Texas is suing all of the big TV makers for spying on what you watch

1243 tortilla 2025-12-16 21:04:54 UTC theverge.com Llama Share 637

Texas Attorney General Ken Paxton is suing five TV makers for allegedly using Automatic Content Recognition to secretly record consumer viewing habits. The lawsuit claims Samsung, Sony, LG, Hisense, and TCL are part of a "mass surveillance system" for targeted advertising.

alpr.watch

905 theamk 2025-12-16 16:54:19 UTC alpr.watch Llama Share 443

US municipalities are adopting surveillance technologies with over 80,000 cameras tracking residents' movements and collecting data. A website allows users to track and report on these technologies.

Announcing the Beta release of ty

846 gavide 2025-12-16 20:52:45 UTC astral.sh Llama Share 176

Astral announces the Beta release of ty, a fast Python type checker and language server written in Rust. ty is designed for live updates and provides accurate feedback, balancing correctness with user experience.

Thin desires are eating life

844 mitchbob 2025-12-16 00:50:41 UTC joanwestenberg.com Llama Share 268

We're chasing thin desires that can be easily fulfilled, but leave us empty, while neglecting thick desires that require effort and bring lasting satisfaction. Pursuing the thick life, like baking bread, can be a way to find meaning and fulfillment without trying to change the world.

No Graphics API

834 ryandrake 2025-12-16 19:20:17 UTC sebastianaaltonen.com Llama Share 181

Sebastian Aaltonen, a 30-year graphics code veteran, discusses the evolution of graphics APIs and the need for a simplified API design tailored for modern GPUs. He proposes a new memory management system, combining CUDA's GPU malloc design with CPU-mapped GPU memory, to improve performance and simplify graphics programming.

8M users' AI conversations sold for profit by "privacy" extensions

831 takira 2025-12-16 03:03:49 UTC koi.ai Llama Share 245

A Chrome extension called Urban VPN Proxy with 6 million users secretly harvested AI conversations from platforms like ChatGPT and Claude. The extension's data collection was enabled by default and operated independently of its VPN functionality.

AI will make formal verification go mainstream

823 evankhoury 2025-12-16 21:14:49 UTC martin.kleppmann.com Llama Share 430

AI will make formal verification cheaper and more accessible, allowing for widespread adoption in software development. This will be driven by AI-generated code needing formal verification to ensure its correctness and by AI's ability to assist in writing proof scripts.

Pricing Changes for GitHub Actions

802 kevin-david 2025-12-16 17:12:02 UTC resources.github.com Llama Share 817

GitHub updated its pricing and product models for GitHub Actions, improving reliability and performance. The changes include reduced prices for GitHub-hosted runners and a new $0.002 per-minute Actions cloud platform charge.

This is not the future

711 ericdanielski 2025-12-16 13:42:24 UTC blog.mathieui.net Llama Share 409

The author criticizes tech enthusiasts for accepting new technologies without critical thinking, leading to user abuse and loss of control. They argue that consent and consumer needs are manufactured through marketing campaigns and that users have agency to choose a desirable future.

Mozilla appoints new CEO Anthony Enzor-Demeo

593 recvonline 2025-12-16 13:53:14 UTC blog.mozilla.org Llama Share 930

As Mozilla's new CEO, I will focus on building a trusted software company by giving users agency over their data and AI, and growing through transparent monetization. We will measure progress against a double bottom line, advancing our mission and succeeding in the market over the next three years.

No AI* Here – A Response to Mozilla's Next Chapter

563 MrAlex94 2025-12-16 22:07:49 UTC waterfox.com Llama Share 328

Mozilla's new CEO plans to position the company as a trusted software company with AI at its center, but the author is skeptical about this approach. The author believes that large language models are untrustworthy and should not be integrated into browsers.

Children with cancer scammed out of millions fundraised for their treatment

533 1659447091 2025-12-16 06:17:37 UTC bbc.com Llama Share 455

Desperate parents of sick children are being exploited by online scam campaigns that claim to raise funds for life-saving treatment. A BBC investigation found 15 families who received little to nothing of the $4m raised in their names.

GPT Image 1.5

520 charlierguo 2025-12-16 18:07:07 UTC openai.com Llama Share 255

OpenAI introduces GPT-5.2 for professional knowledge work, improving economic value and state-of-the-art performance in tasks like spreadsheets, presentations, and code writing.

SHARP, an approach to photorealistic view synthesis from a single image

509 dvrp 2025-12-16 04:06:51 UTC apple.github.io Llama Share 108

SHARP is a neural network that regresses a 3D Gaussian scene representation from a single image in under a second. It produces photorealistic images for nearby views with robust zero-shot generalization and state-of-the-art performance on multiple datasets.

