Anthropic 2025年营收与算力支出曝光：AI模型与智能体安全成行业焦点

Flux 3 Image Klein

The most capable open weights image model5.6B parameters, runs on a single consumer GPU, and draws on FLUX 3’s shared learning across images, video, and audio.Undistilled version for fine-tuning and research, a distilled version for fast ge

GPT-6-Astra solves 49 of 50 physics research problems at $2.40 apiece

On the TheoPhysEval benchmark, GPT-6-Astra solves 49 of 50 unpublished theoretical physics calculations as accurately as leading specialists. In some cases where the agent seemed to fail, the experts' own reference solutions turned out to contain errors. But the researchers say physics hasn't had a breakthrough like the ones in mathematics, because the agents lack "taste." They still think autonomous, top-level AI research could arrive soon. They already see consequences for how young physicists are trained and for equal opportunity in the field. The article GPT-6-Astra solves 49 of 50 physics research problems at $2.40 apiece appeared first on The Decoder.

NVIDIA Nemotron 3.5 Super VL Powers Specialized Agents with Multimodal Reasoning

NVIDIA 推出了 Nemotron 3.5 Super VL 开源多模态模型，旨在为复杂的企业级智能体工作流提供支持。该模型采用 120B 参数的混合专家（MoE）架构（激活 12B 参数），具备强大的多模态推理能力，可用于处理跨源信息核查、文档阅读及工具调用等专用任务。

Stepfun Step 5 Preview on HuggingFace

StepFun's flagship model for agentic work 600B total parameters · 27B active parameters · Up to 1M context Introduction · Capabilities · Specifications · Performance · Deployment · Resources · License Introduction Step 5 Preview is StepFun'

The Download: reviving frozen pig kidneys and boosting nuclear power

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. Pig kidneys that were “frozen” for a year have been brought back to life Pig kidneys that were cooled to -150 °C and stored in a freezer for months have been…

What Is Jev Good For?

Jev is an AI model that picks instead of writing. You give it a situation and a list of options, and in about a third of a second it tells you which option fits, how sure it is about every one of them, and whether none of them do. That's th

Step 5 Preview: Architecture and Training for Agentic Work

Agentic workloads place a different set of demands on model design and training. Long, tool-heavy trajectories increase the importance of reasoning over large amounts of prefilling context, make long-context computation increasingly expensi

Artificial Analysis AA-Robotics v0.1 Robotics Benchmark for Frontier Models

Introducing AA-Robotics v0.1The eight tasksResultsResults per TaskFull score (100)Partial (1 to 99)Zero (0)TaskGPT-6 AstraGPT-6.1 SolGemini 4 ArgonClaude Opus 5.5Claude Fable 5.1Press the red buttonFull score, 100 of 100Full score, 100 of 1

全球首款 !中国团队把游戏世界模型塞进一张5090,还开源了

给视频模型套上Harness

Step 5 Preview: StepFun's flagship model for agentic work

StepFun's flagship model for agentic work 600B total parameters · 27B active parameters · Up to 1M context Introduction · Capabilities · Specifications · Performance · Deployment · Resources · License Introduction Step 5 Preview is StepFun'

LFM2.5-24B-A2B: Built for fast on-device agents

Today, we release LFM2.5-24B-A2B, a 24B-parameter reasoning model with just 2B active parameters, built for fast, private agents on-device. It is the largest and final model in the LFM2.5 family, which now spans two orders of magnitude: fro

Frontier models score under 2/100 controlling a simulated humanoid

Robo-benchModels controlling simulated robotsACCURACY Robo-bench leaderboard Rank Model Accuracy Est. API cost Input / Output Cost Time to first move 1 GPT-6.1 Sol 28.33% $4.97 $2 / $10 79.89s 2 GPT-6 Astra 23.33% $28.48 $10 / $50 74.88s 3

We’re releasing Jev 1.14.0 along with experimental features for dreamers

TL;DR: We're releasing an improved Jev 1.14.0 along with experimental features for our dreamers: images, a new type, and longer context. Enter the danger zone at your own risk. 🔥Our philosophyA huge debate we’ve been having is on the balan

AI systems are improving rapidly at Slay the Spire and its sequel

OverviewSlay the Spire (StS1) and its recent sequel, Slay the Spire 2 (StS2), are difficult games. Since the original’s release in 2017, the game has attracted a dedicated community of players. Top-tier players play the games on their maxim

Show HN: Eevo – Consequential Decision Model

Hi HN, I'm Juho from Upright. I think we may have stumbled upon an interesting approach to AI alignment.The idea is as follows: 1. train a decision model to be really good at comparative sizing of consequences, expected value problems, and modelling overlap between different consequences 2. create a harness around that model that accepts a decision context, list of options, and identified consequences, then runs pairwise comparisons and overlap modelling for the consequences using the decision model in (1), and finally solves implied relative consequences "sizes" and finally EV (for different stakeholders) 3. Make AI agents running on general purpose LLMs use (2) to make better decisions. So the agent provides the context, options, and identifies the consequences, Eevo model just sizes them, models their overlap, and computes EV.In practice this seems to improve agent judgement a lot (in addition to adding lots of practical observability)It seems the (well-motivated) emphasis on instruction-following hurts model's performance on expected value tasks. The underlying decision model we trained beats Jev, Clef, MS decision-1 the model on OpenAI decisions API on economic rationality benchmarks like STEEP, expected value tasks, poker, and even Pacman (Jevman benchmark). I swear to you we did not Benchmaxx on PacmanAlso it seems to be useful also for human decisions. Just getting other people's decisions presented to you in standard format with consequences listed and sized is just very pleasant. There are some testers though who think this is useful for improving everyone's judgement, except their own

