Veo 3 and Imagen 4, and a new tool for filmmaking called Flow

815 youssefarizk 2025-05-20 17:46:36 UTC blog.google Llama Share 525

Google announces new generative media models Veo 3, Imagen 4, and Lyria 2, which create high-quality images, videos, and music, empowering artists to bring their creative vision to life. These models are available in various apps and platforms, including Gemini, Flow, and Vertex AI, and are designed to unleash human creativity and enable artists to bring their ideas to life faster and more easily.

DDoSecrets publishes 410 GB of heap dumps, hacked from TeleMessage

663 micahflee 2025-05-20 00:52:27 UTC micahflee.com Llama Share 185

Distributed Denial of Secrets published 410 GB of hacked data from TeleMessage, an Israeli firm that archives encrypted messages. The data includes plaintext messages and metadata, and is being shared with journalists and researchers.

Making video games (without an engine) in 2025

562 selvan 2025-05-20 05:54:58 UTC noelberry.ca Llama Share 244

The user prefers making games without commercial game engines, finding it more fun and less overhead, and has developed a custom workflow using C#, SDL, and other open-source libraries. They believe that making games "from scratch" is not an impossible task, especially with the state of open-source frameworks and libraries, and encourages others to try it if it sounds fun.

Deep Learning Is Applied Topology

499 theahura 2025-05-20 13:54:54 UTC theahura.substack.com Llama Share 180

The author discusses how neural networks can be viewed as topology generators, creating a high-dimensional space where data has certain properties. They argue that reasoning itself lives on a manifold and can be trained using neural networks to sort out the topology for good and bad reasoning.

Litestream: Revamped

448 usrme 2025-05-20 19:58:27 UTC fly.io Llama Share 96

Litestream is an open-source tool that makes SQLite applications recoverable from object storage, and it has been updated with new features from LiteFS, including transaction-awareness and a time-based lease. The updates will enable features like read-replicas, failovers, and rollbacks, making Litestream a more robust and efficient solution for SQLite applications.

Gemma 3n preview: Mobile-first AI

436 meetpateltech 2025-05-20 18:03:32 UTC developers.googleblog.com Llama Share 160

Google introduces Gemma 3n, a new open model for on-device AI, offering faster performance, reduced memory usage, and multimodal understanding. Gemma 3n enables developers to build interactive experiences and power deeper understanding with combined audio, image, and text inputs, all processed privately on-device.

AI in my plasma physics research didn’t go the way I expected

363 qianli_cs 2025-05-20 04:57:00 UTC understandingai.org Llama Share 297

Nick McGreivy, a physicist, questions AI's potential to accelerate science due to overoptimistic results and methodological errors. He argues that AI adoption in science is driven more by benefits to scientists than benefits to science itself.

Show HN: 90s.dev – Game maker that runs on the web

348 90s_dev 2025-05-20 14:58:03 UTC 90s.dev Llama Share 111

The user created an API for building and sharing apps, including game maker tools and games, with a 320x180 canvas. The API has a simple GUI, auto-layout system, and refs for watchable properties, allowing for extensibility and community-driven app development.

The emoji problem (2022)

330 mtsolitary 2025-05-20 10:18:15 UTC artofproblemsolving.com Full Share 62

Something appears to not have loaded correctly.

Google AI Ultra

317 mfiguiere 2025-05-20 18:20:47 UTC blog.google Llama Share 326

Google AI Ultra is a new AI subscription plan offering highest usage limits and access to premium features for $249.99/month, available in the US with a special offer for first-time users. It includes Gemini, Flow, Whisk, NotebookLM, Gemini in Google apps, and more.

Why does the U.S. always run a trade deficit?

317 jnord 2025-05-20 11:42:24 UTC libertystreeteconomics.newyorkfed.org Llama Share 776

The US runs a trade deficit due to insufficient domestic saving to finance investment spending, requiring foreign borrowing. The deficit is equal to the gap between saving and investment spending.

