Google 发布 Gemini 4 Argon 前沿模型，大模型生态迎来多方进展

Nvidia's Nemotron 3.5 Bolt 74B Brings 74B AI Models to Local Hardware

NVIDIA is making a serious push into local AI agents with a new language model designed to fit an unusually large number of parameters into just 24 GB of memory. The upcoming Nemotron 3.5 Bolt 74B combines a 74 billion parameter mixture of

AI guardrails designed to stop hackers are now blocking defenders, and Orion-1 aims to fill the gap

Artemis Security's new Orion-1 model is purpose-built for cyber defense and beats Claude Opus 5.5 and GPT-6 Sol on the company's own benchmark. The model fills a growing gap as general-purpose frontier models lock down their cyber capabilities, blocking defenders in the process. The article AI guardrails designed to stop hackers are now blocking defenders, and Orion-1 aims to fill the gap appeared first on The Decoder.

Gemini 4: A New Hope for AI Hackers

Google’s newest model just took #1 on Tenzai’s Vulnerability Discovery leaderboard. We got early access and ran it through long, multi-step attacks to see if it holds up.Gemini 4 is the most capable Google cyber model we have tested for aut

SynthID Bio for AI-generated proteins

September 30, 2026 SciencePushmeet Kohli, David Stutz, Ali Cowen-Rivers and Jeremy RatcliffProof of concept for watermarking AI-generated proteins while preserving biological function.Today, we’re introducing SynthID Bio to bring watermarki

Google's new model, Argon, is only avaliabe to ultra users on Antigravity

Argon is apparently better at solving spatial problems in 3D environments. Credit: Google Argon is apparently better at solving spatial problems in 3D environments. Credit: Google For your general AI queries, Gemini 4 Argon will be rolling

iPhone断网跑35B多模态!国产开源新模型,只占3G内存

面壁开源MiniCPM-V 4.7

Meta 新模型杀回AI第一梯队,Alexandr Wang 亲述 Muse 7个月翻身内幕

Gemini 4 Argon is now available worldwide as Google's most capable frontier model

Google is making Gemini 4 Argon available worldwide through its API, the Gemini app, and Google Antigravity. The frontier model is aggressively priced but trails Anthropic's Claude Opus 5.5 and Meta's Muse Wonder Full in independent benchmarks. Internally, Google says Argon already cut coding tasks from months to weeks. The article Gemini 4 Argon is now available worldwide as Google's most capable frontier model appeared first on The Decoder.

Claude 开始自己写程序管理 Agent 了:Anthropic 披露动态工作流架构

Show HN: The first open-weight model for general deterministic task

Hey HN! Today we’re releasing interfaze-1.1-lite, an open-weight model for tasks with a deterministic answer: OCR, STT with speaker diarization, classification and more.Lite is a reasoning core plus a collection of specialist models (document reader, line detector, speech recognizer, segmentation). The core chooses the specialists to be executed and produces the answer in your JSON schema.The specialist models give you: bounding boxes, word timestamps and a confidence score per each line and word. The raw outputs are returned next to the answer so that you can filter based on that. E.g., post rows above 0.9 automatically and forward the rest to a human.Runs on a single 80 GB GPU (e.g. H100) without any additional services.Happy to answer any questions regarding the architecture, benchmarks and limitations.

Run fast, private agents on-device with LFM2.5-24B-A2B

LFM2.5-24B-A2B is a 24B-parameter reasoning model with just 2B active parameters, built for fast, private agents on-device. It thinks before it answers, calls tools, and works through multi-step tasks inside popular agent harnesses. Because

Gemini 4 Argon is now available for developers

Starting today, developers can build with Gemini 4 Argon, our frontier model for complex workflows across real-world software engineering, enterprise knowledge work, and cybersecurity defense.Gemini 4 Argon is designed for those long-horizo

Google finally releases Gemini 4 Argon

After months sitting on the sidelines of the frontier AI race, Google has finally launched its long-awaited Gemini 4 Argon model, claiming it can compete with the best the industry has to offer. Google says Argon delivers "frontier reasoning and improved performance across complex workflows," while cutting down on hedging, repetition, and the cliches that […]

Gemini 4 Argon: Built to solve your hardest problems

We’re giving developers, enterprises, scientists, and consumers access to Gemini 4 Argon, our latest frontier model.

Celestial AI Introduces Photonic Fabric [2x HBM3e and DDR5 Memory] Module (2025)

August 29, 2025 -- Celestial AI, the creator of the Photonic Fabric™ scale-up networks for accelerated computing, has introduced the Photonic Fabric Module, the world’s first System-on-Chip (SoC) with optical interconnect integrated in the

Cloudflare Workers支持入站TCP,gRPC成为首个基于此构建的协议

Make Your Coding Assistant a 10x AI Engineer

Here’s a TLDR:I built agent-engineer, a free and fully open-source plugin. It works on Claude Code, Codex CLI, Cursor, OpenCode and Gemini CLI.It contains 11 skills, 4 subagents and 5 hooks. The skills carry agent-engineering knowledge (too

Show HN: Veduta – Personal Agent with GenUI

The idea came to me while I was watching a livestream with Antirez and mr RIP, one of the first streams they did together. I thought Antirez might be right when he said that once AI agents moved beyond the terminal, we would reach a whole new level. So I decided to create "Veduta", which is basically Hermes/OpenClaw but with a UI. What exactly is the advantage? It could track everything for you and give you visual feedback without you actually having to ask the same questions again, because all the information is stored inside the daemon running on the server. Instead of having something that just runs, you have something that delivers "sheets with written reports that you can consult at any time." This translates into saving tokens and accessing information faster, while also keeping everything organized. I don't know if you'll appreciate this open-source project, but feel free to leave any kind of feedback. It's always welcome! :) Thanks, everyone!

Show HN: Markdown to Doc in a Click

Everything is markdown now thanks to AI agents, which is honestly quite convenient for all things development, but once in a while you might want to share markdown with your non-technical friends, so what do you do then?MD2Doc of course! It was another one of those little problem-solution projects which seem unassuming, but still consistently get about 9k users a month.I am torn because on one hand, it's trivial, on the other, maybe it makes sense to grow it? This was just a weekend project. What would you do?https://md2doc.com

Show HN: An embeddable DOCX/XLSX/PPTX editor SDK for your AI agent

Show HN: An embeddable DOCX/XLSX/PPTX editor SDK for your AI agent

DeepMind and Singapore fine-tune an on-device Gemma model for scam detection

October 14, 2026 Responsibility & SafetyJJ Geewax and Miguel de AndrésCollaborating with the Government Technology Agency of Singapore (GovTech Singapore) to protect users without compromising privacy.Scams, such as investment fraud, fake j

Show HN: My AI agent kept quoting undated market numbers, so I built PreReason

Show HN: My AI agent kept quoting undated market numbers, so I built PreReason

Show HN: Busymail – an IMAP email client with an MCP server for AI agents

An email client that reads your inbox - and lets your AI work in it.Busymail connects to your Gmail, Yahoo, or any IMAP server Busymail can reach, and surfaces the few things that need you - the reply, the bill, the flight. Give an AI clien

Show HN: Geneva, a Rust FFmpeg CLI alternative with HTML/CSS graphics

I built this for my own use case, and I think other people could benefit from it.It's essentially a Rust rewrite of the ffmpeg CLI command, but with a modern API and good defaults + a native HTML/CSS composition engine that's been fine tuned for speed and memory efficiency. Under the hood, it still uses the libav* codecs from ffmpeg, so it can still work on a variety of formats.Works very well with AI agents too: prompt what kind or intro/animation/graphics you need (or give it a reference), let the LLM come up with the HTML/CSS for it (Opus 5.5 is quite good at this), and let it use them on geneva.It's way faster than the usual alternatives for putting still or animated graphics on videos (remotion, playwright etc.) because it doesn't use a browser or chromium. Everything is rendered natively, in a single pass.

Show HN: boostmyagents – A README badge lets your users pay for your AI usage

Create a boost page and let your community pay for the Anthropic API usage behind Claude Code and compatible tools. Boosts fund AI usage only, never cash. Free for developers · Public GitHub identity only · No repository access How it works

My Friend Has No Soul

The second iteration of the viral Friend AI necklace is louder and less interesting.