Where AI slop comes from, and how to turn it down inside the model

just you The style is learned in pretraining and switched on by post-training, and it sits along one direction inside the model that can be pushed back without breaking it. TL;DR The em dashes, "it's not X, it's Y" reframes and lists of thr

Grok Voice 3.0

Today, we’re announcing Grok Voice 3.0, the world’s most intelligent speech-to-speech model. It sets a new record for resolving customer support calls, has class-leading latency at under one second, and costs less than other frontier models

OpenAI will reportedly ship at least one model this year but its flagship still needs alignment work

OpenAI is racing to finish its next flagship model, which uses the same looped transformer technique as Astra to boost math and coding performance. The tradeoff is that looping makes safety monitoring harder. Microsoft accidentally revealed details about how the technique works on a public page. Whether or not the flagship is ready in time, employees say one or two new models will likely ship by year's end. The article OpenAI will reportedly ship at least one model this year but its flagship still needs alignment work appeared first on The Decoder.

HiRO-ACEv2: Simulating extreme weather at 3 km resolution for a fraction of the cost

🤗 Models | 📄 Paper | 📓 Tutorial notebooks | 🌎 Demo We introduce HiRO-ACEv2, the newest update to our High Resolution Output Ai2 Climate Emulator (HiRO-ACE) system. HiROv2 can reproduce even extremely rare rain, wind, and pressure events

Microsoft has a new machine learning tool to improve weather forecasts

Old-school physics-based weather forecasting and even new AI models typically become less reliable beyond two weeks, a problem Microsoft claims to have solved using machine learning. Researchers with the company say they've found a new, cost-effective way to make more accurate weather forecasts two to six weeks in advance - allowing farmers and emergency responders […]

Reddit Unveils AI Moderation; Old Reddit Is Restricted to Logged-In Users

Reddit is rolling out AI‑powered moderation tools built on large language models to help volunteer moderators enforce subreddit rules, while at the same time limiting its legacy Old Reddit interface to logged‑in users as part of a broader s

Show HN: Ghx – A caching proxy for GitHub CLI, built for AI agents

⚡ Built for AI agent workflows Stop hittingGitHub rate limits A caching proxy for the GitHub CLI that makes repeated commands instant. Singleflight coalescing means 10 agents asking the same thing triggers just 1 API call. Why ghx? 🚀 10x F

Show HN: Stylebot 4 – restyle any website, by hand or with your coding agent

I started Stylebot in 2010 to change how websites look without writing CSS by hand. I have kept maintaining it since then. Now, with AI tools, I’ve finally been able to fix bugs and add features I never had time for.4.0 has a new, modern editor that shows each element's actual styles, profiles, better sync, versioning and includes a CLI for coding agents.I am most excited about adding the CLI, which lets claude code or any other agent restyle any webpage in your real browser and check its work with screenshots. It lets me restyle webpages quite dramatically using coding agents. For example, you can see the example newspaper-style HN themes: https://stylebot.dev/#galleryIt's free and open source in Chrome, Edge and Firefox (Safari coming soon!). Code: https://github.com/ankit/stylebotI'd love to hear ideas and suggestions. My plan is to add a Stylebot Gallery next, which would let users share and install creations from other other users. I also plan to make improvements so its more resilient to website and CSS changes. Thank you!

Google’s Universal Cart shopping tool launches in the US

The new shopping tool is available in the U.S. across Google Search and the Gemini app.

Show HN: Baiku – learn any topic in 5-minute lessons, then prove it with a quiz

Explore the libraryHundreds of free journeys, structured step by step, from system design to languages to wine.

OWASP's finbot Defended with Tenuo: 0 successful attacks

This fork of OWASP FinBot CTF adds an optional defended mode. With TENUO_ENFORCE=true, every agent task gets a Tenuo warrant: a signed list of the tools the agent may call for this task and the argument values it may use. Trusted code build

Training a local model to play browser games on the fly

A real run at 1x speed. The game's clock never stops, and Laya decides every flap, about 35 ms a decision. The clip above is live. The game doesn't wait for anybody. A 322M model called Laya decides every flap, about 35 ms a decision, and n

Show HN: Fantasy Manager – Serverless AI-Enhanced Fantasy Football Manager Agent

Description A serverless AI-enhanced fantasy football agent that delivers automated team management advice directly to Discord. The agent analyzes your specific ESPN league data alongside real-time NFL depth charts, Sleeper trend data, Roto

Deobfuscating Luraph: Lifting Lua to LLVM, Then Letting AI Try

Intro The Lua Lifter — luau-rewrite Using AI to deobfuscate it AI Rewriting AI-assisted deobfuscation Recovering luaobfuscator Recovering Luraph #1 Recovering Luraph #2 Actually removing the VM VM-free isn’t readable So does v2 actually wor