AI's energy footprint

308 pseudolus 2025-05-20 10:07:55 UTC technologyreview.com Full Share 341

The emissions from individual AI text, image, and video queries seem small—until you add up what the industry isn’t tracking and consider where it’s heading next.

Is-even-ai – Check if a number is even using the power of AI

300 modinfo 2025-05-20 00:19:44 UTC npmjs.com Llama Share 159

The code uses the IsEvenAi library to perform basic math operations using AI. It imports functions for checking even/odd numbers, equality, and comparisons.

The NSA Selector

299 anigbrowl 2025-05-20 18:30:18 UTC github.com/wenzellabs Llama Share 82

The NSA selector is a Eurorack module that converts network traffic into 4-bit audio. It taps into network traffic and converts it to audio using a delta-sigma modulator.

Show HN: A Tiling Window Manager for Windows, Written in Janet

291 agentkilo 2025-05-20 15:08:42 UTC agent-kilo.github.io Full Share 103

Jwno is a highly customizable tiling window manager for Windows 10/11, built with Janet and ❤️. It brings to your desktop magical parentheses power, which, I assure you, is not suspicious at all, and totally controllable. Note: This documentation is work-in-progress. Some links may return 404, because I have not finished writing those pages yet 😅.

A simple search engine from scratch

286 bertman 2025-05-20 09:58:56 UTC bernsteinbear.com Llama Share 59

A search engine was built from scratch using word embeddings, ranking posts by cosine similarities to query embeddings. The search engine was optimized for web use by splitting the word2vec dictionary into smaller files and using HTTP Range requests to download only needed parts.

What are people doing? Live-ish estimates based on global population dynamics

280 willbc 2025-05-20 01:36:15 UTC humans.maxcomperatore.com Full Share 101

Est. Births/sec: 0.0 Est. Deaths/sec: 0.0 Est. Net Gain/sec: 0.0

“ZLinq”, a Zero-Allocation LINQ Library for .NET

267 cempaka 2025-05-20 22:29:12 UTC neuecc.medium.com GoogleT5 Share 88

Zlinq implements 100% coverage of all methods and overloads in.net 10 (including new ones like shuffle, rightjoin) implementing optimizations beyond just allocation reduction to outperform in most scenarios. a superior alternative to the standard library, we have no plans for commercialization. maintenance has become challenging due to growing scale. when developing large libraries, i need ...

Reports of Deno's Demise Have Been Greatly Exaggerated

221 stephdin 2025-05-20 11:33:14 UTC deno.com Llama Share 200

Deno is not fading away, its adoption has more than doubled since Deno 2's release, and it's now used more widely and seriously than ever. The company is addressing criticism by sharing its plans for Deno Deploy, KV, and Fresh, and is committed to improving communication and transparency.

Red Programming Language

210 hotpocket777 2025-05-20 18:14:02 UTC red-lang.org Llama Share 125

Red is a next-gen programming language inspired by REBOL, aiming to be a full-stack language for system programming and high-level scripting. It offers a single executable for any platform, literate code, and top-flight performance.

The behavior of LLMs in hiring decisions: Systemic biases in candidate selection

191 hunglee2 2025-05-20 09:27:20 UTC davidrozado.substack.com Llama Share 181

Large Language Models (LLMs) consistently favored female-named candidates over male-named candidates when selecting the most qualified candidate for a job. This bias was observed across 22 LLMs and 70 professions, with female candidates selected in 56.9% of cases.

The Dawn of Nvidia's Technology

186 wmf 2025-05-20 17:04:00 UTC blog.dshr.org Llama Share 77

The user, a former Nvidia engineer, compares two books about Nvidia's rise to success, highlighting the accuracy of one author's account of the company's early history and criticizing the other for lacking technical understanding. The user then explains the technical innovations behind Nvidia's early success, including the use of quadric patches and a virtualized objects architecture that ...

OpenAI Codex hands-on review

168 fragmede 2025-05-20 14:29:00 UTC zackproser.com Llama Share 120

Codex is a chat-first experience that allows users to spin up multiple tasks in parallel, but its current limitations make it more suitable for maintenance-level updates and small chores rather than significant refactoring or feature-building. The user is excited about Codex's potential and plans to use it mainly for quick wins, but hopes for future improvements to make it a more effective ...

