Magistral — the first reasoning model by Mistral AI

923 meetpateltech 2025-06-10 14:08:01 UTC mistral.ai Llama Share 418

Mistral AI announces Magistral, a domain-specific reasoning model with transparent and multilingual capabilities. Magistral excels in complex problem-solving, decision-making, and creative tasks with high-fidelity reasoning across multiple languages.

Launch HN: Vassar Robotics (YC X25) – $219 robot arm that learns new skills

569 charleszyong 2025-06-10 19:12:17 UTC news.ycombinator.com Llama Share 218

Charles from Vassar Robotics is launching a $219 desktop robot-arm kit with improved design and two integrated cameras. The kit will be released under an MIT license, and a software update will allow agentic LLMs to control the robot through natural language instructions.

"Localhost tracking" explained. It could cost Meta €32B

565 donohoe 2025-06-10 11:29:29 UTC zeropartydata.es Llama Share 268

Meta used a loophole to track users' online activity even when they used a VPN, incognito mode, and refused cookies. Meta faces up to €32 billion in fines under GDPR, DSA, and DMA for violating user consent and data protection regulations.

OpenAI dropped the price of o3 by 80%

503 mfiguiere 2025-06-10 17:41:29 UTC twitter.com/sama Full Share 489

Something went wrong, but don’t fret — let’s give it another shot. Some privacy related extensions may cause issues on x.com. Please disable them and try again.

Show HN: I made a 3D printed VTOL drone

412 tsungxu 2025-06-10 20:47:11 UTC tsungxu.com Llama Share 143

You built a 130 mile VTOL drone in 90 days with 3 hour flight time, one of the longest range 3D printed VTOLs. You overcame numerous challenges as a beginner in CAD, 3D printing, and aerodynamics to achieve this feat.

Low-background Steel: content without AI contamination

404 jgrahamc 2025-06-10 17:55:08 UTC blog.jgc.org Llama Share 266

You created lowbackgroundsteel.ai in March 2023 as a clearinghouse for pre-AI-generated content. The site points to sources like Wikipedia, Arctic Code Vault, and Project Gutenberg from before 2022.

Show HN: Chili3d – A open-source, browser-based 3D CAD application

398 xiange 2025-06-10 15:39:57 UTC news.ycombinator.com Llama Share 116

Chili3D is an open-source, browser-based 3D CAD application with near-native performance. It offers modeling, editing, and rendering tools, document management, and a user-friendly interface.

Successful people set constraints rather than chasing goals

395 MaysonL 2025-06-10 05:01:49 UTC joanwestenberg.com Llama Share 218

The author questioned the traditional approach to ambition and goal-setting, realizing that it often leads to working in the absence of inner alignment. They found that working within constraints, rather than setting explicit goals, can lead to clarity and meaningful progress.

Android 16 is here

327 nsriv 2025-06-10 18:29:43 UTC blog.google Llama Share 342

Android 16 is rolling out to supported Pixel devices with new features like live updates, notification grouping, and improved accessibility for hearing aid users. It also includes desktop windowing, custom keyboard shortcuts, and enhanced security with Advanced Protection.

Mikeal Rogers has died

317 neom 2025-06-10 13:43:21 UTC b.h4x.zip Llama Share 52

Mikeal Rogers, a close friend and colleague, died leaving a lasting impact on communities and shaping open source projects like Node.js. He modeled technical brilliance, humanity, and a promise to lift others through his work and legacy.

Malleable software: Restoring user agency in a world of locked-down apps

286 jessmartin 2025-06-10 15:16:27 UTC inkandswitch.com GoogleT5 Share 114

We need spaces that let us each express our unique potential. we can start out as a passive user, and gradually become an editor and creator. you don’t need to build airtight, industrial-grade software if you are in direct contact with your users, says daniel saunders. malleable systems collective catalog has some great books on the topic - check it out! cnn's john mc

OpenAI o3-pro

285 mfiguiere 2025-06-10 20:15:47 UTC help.openai.com Llama Share 196

OpenAI has released new models including o3-pro, a more reliable version of o3, and GPT-4.1, a specialized model for coding tasks.