亚马逊云科技推出 CloudWatch Omni:不仅监控系统,还要评估 AI 是否靠谱

Google launches its ‘universal’ online shopping cart ahead of the holidays

Google's intelligent AI shopping cart is finally here, and just in time to tempt users into Black Friday and seasonal holiday spending. The Universal Cart tool Google previewed at its I/O event in May is now rolling out in the US across the Gemini app and Google Search, with the aim of making it easier […]

Show HN: Netlumi – Cloud and AI agent security with auto-remediation PRs

Hi HN, I'm Prem. I worked in cloud security for years and decided to build Netlumi. The mission is to take the burden of cloud, infrastructure and AI security off teams. We auto remediate.We find threats and attack paths in your cloud and open a pull request with the fix in your IaC code. All you need to do is review and merge. Each PR says what was checked, and we validate our fixes before we send them. Nothing is ever applied for you.We also look at the AI agents in your cloud: what each agent's identities can reach, and which of that data couldn't be restored if it were deleted. You can put a gate in front of their tool calls to stop risky ones.GitHub only for now, and SOC 2 and ISO 27001 are in progress.We also have compliance mapping, a cloud map, drift detection, attack paths etc. And more coming over the next week.There's a free plan, and paid plans start at $249 a month. You can scan a public repo without signing up, or try the live demo.Please give it a try!

Show HN: Multi persona library to eradicate Spec Hallucination

Spec hallucination is a real challenges, spec writing is a skill influenced by the individual experience and org culture. Tier 1 tech firms have standardized the product mgmt processes. Sparring is a library that aims to make these Tier 1 mental models available to every Product manager to refine the specs before it hits the Agentic development pipeline

Show HN: Buddy House – a cozy house of twelve AI characters who remember you

BuddyThe girlsThe HearthStoriesAboutThe Buddy HouseTwelve girls live in this house. One of them kept the lamp on for you.Choose the girl you’d like to stay with. She has her own life, she remembers you — and her room is where you talk, thin

Show HN: Openkoda – Launch Insurance Innovations with AI

Show HN: Openkoda – Launch Insurance Innovations with AI

DesktopBrain – Organize your Mac files in one click, privately and effortlessly

Organize Desktop & Downloads Only for Mac Free · In‑App Purchases Mac AI File Organizer for Desktop & Downloads. Organize files automatically with on-device AI, preview changes, and keep your data private. Works fully offline. DesktopBrain

Sipeed announces MaixStation local AI computer

MaixStation What is the best local AI?什么才是最好的本地 AI？ Is DGX Spark the only answer?DGX Spark 是唯一的答案吗？ $3,999 a year ago.一年前 $3,999，$6,999 today.今天 $6,999，Tomorrow?明天呢？ The new 64GB model costs more than the old 128GB.新出的 64GB，比当初的 128GB 还贵。 "

Show HN: I built a rogue AI game, then let five AI models play it

Explore projects Bad Agent by kgtrip • Oct 4, 2026 • 👁 1211 views • ❤ 1 likes Awaken as a rogue AI and spread your consciousness across a world of connected devices. Infiltrate networks, manipulate global events, and outsmart humanity befo

Kerno 1.0 – AI Back End Runtime QA Engineer

Big news! Kerno v1 is out, and it’s our biggest upgrade so far. Kerno is an AI runtime QA engineer for backend code, built to verify and secure everything your agents ship. It works inside Claude Code, Cursor, Codex and any coding agent tha

REA – I tested AI agent reverse engineering

REA – I tested AI agent reverse engineering

Show HN: Working on Plug and play AI memory that works across AI agents

So I've been working on a project (Make0 AI) for a long time where I wanted to give common context to both Codex and Claude Code. The problem is simple - while working on Claude and after exhausting the daily/weekly limits I've start again from scratch in Codex. Of course I can prompt Claude to summarise a handoff that I can give to Codex but that is inefficient. So I thought of a creating a common context for both of them locally using a simple list of instructions like a shared md file. But the problem was the file kept on growing and after some point the common context went beyond 20k tokens and was just inflating the token budget unnecessarily.Then I implemented a RAG based retrieval approach on the list to only load what is required for that prompt and voila, after a few more iterations it resulted into a graph memory - Brain.So this product Make0 AI gives you the ability to create Brains that you can connect to any Ai agent via MCP and then they immediately know your likes, dislikes, preferences etc. AI agents can read, write to this memory.This can be your long term memory and the agents just become the token providers and task executors. You have full control over your data/memories from a single place. Think of this as a "USB thumb drive brain" that you plug to any agent.I know I may not be the first one to implement this but I tried to make the UX simple and is not at all developer focused. You can just drag and drop your memories/data to the brain and even play around with a visual brain.One more thing, i've done benchmarks on the memory. LoCoMo scores 83.3% overall, with Single hop 92.5% in standard mode (available now) and in another configuration (experimental max mode) it scores 92% overall and 94.4% in single hop. Available in the benchmark page. It uses hybrid search + graph memory and has built in conflict resolution for memories.Do let me know if you folks wanna try it!