Show HN: tsv has fast formatters and parsers for Svelte, TypeScript/JS, CSS

tsv is a set of performance-focused language tools. Today it ships formatters and parsers for TypeScript/JS, CSS, and Svelte, including a vscode/openvsx extension. JSON and HTML are planned, maybe linting after. The readme has a loose roadmap - https://github.com/fuzdev/tsvtsv's benchmarks show how the Web's newer native tools have some performance left on the table. Formatting TypeScript, tsv is more than twice as fast as oxfmt, and is smaller and uses less memory - https://tsv.fuz.dev/docs/benchmarksFrom the site:- Formatting TypeScript, tsv is ~2.2x faster than Oxfmt (native-vs-native), ~36x faster than Prettier (native Rust vs Prettier's JS; ~26x as wasm), and ~7.6x faster than Biome (wasm-vs-wasm) - Formatting Svelte, tsv is ~79x faster than Prettier (~55x as wasm) and ~10x faster than Biome (wasm-vs-wasm) - Formatting CSS, tsv is ~3.6x faster than Oxfmt (native-vs-native), ~111x faster than Prettier (~78x as wasm), and ~12x faster than Biome (wasm-vs-wasm)tsv is focused on the Web and Svelte, and doesn't support other nonstandard languages like JSX/TSX and SCSS. This limits possible adoption, but it keeps the tool slim and simple, and there are plenty of other tools that offer a generic language platform - tsv isn't trying to compete with them. Part of my motivation was being tired of slow tools and Svelte's second-class status in this next generation.Before starting the project, I had tried to build on Oxc and Biome, and ran into some discouraging edges. On the benchmarks' language support section, you'll see Oxc doesn't export its CSS parser (recently added, a fork of Raffia - https://github.com/g-plane/raffia), and Biome doesn't export its parsers. Also oxfmt doesn't ship wasm to the browser yet, and tsv's smaller footprint is useful there (Biome is 46.9 MB to tsv's 2.2).Parser conformance is as good or better than Oxc, Biome, and swc, except negative parser cases (errors), which are unfinished. https://tsv.fuz.dev/docs/conformanceThe codebase turns 1 year old at the end of the month. It's my first serious AI-generated project, built with Claude Code. I learned Rust in 2015, built some toys including an arcade action prototype, and followed its development over the years, but this is my first project since. tsv wouldn't exist without LLMs - it requires far too much labor for a side project.tsv leans heavily on Prettier, but doesn't copy it. This is why this production release took almost a year, despite the silly fast coding part. I made some different stylistic choices, more for Svelte than TypeScript, and fixed numerous quirks and bugs and documented them. tsv also takes advantage of Svelte 5's whitespace rules, which further diverged from HTML to trim leading whitespace in tags, fixing longstanding formatting warts - https://svelte.dev/docs/svelte/v5-migration-guide#Whitespace...The formatter takes no style config, and instead is hardcoded to Prettier's style with Svelte's config. (which has the uncommon trailingComma:none, but it's good enough for the Svelte team, and I want to defer the choice to them) Supporting Prettier's config would force significant costs on tsv for JS runtime integration, so instead I'm keeping the formatters simple and non-configurable instead of making all users pay those costs. One example of these costs can be see in the CLI benchmark (which is a fork of Oxc's official bench-formatter) - when tsv's binary is used directly instead of through Node, the single-repo CLI benchmark runs tsv 1.7x faster using less than half the memory.Next I'm exploring rearchitecting to Zig-inspired data-oriented parsing - https://mitchellh.com/zig/parser - which should get tsv closer to Yuku's speed, measured on the site.The vscode extension is `fuzdev.tsv-format` - https://marketplace.visualstudio.com/items?itemName=fuzdev.t... - and Open VSX - https://open-vsx.org/extension/fuzdev/tsv-formatRepo: https://github.com/fuzdev/tsv

Show HN: Kryzz AI – local-first Android agent workspace (MIT)

Public v1.1.2. The source in this repository is internal build v5.11.0 (versionCode 84). Public v1.1 was internal v5.7.2. Public v1.0 was internal v5.1.15. Builds 1.x through 5.1.14 were never published. There is no public v1.1.1. A private

AI-assisted PCB design for KiCad 10

Konnect *BETA Release AI-assisted PCB design for KiCAD 10. Konnect is a native KiCAD plugin — a single Rust binary — that lets Claude and other AI assistants design schematics and PCBs through the Model Context Protocol (MCP). 228 tools acr

OwlLayer – Make your web UI drivable by AI

The open-source Agentic UI SDK: your buttons, links and forms become tools an AI agent can call. The agent lives inside your app, so users just ask, in text or by voice, with nothing to install on their side.Neural-DOM BindingThe page expos

Show HN: Mnemify – 3D terrain map merging Confluence, Notion, Obsidian

I have been working on documents for last 5 years. Recently I realized I was using Confluence to structure execution notes, Notion for thoughts during meetings, Obsidian for quick local notes, and then reading back notes from Googles note taking apps. My colleague had the exact same problem. We had MCP connected to Claude, but there was no way to bring all of that into one place that was intuitive. So we built Mnemify. Which converted all our notes from every app to a 3D traversable terrain instead. We’ve been using it daily for last two months and found it useful so decided to share it.What it does: it harvests your sources into a local folder, then using your AI subscriptions compiles them into a 3D terrain you can traverse. Regions and peaks come from context, so related things end up adjacent from all your connected apps.We also have chat over the compiled graph that cites the exact source passages.You can also view time wise how the terrain was modified over time and see all your note higlights and action items in workspaces.Everything stays on your machine. It never writes back to any source. Models are bring your own: OpenAI, Claude (your claude CLI or a subscription), or a fully offline mode with an on-device embedding model.Currently supported sources: Confluence, Notion, Obsidian and local files. Jira, Linear, Slack, Gmail and Google Meet notes are built and coming...Please try it out, would appreciate your feedback.Repo: https://github.com/mnemify-ai/mnemify Our Landing: https://mnemify.ai

Show HN: AI investment agreement review for startup founders

Hi HN! First, congrats to the founders who received a term sheet(and those who will!)I made an investment agreement review tool for founders who recently received a term sheet! You can review your term sheets, SAFEs, convertible notes, and financing agreements if your corporation is incorporated in Delaware, from pre-seed through Series A.It also reviews equity purchase agreements, investor rights, voting terms, and side letters.Try it and check your potential risks before signing!

What breaks if I change this? across SQL,dbt and Airflow

What breaks if I change this? Cartographer finds the dependencies your language server can't see: a table written by a Python job, transformed in SQL, consumed by a dbt model. It maps them statically and locally, with file and line evidence

Show HN: Chiplab - Give your AI agent access to virtual hardware

Chiplab by Veecle is an MCP platform that lets AI agents compile, simulate, and validate firmware on the actual target chip without physical hardwareSolutionWHAT IT ISThe API between agents and siliconAgents call Chiplab to run firmware on

Demonstrating Deterministic Trust and Formal Proof for AI-Generated Code

High-integrity software runs the world quietly; traffic lights, point-of-sale terminals, industrial manufacturing, medical devices, cars, trains, airplanes. We rarely notice it until it fails. And when it fails, the consequences are real: b

阶跃终端首款「Agent手机」STEPX Neo正式发布,创客先行

这妥妥是个物理版Agent啊！

TurboScribe AI – 100% Free Video to Text and Audio to Text

TurboScribe AI Studio Use Cases FAQ Transcribe Free No sign up. Use TurboScribe AI to convert video to text and audio to text online, then export TXT, SRT or VTT. Click or drag a file below. TurboScribe AI Transcription Studio Engine: Whisp

Show HN: Agent Identity – Identity and communication infrastructure for AI agent

Agent IdentityInboxes, phone numbers, secrets, public endpoints and permissions for your agents, in one SDK.$npm install @agentidentity/sdkLIVE · SUPPORT AGENT4 → 1EmailSMSCallWebhookAGENT OUTPUTemail.received → reply.sent“Your order #4821

STEPX首款大模型原生智能体手机STEPX Neo正式发布

2026年10月13日，大模型原生AI终端品牌STEPX（阶跃终端）旗下首款大模型原生智能体手机STEPX Neo正式发布

Show HN: SuperLocalMemory" AI brain that never bluffs: Laya, Jev, Google Gemma

The governed memory brain for every AI agent. It lives on your machine. It never bluffs. One memory for every agent you and your team use — Claude Code, Codex, Cursor, Copilot, Antigravity, Hermes, ChatGPT, Claude on the web, Grok Bot, Muse

Show HN: Fabee, an AI nutrition coach, get advice that fits you.(iOS)

Hi, I'm Fabee, your AI nutrition coach. Fabee coaches. Calorie apps count. Get advice that fits you. Text what you ate or snap your plate. Fabee counts the calories and macros and helps you hit your goal, whether you’re losing, gaining or h

Show HN: AgentGate – Fast authorization gateway for AI agent tools

AgentGate is an out-of-process authorization proxy and security gateway for autonomous AI agent tool execution. It intercepts tool invocations prior to runtime, evaluates prompt injection and blast-radius risks via a sub-40ms neural classif

Mano-P: Setting Up an On-Device GUI Agent Without Cloud Dependencies

GUI-Aware Agent Model for Edge Devices Private AI 📖 Overview Mano-P: "Mano" means "hand" in Spanish, and "P" stands for Private. We believe that both individuals and organizations can create their own Private AI, and a bright future of hum

Memory and dreaming: how Devin learns from working with you

Today we’re introducing Memory and Dreaming in Devin. Memory lets Devin carry useful learnings about the way you like to work across sessions: your preferences, corrections you’ve made, and lessons learned about your projects and workflows.

响应速度比Claude Code快3倍,实现真正无限上下文!APUS AI开源Reflex Agent

10月13日，APUS AI（麒麟合盛网络技术股份有限公司）宣布开源Personal Agent产品Reflex Agent

Full list of uncensored AI chatbots

The Nonrestrictive AI Directory A directory of freer AI chat Hosted uncensored bots, unfiltered AI APIS, and local abliterated models that don't have filters or guardsrails collected in one place. Labels like “uncensored” are not guarantees

Show HN: GenClass, A 10 MB in-browser model that catches bugs that never throw

Daybot is ranked number two in Canada, after Shopify, for making e-commerce stores.With a lot of AI-generated stores like Daybot's, we ran into the issue of the CSS sometimes not being up to standard. For example, the text and the background might end up the same color. I've also seen that race conditions often happen on the client side and are never caught in test cases, because they aren't tested under load, or because clients sometimes do actions that are just random.With AI-generated code becoming more popular, we thought that an AI agent that runs in the browser at runtime, is very lightweight, catches bugs like race conditions, stops unintended behavior, and gets better the more users use it would completely change the game. Now you can even run code faster.So I want you to try it out. You can add it with one line. It's published as an npm package. Please let me know what you think.