Rust compiler performance

269 mellosouls 2025-06-10 08:24:44 UTC kobzol.github.io GoogleT5 Share 240

The most often repeated complaint about rust is its slow feedback loop and long compilation times. a number of reasons are preventing us from making progress faster, says sam taylor, director of the rust project. the main reason is that the language and compiler are getting new features all the time. it is also important that making improvements is not just about doing the implementation ...

The Gentle Singularity

259 firloop 2025-06-10 21:17:08 UTC blog.samaltman.com Llama Share 520

AI advancements will significantly improve quality of life through faster scientific progress and increased productivity. The future will see abundant intelligence and energy, enabling humans to create and achieve new wonders, but also requiring careful management of safety and distribution.

Europe needs digital sovereignty – and Microsoft has just proven why

257 01-_- 2025-06-10 06:51:14 UTC tuta.com Llama Share 224

Europe needs digital sovereignty due to a Microsoft email shutdown affecting the International Criminal Court's Chief Prosecutor, Karim Khan, following US sanctions. This incident highlights the risks of relying on US-based tech services and the need for European institutions to choose European-based alternatives.

The Danish Ministry of Digitalization Is Switching to Linux and LibreOffice

255 nogajun 2025-06-10 09:30:44 UTC politiken.dk Llama Share 149

Digitaliseringsminister Caroline Stage vil fjerne Microsoft fra sit ministerium og erstatte det med Libre Office. Planen er at alle medarbejdere skal bruge open source-løsninger i løbet af efteråret.

AI Saved My Company from a 2-Year Litigation Nightmare

238 anitil 2025-06-10 03:55:30 UTC tylertringas.com Llama Share 163

The US legal system is broken for defendants, favoring wealthy opponents who can bleed them dry through expensive procedural steps, but AI can level the playing field by empowering entrepreneurs to take charge of their legal issues. By using AI as a coach, entrepreneurs can proactively set budget constraints, identify risks, and develop strategic decisions to turn cost efficiency into a ...

Discrete Mathematics: An Open Introduction [pdf]

235 simonpure 2025-06-10 01:48:07 UTC discrete.openmathbooks.org GoogleT5 Share 45

A large portion of discrete mathematics is about studying particular types of relations. if we want to find the number of hot dogs eaten, we can use sequences of integers x y z d e r t i n g l c u m o p h b q w j k s, f (x) = 2x2x1...

Xeneva Operating System

232 psnehanshu 2025-06-10 19:08:29 UTC github.com/manaskamal Llama Share 77

Xeneva is an open-source OS for x86_64 and ARM64, built from scratch with a hybrid kernel. It welcomes contributions from developers and enthusiasts for kernel, driver, and application development.

Show HN: A “Course” as an MCP Server

202 codekarate 2025-06-10 20:36:55 UTC mastra.ai Llama Share 31

This course teaches AI development using Mastra, where a code agent guides you to build working agents from scratch. You'll learn to equip agents with tools, memory, and MCP, and deploy them to production.

Chatbots are replacing Google's search, devastating traffic for some publishers

199 jaredwiener 2025-06-10 21:03:27 UTC wsj.com Full Share 236

Please enable JS and disable any ad blocker

Detection of hidden cellular GPS vehicle trackers

187 gnabgib 2025-06-10 19:10:12 UTC researchgate.net Full Share 113

First thoughts on o3 pro

186 aratahikaru5 2025-06-10 20:42:19 UTC latent.space Llama Share 201

OpenAI cut o3 pricing by 80% to $2/$8 per mtok, matching GPT 4.1 pricing. o3-pro reports a 64% win rate vs o3, excelling at task-specific models and integration with humans and tools.

Fine-tuning LLMs is a waste of time

184 j-wang 2025-06-10 23:44:05 UTC codinginterviewsmadesimple.substack.com Llama Share 82

The author argues that fine-tuning large language models is a waste of time and can cause unintended consequences due to overwriting existing knowledge. Instead, they recommend using modular methods like retrieval-augmented generation, adapters, or prompt-engineering to inject new information without damaging the model's ecosystem.

Helion: A modern fast paced Doom FPS engine in C#

180 klaussilveira 2025-06-10 22:56:51 UTC github.com/helion-engine Llama Share 65

Helion optimizes Doom map rendering for high framerates and old hardware using static rendering and state management. It's available for download on Windows and Linux with optional .NET 9 and additional library requirements.