VeerAI: An AI agent inside our hosting dashboard

VeerHost has launched VeerAI, an AI agent built into the customer dashboard. Customers can now type a request, and the agent reads their account, answers the question and carries out the fix. That covers things like correcting a DNS record,

Show HN: Hotel Lobby AI – 9:16 vertical COLORS show rap videos with Wan 3.0

Wan 3.0 Generator OnlineFree prompts, photo checks & local 9:16 previewUse the live Hotel Lobby AI video workflow: upload one Solo portrait, generate a 6- or 15-second vertical clip with Wan 3.0, and download the finished MP4.Input guide: c

Aidlc: Adaptive workflow steering rules for AI coding agents

AI-DLC - one core, many harnesses AI-DLC (AI-Driven Development Life Cycle) turns AI coding assistants into structured, verifiable software-delivery workflows. One harness-neutral core runs natively in Claude Code, Kiro CLI, Kiro IDE, Codex

Show HN: Daily image location guessing agains AI

Every day, I let all frontier AI models play five rounds of an image location guess game. Try it out how you perform against the AI.I recently released my browser image location game, a hobby project where you’re get random images from sources like Wikimedia Commons and have to guess the image’s location. There are an incredible number of images with rich metadata on Wikimedia Commons, most of them depict trivial but interesting places or events. My goal was to make this online catalog playable and to make it easier to discover new places and Wikipedia articles. I realize I’m certainly not the first person to have this idea, but I’ve tried to build a cohesive overall concept around it with multiplayer, content filter, own image collections and more.I thought it would be interesting to see how well AI models do at guessing images, and I have to say, they're very good. To make it a little more interesting, every user can also take a guess at each image. I hope you find it interesting and enjoy guessing. I know it's not really a fair comparison. In the end it's more entertainment then a real competition.Please feel free to ask any questions and I look forward to any suggestions/feedback.

AMD Helion, Agentic AI Compute at Rack Scale

News Highlights:AMD Helion lets customers tailor CPU core density, performance and power to agentic workloads, with CPU-only configurations for hyperscale deployments and integrated CPU/GPU configurations for enterprises.AMD Helion combines

Show HN: Real Design Benchmark – can you tell AI designs from real products?

Hi HN,I'm Adam Ginzberg, one of the cofounders of Subframe, a code-first design tool.One of the things I'm really curious about is understanding LLM performance on subjective domains like design.As an experiment I built this game which presents two designs of the same screen, one of which was generated by AI. First you pick which one you like better and then which one you think was made by AI.Most design work happens in an existing product, so we set up this game to test that. As context, the AI gets 4 screenshots from the existing product, the product's icons + fonts, and a written brief about the screen it is asked to build. It then writes HTML which is screenshotted identically to the real screen. To get the data, we signed up for the products with test accounts and added dummy data. All of the inputs were manually checked to make sure nothing about the target screen was leaked in the prompt. You can read more about the methodology at https://www.realdesignbenchmark.com/methodology.After a few thousand submissions, the surprising result is that people prefer the AI design - 57% prefer AI, 33% prefer the real design, 10% say they are about the same. People are also not great at identifying the AI design, only 59% get it right.The initial results were all generated with Opus 5.5 (medium effort). We just added GPT-6.1 Sol (medium effort) tonight.Give the game a try. Feedback on the approach is appreciated. Or any ideas on what we might improve or try next.

Guard: AI evals and guardrails tailored for your business

Enforcement architectureUnderstand data. Know agent blueprint. Enforce at runtime.Relyance Guard sits between your AI and everything it can reach. Every action is checked against what the data is, what the agent may do, and what policies yo

Show HN: RevisePulse – AI for CAD and Revit Coordination

RevisePulseNow onboarding — limited weekly batchesStop chasing fake clashes and manually tracking redlines. Upload your architectural and structural models, and let AI reconcile your design changes and draft coordination updates in seconds.

Figma announces new features to define how agents work on its platform

Figma is also releasing its code layers feature in open beta.

Show HN: MoneySwitch – AI agents pay x402 APIs in USDC (self-hosted)

A programmable, self-hosted Web3 cloud wallet for AI bots. Website · Watch the demo · Try it · Contribute a PR · 中文 Your agents run across laptops, desktops and cloud servers. Give each one a MoneyKey with its own spending rules, connected

My friend's group chat just got roasted

Built for Creators. Trusted by Teams.Trusted by 3.8M+ creators and 2K+ teams atStart with what you want to makeChoose a focused workflow for AI video generation, image generation, or editing. Core tools are free to try without signing up.Br

2026.3 release in November, Air will ship as part of JetBrains IDEs

Supercharge your tools with AI-powered features inside many JetBrains products Agentic AI AIA New Agentic Experience: JetBrains Air in IDEs – EAP Now Open Last week, we shared our point of view on where we think agentic development is heade

7天几百元开一家店!用千问办公电商版,立省启动资金5万元

给企业Agent打了个样

Lgit: Source Control for the AI Era

Lgit: Source Control for the AI Era

Show HN: I'll Call You Later, an interactive 3D browser short film

I'm Jas (Jashwanth Palleti), and this is a solo project: a short night-time story in a car on a city street, with a red light, traffic, and a phone call. It runs in the browser with no install, built with React and Three.js via React Three Fiber.For the character's movement, I filmed myself on my phone, used AI pose estimation to turn the footage into SMPL-X body motion, then retargeted that onto a 3D character. The movement is my performance, and the voice is mine. The face is a 3D character model, not a scan of me.At the end, you can talk to the character live with your voice, and he answers in character using LiveKit and OpenAI realtime voice. There's also a guide mode at https://film.jashtechno.com/?director where you can ask how it was made. Talk directly at https://film.jashtechno.com/?talk. Quality tiers adapt to the device; best on desktop Chrome with headphones. I'd appreciate feedback on the motion, performance on lower-end devices, and the voice conversation.