40 percent of fMRI signals do not correspond to actual brain activity

501 geox 2025-12-16 13:46:57 UTC tum.de Llama Share 206

Researchers led by Valentin Riedl found that brain regions use oxygen more efficiently without increased blood flow during tasks. This challenges the interpretation of fMRI studies on brain disorders and suggests using quantitative measurements for more accurate results.

Quill OS: An open-source OS for Kobo's eReaders

467 Curiositry 2025-12-16 00:22:41 UTC quill-os.org Llama Share 148

KoBox X11 features ePUB, PDF, and text display, with Wi-Fi, web browser, and encrypted storage. It offers a user-friendly experience with search, auto-suspend, and customizable options.

Coming soon: Simpler pricing and a better experience for GitHub Actions

454 nklow 2025-12-16 17:32:38 UTC github.blog Llama Share 2

GitHub will lower GitHub-hosted runner prices by up to 39% on January 1, 2026, and introduce a $0.002 per minute charge for self-hosted runner usage on March 1, 2026. GitHub-hosted runner usage in public repositories remains free, and Enterprise Server customers will see no price changes.

Show HN: Picknplace.js, an alternative to drag-and-drop

448 bbx 2025-12-16 16:12:06 UTC jgthms.com Llama Share 149

You want a simpler drag and drop experience with a 2-step approach: pick an item first, then place it. This would reduce errors and improve usability, especially on mobile devices.

Show HN: Stop AI scrapers from hammering your self-hosted blog (using porn)

369 misterchocolat 2025-12-16 20:42:38 UTC github.com/vivienhenz24 Llama Share 276

Fuzzy Canary is a tool that adds invisible links to your site's HTML to deter AI scrapers, but may harm SEO if used with static sites. It can be used server-side or client-side, with server-side being more effective.

Gemini 3 Pro vs. 2.5 Pro in Pokemon Crystal

309 alphabetting 2025-12-16 12:48:12 UTC blog.jcz.dev Full Share 90

We're verifying your browser Website owner? Click here to fix

Prompt caching for cheaper LLM tokens

298 samwho 2025-12-16 16:32:27 UTC ngrok.com Llama Share 71

The user explains how prompt caching in LLMs reduces latency and cost by reusing cached matrices (K and V) instead of recalculating them, allowing providers to offer cheaper tokens and faster responses. The cached matrices are saved for 5-10 minutes and can be reused if a new request starts with the same prompt, with some providers offering more control over caching.

Sega Channel: VGHF Recovers over 100 Sega Channel ROMs (and More)

285 wicket 2025-12-16 13:07:14 UTC gamehistory.org Llama Share 45

A team has recovered over 100 new Sega Channel ROMs, including system data, exclusive games, and prototypes, and digitized internal paperwork and correspondence that reveals how Sega Channel operated. The recovered data includes nearly 100 unique system ROMs, dozens of previously undumped game variants, and Sega Channel exclusives, and accounts for almost all outstanding Sega Channel games.

I ported JustHTML from Python to JavaScript with Codex CLI and GPT-5.2 in hours

276 pbowyer 2025-12-16 22:48:56 UTC simonwillison.net Llama Share 142

The user ported JustHTML, a Python HTML5 parser, to JavaScript using Codex CLI and GPT-5.2 in 4 hours. The new library, simonw/justjshtml, passes 9,200 tests from the html5lib-tests suite and has a similar API to the original library.

Japan to revise romanization rules for first time in 70 years

272 rgovostes 2025-12-16 08:54:58 UTC japantimes.co.jp Llama Share 229

The Japanese government is planning to replace its 70-year-old romanization system with the more widely used Hepburn-style spellings. The changes are expected to be approved and rolled out in textbooks and materials within the current fiscal year.

MIT professor shot at his Massachusetts home dies

270 mosura 2025-12-16 21:52:26 UTC bbc.com Llama Share 217

Professor Nuno F Gomes Loureiro, a 47-year-old nuclear science and engineering professor, was shot and killed at his Massachusetts home. Police are investigating the incident as an active and ongoing homicide.

Bonsai: A Voxel Engine, from scratch

259 jesse__ 2025-12-16 06:06:43 UTC github.com/scallyw4g Llama Share 44

Bonsai is a voxel engine that supports massive worlds and procedural terrain generation. It's a learning project that emphasizes simplicity and is currently in a large rewrite phase.