Apple has announced its final version of macOS for Intel

171 mdp2021 2025-06-10 03:51:56 UTC tedium.co Llama Share 218

Apple is ending its 20-year run of Intel-based Macs, giving users one more year of new versions and security fixes. The transition to Apple silicon will allow for more innovation, but it also means good hardware will be put to pasture.

Finding Atari Games in Randomly Generated Data

170 wanderingjew 2025-06-10 13:58:50 UTC bbenchoff.github.io Llama Share 37

A project generated 30 billion random 4kB files and used heuristics to narrow it down to 10,000, then used an Atari 2600 emulator to find interesting ROMs. The project successfully generated a ROM that displays game-like behavior, including input translation into visual output, using a combination of random data and basic heuristics.

Las Vegas is embracing a simple climate solution: More trees

160 geox 2025-06-10 00:21:40 UTC npr.org Llama Share 139

Las Vegas is taking steps to mitigate the urban heat island effect by planting more trees and providing shade, especially in lower-income neighborhoods. The city aims to plant 60,000 trees by 2050, which can lower temperatures by at least 10 degrees and help reduce heat-related deaths.

The curious case of shell commands, or how "this bug is required by POSIX" (2021)

154 wonger_ 2025-06-10 12:56:36 UTC notes.volution.ro Llama Share 124

The user discusses the issue of shell injection vulnerabilities in tools that use the system(3) function to execute shell commands, which can lead to security issues like shellshock. To mitigate this, the user suggests avoiding tools that use system(3), providing full or relative paths to executables, and using alternative tools or libraries that execute commands directly via execve(2).

Faster, easier 2D vector rendering [video]

146 raphlinus 2025-06-10 13:17:01 UTC youtube.com Embed Share 43

Thiings

141 pentagrama 2025-06-10 15:30:23 UTC thiings.co Full Share 65

A growing collection of 2800+ free 3D icons, generated with AI. Download PNGs with transparent backgrounds, browse by theme, or get lifetime access for just $29. Perfect for designers and creative projects.

Show HN: PyDoll – Async Python scraping engine with native CAPTCHA bypass

133 thalissonvs 2025-06-10 14:01:35 UTC github.com/autoscrape-labs Llama Share 39

Pydoll is a Python automation library that connects directly to the Chrome DevTools Protocol, eliminating the need for external drivers and providing seamless handling of modern web protection systems. It offers advanced features like intelligent captcha bypass, human-like interactions, and async performance for lightning-fast automation.

Another Crack in the Chain of Trust: Uncovering (Yet Another) Secure Boot Bypass

124 vitplister 2025-06-10 19:55:18 UTC binarly.io Llama Share 130

A Secure Boot bypass vulnerability, CVE-2025-3052, was discovered in a UEFI module signed with Microsoft's third-party certificate, allowing attackers to run unsigned code during boot. Microsoft added 14 new hashes to the Secure Boot dbx as a mitigation measure in Patch Tuesday on June 10, 2025.

Show HN: High End Color Quantizer

120 big-nacho 2025-06-10 11:50:15 UTC github.com/big-nacho Llama Share 34

The library patolette implements a weighted variant of Xiaolin Wu's PCA-based quantizer with optional saliency maps for visually prominent areas. It's still in development and not ready for production use.

Implementing DOES> in Forth, the entire reason I started this mess

120 todsacerdoti 2025-06-10 01:36:39 UTC boston.conman.org Llama Share 30

The user implemented the DOES> word in Forth, which is an immediate word that modifies the newly created word to execute code after itself, and explained its three temporal aspects during compilation and runtime. The implementation involves creating a new xt that words created by the SHAPE word will use, which is done by laying out a single instruction, JSR forth_core_create.does_hook, and ...

A Blacklisted American Magician Became a Hero in Brazil

118 bookofjoe 2025-06-10 17:10:51 UTC wsj.com Full Share 59

Please enable JS and disable any ad blocker

WWDC25: macOS Tahoe Breaks Decades of Finder History

118 syx 2025-06-10 10:52:29 UTC 512pixels.net Llama Share 75

The macOS Tahoe Finder icon has been reversed, changing the traditional dark side placement. This change is meant to fit with the new Liquid Glass UI, but some users prefer the original design.