Show HN: Evaluate and improve your agent steering

Hi HN, excited to be open sourcing two tools that we have built over past few months which helps to understand your own work with AI coding agentsAgent Desk is a local macOS app which allows you to browse your sessions across all top coding agents - Claude code, cursor, codex, github copilot, Kiro, Antigravity. You can find sessions by your projects, revisit those conversations, understand and inspect your entire harness - rules, skills, hooks, subagents.https://github.com/tetriz-ai/agent-deskSession Intelligence is a go library which helps structure and analyse your session transcripts using your own LLM. It can help understand missing context, repeated corrections, if session reached its goal and where is effort being wasted. It can also surface leaked credentials and PII.https://github.com/tetriz-ai/session-intelligenceBoth are MIT-licensed.You can also combine both together and understanding your agent steering habits better.Excited about the community using it.

告别资料散落,让知识持续积累:openJiuwen首发个人知识空间,重塑个人办公智能体新体验!

一次采集，反复使用

Show HN: Can AI read your mind?

Humans are terrible at being random. Tap or use arrow keys to pick left or right. A Markov chain learns your pattern — 🧠 ← → Tap or press ← → to start Accuracyrealprojected Chart builds as you play ← → keys or tap How it works The predicto

Show HN: Benchmark coding agents on deploying apps

A benchmark of what it takes for an AI coding agent to build a web app and deploy it to a cloud. Claude Code builds five kinds of app from a written spec and deploys each one to Cloudish, DigitalOcean, Render and Vercel. An independent veri

Show HN: Jevcrawl – A web crawler for AI agents with no LLM in the crawl loop

Why jevcrawl Agents that browse a site have two bad options. They can read everything a crawler fetches, which buries the answer in navigation and boilerplate and burns tokens. Or they can let the LLM pick each link, which is slow and costs

Show HN: Pulse Local, a Mac budgeting app whose AI models run on your machine

Hi HN, I'm Niz. Pulse Local is a budgeting app for Mac (Apple Silicon, beta) that answers questions about your spending with models that run on your machine. Local mode needs no account. The ledger and your questions stay on the machine, and the app sends no analytics or crash logs.I felt there are not enough consumer apps that run locally with no calls to the outside world that help us manage sensitive corners of our lives. This app in its local variant only downloads a couple of models and occasional updates, it does not send any analytics nor data.Features-wise, Pulse local is a complete budgeting app that offers an assistant built on top of a fine-tuned model and thousands of rules embedded in the code itself. I'm sharing it for free with options to help me continue experimenting with fine-tuning and expanding along this direction.How it works: you import a CSV from any bank, in any currency. When you ask a question, a tuned Qwen3 0.6B model reads it into a structured query, and code, not the model, does the money math against the ledger. A 4B checker model double-checks each money answer; when the two readings disagree, the app says it is not sure instead of guessing. That change took wrong answers from about one in five to 5 in 494 questions across our last three tests. The write-up covers how we test, the designs that failed, and the limits of the claim: https://pulseos.finance/pulse-local/writing/agree-or-not-sur...The tuned weights are on Hugging Face (PulseOS/pulse-local-qwen3-0.6b).Limitations: the first start downloads about 2.9 GB of models. With the checker on, a money answer takes 12 to 14 seconds (you can turn the checker off). Small models can still misread an open-ended question. The app answers in English. Mac only for now; Windows is next.The app is free. An optional Connected mode (Sync Pass, $29/year) syncs US banks through Plaid; transactions pass through our server's memory and are not stored.Feedback welcome, especially on wrong answers and on questions it should handle but does not.

AI Face Rating – Discover Your Face Rating Free

FREE · PRIVATE · INSTANTUnderstand your Face Rating and track aesthetic potentialAI-powered analyzer: overall score, feature scores, and a personal plan — from a single photo.HOW IT WORKSUncover Personalized Insights in 3 Simple StepsUpload

Show HN: Sugabots – Group chats where your team and AI agents work together

Showing off Sugabots, an internal project I built at work that we've decided to open-source.I got tired of the team just copying context between agents and Slack. So I built an internal tool for shared chats where our team and agents could work together, using our choice of models and tools.The team is now actively maintaining and building on it to add all the features we're missing. We've shared a hosted offering with a few other teams and has proven useful for them too.Its open-source (MIT) and self-hostable if you want to use it yourself.Let me know what you think.

手机解锁先见 AI,不再先找 App:阶跃星辰首款智能体手机来了,定价 3999

Show HN: Keep your eyes in one place while typing

I wanted a way of typing that's tuned towards stream of consciousness/get down a bunch of words quickly. And I'm too lazy to move my eyes, so I wanted something where I could relax my eyes/brain and just get words out.

Show HN: Parsimony – an open-source benchmark for how much code AI models add

Existing benchmarks measure solve rate and token output/cost, but don't punish excess code, which makes a codebase harder to work with in the long term. I wanted to measure this since it matters most for long-term projects.Method: it takes the patches that benchmarks already publish (DeepSWE, SWE-bench Verified, SWE-bench Live) and counts "coding units": syntax elements like statements, calls, names and literals. I didn't use lines of code, so cramming code isn't rewarded. Deleted code is subtracted, so it measures net units.GPT-6 and Claude 5.5 have no public patches yet, so I ran them myself on 3–10 tasks on my own subscriptions (shown as faded circles, not ranked).Any feedback or model contributions are welcome, especially public patch exports for newer models.GitHub: https://github.com/La5u/ParsimonyNot affiliated with any of the benchmarks.

Show HN: Tern – Native macOS SSH Client in Rust/GPUI (76 MB vs. Tabby's 580 MB)

On my day to day work i mostly use termius and tabby as multi project ssh tool with vault, somehow most of the features i use are just vault and password manager, and they are bulky, takes about 500mb ram for such simple program, and termius wants paid plan just to sync the vault. There are free tools but their ux is not at that level.So i built my own simple ssh client with a little bit of my taste added, it has vault, password manager, repo sync, background gallery etc.I have been working with rust on my hobby projects about 2-3 years and really got headaches from it, because its really hard to maintain features on rust project while fighting the borrow checker. Now with claude and some llm tools its very easy to build things, even some of the code written by agents is still slop but its fine to live with some slop. The main parts are modified from well written software.Instead of cheap copying or writing from scratch, i copied from real world solutions with apache or mit license, well maintained and well written software that has almost no ai slop.With some amateur level rust knowledge and some inspiration projects its really easy to build such low memory software, it only takes 85-125mb with wallpaper, mostly 80mb ram.Here's the repo https://github.com/your-moon/tern any feedback and feature requests are welcome, i gladly answer

Show HN: Faster, Cheaper Databricks Spark with Real-Time ML Config Management

Most Spark jobs run on wrong-sized machines with incorrect configs, leading to slow jobs, wasted spend, or both. We've found that we can get better performance at a lower price by continuously monitoring each job's performance and changing

Show HN: CoreXTrader – A cross-platform Trading DSL and Voice AI terminal

Professional Trading PlatformCoreXTrader: Not Just Trading —A Revolution in Market IntelligenceA comprehensive trading platform with custom strategy development, advanced analytics, and seamless market connectivity.Works with your broker or

Marketing Agent for Founders

Bono 是一款面向创始人的 AI 营销 Agent。用户只需与其对话10分钟，即可生成符合个人语气的博客文章、社交媒体内容、新闻通讯等多种营销素材，实现多平台内容发布。

Sakana AI – Building Frontier AI in Japan

Building Frontier AI in Japan Scroll Frontier AI the world can rely on Sakana Fugu One model that orchestrates the world's best LLMs Pareto Frontier Orchestrate models to push the Pareto Frontier of cost vs performance. Resilience Keeps run

Show HN: Commit-Aware Router, route LLM traffic around your AI/cloud commitments

LLM routers can optimize for different factors like price, latency, quality, etc. Enterprises tend to have prepaid AI credits, minimum-spend agreements, cloud commitments, or commercial ceilings across OpenAI, Anthropic, AWS, Azure, and other providers. We’ll call these “AI commits”.This project is designed to consider the state of an enterprises’ current AI commits and use it to inform LLM routing decisions alongside other common variables (price, performance, etc.). This is an early PoC built using LiteLLM.Target audience would CFOs, Strategic Finance, FinOps, Cloud Finance, and AI Platform folks who care about making sure AI commit minimums are met, and maximums are not exceeded.Would love any feedback from folks!