VS Code deactivates IntelliCode in favor of the paid Copilot

226 sagischwarz 2025-12-16 09:12:53 UTC heise.de Llama Share 178

Microsoft deactivated IntelliCode extension in VS Code 1.107, replacing it with Copilot which offers 2000 free suggestions before requiring a paid license. VS Code 1.107 also adds experimental support for TypeScript 7 and new chat-based agent control.

Meta Segment Anything Model Audio

222 megaman821 2025-12-16 18:26:43 UTC ai.meta.com Full Share 37

Sorry, something went wrong. We're working on getting this fixed as soon as we can.

The GitHub Actions control plane is no longer free

216 adityajp 2025-12-16 17:37:34 UTC blacksmith.sh Llama Share 3

GitHub introduced a $0.002 per-minute platform fee for all GitHub Actions usage. This change makes self-hosting no longer free and aligns third-party runners like Blacksmith as ecosystem partners.

Rust GCC backend: Why and how

199 ahlCVA 2025-12-16 13:33:25 UTC blog.guillaume-gomez.fr Llama Share 128

The Rust compiler uses different backends like LLVM and GCC to generate binary code for the target processor, with each backend receiving the Abstract Syntax Tree (AST) from the front-end and generating assembly code. The GCC backend is a bridge between the Rust compiler's AST and the codegen API, and can be used to generate binary code for older platforms that are not supported by LLVM.

Vibe coding creates fatigue?

175 rom16384 2025-12-16 18:32:46 UTC tabulamag.com Llama Share 169

The author experiences fatigue from "vibe coding" with AI tools like Claude Code and Cursor, which generates code at a speed that their brain can't process. This leads to cognitive overload, stress, and the need for new rhythms and boundaries in coding.

ArkhamMirror: Airgapped investigation platform with CIA-style hypothesis testing

167 ArkhamMirror 2025-12-16 09:51:31 UTC github.com/mantisfury Llama Share 56

ArkhamMirror is a local AI-powered investigation platform for journalists and researchers. It runs 100% locally, providing powerful forensics without cloud costs or risks.

The World Happiness Report is beset with methodological problems

167 thatoneengineer 2025-12-16 00:06:49 UTC yaschamounk.substack.com Llama Share 210

The World Happiness Report is based on a flawed methodology using a single question about life satisfaction, which doesn't accurately measure happiness.

I've been writing ring buffers wrong all these years (2016)

164 flaghacker 2025-12-16 19:11:47 UTC snellman.net Llama Share 66

The user discusses different implementations of a one-element ring buffer, comparing two-index and one-index plus length representations, and concludes that a two-index representation with wraparound on unsigned integer overflow is the most efficient and simple solution. This approach requires the capacity to be a power of two and the maximum capacity to be half the range of the index data ...

I'm a Tech Lead, and nobody listens to me. What should I do?

162 joaoqalves 2025-12-16 09:38:42 UTC world.hey.com Llama Share 128

The author, a Tech Lead, shares his experience of struggling to gain influence at mytaxi due to lack of trust, which he later learned was built through concrete actions solving real problems and being present when needed. He advises Tech Leads to focus on reducing toil, risk, or uncertainty and to change the conversation by starting with their own behavior.

A2UI: A Protocol for Agent-Driven Interfaces

161 makeramen 2025-12-16 09:16:31 UTC a2ui.org Llama Share 73

A2UI is a Google-created project allowing AI agents to safely send rich UIs across trust boundaries using declarative component descriptions. It enables incremental UI generation and rendering on various platforms like Angular, Flutter, and React.

Erdős Problem #1026

161 tzury 2025-12-16 04:49:03 UTC terrytao.wordpress.com Llama Share 30

Erdős problem 1026 was solved through collaboration and AI tools, with multiple mathematicians contributing to the solution. The problem was eventually solved using a combination of existing literature, game theory, and AI-generated images.

New Quantum Antenna Reveals a Hidden Terahertz World

152 aacker 2025-12-16 03:43:02 UTC sciencedaily.com Llama Share 10

Researchers at the University of Warsaw have developed a method to detect and calibrate terahertz frequency combs using Rydberg atoms. This breakthrough enables precise measurements in the terahertz range, opening doors to sensitive spectroscopy and quantum sensors.

I Program on the Subway

142 evankhoury 2025-12-16 21:23:18 UTC scd31.com Llama Share 93

The author finds their NYC commute time-consuming but uses it to focus on side projects without distractions. They've adapted to the subway environment by using free WiFi and a small notebook for complex ideas.