Denmark: Minister for Digitalization wants to phase out Microsoft

112 doener 2025-06-10 08:56:30 UTC nordjyske.dk Llama Share 54

Digitaliseringsminister Caroline Stage planlægger at mindske brugen af Microsoft-programmer i ministeriet. Ministeren ønsker at gøre Danmark mere fri af techgiganter, særligt med hovedsæde i USA.

The hunt for Marie Curie's radioactive fingerprints in Paris

111 rmason 2025-06-10 01:02:13 UTC bbc.com GoogleT5 Share 96

The curie museum in paris is 'a link between the past and the future' it contains traces of radioactivity from the laboratory notebooks of the curies. the museum is also home to active research labs, including a nuclear power plant et al. it is open to the public, but visitors must register to access it. if you want to visit, you need to be accompanied by an interpreter. click here for more ...

Dubious Math in Infinite Jest (2009)

102 rafaepta 2025-06-10 15:04:38 UTC thehowlingfantods.com Llama Share 118

David Foster Wallace fan Mike Strong documents four mathematical errors in Infinite Jest, including a probability mistake on page 259 and a Mean Value Theorem error on page 1024. Strong notes these errors may be typographical, but is curious about their presence in DFW's work.

Spoofing OpenPGP.js signature verification

96 ThomasRinsma 2025-06-10 13:58:30 UTC codeanlabs.com Llama Share 30

A vulnerability in OpenPGP.js (CVE-2025-47934) allows attackers to spoof signatures by appending malicious packets. This was patched in v5.11.3 and v6.1.1.

Modern Minimal Perfect Hashing: A Survey

86 matt_d 2025-06-10 21:46:44 UTC arxiv.org Llama Share 30

Perfect hash functions map keys to integers without collisions, with varying trade-offs between space consumption, construction time, and query time. They have applications in databases, bioinformatics, and stringology, and have made significant progress since the last comprehensive survey in 1997.

A Family of Non-Periodic Tilings, Describable Using Elementary Tools

78 joshu 2025-06-10 19:47:37 UTC arxiv.org Llama Share 9

Researchers construct non-periodic tilings using modular arithmetic and vector geometry, exhibiting modulo-staggered rotational symmetry. This construction is self-contained and does not rely on previous tiling theories.

Show HN: A MCP server and client implementing the latest spec

72 init0 2025-06-10 06:30:03 UTC github.com/hemanth Llama Share 27

A Model Context Protocol (MCP) server implements the latest MCP specification with features like HackerNews and GitHub API integrations, AI-powered analysis, and enhanced sampling. The server demonstrates its capabilities through various tools, resources, and prompts, including code review and technology analysis.

Mozilla shuts down even more Firefox services

71 bundie 2025-06-10 08:38:11 UTC neowin.net Llama Share 99

Mozilla is shutting down several services including Pocket, Fakespot, Deep Fake Detector, and Orbit by June 26 or July 8, 2025. The company is focusing on Firefox due to limited resources and competing with tech giants.

Plato got virtually everything wrong (2018)

68 pfdietz 2025-06-10 14:05:25 UTC prospectmagazine.co.uk Llama Share 105

Plato's ideas have had a lasting impact on Western philosophy, but many of his arguments were wrong and have caused lasting harm. His influence on Western thought has been so profound that we're still paying the price for his brilliant but misguided ideas.

Reinforcement Pre-Training

68 frozenseven 2025-06-10 05:30:22 UTC arxiv.org Llama Share 18

We introduce Reinforcement Pre-Training (RPT) for large language models, reframing next-token prediction as a reasoning task with verifiable rewards. RPT improves language modeling accuracy and provides a strong pre-trained foundation for further fine-tuning.

Ada Continues to Climb in June Tiobe Index and PYPL

67 DragonSpiritWTP 2025-06-10 06:12:52 UTC forum.ada-lang.io Full Share 45

[IMG_6711] [IMG_6712]

Scientific Papers: Innovation or Imitation?

65 tapanjk 2025-06-10 04:04:41 UTC johndcook.com Llama Share 30

Pioneering research papers often fail to spark significant new directions due to imitative works and stovepiping. This leads to missed opportunities for breakthroughs and connections across fields.