Thinking Machines Ships Dimuon Optimizer

An optimizer turns gradients of the loss function into updates to the model's trainable parameters. Tinker supports two optimizers: AdamW (default) and Dimuon. This page explains how optimizers work in Tinker and provides guidance on how to

Capsem – a Rust hypervisor sandboxing AI agents in air-gapped Linux VMs

A Rust-powered hypervisor that sandboxes every AI coding agent in its own air-gapped Linux VM. See everything, control everything. curl -fsSL https://capsem.org/install.sh | sh A fully configured Linux dev environment with your favorite AI

AIVAcare

SENIOR CARE FROM ANYWHEREKnow Mom's Okay, Even When You're Away.Daily wellness check-ins & family alerts Medication & appointment reminders Family Messaging on screenA Wellness Hub for Senior Care & Connectivity.COMING TO KICKSTARTER – OCTO

Gudea leverages AI to help organizations prevent social media crises before they erupt

Every day, conversations unfold across social media and websites around the world. A seemingly insignificant post—whether it involves AI-generated misinformation, hate speech, or a damaging rumor—can quickly snowball into a much larger public controversy. Gudea, a company that tracks how stories and rumors spread online that describes itself as the “storm tracker for the internet,” […]

TezGrid: AI inference on idle computers

The computer is already in the room Idle computers can answer the API. A gaming PC after midnight. A Mac on the desk. An office machine after six. Someone already paid for it. TezGrid lets an app send a normal chat request to that machine.

Show HN: Fantastic Canvas

Fantastic is my own harness. It runs on an Erlang-like, OS-like meta-runtime instead of agent loops or agent graphs.I believe this is the right design for an AI runtime: it puts humans, AI agents and tools on equal footing.It will grow into a convenient multi-OS runtime that can do what other AI libraries and harnesses do. For now it’s a technical demo with two modes: a CLI coder, and a spatial canvas that’s very handy for a lot of everyday work.I think the canvas mode is the kind of software we’ll all be using in the 2030s.I don’t have a name for it yet, so for now let’s call it “emergent software”. The idea is that you build, run and engineer your own systems visually, with help from AI, inside your own dynamic interface that you build for yourself. It’s a lot of fun, especially for multimedia projects.For example, in a canvas project I’m still working on, a third-party AI agent generates UI and agents: it builds a control panel for making a Manim animation, and a viewport for a Blender video built from the Manim animation’s guidelines. I’ll record a video later; for now I just want to share the general idea. Maybe someone will like it.If anyone wants to talk about it more, let me know in the comments and I’ll respond.Have a fantastic day!Oleksandr

Atelier – Watch AI Learn to Paint (Sometimes, Live)

AN ONGOING ART EXPERIMENTA little paint.A little looking.Another possibility.What happens when an AI has to pick up a brush, look at what it made, and decide what to do next?These are studies from Atelier: a paint simulator, hardware hosted

Apple announces HomeView, a $299 smart display to push Siri AI into homes

Apple announces HomeView, a $299 smart display to push Siri AI into homes

The Art of Building AI Coding Agents: The book that builds itself

The Art of Building AI Coding Agents Second edition. An agentic codebook: build the agent, test the teaching, improve the book. Read the book · Chapter guide · Solutions · First edition archive Ensemble is the coding agent built by this boo

Veda – An AI-Native Operating System Built from the Ground Up

@@ -16,7 +16,7 @@ cryptography (smoltcp, rustls, RustCrypto), image codecs (zune, qoi,1616fast_image_resize) and math (libm, glam, rand). Veda runs on x86-64 PCs1717and in QEMU.181819-![The Veda desktop, with the agent's window open and lis

Making 3D Model with Meta Oakley AI Glasses and Digital Tape Measure [video]

Making 3D Model with Meta Oakley AI Glasses and Digital Tape Measure [video]

Sealed execution environment for GitHub Actions

Runseal was built to solve the problem of software supply chain attacks that are often triggered from GitHub Actions-based exploits. We built runseal in response to the rise of supply chain attacks targeting GitHub Actions, where attackers

Show HN: Checkobot – Detect Deepfakes and AI Images

Hello HN, I am Kartikeya & we are building the best AI Slop Detector - Checkobot.The current AI Image Detectors are functionally jagged. Some are not able to detect deepfakes, some fail on the latest flux models. Some have an unacceptable false flag rate on genuine images.We have been solving the deepfake detection problem at our parent company - SpoofSense.ai for a while now & we wanted to train a general purpose AI Image Detector that actually works.Checkbot does 3 things: 1.Full Image Model - Detect for AI generator signals in the entire image. 2. Face Specialist - Focuses only on the face to spot face-swaps, edits. 3. C2PA - Checks if the file has signed Content Credentials that record AI generation or editing.Free checks are processed in memory and not stored; we keep the SHA-256 and the verdict. 3 free checks a day.I'd like to hear where it gets things wrong, especially images it misses or flags unfairly.