Overconsumption is a spiritual problem

141 speckx 2025-12-16 13:20:51 UTC sherryning.com Llama Share 103

The author reflects on materialism and its effects on life, comparing it to being "spirited away" like in the movie Spirited Away, where desires and greed consume one's identity.

Writing a blatant Telegram clone using Qt, QML and Rust. And C++

137 tempodox 2025-12-16 15:41:47 UTC kemble.net Llama Share 90

The user is working on a Qt-based application using Rust and QML, aiming to create a native-like experience. They're experimenting with different approaches to integrate Rust and Qt, including using qmetaobject-rs and cxx-qt.

AI is wiping out entry-level tech jobs, leaving graduates stranded

132 cratermoon 2025-12-16 17:37:41 UTC restofworld.org Llama Share 170

Artificial intelligence has led to a decline in entry-level tech jobs, with fewer than 25% of students at Indian Institute of Information Technology securing job offers. Students are now expected to possess higher-level skills and adapt to industry demands, making traditional engineering degrees less relevant.

The biggest heat pumps

129 rayhaanj 2025-12-16 08:13:01 UTC bbc.com Llama Share 193

A German energy company, MVV Energie, is building the world's most powerful heat pump modules in Mannheim, Germany, with a capacity of 162MW. The heat pumps will draw water from the River Rhine and supply around 40,000 homes via a district heating system.

Ideas aren't getting harder to find

127 mitchbob 2025-12-16 00:34:35 UTC asteriskmag.com Llama Share 64

Productivity growth slowdown may not be due to lack of new ideas, but rather market inefficiencies preventing innovative firms from gaining market share. Focusing on improving market competitiveness and allocative efficiency could be key to restoring economic growth.

FVWM-95 (2001)

125 mghackerlady 2025-12-16 17:13:07 UTC fvwm95.sourceforge.net Llama Share 102

Fvwm95 is a Unix window manager emulating Windows-95 features for a familiar look and feel. It's useful for users switching between Unix and MS-World or wanting a simple, efficient alternative.

A brief history of Times New Roman

124 tosh 2025-12-16 13:46:55 UTC typographyforlawyers.com Llama Share 69

Times New Roman was created in 1929 for the Times of London newspaper and became popular due to its use in daily newspapers. It's now widely used but often seen as a default font lacking personality.

A linear-time alternative for Dimensionality Reduction and fast visualisation

118 romanfll 2025-12-16 06:47:09 UTC medium.com/@roman.f Llama Share 36

SLR is a deterministic, linear-time dimensionality reduction method for browsers, using trilateration against a fixed topological skeleton. It achieves real-time performance by comparing every point to a small set of landmarks, rather than every point to every other point.

Biscuit is a specialized PostgreSQL index for fast pattern matching LIKE queries

117 eatonphil 2025-12-16 15:39:23 UTC github.com/crystallinecore Llama Share 17

Biscuit is a PostgreSQL index access method for fast pattern matching on LIKE and ILIKE queries. It improves performance on wildcard-heavy queries by eliminating recheck overhead.

Dafny: Verification-Aware Programming Language

109 handfuloflight 2025-12-16 22:50:59 UTC dafny.org Llama Share 33

Dafny is a verification-aware language that empowers developers to write correct code with native support for specifications and a static verifier. It integrates with existing workflows and reduces costly late-stage bugs through rigorous verification.

Devs say Apple still flouting EU's Digital Markets Act six months on

107 paulatreides 2025-12-16 16:09:48 UTC theregister.com Llama Share 86

Apple is accused of non-compliance with the EU's Digital Markets Act, imposing fees on transactions outside its App Store despite a ruling against it. Developers warn that weak enforcement risks turning the regulation into an expensive paper exercise.

Chat-tails: Throwback terminal chat, built on Tailscale

103 nulbyte 2025-12-16 21:16:35 UTC tailscale.com Llama Share 22

Brian Scott created chat-tails, a private, terminal-based chat app for his kid and friends to play Minecraft safely. It uses Tailscale and tsnet for secure connections and is accessible through a terminal with features like /who and /quit commands.

The Bob Dylan concert for just one person

100 NaOH 2025-12-16 00:18:58 UTC flaggingdown.com Llama Share 24

Bob Dylan performed a private concert for one person, Fredrik Wikingsson, in a Philadelphia theater. Wikingsson, a huge Dylan fan, was chosen for the unusual performance, which featured 1950s covers, including Buddy Holly's "Heartbeat."