Clojuring the web application stack: Meditation One

164 adityaathalye 2025-05-20 13:15:52 UTC evalapply.org GoogleT5 Share 37

A minimal web application stack must, at the very least, facilitate the drudgery of interpreting http requests and creating http responses. ring-jetty-adapter is our interface, which wraps over jetty's java apis. this pattern is nothing but the poor man's object-oriented factory pattern. we can't have an eta-reduction in lambda calculus terms. but it is

Robin: A multi-agent system for automating scientific discovery

148 nopinsight 2025-05-20 16:21:03 UTC arxiv.org Llama Share 19

Robin is a multi-agent system that automates the scientific process, generating hypotheses, proposing experiments, and analyzing data. It discovered a novel treatment for dry age-related macular degeneration using ripasudil, a clinically-used rho kinase inhibitor.

Semantic search engine for ArXiv, biorxiv and medrxiv

147 0101111101 2025-05-20 21:49:36 UTC arxivxplorer.com Full Share 31

Thank you to arXiv, bioRxiv and medRxiv for their open access interoperability.

The value isn't in the code (2022)

139 fragmede 2025-05-20 23:33:27 UTC jonayre.uk Llama Share 76

You argue that software code is not as valuable as people think, citing the time and skill required to create it. You claim that the real value lies in the team, business logic, design, and not the code itself.

Biff – a batteries-included web framework for Clojure

136 TheWiggles 2025-05-20 03:13:20 UTC biffweb.com Llama Share 23

Biff is a backend framework that enforces schema with Malli and allows for interactive UIs with hyperscript. It supports live development and deployment with Docker or an Ubuntu VPS.

A shower thought turned into a Collatz visualization

136 abstractbill 2025-05-20 01:04:14 UTC abstractnonsense.com Llama Share 25

The author had a shower thought about the Collatz Conjecture and came up with a way to visualize it using binary fractions. He created an interactive plot that shows the distribution of Collatz fractions for the first N positive integers.

Gemma 3n preview: Mobile-first AI

129 Philpax 2025-05-20 18:29:05 UTC developers.googleblog.com Llama Share 1

Google introduces Gemma 3n, a new open model for on-device AI, offering faster performance, reduced memory usage, and multimodal understanding. Gemma 3n enables developers to build interactive experiences and power deeper understanding with combined audio, image, and text inputs, all processed privately on-device.

Instagram Addiction

124 gregsadetsky 2025-05-20 22:25:56 UTC blog.greg.technology Llama Share 102

The writer was addicted to Instagram for a month, realizing it was an undesirable habit, and left the app for three to four months. They struggled with self-control and found it hard to escape the app's addictive nature, even after reinstalling it.

LLM-D: Kubernetes-Native Distributed Inference

119 smarterclayton 2025-05-20 12:37:47 UTC llm-d.ai Llama Share 15

llm-d is a Kubernetes-native framework for high-performance distributed LLM inference. It leverages optimizations like disaggregated serving and prefix-aware routing to achieve high performance per dollar.

Show HN: Juvio – UV Kernel for Jupyter

115 okost1 2025-05-20 16:32:58 UTC github.com/okua1 Llama Share 26

This script requires Python 3.10.17 and installs dependencies like numpy and pandas in an ephemeral virtual environment. It converts notebooks to script-style format for easy version control.

GPU-Driven Clustered Forward Renderer

114 logdahl 2025-05-20 15:57:01 UTC logdahl.net Llama Share 28

The user developed a GPU-driven forward renderer using clustered shading that can render 27,000 Stanford dragons with 10,000 lights at 1080p in over 60fps on a GTX 1070 GPU. The renderer uses indirect multi-draws, a contiguous buffer for entity data, and clustered shading to reduce the number of draw calls and lighting calculations.