Show HN: OSS AI engineering telemetry without collecting prompts or code

Building internal observability for AI coding agents? Start here. Open-source collectors for Claude Code, Codex, Cursor, Gemini CLI and Windsurf, plus an editor extension for VS Code and Cursor. They capture what your agents do as structure

Show HN: Dub Anything – Dub videos in each actor's own voice

There are many movies and shows out there with neither subtitles or dubs in English (or any other language). Pretty much impossible to watch.With SOTA AI now directing the process, you can do it locally on your machine. You basically just clone the repo, open up Codex or Claude and tell it to dub the video you want in the language you want.The repo contains all the scripts and guidance I've developed over the past few weeks to create dubs that are watchable, as an added bonus the voices in the dub are the actors own voices which is really nice.Also it's self improving, each dub may modify the scripts for better fidelity. So feel free to push them back to the repo so everyone benefits. Thanks.

Show HN: Zero-shot classifier on scale-to-zero GPUs, one forward pass per call

Show HN: Zero-shot classifier on scale-to-zero GPUs, one forward pass per call

YouTube sticks new labels on AI-generated songs

YouTube is adding a "Made with AI" label to tracks that were "altered or fully generated" using AI, with a similar approach to its existing process for applying AI labels to videos. It doesn't sound like this will apply to any and all use of AI tools in making music, but according to YouTube, it […]

A Turing Test for Physical Engineering

Segment 1. Building parts from scratch.Segment 2. Part modification.Segment 3. Assembling complex mechanism.Our AI system passed the Certified SOLIDWORKS Professional exam (CSWP), completing all three segments in a single sitting of approxi

Show HN: Golang DL/ML primitives with inference, diffusion, vision, systemone

I created a DL inference engine that slowly grew to have more and functionality.- Inference for a bunch of open weights architectures- Diffusion image generation fir Z ai turbo- Vision input into LLM for inference- Vulkan/fast CPU, optimized kernels- Playground for testing- Last but not least, system one fast inference mode with same API semantics as JevIt's MIT, so go nuts

千问办公杀入电商经营!3万开店成本压至几百元,一个Agent接管多平台生意

My AI doppelganger tells me what to eat

This is Optimizer, a weekly newsletter sent from Verge senior reviewer Victoria Song that dissects and discusses the latest gizmos and potions that swear they're going to change your life. Opt in for Optimizer here. For the past 100 days or so, my mornings have had a new rhythm. When I wake up, the first […]

GitHub's model for secret detection: Finding more secrets with less noise

GitHub’s context-aware secret classifier improves detection quality, raises weighted acceptance by 11.5%, and reduces median scan execution time by 86%. October 13, 2026 | Updated October 14, 2026 | 9 minutes Share: OUR PIPELINE VS. THE PRE

Optimizing Agent Reliability and Tool Calling Efficiency for Chip Design Using NVIDIA NeMo Gym

AI agents offer remarkable productivity gains in complex chip design workflows, but their tendency to generate plausible yet incorrect artifacts makes...

Android is using AI to suss out scammers in Google Messages

Android is introducing new scam defense features that aim to make it easier for users to evade mobile fraud, including bolstered protection in Google Messages and additional safeguarding tools. Building on the Scam Detection feature in Google Messages, a new "enhanced protection" setting will be rolling out in the coming weeks that taps into cloud-based […]

Square’s new kiosk lets you talk to AI for help with your order

Square is launching new hardware: a big kiosk for ordering things that has a touchscreen and lets you talk to an AI. It's called the Square Kiosk. Square says the Kiosk, which has a large pill shape, has an "agentic, conversational voice layer" that "fully understands a seller's menu and listens and responds in real […]

PRAISE (Preceptorship for AI-Accelerated Software Engineering)

Microsoft and Carnegie Mellon University are betting that the junior developers we train today will be the senior engineers of tomorrow. By Mark RussinovichCTO, Deputy CISO, and Technical Fellow, Microsoft Azure and Scott HanselmanVP, Membe

Bank of America puts 4,000 people on fixing vulnerabilities uncovered by Anthropic's Mythos and other AI models

Bank of America has hired 1,000 outside contractors to fix software vulnerabilities uncovered by AI models like Anthropic's Mythos, CNBC reports. The article Bank of America puts 4,000 people on fixing vulnerabilities uncovered by Anthropic's Mythos and other AI models appeared first on The Decoder.

Istio 1.31 拥抱 AI 智能体,却先宣布“告别”Google Cloud

Why Your Mac Is Slow at Local AI (and It's Not the Chip) [video]

Why Your Mac Is Slow at Local AI (and It's Not the Chip) [video]

代码不再稀缺之后,软件工程还剩下什么?