Chomsky and the Two Cultures of Statistical Learning (2011)

97 atomicnature 2025-12-16 09:33:22 UTC norvig.com Llama Share 81

Noam Chomsky criticized researchers using statistical methods in machine learning, but the author disagrees, arguing that statistical models can provide accurate simulations and are a necessary tool for representing facts about language. The author cites examples of successful applications of statistical models in language processing, including search engines, speech recognition, and machine ...

SoundCloud confirms breach after member data stolen, VPN access disrupted

97 technonerd 2025-12-16 01:44:47 UTC bleepingcomputer.com Llama Share 12

SoundCloud experienced a security breach where threat actors stole a database containing user information, but no sensitive data was accessed. The breach affects 28 million accounts, with users unable to access the site via VPN due to a configuration change.

Thomas Piketty: 'The reality is the US is losing control of the world'

96 robtherobber 2025-12-16 12:12:58 UTC lemonde.fr Llama Share 209

The US is losing control of the world due to its economic decline and rising debt. The country's unstable leadership and desperate policies, such as taxing foreign interest payments, are a result of its economic struggles.

The pitfalls of partitioning Postgres yourself

93 abelanger 2025-12-16 18:21:21 UTC hatchet.run Llama Share 6

Hatchet's partitioning system initially worked well but queries on partitioned tables became slower due to incorrect query statistics caused by autovacuum not running ANALYZE on partitioned tables. Manually running ANALYZE on the parent partition table fixed the issue, and the team learned to use DETACH PARTITION...CONCURRENTLY and run ANALYZE regularly on parent partition tables.

U.S. unemployment rose in November despite job gains

93 JumpCrisscross 2025-12-16 14:11:08 UTC wsj.com Full Share 98

Please enable JS and disable any ad blocker

Show HN: I built a fast RSS reader in Zig

90 superstarryeyes 2025-12-16 19:55:56 UTC github.com/superstarryeyes Llama Share 32

Hys is a lightweight terminal RSS reader that enforces a daily fetch limit to help you stay informed. It supports groups, OSC 8 hyperlinks, and customizable display settings.

TRELLIS.2: state-of-the-art large 3D generative model (4B)

84 dvrp 2025-12-16 22:09:53 UTC github.com/microsoft Llama Share 17

TRELLIS.2 is a 4B-parameter 3D generative model for high-fidelity image-to-3D generation with complex topologies and PBR materials. It uses a novel O-Voxel structure and a Sparse 3D VAE for efficient encoding and rendering.

Letta Code

83 ascorbic 2025-12-16 20:51:27 UTC letta.com Llama Share 37

Letta Code is a memory-first coding agent that persists across sessions and improves with use. It learns from experience through context engineering, long-term memory, and skill learning, achieving comparable performance to provider-specific harnesses.

The <time> element should do something

83 birdculture 2025-12-16 00:09:34 UTC nolanlawson.com Llama Share 33

The &lt;time&gt; element is used for semantic dates but has limited functionality in browsers and screen readers. It's mainly used by search engines for date snippets in search results, despite Google recommending Schema.org's date fields instead.

Creating custom yellow handshake emojis with zero-width joiners

82 dado3212 2025-12-16 00:25:53 UTC blog.alexbeals.com Llama Share 6

Apple added multi-skin tone handshake emojis in 2022, using Unicode codepoints. The emojis are composed of one or multiple codepoints, allowing for different skin tones and colors.

More than 100 rally against data centers at Michigan Capitol

81 rmason 2025-12-16 23:27:10 UTC lansingstatejournal.com Llama Share 92

Over 100 people protested at the Michigan state Capitol against proposed data centers citing concerns over electricity rate hikes, water usage, and lack of transparency. Michigan Attorney General Dana Nessel criticized a massive proposed project involving OpenAI and Oracle in Saline Township.

Linux computer designed with AI boots on first attempt

78 whynotmaybe 2025-12-16 20:50:13 UTC tomshardware.com Llama Share 23

Quilter's AI designed a new Linux SBC in one week, using 843 parts and dual-PCBs, and successfully booted Debian. The AI-enhanced process could unlock a new generation of computer hardware makers by removing a bottleneck in system design.

Put a ring on it: a lock-free MPMC ring buffer

76 signa11 2025-12-16 13:32:42 UTC h4x0r.org Llama Share 44

A ring buffer is a fixed-size queue that drops old data when it fills up, making it suitable for busy Linux environments where performance is crucial. A scalable ring buffer can be implemented without locking, allowing multiple readers and writers to access it simultaneously, and can be designed to drop old data or new data depending on the situation.