If an AI agent can't figure out how your API works, neither can your users

97 mattmarcus 2025-05-20 14:52:21 UTC stytch.com Llama Share 52

Improving API developer experience for AI agents requires clear documentation, consistent patterns, and actionable error messages to help them learn and use the API efficiently. By treating agent failures as usability bugs and merging DX and AX, developers can create a seamless experience for both humans and machines, leading to reduced support tickets and increased word-of-mouth.

The Lisp in the Cellar: Dependent types that live upstairs [pdf]

87 todsacerdoti 2025-05-20 13:38:07 UTC zenodo.org Llama Share 19

Dependent types allow code to compute types, enabling powerful patterns. The Deputy system is a Clojure-hosted language for dependently-typed programming with inductive datatypes.

The Last Letter

84 HR01 2025-05-20 17:10:15 UTC aeon.co Llama Share 33

The author takes refuge in a Bordeaux bookshop and discovers a collection of last letters written by those executed by the Nazis during World War II. These letters offer a raw portrait of the human condition, revealing the emotions, hopes, and regrets of those facing imminent death.

Ogmo – free, open source, 2D level editor

83 aeontech 2025-05-20 13:42:51 UTC ogmo-editor-3.github.io Llama Share 11

OGMO Editor offers a holistic workflow with project data flowing to level files. It features various layers for level creation, saving to a JSON file for easy access.

Ann, the Small Annotation Server

76 todsacerdoti 2025-05-20 03:45:16 UTC mccd.space Llama Share 11

Ann is a decentralized social media built around Web Annotations, allowing users to store and share comments on content. It enables self-hosted annotation servers for various applications, promoting choice, privacy, and control.

Teachable Machine

72 tosh 2025-05-20 16:07:13 UTC teachablemachine.withgoogle.com Llama Share 12

Teachable Machine is a web-based tool that creates machine learning models without coding. It allows users to gather examples, train models, and export them for various projects.

Show HN: Text to 3D simulation on a map (does history pretty well)

71 lukehollis 2025-05-20 11:43:55 UTC mused.com Full Share 48

Cleo, the mathematician that tricked Stack Exchange

68 schaum 2025-05-20 05:47:55 UTC en.wikipedia.org Llama Share 9

Cleo was a mysterious mathematician on Stack Exchange who provided precise answers to complex integration problems without showing intermediate steps. Cleo was revealed to be Vladimir Reshetnikov, a software developer from Uzbekistan, who created the persona to generate interest in mathematical problems.

Show HN: A free, privacy preserving, archive of public Discord servers

67 searchcord 2025-05-20 02:55:32 UTC searchcord.io Full Share 68

I loved this project and I was happy to work on it. A research article is in the works and will be published soon... thanks! ♡

It's Breathtaking How Fast AI Is Screwing Up the Education System

61 prawn 2025-05-20 04:08:50 UTC gizmodo.com Llama Share 67

The AI industry is disrupting education, causing students to rely on AI for assignments and learning, leading to intellectual stagnation. This trend is also affecting teachers, who are using AI to create lesson plans, further exacerbating the problem of intellectual decline.

Show HN: Astra – a new js2exe compiler

60 qwertycodepl 2025-05-20 14:55:25 UTC github.com/astracompiler Llama Share 32

Astra is a tool for compiling servers and CLIs, not replacing Electron. It supports ESM-based applications with Node.js workarounds.

Ask HN: When will managers be replaced by AI?

59 GianFabien 2025-05-20 02:28:29 UTC news.ycombinator.com Llama Share 83

AI may replace middle-management roles that focus on information flow and team facilitation, but top-level management and leadership skills like networking may remain human domains. The real concern is not AI replacing managers, but rather overemphasizing AI-related skills in entry-level workers.

The Fractured Entangled Representation Hypothesis

57 akarshkumar0101 2025-05-20 15:56:04 UTC github.com/akarshkumar0101 Llama Share 22

Researchers Akarsh Kumar, Jeff Clune, Joel Lehman, and Kenneth O. Stanley compared neural networks evolved and trained via SGD, finding internal representation differences. Evolved networks lack "fractured entangled representation" and approach a unified factored representation.

