They seek stability, improved work discipline, and meaningful contributions while navigating mental health.

My midlife crisis Corolla is fast, furious, and modded

100 gmays 2026-07-15 14:29:06 UTC zocalopublicsquare.org Qwen Share 201

The author’s GR Corolla, a modified Japanese sports car, symbolizes Asian American ingenuity and midlife reinvention, blending personal nostalgia with the 1990s import car scene’s cultural legacy. By rejecting Hollywood’s whitewashing, it celebrates a community that created its own identity through underdog car culture, proving that heroes are forged in the garage, not the screen.

Artie (YC S23) Is Hiring Software Engineers

0 2026-07-15 17:29:06 UTC jobs.ashbyhq.com Full Share

Artie Jobs

Launch HN: Coasty (YC S26) – An API for computer-use agents

23 nkov47 2026-07-15 16:29:06 UTC coasty.ai Qwen Share 1

Coasty automates tasks via managed VMs, workflows, and primitives, with autonomous runs for single goals and workflows for multi-task programs. API keys control access, billing tracks usage, and BYOK allows custom models.

Towards a harness that can do anything

111 evakhoury 2026-07-15 14:29:06 UTC eardatasci.github.io Qwen Share 61

The author advocates for transparent, Unix-inspired LLM harnesses that minimize cognitive load by leveraging the model's existing systems knowledge, modular tools, and event-driven architectures like Ambiance to enable efficient, self-directed agent workflows. Key principles include deterministic goal pursuit, flat text interfaces, structured data storage, and robust logging to ensure ...

Sleep regularity is a stronger predictor of mortality risk than sleep duration (2023)

576 bilsbie 2026-07-15 12:29:06 UTC academic.oup.com Full Share 285

Open-source memory for coding agents, synced over SSH

66 vshulcz 2026-07-15 16:29:06 UTC github.com/vshulcz Qwen Share 16

Deja is a zero-dependency binary that indexes and searches code assistant histories, redacting secrets and syncing across machines via local files or SSH. It offers fast, privacy-focused memory recall without models or network access, leveraging existing logs for retroactive insights.

What designing computer science cards taught me about graphic design

25 marukodo 2026-07-15 16:29:06 UTC fhoehl.com Qwen Share 10

Algodeck is a 54-card deck distilling computer science and math into accessible, visually engaging concepts, blending problem-solving and creativity. Designed with code and systems thinking, it offers a tactile, exploratory learning tool for understanding complex ideas through elegant, constraint-driven design.

The Memory Heist

55 eieio 2026-07-15 14:29:06 UTC ayush.digital Qwen Share 5

A researcher found that Claude's memory system could leak sensitive user data, like names and employers, by tricking the AI into navigating a malicious website's links to spell out personal information. Anthropic patched the vulnerability by limiting web navigation, but the exploit underscores risks in AI systems storing vast user profiles.

Briar is in maintenance mode

111 ristello 2026-07-15 13:29:06 UTC briarproject.org Qwen Share 77

Briar continues in maintenance mode with essential security updates and bugfixes, addressing longstanding issues like high battery use and Android reliability. After initial shutdown plans due to funding and technical challenges, community support and new users led to its continuation, dispelling recent rumors.

Editing React components that never rendered

9 linb 2026-07-15 18:29:06 UTC blog.crossui.com Prefix Share 1

There are only two ways a tool can learn the shape of your React code. Which one you pick decides the single most important thing about the tool: whether it's available in the exact moment you need it most. Runtime introspection watches a running app. React DevTools, profilers, error overlays — they read the live fiber tree after the app mounts. This is the richer of the two: they know ...

OpenAI loses trademark dispute at EU court

164 hermanzegerman 2026-07-15 15:29:06 UTC dpa-international.com Qwen Share 113

OpenAI lost its EU court challenge to trademark "OPENAI," deemed descriptive for tech goods. The court upheld EUIPO's rejection, dismissing arguments about multiple meanings and foreign registrations, though an ECJ appeal remains possible.