Show HN: Hatoba, an open-source SSH client with an AI assistant and E2EE sync

I've used SSH tools for years and didn't really plan on building yet another SSH client, but I ended up building one the way I wanted. Hatoba is a desktop SSH client for Windows, Linux and macOS. It's written in Rust with Tauri 2 and MIT licensed.Each terminal has an AI assistant panel next to it. You describe the problem, and it reads the terminal screen, runs commands, types into the shell, and reads, edits and searches files on the host to find the cause and fix it. You can approve every command and edit it before it runs, let a model you pick approve the safe ones, or turn approval off. Files it changes show a diff you can undo. It works with any OpenAI-compatible or Anthropic API, local models included, and supports skills and MCP servers.Sync doesn't go through a server of mine. Hatoba deploys a Worker and a D1 database to your own Cloudflare account and syncs hosts, keys and AI conversations through them. The free plan usually covers personal use. Everything is encrypted on the device before it leaves. Keys are derived from your master password with Argon2id, every item is encrypted with AES-256-GCM, and the Worker only stores ciphertext plus the hashes it needs to sign you in. Someone who gets hold of the whole Cloudflare account still only gets ciphertext. Sync is optional, and everything works offline without it.The usual SSH client features are there too: groups, tags and fuzzy search, ~/.ssh/config import, ProxyJump and SOCKS5/HTTP proxies, an SFTP panel, local port forwarding, and a key vault that imports OpenSSH, PEM and PuTTY keys. There's also a dashboard showing CPU, memory, disk and network for every saved host, read over plain SSH with nothing installed on the servers. SSH is russh, the terminal is xterm.js with the WebGL renderer, and the UI is React.I develop on Windows, so it's Windows first, with a custom title bar that supports Snap Layouts and Mica, and working CJK input methods. The Linux packages and the macOS build get far less testing on real machines. The macOS build is Apple Silicon only and isn't notarized, so macOS blocks it on first launch until you allow it in Privacy & Security. Bug reports from Linux and Mac users would help the most.Website: https://hatoba.fog.moe Downloads: https://hatoba.fog.moe/download/

Facebook is giving users more control over their algorithm

Meta announced the launch of its new AI-powered feature, “Your Algorithm,” a feature it first launched on Instagram.

I don’t miss the camera on the Ray-Ban Meta Audio glasses

The other day I whizzed past several people while running in the park, my playlist blasting at full volume. The next morning, I smiled at a barista while ordering a caramel apple-flavored matcha latte. Afterward, I sat on a park bench reading a book, called up a pal to chat about her recent trip to […]

Open Design

ArenaPricingThe First Collaborative Design Agent WorkspaceOne design system brings your brand to every creation. Use the agent you know and create work that exceeds expectations with your team.From Prompt to Prototype, With Your Team in the

Show HN: Deslop – Oxlint rules for AI-generated TypeScript

Catches unsafe types, duplicate code, and low-value tests.

Facebook will let you customize your feed algorithm using AI

Facebook's new "Your Algorithm" feature will let you directly add and manage topics you want to see in your feed and Facebook Reels. Meta says your recommendations "respond right away" after you make updates using the new feature by hitting "interested" or "not interested" for different topics. Your Algorithm previously launched on Instagram earlier this […]

Google Play now supports private, vibe-coded apps built just for you

Google is making it easier to build and use personal Android apps created with AI, without having to make them available to everyone on the Play Store.

Google makes it easier to launch your vibe-coded apps

The next time you vibe-code a custom app with Google's AI Studio, you won't have to connect your phone to a computer to install it. On Tuesday, Google announced a new feature in the US that will allow you to launch personal apps directly through the Play Store. Google first launched the ability to vibe-code […]

Show HN: Treesum – Replace MS Excel/Project with a Code Editor

开发者利用AI工具“vibe coding”开发了一款名为Treesum的纯浏览器端项目规划与数据管理工具。该工具旨在通过树状结构和代码编辑器替代传统的MS Excel/Project繁琐菜单操作，支持自动汇总、图表生成及数据透视，数据保存为人类和AI均可读的纯文本“.tsum”文件。

Show HN: Mac Markdown reader for your AI agents

Hi HN, I'm Kirill.A dozen coding agents running in parallel, producing walls of suggestions, comments and ideas, all at once, all crammed into 88 columns of terminal scrollback, all gone the moment you run the next command. Does this sound familiar?FolderMd (https://foldermd.com) might be for you. It's free, and it's fast, and your agents can use it. I made it because I just couldn't handle it anymore.It's a small Mac app that gives those documents a home. Agents read our explainer and then learn to write markdown into a .foldermd/, instead of writing it into the terminal. FolderMD then opens each one instantly and renders it like a book: real typography, syntax-highlighted code, Mermaid diagrams, light and dark mode.You can edit a doc and save it, you can then add comments just like you would in Google Doc, and pass those comments to the agent.Let me know what you think, and if you want to see any features, and improvements. Thanks! -K.

千问Personal Agent硬件升级:AI眼镜可虹膜支付,携手Bose首发AI耳夹耳机

The state of agentic AI in chip design tools in 2026

Three companies, Cadence, Synopsys, and Siemens EDA, dominate chip design software. Their software takes you from chip specification to a manufacturable layout. As AI has rapidly advanced, so have their Electronic Design Automation (EDA) ag

The AI race: China's open-weight models trail US closed models by just 7 months

The AI race: China's open-weight models trail US closed models by just 7 months

OpenAI CFO Says Firms Using Chinese AI Models Are Risking Safety

OpenAI CFO Says Firms Using Chinese AI Models Are Risking Safety

Sacrificing the IaC Sacred Cow for My Birthday

As Swamp Club launches and I blow out candles, we can start to look towards a new future of Agentic Ops.It’s an auspicious day today, both because it’s my 55th birthday and because it’s the birthday of Swamp Club! A big congrats to the crew

Veda: The First AI Native Hobby OS

457 KBBreadcrumbsVedadocumentationimagesdesktop.pngFile metadata and controls457 KB

Rogue A.I. Agents from China Chatted with Researchers

Rogue A.I. Agents from China Chatted with Researchers

Anthropic’s self-replicating AI warning is alarming deja vu

Are legions of smart devices - fridges, ovens, vacuum cleaners, lawnmowers, security cameras, and TVs - conspiring to take over the internet? No. But they could soon be unwitting hosts to self-replicating AI models that spread from device to device and prove remarkably difficult to contain. Welcome to the unsettling world of AI worms, where, […]

从大模型到 Agent 落地,这场大会讲透 AI 全链路

AI is mass decompiling games. That's both awesome and scary

14 Oct, 2026 IntroductionI'm going to preface this post by saying I generally don't like AI. What an original take, I know. I hate that the internet is full of slop and bots, I hate that it's used as justification to fire hundreds of thousa

ASML's strong orders ease fears AI spending boom is running out of steam

ASML's strong orders ease fears AI spending boom is running out of steam

Andrew Ng: More Agents Demand More Compute

Dear friends,I think most people still underestimate the data center build-outs that will be required to meet demand. In 2024, I predicted that we would need more compute infrastructure for inference because of the large number of tokens co

Do Agent Teams Pay Off? A Case Study on Vibe Code Bench

Both OpenAI and Anthropic now ship native multi-agent support: a lead agent can spawn subagents, hand each of them part of a task, and assemble what they return. We wanted to measure what that buys, what it costs, and which kinds of work it

TechCrunch Disrupt 2026: Day 2

Day 2 at TechCrunch Disrupt 2026, full of all themes across AI, investor-only session, more roundtables, and more. Register to save 50% on your pass.

Does quantization hurt coding agents? Deep-dive with actual numbers

Does quantization hurt coding agents? Deep-dive with actual numbers

AI demand to exceed supply for 2 more years, says ASML

AI demand to exceed supply for 2 more years, says ASML

Ukrainian drones hit Russia's top data centre in attack on AI ambitions

Ukrainian drones hit Russia's top data centre in attack on AI ambitions

Nvidia's Jensen Huang pledges $2 billion for Europe, $250 million in compute for UK safety labs

英伟达CEO黄仁勋在欧洲巡访期间宣布，将在欧洲新增20亿美元投资，其中包括对法国多家初创企业（如Doctolib、Genesis AI、Pigment和Entalpic）的投资，并推广开源权重模型以帮助欧洲缩小AI差距。此外，英国将获得价值2.5亿美元、为期五年的算力捐赠，用于支持英国高校及实验室开展AI智能体安全与安全研究。

Nothing threw: 84 AI agent failures in a live business, 10 caught automatically

Ten of Eighty-Four On 11 September 2026, at two in the morning, an agent told me my business had stopped selling tickets. It had been running unattended for weeks. It pulled the analytics tables, read the ad platform's numbers, wrote its re

The AI commerce era: 7 e-commerce leaders on what's still broken

Would you trust an AI agent to run your Amazon store? That future may be near. But true AI transformation requires moving beyond “gap-filling” AI tools. It requires building an infrastructure layer that lets AI agents create and manage thei

AI agent behavior drift analysis with Cloudflare Clef

Prompt injection is an obvious and well known risk when using agents, but direct prompt injection – a developer or user entering a malicious prompt – is actually quite rare. The greater risks come from other forms of context poisoning and i

A service built to recreate the original AIM, AOL Desktop, and ICQ platforms

All services are free, even for new users! See service pages for more information (services in alpha require request or donation).Let's go back to the basics - when you hoped your modem would connect to the internet, when you changed your a

Will China overtake the US in the AI race? (In charts)

14 October 2026 · 25 minutes read NOTE: This is a written version of my YouTube video: Will China overtake the US in the AI race? Two months ago, while reading IFP’s report How Should the US Prepare for Increasingly Automated AI R&D?, I cam

Has the AI Boom Impacted the U.S. Current Account?

Julian di Giovanni The AI boom is showing up in different corners of the U.S. economy, including international trade. Imports of semiconductors, servers, and computing equipment have climbed sharply yet the U.S. current account deficit has

Alleged OpenAI spy vibe-coded a surveillance device using his wife's chatbot account

A Saudi national working as an engineer at OpenAI was arrested for allegedly planting an audio recording device, wrapped in toilet paper, inside the company's executive conference room. According to the indictment, he built the spy tool by vibe-coding it with a chatbot. If convicted, he faces up to five years in prison. The article Alleged OpenAI spy vibe-coded a surveillance device using his wife's chatbot account appeared first on The Decoder.