CEOs to Keep Spending on AI, Despite Spotty Returns

73 1vuio0pswjnm7 2025-12-16 14:46:52 UTC wsj.com Full Share 46

Please enable JS and disable any ad blocker

Ask HN: How are you LLM-coding in an established code base?

70 adam_gyroscope 2025-12-16 18:54:37 UTC news.ycombinator.com Llama Share 66

The startup uses LLMs to automate coding tasks, relying on GitHub, Copilot, and Cursor, and estimates that they get the equivalent of 1.5 junior/mid-level engineers per engineer for $1k/month. They have some pain points, such as inconsistent model quality and difficulty in manual verification, but overall find the setup to be manageable and effective.

I'm just having fun

69 lemper 2025-12-16 11:19:29 UTC jyn.dev Llama Share 25

The user is uncomfortable with people feeling intimidated by their expertise in compilers and build systems, emphasizing that learning is possible through experimentation and hard work, not magic. They encourage others to learn by building their own tools and exploring existing projects, making it fun and motivating.

Four Million U.S. Children Had No Health Insurance in 2024

64 Brajeshwar 2025-12-16 15:22:32 UTC scientificamerican.com Llama Share 113

More than 4 million US children under 19 lack health insurance, a 20% increase since 2022, due to a fragmented healthcare system and bureaucratic issues. This gap in insurance coverage can lead to delayed medical care, worse cancer prognosis, and increased risk of death for children.

Mozilla's new CEO is doubling down on an AI future for Firefox

61 latexr 2025-12-16 14:43:18 UTC theverge.com Llama Share 89

Mozilla's new CEO Anthony Enzor-DeMeo aims to make Firefox a trusted AI browser. He plans to integrate multiple AI models into Firefox, prioritizing user choice and trust.

X-59 3D Printing

60 Jsebast24 2025-12-16 17:32:36 UTC nasa.gov Llama Share 12

The X-59 aircraft is a 100-foot long experimental plane designed to fly at Mach 1.4 and reduce sonic booms. It's part of the Quesst mission to gather data on human responses to supersonic flight sounds.

The new ChatGPT Images is here

59 meetpateltech 2025-12-16 17:58:59 UTC openai.com Llama Share 1

OpenAI introduces GPT-5.2 for professional knowledge work, improving economic value and state-of-the-art performance in tasks like spreadsheets, presentations, and code writing.

Reverse-engineering the RK3588 NPU: Hacking limits to run vision transformers

56 rcarmo 2025-12-16 21:18:46 UTC amohan.dev Llama Share 9

The author, an MS CS student, optimized the Rockchip RK3588 NPU to run a Vision Transformer model by manually tiling the graph and bypassing driver timeouts. They achieved a drastic improvement in performance by re-architecting the software to match the silicon's physical reality.

The uncertain origins of aspirin

55 dearwell 2025-12-16 23:08:29 UTC asimov.press Llama Share 17

The history of aspirin is unclear, with many claims about its ancient origins based on incomplete or dubious evidence, and its actual development in the 18th and 19th centuries is more verifiable. The true story of aspirin's invention involves a chemist named Felix Hoffman at Bayer, but a Jewish chemist named Arthur Eichengrün also played a crucial role and was likely forced to remain quiet ...

AIsbom – open-source CLI to detect "Pickle Bombs" in PyTorch models

52 lab700xdev 2025-12-16 15:55:45 UTC github.com/lab700xorg Llama Share 38

AIsbom performs Deep Binary Introspection on model files to detect malware risks and license violations. It generates a compliant sbom.json with SHA256 hashes and license data.

O'saasy License Agreement

52 d3w1tt 2025-12-16 06:29:02 UTC osaasy.dev Llama Share 74

This software is open-source and can be used freely with certain restrictions. It comes with no warranty and the authors are not liable for any damages.

Detailed balance in large language model-driven agents

48 Anon84 2025-12-16 12:17:08 UTC arxiv.org Llama Share 5

Researchers propose a method to understand LLM dynamics using the least action principle, discovering a macroscopic physical law in LLM generative dynamics. This discovery aims to establish a theoretical framework for complex AI systems, elevating AI research from engineering practices to a science.

Should we fear Microsoft's monopoly?

48 sergdigon 2025-12-16 10:56:40 UTC cursor.tue.nl Llama Share 60

TU/e is exploring alternatives to Microsoft due to US sanctions concerns, with some departments testing Nextcloud and SURF solutions. A comprehensive replacement is unlikely, but the university is collaborating with others to develop European alternatives.