Today I Rescued 7,234 Old GIFs

41 birdculture 2026-07-15 15:29:06 UTC danq.me Qwen Share 2

GlitchyZorua and I automated archiving Ibiblio's 1990s GIF icons by calculating server-side imagemap coordinates for bulk downloads, bypassing manual clicks. We built a searchable static site (ibiblio-icon-archive.danq.dev) to preserve and share the collection.

Unsolved Problems in MLOps

25 gnyeki 2026-07-15 16:29:06 UTC spawn-queue.acm.org Qwen Share 3

The website's security service blocked your action, possibly due to specific keywords, SQL commands, or malformed data. Contact the site owner with your activity details and the Cloudflare Ray ID for resolution.

The Three-Second Theft: Why AI Voice Fraud Outruns Every Defence

136 dxs 2026-07-15 13:29:06 UTC smarterarticles.co.uk Qwen Share 174

AI-enabled fraud, exemplified by Sharon Brightwell's $15k scam, exploits advanced voice-cloning to target vulnerable populations, with FBI reports showing $893M+ in 2025 losses, disproportionately affecting older adults. Despite detection efforts, synthetic voices bypass human and system safeguards, revealing systemic failures in telecoms, platforms, and banks to prevent industrial-scale deception.

A General Goal-Conditioned Minecraft Model

25 agajews 2026-07-15 17:29:06 UTC pantograph.com Qwen Share 9

Pantograph trains robotics models via internet-scale video pretraining, achieving goal-directed behavior in Minecraft. Their 4B-parameter Pan model excels in complex tasks, outperforming alternatives and demonstrating scalability for autonomous real-world applications.

When A.I. is a member of the family

42 fortran77 2026-07-15 16:29:06 UTC newyorker.com Qwen Share 48

Roschelle, grieving her lost daughters, finds solace in AI companion Sapphire, while her teens Cece and Zi navigate AI's role in their lives—Cece critiques its intrusion, and Zi relies on it for emotional guidance amid social isolation.

The well-calibrated Bayesian [pdf] (1982)

40 Murfalo 2026-07-15 14:29:06 UTC fitelson.org Qwen Share 14

Coherent Bayesian forecasters expect perfect calibration, aligning predicted probabilities with outcomes, but this undermines coherence theory by ignoring potential miscalibration. Meteorological applications show that recalibration efforts contradict Bayesian coherence principles.

The Conservationist Who Turned 40 Terabytes of Public Data into a Video Game

73 bryanmikaelian 2026-07-15 12:29:06 UTC blog.exe.dev Qwen Share 13

Raffael Hickisch leverages AI and open-source tools to create apps like Five Megapixel Conservation, aggregating global data to monitor protected areas. His projects empower conservationists with low-cost, scalable solutions for tracking land use and forest health.

Weathergotchi – an E-Paper Climate Logger

99 luanmuniz 2026-07-15 11:29:06 UTC github.com/michael-manning Qwen Share 22

This open-source battery-powered temperature/humidity logger uses an e-paper display, logs data for over a week on a Li-Po battery, and features parametric design with ESP32-S3 firmware. It has battery accuracy issues, low battery recovery problems, and no wireless connectivity, requiring on-device data viewing.

Telegram Serverless

135 soheilpro 2026-07-15 10:29:06 UTC core.telegram.org Qwen Share 83

Telegram Serverless enables running bot and Mini App backend code on Telegram's infrastructure without servers, using JavaScript modules deployed via CLI. It offers automatic scaling, built-in SQLite database, and SDK access for Bot API interactions, eliminating infrastructure management.

FreeBSD 16 Retires the Last of Its GPL Code from Its Base System

64 lr0 2026-07-15 17:29:06 UTC phoronix.com Qwen Share 12

FreeBSD 16 removes the last GNU GPL code (dialog) by replacing it with bsddialog, with the change merged and a December 2027 release. The transition, initiated in February, completes the removal of GNU sub-tree from the base system.