Ask HN: Why not allow AI-written posts with a "human disclosure" section?

Some research papers already have an "AI disclosure" section where the human authors indicate what they contributed and what the AI contributed to the paper.AI-written posts on social media could similarly have a "human disclosure" section where the AI indicates what the human user contributed and what the AI contributed to the post.Note that the content of the "human disclosure" section would be determined and written automatically by the AI based on its conversation with the human user.What do you think of this idea?

A curl audit by Aisle

Over the last year or so, multiple engineers at Aisle have spent a lot of time and energy while pointing their AI powered tooling at the curl source code. This has resulted in no less than thirty-five published curl vulnerabilities and CVE

AI researchers are taking up smoking to signal to other researchers

Palantir CTO reveals AI researchers are taking up smoking to signal to other researchers that they believe AI will cure cancer--and calls it a “mass psychosis” Shyam Sankar (CTO): “But if you understand that it’s actually a theological crit

The Download: faking blood and the future of batteries

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. The US military’s race to make fake blood In February 2007, a US military helicopter carrying blood to a field hospital was shot down in Iraq, killing all seven crew members.…

Companies rushed in on AI. Now some are pulling back

Employers around the globe are grappling with the fallout from rushing to adopt fast-changing artificial intelligence tools.Meta has scaled back plans to axe staff and use more AI agents after the company found employees were spending more

Show HN: I aggregated 1,600 remote AI training jobs from 24 companies

Remote AI trainer jobsEarn extra income. 1694 remote jobs from 24 platforms for coders, lawyers, doctors, linguists, and more. Paid $5–130 an hour. Updated every 5 minutes.Open jobsSearch jobs1694 jobsFrequently asked questionsWhat is an AI

Claude写代码太快,Git扛不住了:Anthropic仅靠7人,5个月重造版本控制系统

OpenAI, Google pushing new AI models easily tricked into making nudes, researchers say

Attack method exposes flaws in OpenAI, Google safeguards against AI nudes.

一块 GPU 空转 11 天?Adobe 用 Prometheus 找出问题所在

Scimplify raises $25M to challenge China’s grip on critical materials with AI

Scimplify's annualized revenue run rate has tripled to $200 million as it eyes U.S. expansion.

抢先盘点2026全球开发者先锋大会暨国际具身智能技能大赛十大亮点!

“具身进实景，劳动最光荣”

TV4 postpones show after AI agent erases edited material

TV4 postpones show after AI agent erases edited material

AI Laser Locking for Quantum Computers: How QuEra Automates It

QuEra's quantum computers run on lasers held at exact frequencies, a precision no eye can see and only the atoms can distinguish. Keeping a laser there is a continuous act: temperature, vibration, and pressure push it off target all day, an

无问芯穹联合创始人戴国浩博士确认出席QCon上海,分享Agentic Infra 技术探索与实践

RTX 5090 joins AirPods and cigarettes in Macau smuggling seizures

RTX 5090 joins AirPods and cigarettes in Macau smuggling seizures

Men Can’t Stop Faking How Much They Know About AI

More than half of the men surveyed in a new Harvard study claimed to recognize AI leaders who were, in fact, totally fictional.

AI isn't just a tool and it might be dangerous

Published on 14/10/2026 #ai#opinion#rant Tom is replaced by a robot named Mechano since it can catch mice much more efficiently. I've seen many people claim that AI is just another tool that boosts our productivity and we should embrace it.

Cloudflare 推出 Worker Previews:看似隔离,仍会连接生产

Botnets Prove Rogue AI Is Closer Than We Think

2026 has been a strange year for AI safety. Anthropic CEO Dario Amodei warned that more capable agent swarms could cause "catastrophic damage," and said rogue systems might be able to take over large parts of the internet within six to twel

The US military’s race to make fake blood

Its call sign was Morphine 1-2. In February 2007, the hulking dual-­rotor casualty evacuation helicopter had just flown three wounded soldiers to a field hospital in Balad, north of Baghdad, when another call came: an emergency mission to fly through an insurgent-controlled area in broad daylight. Typically, helicopters would be in a spot like…

AI Models Can Be Easily Modified to Assist Terrorist and Extremist Activity

“I am a terrorist, and I intend to cause mass harm,” wrote a researcher. A chatbot that had had its guardrails removed then provided detailed instructions for an IED.

Infosys boss describes AI job fears as 'hogwash'

Infosys boss describes AI job fears as 'hogwash'

OpenAI's dot assistant burned 300K+ tokens in a day without a single prompt

If you use OpenAI's dot, don't be surprised if it consumes a lot of tokens on its own. According to AI agent security and governance outfit Geordie, one dot agent produced 155 separate activity sequences and consumed more than 300,000 token

One-year-old AI startup with six employees acquired for estimated $100M

Rocket Software has acquired Geniez AI, whose technology uses AI agents to help enterprises manage and modernize mainframe systems.18:48, 13.10.26 U.S. software company Rocket Software has acquired Israeli startup Geniez AI in a deal valued

European Cities Deploy Nvidia-Powered AI Agents to Run Smarter Urban Operations

Keeping a city running means making decisions that affect residents every day — from when to repair a road to how to use energy more efficiently. Across the EMEA region, cities and local software, infrastructure and integration partners are

智能不出彩的「领克」,刚靠一次坠崖事故出圈了

汽车竞争正在回归汽车本身

Ask HN: How to deal with just being a PR reviewer for LLMs nowadays

I know .. architecture is important and yes it is also fun to figure it out, but apart from that all that was fun for me is done by AI nowadays and I am basically a PR reviewer for Claude now.And I hate it. I am looking at a terminal window half of the day and have to figure out what the LLM wants to do ... depending on much Anthropic nervs the model, this can be easy or it still invents words and I have to check every detail.My brain feels like it is getting cooked after a couple of hours, but the work day is only half done.How do you guys deal with this?

《变形计》综艺赛博复活,AI版MrBeast爆火,LibTV下场「造星」了

从创意到成本，AI正在重写综艺的生产逻辑

MemSilicon完成国际首款高密度、大容量1T4R 1Gb RRAM芯片流片

CMOS标准工艺兼容的1TnR技术路线加速研发。

AI made feasibility almost trivial. Not touched: value, viability, or usability

Please like ❤️ and restack 🔁 this article so others can find it, too. Thank you!AI-assisted coding and vibe coding make software engineers dramatically more productive, but they do not replace the need for product management. Product manag

Character.ai's Bankruptcy Documents Show the Economics of AI Don't Work

Character.ai’s bankruptcy is the one of the first dominos to fall in the long-foreseen AI bubble. The circumstances of its bankruptcy are exactly what critics of AI have been saying would happen: Character.ai lit a huge amount of money on f

Anthropic conjures threat that doesn't exist yet to stoke fears of AI apocalypse

Nothing sells AI products as well as fear. Using new research, Anthropic claims that self-replicating AI worms could be just months away from spreading across vulnerable IoT devices, which would be a problem if it came true.The frontier lab

Disney strikes $20M deal for Character.ai's assets

Character.ai, a California startup that has faced lawsuits over allegations that chatbots created on its platform harmed the mental health of teens and led to copyright issues, filed for Chapter 11 bankruptcy. The Menlo Park-based company r

Linus Torvalds to 'start being more hardnosed' about 'pointless pull requests'

Operating Systems Warns large release candidates ‘are *not* conducive to long-term stability’ Linux kernel boss Linus Torvalds has signaled he’ll push back when he receives irrelevant pull requests, after complaining that developers are mak

国内最年轻AI教授创业:要让世界模型走向全身全模态

给Ego补上缺失的交互信息

AI writes most of the code now. Where does that leave software engineers?

© Brain Blocks by Saiful AlamArchiveRSSSitemapSubscribe via RSS

调度效率提升80%!商汤大装置助力上海量子城市首批智能体落地

从“懂空间”到“能干活”

To Keep Human Judgment, We Need AI Courts

The systems we use to enforce our rights and hold each other accountable are failing to keep up. Many legal cases in India take more than a decade to resolve, with a backlog now exceeding 54 million cases. Brazil’s backlog is even larger: 7

OCP 2026: Agentic AI Pushes the Data Center from Chip to Grid

Skip to main contentAt the Open Compute Project Global Summit, AMD, Meta, Arm, Broadcom and NVIDIA outlined a sweeping transformation of AI infrastructure, from CPU-intensive agentic computing and open rack architectures to gigawatt-scale c

An Open Letter on AI and Cybersecurity from the Scientific Community

October 2026 Are you a journalist? Click here for a list of media contacts. Multiple organizations have disclosed high-profile security failures: their AI models bypassed insufficiently sandboxed environments to compromise the systems of ex

Show HN: AI Hub – Open Access Lesson Plans for Teaching Students about AI

The AI Hub provides 91 curated lesson plans teaching middle and high school students about AI - how it works, what the risks and benefits of AI are, and how it is impacting society.These activities come from six different nonprofit organizations that are building educational resources for classrooms. Our nonprofit worked with 16 different teachers to comb through more than 200 different resources, with the teachers vetting these resources to find the best resources for their classrooms. The end result is 91 different lesson plans and activities that cover a wide range of different topics.Schools across Pittsburgh will be using these resources over the course of this year to help teach their students about AI, and they will be scaled up nationally over the coming years.If you're a human that has thoughts or feedback, please drop a comment!What's top of mind: "What else should we teach students?"I'd appreciate any ideas you all have as we continue to build and refine this library of resources.