Memory Consistency Models: A Tutorial

57 tanelpoder 2025-05-20 01:25:01 UTC jamesbornholt.com Llama Share 5

Memory consistency is a challenge in computer science where multiple threads access shared memory, and different models like sequential consistency and total store ordering define how these threads can observe their shared memory state. Modern architectures often use store buffers and weak memory models to optimize performance, but this can lead to unintuitive behaviors and the need for ...

Hypervisor as a Library

54 ingve 2025-05-20 06:07:19 UTC seiya.me Llama Share 3

The author created a Linux compatibility layer for the Starina OS by running the real Linux kernel in a lightweight virtual machine. This approach allows for seamless integration with Linux and provides a simple API for running Linux binaries on Starina.

Quantum Picturalism

51 mathgenius 2025-05-20 20:32:08 UTC quantuminpictures.org Llama Share 19

Quantum Picturalism breaks down complex quantum concepts into simple visual learning experiences using addition, subtraction, and angles. This approach makes quantum math accessible and inclusive for teachers and learners of all ages.

Show HN: A Simple Server to Match Long/Lat to a TimeZone

48 ChrisMarshallNY 2025-05-20 21:43:50 UTC github.com/littlegreenviper Llama Share 27

This project is a simple PHP server that accepts GeoJSON output and provides a basic API to match longitude/latitude locations with timezones. It uses a shapefile to locate points within polygons and returns the standard TZ time zone designator of the timezone that covers the point.

Questioning Representational Optimism in Deep Learning

46 mattdesl 2025-05-20 06:54:27 UTC github.com/akarshkumar0101 Llama Share 7

Researchers Akarsh Kumar, Jeff Clune, Joel Lehman, and Kenneth O. Stanley compared neural networks evolved and trained via SGD, finding internal representation differences. Evolved networks lack "fractured entangled representation" and approach a unified factored representation.

ChatGPT Helps Students Feign ADHD: An Analogue Study on AI-Assisted Coaching

44 paulpauper 2025-05-20 17:57:20 UTC link.springer.com GoogleT5 Share 46

A preregistered study aimed to assess whether ai-generated coaching helps students to successfully feign attention-deficit/hyperactivity disorder (adhd) in adulthood. researchers have found that examinees seek out online information to help them produce test results that mimic those of individuals genuinely diagnosed with adhd while evading detection. the scientific examination of coaching’s ...

Show HN: JavaFactory – IntelliJ plugin to generate Java code

44 javafactory 2025-05-20 11:29:45 UTC github.com/javafactoryplugindev Llama Share 26

JavaFactory uses LLMs to generate repetitive Java code with predictable results. It operates based on pattern definitions and user-configurable prompts to automate code generation.

Old Growth Wood

38 ksec 2025-05-20 00:32:37 UTC brenthull.com Llama Share 14

Historic windows made with old growth wood last longer due to better stability and durability. New growth wood, used in most modern windows, begins to rot and warp after 20 years.

Show HN: Olelo Foil - NACA Airfoil Sim

35 rbrownmh 2025-05-20 16:44:41 UTC foil.olelohonua.com Full Share 19

NACA airfoils are mathematical representations of airfoil shapes developed by the National Advisory Committee for Aeronautics (NACA). These airfoils are widely used in aerodynamics for designing wings, blades, and other aerodynamic surfaces. You can learn more about airfoils here.

Launch HN: Opusense (YC X25) – AI assistant for construction inspectors on site

33 rcody 2025-05-20 15:35:29 UTC news.ycombinator.com Llama Share 19

Roya and Michael co-founded Opusense AI to help engineers generate construction site reports automatically from typed or voice notes, photos. The tool uses LLMs to transform structured inputs into structured outputs, reducing tedious reporting work.

Photos in a Similar Style Aren't Copyright-Infringing–Woodland vs. Lil Nas X

31 hn_acker 2025-05-20 20:40:36 UTC blog.ericgoldman.org Llama Share 11

A court case involved a freelance artist, Rodney Woodland, suing Lil Nas X for copyright infringement over semi-naked photos. The court ruled that Woodland failed to prove access and similarity, dismissing the case due to lack of evidence.