Telegram, the FSB, and the Man in the Middle

63 xoredev 2025-06-10 10:41:00 UTC istories.media Llama Share 27

Telegram's infrastructure is controlled by Vladimir Vedeneev, a Russian network engineer with ties to Russian intelligence services. His companies have worked with sensitive clients, raising concerns about user privacy and security on the popular messaging app.

Micrographia (1665) [pdf]

62 andsoitis 2025-06-10 04:57:52 UTC arhipa.org GoogleT5 Share 20

'from whence there may arise many admirable advantages, towards the increase of the operative, and the mechanick knowledge, that this age seems so much inclined'. perhaps we may also judge (as other creatures seem to do) what is wholsome, what poyson; and in a word what are the specifick properties of bodies' the brass box cc in the middle was shap'd very much like the

Always On, Always Connected, Always Searching, Always Distracted

61 leejo 2025-06-10 14:29:29 UTC leejo.github.io GoogleT5 Share 6

A week to look at our own work, other’s work and to critique, to inspire. the key attraction is the portfolio reviews for the main project we’re working on - i was only in shanghai for three days but it was enough. an underground city more populous than many countries. if you’ve never been to the festival before, you may not have heard of it. but oh well, it is still worth it to go.

Launch HN: BitBoard (YC X25) – AI agents for healthcare back-offices

60 arcb 2025-06-10 15:09:30 UTC news.ycombinator.com Llama Share 25

Connor and Ambar from BitBoard are building AI agents to automate repetitive tasks in healthcare clinics, reducing errors and freeing up clinicians' time. Their software uses clinic SOPs to create AI agents that perform tasks like filling out intake forms and managing referrals.

Show HN: MidWord – A Word-Guessing Game

55 minaguib 2025-06-10 18:42:52 UTC midword.com Full Share 36

World fertility rates in 'unprecedented decline', UN says

54 mmarian 2025-06-10 05:17:15 UTC bbc.co.uk Llama Share 220

Namrata Nangia and her husband face financial struggles with their five-year-old daughter, echoing a global trend of people unable to have the number of children they want due to high costs. A UNFPA report finds that hundreds of millions of people worldwide are undershooting their fertility ideals due to financial limitations and lack of time.

Teaching National Security Policy with AI

49 enescakir 2025-06-10 13:54:25 UTC steveblank.com Llama Share 21

Stanford's international policy class integrated AI to prepare students for an AI-enabled world. Students used AI tools to summarize content, identify key themes, and synthesize insights to solve policy problems.

Apple Fails to Clear a Low Bar on AI

47 codexy 2025-06-10 11:58:22 UTC wsj.com Full Share 86

Please enable JS and disable any ad blocker

Web-scraping AI bots cause disruption for scientific databases and journals

30 tchalla 2025-06-10 20:25:01 UTC nature.com Llama Share 25

Website traffic surged due to bots scraping content, causing strain on systems and disrupting genuine users. Publishers and researchers are working to block the bots, but it's a challenging task, especially for those with limited resources.

CompactLog – Solving CT Scalability with LSM-Trees

30 Eikon 2025-06-10 10:45:49 UTC github.com/barre Llama Share 9

CompactLog is a Certificate Transparency (CT) log implementation that addresses scalability challenges by incorporating certificates before issuing SCTs and reducing versioned storage. It uses SHA-256 hashes of certificates and deduplicates entries referencing the same certificate.

JavelinGuard: Low-Cost Transformer Architectures for LLM Security

29 sharathr 2025-06-10 15:59:42 UTC arxiv.org Llama Share 2

JavelinGuard is a suite of low-cost, high-performance models for detecting malicious intent in LLM interactions, optimized for production deployment. It includes five transformer-based architectures offering unique trade-offs in speed, interpretability, and resource requirements.

High-speed fluorescence light field tomography of whole freely moving organisms

27 PaulHoule 2025-06-10 21:43:33 UTC opg.optica.org Full Share 0

Please wait...

Wharton Esherick and the Armstrong Linoleum Company

25 thomasjb 2025-06-10 13:31:02 UTC whartonesherickmuseum.org Llama Share 6

Wharton Esherick created iconic Three-Legged Stools from leftover wood scraps in the 1950s. He sold around 300 stools over 20 years, with prices ranging from $25 to $300, and they are now highly valued in the secondary market.