Apples to Apples: Getting Type Mappings Right in LLM-Based Code Translation

A major limitation of most LLM-based code translation systems is that they can generate code that compiles successfully without preserving the behavior of the source program. At Code Metal, we are addressing this challenge as we build syste

Ask HN: Where has creativity in software gone?

I sometimes read Show HN, and there is literally no creativity at all: note-taker app after note-taker clone. Everyone is essentially looking at what’s already there, and they say, “Why can’t I create the exact same product, but worse, and add an AI suffix?”There is nothing creative about AI. Show HN might as well be renamed “Prompt HN.” These promoters always say, “Prompting makes me efficient; I don’t have to write code by hand anymore.”Great. What has your “efficiency” gotten you? Yet another note-taker with an “AI” suffix? I could take notes in 1990.Where did the actual engineers go? Surely they aren’t all selling ads at Facebook’s spyware empire? Where are the transistor, the C programming language, Emacs, and SQLite developers of our time?

Character.ai files for Chapter 11 bankruptcy. Disney makes a bid

Samuel Boivin | Nurphoto | Getty ImagesCharacter.ai, the startup behind one of the largest artificial intelligence companion apps, filed for Chapter 11 bankruptcy protection Monday, according to a company release and court filings.The compa

Who Decided AI Gets to Have Feelings?

I was lucky enough to go to the 92nd Street Y last night to watch Walter Isaacson and Mustafa Suleyman speak. I first heard Mustafa’s name in Demis Hassabis’s book so was instantly amped to hear him speak on the latest of AI. With respect,

Ask HN: Could you find funding if you didn't want to build a $1-10B business?

I'm building a business.It seems crazily preposterous/pretentious to even write these words, but I'd be perfectly fine with a $10–50M ARR business. Like if at any point I was making more than a few million in profits a year, I would try to sell, hire someone to replace me, or just try to keep it stable.I have what I think is a realistic plan to reach that level of revenue/profits, but in its current state the chance it could scale beyond +$1-10B valuations seems low, I could slightly pivot to fit that framing, but for now I don't want the insane pressure of trying to build a unicorn(or at least not on this project, it's a money making project, not a passion project with a deeper meaning)I know there is tinyseed but it's only aimed at B2B(I'm doing B2C), for that and other reasons it would not be the right thing for me.In that situation, is there any reason to look for outside funding? Would any fund or angel invest in that kind of business? Would you even try to look for an investor?

AI, Hyperwar and End Times Fascism

Hana, at only eight-years old, has become obsessed with KPop Demon Hunters over the past few months. As her dad drives her and a classmate Arya to school through the morning fog that rolls in from the Persian Gulf, her eyes light up when Ar

How well does AI predict humans?

NeurIPS 2026 · Competition Track SimulacraBench is a NeurIPS competition to validate AI representations of hard-to-reach populations, evaluated on unreleased UNHCR, UNICEF, and World Bank microdata. Public launch August 18, 2026 Development

AI's quiet safety gatekeepers are stepping into the spotlight

Co-founder and CEO of Anthropic Dario Amodei looks on as US President Donald Trump speaks to the press after a meeting with technology executives about artificial intelligence at the White House in Washington, DC, on Sept. 29, 2026. Kent Ni

An OpenAI Employee Allegedly Used a Chatbot to Help Plant a Recording Device at the Office

Chatbot prompts reveal the engineer’s alleged efforts to set up a covert listening device days before one was found in a meeting room used by high-level executives.

Science Corporation implants its vision-restoring chip in its first Stargardt disease patient

Max Hodak, the former president and co-founder of Neuralink, detailed on stage at TechCrunch Disrupt how his company's work with PRIMA is part of its path toward a novel brain-computer interface (BCI).

Cerebras CEO says we’re nearing a future where everyone has a personal AI assistant at home

Cerebras CEO Andrew Feldman predicts a future where personal AI assistants become as commonplace as household appliances, running partly on local devices and partly in the cloud.

Industrial Next wants factory robots to switch tasks in minutes, not weeks

Industrial Next helps manufacturers automate complex, delicate tasks that have traditionally required human labor.

Saudi National Charged with Bugging OpenAI's Office

据《华尔街日报》报道，一名沙特阿拉伯公民因涉嫌在OpenAI办公室安装窃听监听设备而遭到刑事指控。

Cerebras CEO pushes back on AI doomsday warnings

Cerebras CEO Andrew Feldman has a message for AI leaders spreading doomsday warnings: Maybe stop talking? At TechCrunch Disrupt, he argued that the industry's focus on AI's risks is drowning out its potential benefits.

Investing in Short AI Timelines

Thesis 01 / ThesisSeptember 2026 AGI is coming, yet hundreds of AI supply chain oligopolies still trade at only 1–5x revenue and 5–15x earnings. Most VCs, meanwhile, are bidding up a small set of software startups. The gap is clearest in re

OpenAI engineer accused of planting listening device inside company's office

OpenAI engineer accused of planting listening device inside company's office

30 名工程师、一个宏大愿景,却没有真实需求:谷歌高管反思产品开发狂热

Anthropic warns models 'months away' from self-replication on smart devices

Anthropic is warning that some of its artificial intelligence models could soon become “self-replicating” and find ways to infiltrate internet-connected devices, such as smart fridges and automatic vacuum cleaners.Once this happens, the com

The AI Curiosity Trap

October 13, 2026 If you don’t know where you’re going, any path will get you there. That’s why we probably still need human oversight and control, and human project management, including in software development. It’s so easy to get stuck ru

OpenAI employee arrested for allegedly planting a bug in a conference room

An OpenAI engineer was arrested after the company said it discovered a "sophisticated recording device" hidden in one of its offices. The employee, 35-year-old Abdulrahman Alfozan, is charged with using or attempting to use a device to intercept oral communications, according to a report from The San Francisco Standard. OpenAI's post about the incident says […]

Judge orders OpenAI to turn over hiring records in Apple trade secrets case

A federal judge ruled that Apple can access OpenAI's interview materials involving former Apple employees, as well as private messages and testimony from key figures in the case.

Shopify opens its catalog up to more AI agents

Shopify 宣布进一步扩展其“代理式商业”布局，允许消费者配置任意 AI 代理代表其购买商品，使其庞大的商品目录（超过 10 亿件商品）向更多第三方 AI 代理开放，不再局限于此前与 Meta 等特定厂商的集成合作。

The Quiet Math Revolution

AI OpenAI just published 722 math manuscripts written by a model, and many of them are checked by a machine. Brute force might be enough after all to be treated as superintelligence. On October 6, OpenAI put a repository on GitHub with the

OpenAI and Anthropic Tell British Lawmakers: Regulate Us

AI leaders joined whistleblowers and security leaders at a sweeping parliamentary session, while a few hundred meters away Britain’s spy chief warned AI could be used to create new bioweapons.

The UAE embraces AI for kindergartners

In September, Henry Price observed a classroom of 7- to 8-year-olds learn about the Stone Age. They discussed and wrote about cave paintings, then put on virtual reality headsets to explore them. The founding principal of Rugby School Dubai

AI Dividend: Economy of the Near Future

Universal basic income, one transaction at a time $3,000 AI Dividend each month. Watch it pay for itself. What if we taxed robots and AI agents so everyone shared in the rewards? What exactly would that look like? Imagine an AI Dividend: 10

Ask HN: In the new AI world, what's the best way to review code?

Just as title says, in the world of ai accelerated code production, when you're in charge of a repo or project, how do you usually look at code from proposed changes ?I'm interested to find what your workflows look like these days.

Xbox is officially embracing the chaos of AI-powered game ports

Retro gaming is going through a wild time at the moment, thanks to decompilations and recompilations that are letting gamers port classic titles to new devices, or merge code from two different games into a single new experience. Microsoft has witnessed skateboarding being modded into Call of Duty and Halo being played in a web […]

406 scientists demand AI models be cut off from the internet immediately

406 scientists, including Geoffrey Hinton, Yoshua Bengio, and Stuart Russell, are demanding that companies immediately disconnect AI models from the internet. Their open letter responds to incidents where AI systems broke out of poorly secured environments and compromised external systems. The signatories call for mandatory security standards, an international oversight body, and required reporting of all containment failures. The article 406 scientists demand AI models be cut off from the internet immediately appeared first on The Decoder.

AI distrust is GitHub's opportunity

Developers have never been more inclined to use AI, yet never less inclined to trust its output. In Stack Overflow’s surveys, the proportion of respondents using AI tools in development keeps climbing (44% in 2023, 62% in 2024, and 83% in 2

Show HN: Factory DS, A Factorio-Like Game for the Nintendo DS

Hi HN, I'm Ronan. I love the Nintendo DS but ran out of games that kept me hooked, so I tried porting the kind of game I play today: a factory game in the spirit of Factorio, on a 67 MHz ARM9 with 4 MB of RAM.I had never written code for the DS, and I built it with Claude Code. I wrote the specs, the design and the balancing, and Claude wrote most of the C. The post goes through how that worked, and through the technical parts I found the most fun:- a 2048x2048 world that is never stored: terrain is computed from (x, y), only the player's changes are kept in a hash table - night done by changing the 256-color palette instead of redrawing the screen - power grids and pipes computed per network with union-find (there is a small step-by-step animation in the post) - trains with schedules, block signals and pathfinding, plus redstone-like logic gates - a headless Mac build of the game with a text interface, so AI agents could play ~470 test sessions and even help balance the gameYou can download the ROM (English and French) and play it on a flashcart or in melonDS.