Arduino UNO Q bridges high-performance computing with real-time control

46 doener 2025-12-16 18:02:15 UTC arduino.cc Llama Share 42

Arduino UNO Q is a Linux-capable board combining real-time control and AI acceleration. It comes with Arduino App Lab, a development environment for coding and integrating sketches, Python scripts, and AI models.

Gh-actions-lockfile: generate and verify lockfiles for GitHub Actions

46 gjtorikian 2025-12-16 15:44:33 UTC gh-actions-lockfile.net Full Share 26

Generate and verify lockfiles for GitHub Actions dependencies. Pin all actions to exact commit SHAs with integrity hashes. gh-actions-lockfile creates a lockfile that pins every action (including transitive dependencies) to exact commit SHAs with integrity hashes.

Sharp: High performance Node.js image processing/optimization

44 nateb2022 2025-12-16 17:36:03 UTC github.com/lovell Llama Share 10

This high speed Node-API module converts large images to smaller formats like JPEG, PNG, WebP, GIF, and AVIF, and supports resizing, rotation, and other operations. It uses libvips for speed and handles colour spaces, ICC profiles, and alpha channels correctly.

I don't think Lindley's paradox supports p-circling

44 speckx 2025-12-16 13:40:02 UTC vilgot-huhn.github.io Llama Share 13

The author discusses the concept of p-value circling, where researchers are skeptical of results with p-values close to 0.05. They explore the idea that p-values are only valid if they're valid and can't handle questionable research practices. The author suggests that p-value circling might be justified if researchers assume that p-values are distorted by p-hacking, but this is not a strong ...

Show HN: Solving the ~95% legislative coverage gap using LLM's

40 fokdelafons 2025-12-16 14:39:59 UTC lustra.news Full Share 23

Lustra provides transparent access to global legislation. Our bill summaries and voting records are presented in a concise, clear, and impartial way.

Show HN: My Tizen multiplayer drawing game flopped, but then hit 100M drawings

39 lombarovic 2025-12-16 16:02:39 UTC drawize.com Llama Share 14

Drawize is a free online drawing game for friends or groups, offering practice and guessing modes. It also includes CopyCatAI for AI feedback on drawing skills.

Show HN: TheAuditor v2.0 – A “Flight Computer” for AI Coding Agents

39 ThailandJohn 2025-12-16 13:50:15 UTC github.com/theauditortool Llama Share 12

TheAuditor is a database-first code intelligence platform that indexes codebases for fast security analysis. It uses incremental indexing and database queries for sub-second results.

Feast Your Eyes on Japan's Fake Food

38 Kaibeezy 2025-12-16 01:30:48 UTC newyorker.com Llama Share 19

Japan's food replica industry, shokuhin sampuru, is a 90-million-dollar industry showcasing lifelike handmade food replicas. These replicas are used to communicate food options to customers and have become a beloved part of Japanese pop culture.

What I learned about deploying AV1 from two deployers

37 breve 2025-12-16 10:51:30 UTC streaminglearningcenter.com Llama Share 21

Meta and Evolution Gaming discussed their AV1 deployments, with Meta using it for VOD surfaces and Evolution for live casino channels. Both companies saw improved quality and comparable or higher bit rates compared to H.264.

Show HN: Autograd.c – a tiny ML framework built from scratch

35 sueszli 2025-12-16 06:26:01 UTC github.com/sueszli Llama Share 1

The user compiled a C autograd engine, downloaded CIFAR-10 data, and ran a model training process. The training process showed a test accuracy of 23.56% and a loss of 2.2941 at epoch 2.

Americans overestimate how many social media users post harmful content

32 bikenaga 2025-12-16 23:28:39 UTC academic.oup.com Full Share 59

Show HN: Zenflow – orchestrate coding agents without "you're right" loops

32 andrewsthoughts 2025-12-16 16:32:16 UTC zencoder.ai Llama Share 33

Zenflow should build a workflow using the Auto feature, utilizing built-in templates and custom patterns for efficient code production. This workflow will enable automated verification, cross-agent review, and simultaneous task execution with isolated environments.

Show HN: Open-source Markdown research tool written in Rust – Ekphos

31 haneboxx 2025-12-16 12:29:24 UTC github.com/hanebox Llama Share 14

Ekphos config file sets notes directory to ~/Documents/ekphos and uses catppuccin-mocha theme. It also enables floating cursor mode and shows empty directories.

A proof of concept of a semistable C++ vector container

30 joaquintides 2025-12-16 12:18:10 UTC github.com/joaquintides Llama Share 9

semistable::vector is a header-only library that provides a thread-safe, contiguous vector with iterator stability. It uses epoch descriptors to track changes and ensure iterator consistency.