I trapped an AI model inside an art installation [video]

31 handfuloflight 2025-05-20 07:52:25 UTC youtube.com Embed Share 2

Our Journey Through Linux/Unix Landscapes

19 alekq 2025-05-20 21:16:27 UTC blog.kalvad.com GoogleT5 Share 17

Cnn's john sutter recently visited a university in dubai. he was looking for fresh talent and decided to switch to freebsd and alpine linux instead of rpm based distros. the team is impressed by the number of vulnerabilities they've been able to reduce. check out the full list of winners here. and don't forget to share your thoughts on the distributions you're considering.

Can Sam Altman Be Trusted with the Future?

19 wouterjanl 2025-05-20 11:41:27 UTC newyorker.com Llama Share 13

OpenAI's Sam Altman led a team that created the breakthrough language model GPT-2, but his leadership was marked by controversy and power struggles. Altman's vision for AI was shaped by his Silicon Valley upbringing and his desire to create a technology that would "belong to the world."

What happened to American manufacturing is not unusual

18 nabla9 2025-05-20 12:32:35 UTC blog.waldrn.com Llama Share 19

President Trump's protectionist trade policies won't reverse US de-industrialization. Economists recommend public investment in declining communities through education and job training.

Firewalls and BIOS's and Coreboot (2020)

16 transpute 2025-05-20 12:21:58 UTC hagensieker.com Llama Share 6

You're a computer security enthusiast who values Open Source software and believes it's essential to have Open Source firmware for maximum security. You've taken steps to secure your hardware firewall appliance by installing Coreboot and updating the firmware.

The Entropy of Thought: Why Our Minds Gravitate Toward Divided Realities

16 victorasao 2025-05-20 07:37:00 UTC victoranastasiu.medium.com Llama Share 5

Beliefs form through emotional associations with external stimuli, stored in an unconscious library, influencing our perception of the world. This process, called the intelligence of rhythm, affects both physical and digital patterns, impacting our mental state and behavior.

Show HN: Output – An AI Agent That Controls Your Computer (Mac and Windows)

15 sudhar172 2025-05-20 10:44:42 UTC theoutput.co Full Share 14

Output is the first AI that truly uses your computer for you. Everything you'd do, now done for you.

On-device small language models with multimodality, RAG, and Function Calling

14 cuuupid 2025-05-20 18:40:35 UTC developers.googleblog.com Llama Share 1

Google AI Edge introduces RAG for on-device Retrieval Augmented Generation and function calling for interactive language models. These features enable developers to create intelligent features grounded in user-relevant information and allow language models to call predefined functions or APIs within applications.

KumoRFM: Gen-purpose model for making instant predictions over relational data

14 agold97 2025-05-20 18:02:14 UTC kumo.ai Llama Share 0

KumoRFM is a Relational Foundation Model that makes accurate predictions on relational databases without task-specific training. It outperforms conventional approaches by 2-8% and is orders of magnitude faster.

Center for AI Safety's new spokesperson suggests „burning down labs"

14 MrBuddyCasino 2025-05-20 07:52:31 UTC twitter.com/drtechlash Full Share 25

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.

A Scientist Fighting Nuclear Armageddon Hid a 50-Year Secret

14 greesil 2025-05-20 03:16:23 UTC nytimes.com Llama Share 1

Richard L Garwin, a nuclear scientist, was inspired by Enrico Fermi to speak out on public policy after Fermi's death in 1954. Garwin spent his life countering nuclear horrors, leaving a legacy after his own death at 97.

Still Booting: People Stuck Using Ancient Windows Computers

12 andrewl 2025-05-20 02:46:54 UTC bbc.com Llama Share 2

Microsoft's legacy of outdated software and hardware continues to impact society, with many systems still relying on Windows XP and older operating systems. This phenomenon is due to Microsoft's business strategy of supporting older software and hardware, making it difficult to upgrade and replace outdated systems.