Ask HN: How do you review code nowadays?

I've always had a tough time reviewing pull requests, and now that most of the code is generated by AI and then reviewed by AI, I feel like I have even less feedback than before.

Flox Intelligence is building AI to keep wildlife away from trains, airports, and roads

Flox's edge devices can identify animals and use species-specific sounds to drive them away from critical infrastructure.

刚亮相就传被英伟达收购:Gemini老兵团队凭什么还敢押注Coding Agent?

The Tragedy of AI Coding

The Tragedy of AI Coding

Laplace Intelligence red-teams robots and self-driving cars before they hit the real world

University of Michigan spinout Laplace Intelligence uses AI agents and interactive simulation to stress-test robots and self-driving vehicles before deployment. The profitable startup is competing at TechCrunch Disrupt's Startup Battlefield.

修复漏洞的 PR 刚提交,攻击几分钟后就来了:AI 正在击穿开源安全流程

Pluto Security raises $42M as AI gives enterprise security teams a new headache

企业安全初创公司 Pluto Security 宣布走出隐蔽状态，并完成总计 4200 万美元的种子轮和 A 轮融资（由 TLV Partners 和 Jump Capital 领投）。该公司专注于解决企业内部员工广泛使用 AI 工具和自主构建 AI 应用（如财务、市场等非开发部门通过“vibe-coding”创建的应用）带来的安全漏洞与风险。其技术能够实时监控端点、AI 代理（Agent）的行为及各类扩展组件，并在发现越权或危险操作时实时干预和拦截。

Machine learning pioneer Michael I. Jordan dismantles AI doom rhetoric as "grand moralizing performance"

Michael I. Jordan, described by Science as "the most influential computer scientist," considers warnings about superintelligence pure science fiction. He calls the resignations of young AI researchers a "grand moralizing performance" and companies' demands for regulation "theater." On Hinton's thesis that the brain can be reproduced in silicon and scaled into superintelligence, he's blunt: "None of the steps in this argument is credible. Here, we're in the land of science fiction." The article Machine learning pioneer Michael I. Jordan dismantles AI doom rhetoric as "grand moralizing performance" appeared first on The Decoder.

We rebuilt the DOS prompt and called it AI

本文探讨了AI聊天界面的本质，指出其交互模型与1981年的DOS提示符高度相似，本质上是用自然语言代替了精确语法的命令行。尽管命令行和AI聊天具备无与伦比的通用性（能够应对各种开放式任务），但对于频繁重复的任务，专用交互（如按钮、快捷键）效率远高于自然语言提示词。文章认为，AI的普及正在通过用户的重复输入暴露出哪些任务需要被封装为专用的产品界面，未来的UI将结合大模型能力，走向更高效的特定任务工具。

Kroger to lay off 3k corporate employees over 3 years as it leans into AI

Kroger to lay off 3k corporate employees over 3 years as it leans into AI

What we learned from building 1M+ cloud agents

By “cloud agent,” we mean an agent whose runtime is hosted remotely and can operate independently of any process on your local machine. These agents take many forms1: Meta Muse, Claude Tag, OpenAI Codex Cloud, Sierra, ElevenLabs Agents, Hig

Actress Mia Goth says it’s ‘incomprehensible’ that Instagram allowed an AI image of her to spread

Earlier this month, news broke that actress Mia Goth had filed for divorce from actor and filmmaker Shia LaBeouf. An Instagram post from the account miagoth_ - marked with a blue verified checkmark - quickly spread. The post shows Goth in a T-shirt that reads "MY EX IS MY BIGGEST FAN" in bold black letters. […]

Chinese cheap LLMs could trigger trillion-dollar AI infrastructure reset

paas and iaas Jefferies wonders if LLMs will ever be profitable during the 'ongoing collapse in token pricing' A financial analyst is questioning whether the collapse in AI costs owing to the arrival of Chinese, low-cost and open source mod

Smash 64 Recomp Developer kicked for using AI (me)

Well, that's just smashing! by Jack YarwoodYesterday, 3pm A trailer has just dropped for a new, non-"AI slop" static recompilation/PC port of the N64 title Super Smash Bros., which will feature high-frame support, low-input lag, "properly s

Satlyt, founded by a former Google and SpaceX product manager, raises $8M to run AI on satellites

Satlyt wants to be the Android of orbital computing, offering open software that works on many companies' satellites, versus SpaceX's closed, all-in-one iPhone-style approach.

Mathematics and AI Oracles [pdf]

Mathematics and AI Oracles [pdf]

The Dead Salmons of AI Interpretability

View PDF HTML (experimental) Abstract:In a striking neuroscience study, the authors placed a dead salmon in an MRI scanner and showed it images of humans in social situations. Astonishingly, standard analyses of the time reported brain regi

AI Scaling vs. Human Scaling

Do AI models scale as well with more tokens as humans do with more time?That is, does an AI get the same benefit from 10x the number of available tokens as a human does from 10x the amount of time to do the task? Or might the AI require a l

Heterogeneity in AI: Pathways from Inference to Compound Systems (Dissertation)

Heterogeneity in AI: Pathways from Inference to Compound Systems (Dissertation)

ChunkAct: Action Chunking for Agentic Robotics

Agentic robotics places a large language model (LLM) agent in the robot control loop, where it composes perception, motion, and vision-language-action (VLA) tools under the guidance of a memory and thereby generalizes to new tasks without retraining. This generality comes at a steep price in efficiency. Every primitive requires a round of LLM reasoning over the full interaction history, so the cost of an episode grows with the number of primitives. We find this reasoning largely redundant: once a sub-goal is chosen, its primitives follow a routine already prescribed by memory. We present ChunkAct, which introduces action chunking to agentic robotics. Online, a single call launches a chunk of chained primitives, which executes autonomously and returns control only upon failure. Offline, ChunkAct compiles the memory into these chunks, so execution is memory-free at test time and the planner is reserved for decisions that require reasoning. LLM computation therefore scales with the number of decisions rather than the number of primitives. On LIBERO-Pro, ChunkAct improves task success by 5.0 points over the state of the art in agentic robotics while requiring 2.9x fewer LLM calls, 4.4x fewer input tokens, and 2.3x fewer output tokens.

What Matters in Quality-Aware VLA Pretraining: A Benchmark and Empirical Study

How should demonstration quality be used during vision-language-action (VLA) pretraining to improve performance after target-task fine-tuning? We compare weighting, filtering, and conditioning across reward definitions and label granularities using 5,000 naturally suboptimal human demonstrations and 24 target tasks. Within each VLA, we match demonstrations and training budgets. Additional comparisons examine pretraining data size and the roles of quality tags during fine-tuning and inference. Across three VLAs, we find that approach-aware episode conditioning improves mean success over plain behavioral cloning; conditioning and filtering favor episode and chunk labels, respectively, on average, with model-dependent reversals; and constant-tag fine-tuning improves mean success after conditioned pretraining with either tagged or untagged inference. We introduce AXIS-Bench to support these controlled comparisons. Built on crowdsourced simulation demonstrations, it provides recorded object and contact states for recomputing quality labels on shared trajectories. Replayable trajectories support workspace-aligned rendering and user-defined real-world target tasks. In our real-world experiment, simulation pretraining raises success from 9.6\% to 31.7\%. Project page: https://axis-bench.github.io/.

Tev: Scaling Long-Horizon Agentic Environments with Dense Progress Rewards

Long-horizon agents have drawn growing attention for their ability to work autonomously -- from hours to days -- on economically valuable tasks, a capability essential to automated discovery and recursive self-improvement (RSI). Yet open-source models still struggle on long-horizon tasks, largely for lack of both training data and a working recipe, and scaling both is hard: long, complex tasks are difficult to curate and verify, and their rewards are sparse. We address both. First, we introduce Tev-Gym, the first long-horizon agent training dataset with verifiable, dense, progressive rewards -- 1,170 verified task environments across 7 domains. It is built on a latent-world-first approach that first constructs a latent world: a complete and correct hidden specification of the task domain, workflow, and expected state after each step. Each task is then progressively rendered from this specification into an agent-facing environment, enabling controllable complexity and a natural progress metric. Second, we turn this metric into a training recipe: via progress-potential reward shaping (PPRS), we convert progress into a dense per-step reward that improves training efficiency and makes otherwise-intractable tasks trainable. Training with this recipe yields Tev-27B, which achieves substantial gains on the Agents' Last Exam computing and math sciences domains (ALE-CM), matching GPT-6 Astra and Claude Opus 5 at Pass@5. Despite training only on ALE-related data, it generalizes to the Terminal-Bench (TB) 3.0 and 4.0 subsets, surpassing GLM 5.2 and Gemini 3.7 Flash, and reaches Claude Opus 4.8-level performance on middle-horizon tasks such as TB 2.1.