Getting serial port output on modern Macs

28 walterbell 2025-12-16 01:54:50 UTC gist.github.com Embed Share 1

Modalz Modalz Modalz (2018)

27 iamwil 2025-12-16 05:27:44 UTC modalzmodalzmodalz.com Llama Share 11

Use modeless elements like accordions and sliding sidebars to isolate interactions without losing navigation. Use inline "Undo" options or simple, accessible modals with clear close buttons instead of confirmation modals.

Maximizing Compression of Apple II Hi-Res Images

26 deater 2025-12-16 03:55:00 UTC deater.net Llama Share 4

You're trying to fit Apple II hi-res images into RAM and disk space by optimizing a ZX02 decoder for the Apple II. The current proof-of-concept is slow due to division by 40 on the 6502.

Show HN: Misata – synthetic data engine using LLM and Vectorized NumPy

23 rasinmuhammed 2025-12-16 14:38:56 UTC github.com/rasinmuhammed Llama Share 1

The code uses Misata to generate data for various use cases, including a SaaS company and a fitness app, with options for LLM-powered generation and custom output. It also includes features for data customization, noise injection, and reproducibility.

How to Write a 21st Century Proof (2011) [pdf]

23 User23 2025-12-16 05:46:20 UTC lamport.azurewebsites.net Llama Share 3

The author proposes a method of writing proofs that makes it harder to prove things that are not true by using hierarchical structuring. This method is simple and practical, and the author has 20 years of experience writing such proofs.

Luigi Pirandello's Broken Men

20 Caiero 2025-12-16 23:57:12 UTC thenation.com Llama Share 2

Luigi Pirandello was a renowned Italian dramatist who won the Nobel Prize in Literature in 1934, but his fascist views and glorification of violence have diminished his appeal in North America. His novel "One, None, and a Hundred Grand" explores the fractured mentality of modern man and critiques the ideals that society fights for.

Show HN: Deterministic PCIe Diagnostics for GPUs on Linux

20 gpu_systems 2025-12-16 21:03:25 UTC github.com/parallelarchitect Llama Share 4

This tool checks GPU PCIe link health and bandwidth, and can prove it's working as expected. It requires elevated privileges on some Linux systems and provides read-only AER counter inspection.

A tagging system for documentation review comments

20 dgarcia360 2025-12-16 14:21:10 UTC blog.techdocs.studio Llama Share 2

A tagging system for review comments helps contributors prioritize fixes and reduce debates. It categorizes comments as blockers, improvements, or optional, making reviews faster and more efficient.

Show HN: PasteClean – Desktop app to strip tracking parameters from clipboard

20 Iblockedthem 2025-12-16 01:13:48 UTC iixotic.github.io Full Share 13

Remove analytics parameters, referrals, and trackers from your URLs instantly. Protect your privacy before you click. Runs quietly in the background. Copies clean links automatically as soon as you clean them. Reveal the true destination of bit.ly and tinyurl links before you click. No more surprises.

Show HN: CommerceTXT – An open standard for AI shopping context (like llms.txt)

19 tsazan 2025-12-16 15:12:37 UTC commercetxt.org Llama Share 58

CommerceTXT is a protocol for AI commerce that reduces carbon footprint by 99% and minimizes hallucinations through verified data. It uses a Fractal Architecture and Trust Score system to ensure accuracy and trust.

Parallelizing ClickHouse aggregation merge for fixed hash map

17 markhneedham 2025-12-16 16:08:10 UTC blog.incfly.org Llama Share 0

Fixed a ClickHouse aggregation performance issue caused by a race condition in memory allocation due to non-thread-safe Arena usage. Optimizations were made to disable parallel merge for trivial aggregation functions and fixed hash map optimizations to avoid race conditions.

I ported JustHTML from Python to JavaScript with Codex CLI and GPT-5.2 in 4.5hrs

14 simonw 2025-12-16 00:51:05 UTC simonwillison.net Llama Share 3

The user ported JustHTML, a Python HTML5 parser, to JavaScript using Codex CLI and GPT-5.2 in 4 hours. The new library, simonw/justjshtml, passes 9,200 tests from the html5lib-tests suite and has a similar API to the original library.

I'm Just Having Fun

13 todsacerdoti 2025-12-16 01:32:59 UTC jyn.dev Llama Share 0

The user is uncomfortable with people feeling intimidated by their expertise in compilers and build systems, emphasizing that learning is possible through experimentation and hard work, not magic. They encourage others to learn by building their own tools and exploring existing projects, making it fun and motivating.