How to not use AI to code for you

21 mandaputtra 2025-06-10 08:43:01 UTC mandaputtra.id Llama Share 24

You're using AI for trivial tasks but feel it's not suitable for complex problem-solving, requiring understanding of underlying code and concepts. Experienced developers benefit from AI, but new programmers may rely too heavily on it, lacking fundamental knowledge.

How Long Does It Take to Draw a Picture of Every Pub in London?

20 mooreds 2025-06-10 11:17:31 UTC nytimes.com Prefix Share 13

You have a preview view of this article while we are checking your access. When we have confirmed access, the full article content will load. At 10:30 a.m. on a recent Wednesday in southeast London, the artist Lydia Wood stood next to a dumpster and set up her easel. She rubbed sunscreen on her neck and sized up her subject: The Lord Clyde, a pub in Southwark, just south of the River Thames, ...

1.5 TB of James Webb Space Telescope data just hit the internet

18 johnshades 2025-06-10 14:33:52 UTC theregister.com Llama Share 1

The COSMOS-Web project released 1.5 TB of JWST data for open science, offering a large view of the universe. The dataset includes a catalog of galaxies and an interactive viewer for users to search and analyze images.

Subtype Inference by Example

18 azhenley 2025-06-10 00:04:40 UTC blog.polybdenum.com Llama Share 4

A new type inference system called cubic biunification is presented as an alternative to traditional Hindley-Milner systems, offering improved flexibility and intuitiveness. The system is demonstrated through a simple ML-like language called cubiml, which is designed to be easy to understand and implement.

Drone shows: Will they overtake firework displays?

17 neversaydie 2025-06-10 13:27:00 UTC bbc.co.uk Llama Share 49

A couple's New Year's Eve wedding in New York State was surprised with a 300-drone show displaying lights and images. Drone shows are becoming popular, but their high cost and regulations may limit their use.

News Sites Are Getting Crushed by Google's New AI Tools

14 pondsider 2025-06-10 12:24:44 UTC wsj.com Full Share 4

Please enable JS and disable any ad blocker

OpenAI's Sora is now available for Free to all users through Bing Video Creator

14 ahmetcadirci25 2025-06-10 10:35:49 UTC venturebeat.com Llama Share 3

Microsoft has released Bing Video Creator, a free AI video generation tool, through its Bing mobile app. It allows users to create five-second videos from text prompts, with options for faster or standard generation speeds.

IBM to build first large-scale, error-corrected quantum computer by 2028

13 baruchel 2025-06-10 10:47:41 UTC technologyreview.com Llama Share 3

IBM plans to build a large-scale error-corrected quantum computer named Starling by 2028, with the goal of making it available via the cloud by 2029. Starling will consist of a network of modules, each containing a set of chips, and will be capable of performing 100 million logical operations consecutively with accuracy.

Tim Owens Jazz and Broadcast Collection Digitized by Grammy Museum Grant

12 gnabgib 2025-06-10 23:47:57 UTC library.unt.edu Llama Share 0

The UNT Music Library debuted the Tim Owens Jazz and Broadcast Collection, a digitized collection of 150 hours of jazz interviews and performances. The collection advances music research and preservation through the GRAMMY Museum Grants Program.

EDAN: Towards Understanding Memory Parallelism and Latency Sensitivity in HPC [pdf]

11 matt_d 2025-06-10 22:19:50 UTC spcl.inf.ethz.ch GoogleT5 Share 0

Resource disaggregation is a promising technique for improv- ing the efficiency of large-scale computing systems. edan provides us with the capacity of calculating the theoretical bounds for performance metrics and gaining insight into memory-level parallelism inherent to hpc applications - and it's free to do so offline, if desired ; edan can be downloaded from github or downloaded directly ...

macOS Tahoe 26 makes the Mac more capable, productive, and intelligent

11 doener 2025-06-10 04:24:42 UTC apple.com Llama Share 13

Apple previewed macOS Tahoe 26 with a stunning new design and powerful capabilities, including a personalized Control Center and new color options. The update also brings Continuity features like the Phone app and Live Activities, as well as improved Spotlight and Apple Intelligence.