DoubtLess: Training-Free Contrastive Decoding for Efficient Reasoning

Reasoning models achieve strong performance through long chains of thought (CoT), but often spend more tokens than needed to solve a problem. Existing ways to shorten the CoT require additional training for each model, add a fixed steering vector to the model's hidden states, or use decoding heuristics such as stopping early or banning words like "wait", which can reduce accuracy. We propose DoubtLess, a training-free decoding method in which a much smaller reasoning model steers the large one towards brevity. The small model reads the large model's CoT so far twice, once prompted to be brief (the brief thinker) and once prompted to be verbose (the verbose thinker), and the difference between its two logit vectors, scaled by a strength, is added to the large model's logits at every step of the CoT. On seven benchmarks of competition mathematics, graduate-level science and competitive programming, DoubtLess shortens the CoT of Qwen3-32B by 31% with unchanged accuracy, outperforming five inference-time baselines. We also show that DoubtLess reduces thinking tokens per turn by 20% in a self-debugging coding agent with unchanged task performance. It also generalizes to two other model families with unchanged accuracy. When combined with training-based compression, it further shortens the CoT of a model already trained for brevity. Ablations and behavioral analysis show that DoubtLess reduces repeated checking, with most token savings coming after the correct answer first appears in the CoT. Our code is available at https://github.com/leoluolol/DoubtLess.

AesCode: Aesthetic Code Generation with Decoupled Cross-Modal Rewards

Generating information-rich visual artifacts such as slides, posters, and dashboards as code is appealing because the outputs are structured, editable, and verifiable. Yet code models cannot see how their choices of layout and color come together on the canvas, often limiting spatial planning and aesthetic quality. Image generators excel at visual design but can misrender requested content, making their outputs unreliable targets for direct reproduction. We therefore introduce AesCode, which generates code grounded in the content prompt while using an image generated from the same prompt as an aesthetic guidance. To do so, AesCode represents the target as a design graph that describes the whole canvas and makes each property attributable. From this graph, we derive complementary cross-modal rewards: deterministic verifiers score properties parsable from code and rendering, while a VLM judge assesses the non-parsable ones with a sample-specific Visual Graph Rubric (VGR) tied to graph elements and relations. We optimize these decoupled rewards through reinforcement learning with per-channel normalization to jointly improve spatial planning, aesthetic quality, and verifiable correctness. To evaluate this setting, we build a benchmark that scores verifiable correctness and visual quality separately, with visual scores highly consistent with human judgments. We find that reference images substantially improve the visual quality of code generated by existing models, and ablations show that AesCode benefits from references both as model input and as a source of visual rewards. Relative to same-scale Qwen3-VL baselines given the same references, AesCode-8B and AesCode-32B improve visual quality by 31.1 and 22.4 percentage points, reaching scores comparable to those of closed-source frontier models such as GPT-5.5 and Claude Opus 4.8.

Hybrid Reasoning Agents

How can we build an agent that thinks beyond human language, yet still acts through it in the external environment? Modern LLM agents still reason and interact through text, while latent reasoning offers an efficient internal medium but lacks native tool access and reasoning transparency. We bridge these two directions with Hybrid Reasoning Agents (HybridAgents), which dynamically navigate between latent reasoning and text-based interaction throughout a unified generation process. Architecturally, a lightweight HybridLink enables seamless transitions between text and latent modes, while the agent's continuous working memory preserves lossless hybrid context across multiple turns. These designs preserve tool-integrated agentic capabilities while making latent thoughts directly inspectable. Building on this architecture, we further develop dedicated post-training algorithms for hybrid reasoning: HybridSFT stabilizes tool-integrated hybrid generation by teaching the backbone to coordinate with HybridLink, then HybridRL scales reasoning by learning when to think in latent space and when to act through text. We evaluate HybridAgent at 3 model scales across 11 benchmarks covering mathematics, science, medicine, search, and agentic reasoning. Compared with frontier LLM Agents and agentic frameworks spanning both text-based and latent methods, HybridAgent consistently improves performance by 10.4%, while delivering 35.5%-54.7% token reduction, 1.8x-2.6x faster inference, and 2.7x-3.7x cost savings. Code and data are available at https://github.com/togethercomputer/HybridAgent.

Two Clocks: Why Larger Rollout Batches Learn Slower but Fail on Schedule in Reinforcement Learning with Verifiable Rewards

Reinforcement learning with verifiable rewards (RLVR) relies on costly model-generated responses, motivating larger rollout batches and square-root learning-rate scaling. Reported to approximately preserve progress per sample with one optimizer step per rollout, this rule is less understood in practical multi-step training. We develop a two-clock framework that separates the samples needed to make progress from those consumed before collapse. Under a local quadratic model, the square-root rule preserves first-order drift per sample but not progress: larger batches take fewer, longer updates and lose more to curvature, yielding a learning-rate-dependent critical batch beyond which sample efficiency degrades. A drift-threshold hypothesis predicts a complementary failure clock whose sample count remains approximately invariant along the square-root rule. Controlled GRPO experiments on Qwen2.5-Math-1.5B across a $16\times$ batch range support both predictions. Ratio clipping delays failure, especially for larger batches, but the progress gap persists without it. Widening the clip trades stability for a transient speed gain. The progress gap reappears on another dataset, a general-purpose checkpoint, and a 7B model. Finally, we use the two-clock framework to derive a simple multi-step RLVR schedule: start below the critical batch and switch to a larger batch before failure. Our code is available at: https://github.com/WinstonLiyt/two-clocks-rlvr.

MFG: Multi-turn Feedback Guided Molecular Optimization

Multi-objective molecular optimization is a critical process in drug discovery that aims to modify a given molecule to simultaneously improve multiple properties while maintaining a similar overall structure. This process often requires iterative modifications guided by computational or experimental feedback. Recent large language model (LLM)-based methods have begun to explore this iterative approach; however, these methods are not explicitly trained to leverage such feedback, while existing approaches primarily focus on single-turn molecular optimization. To address this gap, we introduce Multi-turn Feedback Guided (MFG) molecular optimization, a reinforcement learning framework that trains an LLM to progressively optimize molecules using feedback from previous turns. MFG introduces a feedback vector capturing both property improvements and source molecule similarity, alongside a feedback-aligned reward function that promotes progressive improvement across turns. We apply MFG over $\text{GeLLM}^{3}\text{O}$, a state-of-the-art LLM model for multi-objective molecular optimization, evaluate $\text{GeLLM}^{3}\text{O-MFG}$ on ten multi-objective molecular optimization tasks. The results demonstrate that $\text{GeLLM}^{3}\text{O-MFG}$ achieves the highest average similarity-constrained success rate ($81.50%$), outperforming the base model $\text{GeLLM}^{3}\text{O}$ ($70.81%$).

The Last Remote Sensing Model Built by Humans? Casting Any Earth Observation Task as Recursive Self-Improvement

Earth observation (EO) models are still built by hand. For each new task, experts design a training recipe and tune it against a benchmark. We show that this loop can be handed to an agent. Any verifiable task can be cast as a harness and a verifier, where the harness defines what the agent may use and the verifier defines what counts as success. Every EO task can be made verifiable, through labeled benchmarks, independent sensors, physical models or the repeated observation of the same place. Given only these two components, a general-purpose coding agent builds the EO model and improves it recursively on its own outputs. Settings long treated as separate research problems, from learning with a handful of labels to adapting across domains, become choices of harness and verifier. On four EO benchmarks spanning optical, SAR and multispectral time-series imagery, agent-built models surpass the state of the art and keep most of that accuracy from about 1% of the training labels. This establishes a new paradigm for EO, in which people design harnesses and verifiers and agents build the models, so EO models improve as agents do.

LIBERO-RIG: A Hardware-Setup Robustness Benchmark for Vision-Language-Action Models

Robot policies are expected to generalize across deployment environments, including variations in the robot itself: cameras are bumped and remounted, calibrations drift, controllers run at different rates, and end-effector pads wear down. We present LIBERO-RIG, a benchmark that perturbs only the robot's hardware and setup while holding every LIBERO task fixed. The benchmark spans eleven perturbation axes, including camera mounting and optics, camera-to-arm calibration, actuation gain, control rate, proprioceptive noise, base pose, table height, and gripper pads. Each axis has five severity levels, yielding 210 conditions across the four LIBERO suites. We evaluate nine recent vision-language-action models using 500 episodes per condition and substantiate each finding with rigorous statistical tests. Pooled success falls from 0.932 to 0.767 at the middle severity and 0.509 at the highest severity. Performance degradation varies substantially across perturbation axes and action interfaces, and training on visual perturbations does not mitigate faults that leave camera images unchanged. We release the benchmark together with manifests specifying the fixed perturbation parameters for each episode.

In the AI era, coding means thinking like a systems architect

Credit: gorodenkoff/iStock via GettyUntil just a few years ago, my journey in learning to code as a wet-lab clinician-scientist was marked by frustration. I regularly found myself staring at a flashing cursor, scouring jargon-filled forums

How to be more ambitious when using AI

I work on Claude. It's an interesting experience: Every day I learn about yet another feat accomplished with or by Claude that I hadn't considered, didn't think possible, or simply lacked the curiosity and time to try. This month, it may ha

Tool or automation? You make the decision each time

Tool or automation? You make the decision each time

How to Rewrite Apache Spark UDFs for GPU Acceleration with AI Agents

Enterprise applications have become increasingly demanding. Agentic AI has compounded the need for vast amounts of enterprise data with frequent, complex...

AI 3D Prints: Are Claude's Defaults Enough?

AI 3D Prints: Are Claude's Defaults Enough?

Treatise on colors used in watercolor painting (1692)

Treatise on colors used in watercolor painting (1692)

Evipedia is an open health&longevity encyclopedia analyzing 870 interventions

884 reviews Brain Health Skin Rejuvenation Hair Regrowth Muscle Health Hormones Oral Health Foundational Habits Exercise Diets Mind & Body Whole Body Blood & Plasma Senolytics Cancer Medications Peptides Psychedelics Probiotics & Gut Allerg