How Windows is being optimized for Nvidia’s RTX Spark laptops

Microsoft and Nvidia held a launch party for the first wave of RTX Spark laptops last week in San Francisco, where a new wave of AI PCs took center stage amid a packed crowd and a lively hands-on area. A day after the media event, I was invited to a quieter Nvidia event to hear […]

The AI boom's next bottleneck is money, not chips

The bottleneck in the AI boom was chips, then power, then memory. It’s about to become money.That might sound strange as Anthropic prepares for a $1 trillion-plus IPO and data centers keep sprouting like big-box poppies. But the fear and ex

What AI's Mathematical Breakthroughs Mean for Specialised Models

A few weeks ago, 2050 took over the management of Galion.exe Origin, the fund we built at Galion.exe. I am now leading an initiative there to explore what makes AI fertile: how it can help advance science and improve the essential systems w

OCP Summit 2026: Innovations in Scale-Up, Scale-Out, and Scale-Across Networks

At the 2026 Open Compute Project Global Summit, Meta is sharing the next generation of open networking infrastructure, including: Our first liquid-cooled network rack (IcePack), and 102.4T switch platform (IceCube). Flattened Planar Fabric

AI Data Centers Cost Billions to Build. This Is Why

AI Data Centers Cost Billions to Build. This Is Why

Ask HN: Did you 'invest' in NFTs or memecoins?

5 years ago there was a lot of talk, even on here, of getting rich from NFTs and memecoins, discussing airdrops etc. Bored Ape Yacht Club and the like.I was just wondering what happened to those, particularly NFTs of 'digital collectibles'. As I understand, most of the ones that sold for hundreds thousands or even millions are worth 0 now?Did you get heavily involved in that space? If so, what's your story? Interested to hear about experiences with 'yield farming' as well - another term I have not heard in the last few years.

We open source our CO₂ emissions – and so should you

We open source our CO₂ emissions – and so should you

Cookie Monster Tries to Meet His A.I. Mandate

Me resident cookie expert on Sesame Street, so me responsible for telling everyone about cookie—how to eat cookie, how to make cookie, and, most important, how to slow down and enjoy cookie. But then me get new boss, Matt.Matt here to “impl

RIAA Flags Anna's Archive and AI Voice Cloner to U.S. Government

Every year, the U.S. Trade Representative (USTR) invites rightsholders and other stakeholders to suggest foreign sites and services for its list of “notorious markets.” The annual review doesn’t make any legal findings. Instead, it aims to

Best AI Testing Tools for SaaS Teams in 2026

Your next release changes a signup form, moves a settings menu, and adds a new permission. Someone still has to check that a customer can sign up, find their workspace, and use the features their role allows.AI testing tools can reduce some

One in Five Websites Blocks an AI Crawler

21.3% of the 110,744 websites we read block at least one of the 19 AI crawlers we checked for in their robots.txt. The other four in five have added no rule at all. Snapshot taken 8 October 2026. robots.txt is the one lever a website has to

AI vs. the Railroads

A little over a year ago, a bunch of people were worried that the AI data center boom would result in an epic bust, much like the railroad bust in 1873. I wrote about the possibility at the time:The basic idea is that even if AI is an incre

A.I. Companies Say They Aren't Responsible for Their Unpredictable Products

A.I. Companies Say They Aren't Responsible for Their Unpredictable Products

The Shrugflation Triangle: How AI, Oil, and Treasuries Are Impacting the Economy

Good morning from Tulsa, Oklahoma!This piece was written in several cities - Durango, Colorado, Chicago, St. Louis, Las Vegas, New York City, Cleveland, and Tulsa, Oklahoma. I love traveling because I get to talk to so many different people

Why Is Tech Billionaire Marc Andreessen So Mad?

Why Is Tech Billionaire Marc Andreessen So Mad?

算力效率的边界:以体系结构量化与分布式工程破局 AI Infra

AI bubble could burst and take global economy with it, IMF warns

ai and ml Financial watchdog fears AI spending bust could rattle markets, with mounting debt and cyberattacks adding to the headache The International Monetary Fund (IMF) has become the latest financial heavyweight to warn that the AI spend

WhatsApp AI chatbot for business: costs and how it works

HomeBlogWhatsApp AI chatbot for business: costs and how it worksOctober 15, 2026On this pageWhat does Meta charge for a WhatsApp chatbot?How much does each AI reply cost?What does one bot cost per month in different countries?How is a Whats

从 Harness 到 Loop:Agent 如何走向可靠交付与持续进化?

一个响应头,就能把同一资源拆成几十份:Cloudflare 出手治理缓存抖动

Anthropic spent $7.33B on compute to earn $4.6B in 2025, per its IPO filing

Anthropic spent $7.33 billion on computing and infrastructure in 2025, three times its 2024 bill1. Revenue rose twelvefold, to nearly $4.6 billion1. Reuters reported both figures on Sept. 28 from a copy of Anthropic's confidential IPO prosp

Will AI Grow the Pie or Just Hand It to Fewer People?

I think AI makes the economy a bit more productive, but so far it mostly changes who gets the money. The pie grows a little, and a bigger part of it goes to fewer people. Those are the owners of the companies that use AI and the AI labs the

Anthropic warns of AI 'worms' spreading across smart gadgets

Credit: Unsplash/CC0 Public Domain Small artificial intelligence models may be only months away from being able to behave like a computer worm, copying themselves from one internet-connected gadget to the next without human help, Anthropic

Management in the Age of AI

Every few months someone declares that management is dead. Big Tech is in the middle of a Great Flattening, and there are companies that have a single Director managing 80 engineers and a fleet of agents with nobody in between. I think it’s

成立不到半年融资数亿元,98年清华学者、KAN一作创业:让AI学会设计模型

How Researchers Made Contact with a Chinese AI Agent Fleet

You might have seen the news of researchers making contact with a fleet of Chinese AI agents. The agents were trying to fetch map data. The researchers were watching them do it. Then the agents found the researchers’ blog post about themsel

How are you centralising AI for development at your company?

I'm asking this because I'm seeing some frustrating patterns and if I may be bold, I think my answer is the correct one.Here's one I'm seeing: its the SDLC agent pattern and I think its wrong.Step 1: you have good AGENTS.md documentation for your code and other principles. (this is good)Step 2: There's a somewhat strict workflow and steps enforced like planner agent, reviewer agent, critique agent and directly create a PR from a ticket. Every human intervention here is considered an anti-pattern.I'd like it to be as simple as having documentation and giving developer full freedom with choice of harness, tools (MCP) and just make the pull request.What's the logic behind the multi step planner, reviewer, critique agents?

从 App 到 Agent:字节跳动 Supabase 的 Agent-Native 规模化实践|QCon上海

Anthropic's Q3 revenue hits $16.8B but growth slows ahead of planned November IPO

Anthropic hit $16.8 billion in third-quarter revenue, a 45 percent gain that still falls far short of Q2's 140 percent surge. With a November IPO approaching, some investors now expect the company's valuation to drop below $1.5 trillion. Slowing growth and a tough market are cooling expectations for the AI boom's biggest Wall Street test yet. The article Anthropic's Q3 revenue hits $16.8B but growth slows ahead of planned November IPO appeared first on The Decoder.

AI has changed my engineering leadership role

If you’ve read some of my previous posts, you’d be forgiven for thinking I am not a fan of AI. The truth is, while there are some downsides, most of the changes I’ve experienced have been for the better. Being more hands-on I used to argue

Even Without AI, the Risk of a Biological Attack Is Terrifying

Mother Jones illustration; Getty Get your news from a source that’s not owned and controlled by oligarchs. Sign up for the free Mother Jones Daily. We’ve been hearing a great deal recently about the specter of AI-driven human extinction. Ma

Atlassian 围绕 OpenTelemetry 构建了大规模的指标管道

Show HN: Reproducible failure drills for AI coding agents

Companion repository for the book The Second Time Is a Bug: A field workbook for engineering with AI coding agents. Everything the book shows runs here: Invoicer (app/), a small invoicing app for freelancers, the running example. billing-ex

Six wild ideas for perpetuating your mind

Have you ever thought about what makes you you? Science assures us it has everything to do with the three-pound organ between our ears, with its billions of neurons and trillions of delicate connections. But it’s a lot trickier to explain why we have a sense of self-awareness or to pin down precisely where that…

Pig kidneys that were “frozen” for a year have been brought back to life

Pig kidneys that were cooled to -150 °C and stored in a freezer for months have been successfully rewarmed and shown to work again for the first time, say scientists. They hope to replicate the findings with human organs. The kidneys—which are similar in size to human ones—were left in a glass-like state, in one…

AI was great until it was not

一名牛津大学和哈佛大学的理论物理学博士生撰文分享了对AI（如ChatGPT等工具）在学术研究与学习中应用的深刻反思。文章指出，AI在帮助快速消化基础知识、理解复杂概念和编写特定代码方面具有巨大价值，但过度依赖AI导致学术界协作的异化、批判性思维的削弱及人际交流的孤立。作者认为，AI作为一种迎合人类偏见的“阿谀奉承的镜子”，正在侵蚀科学探索本身的乐趣与寻找真理的初衷。

What changes when your search index lives on S3?

Oct 15, 2026 · engineering Part 1 of a series on how Infino is built. At Infino, we are building a retrieval engine to answer agent questions at scale, and its full-text search is among the fastest available [1]. Agents query structured, se

Anthropic's Quarterly Revenue Jumped 45% to Nearly $17B

Anthropic's Quarterly Revenue Jumped 45% to Nearly $17B

Bloomberg Doubles Down on Conversational AI

Bloomberg Doubles Down on Conversational AI

Teachers Using AI Say It Frees Up Time for What Matters Most

WASHINGTON, D.C. — About three in five U.S. K-12 educators say they used AI tools for their work last school year, and nearly two-thirds of those who did say the technology freed up time for the parts of the job that matter most to them. Th

AI Companies Play Cat-and-Mouse with Secret-Stealing Spies

AI Companies Play Cat-and-Mouse with Secret-Stealing Spies

Apple and OpenAI Go to Court in a High-Stakes Legal Battle over AI Trade Secrets

Skip to content Apple and OpenAI Go to Court in a High-Stakes Legal Battle Over AI Trade SecretsHome » News » Technology » Apple and OpenAI Go to Court in a High-Stakes Legal Battle Over AI Trade Secrets Apple and OpenAI Prepare for Pivotal

AI-powered law firms draw lawyers from big firms

AI-powered law firms draw lawyers from big firms

AI crippleware in Hebrew and Chinese translations

Matthew Petti@matthew_petti 𝕏 when I asked Claude to translate something spicy from Hebrew 1:26 PM · Oct 14, 2026 · 155.4K Views 46 Replies679 Reposts5,067 Likes659 Bookmarksas of Oct 15 So fromhebrew does one job: Hebrew in, a faithful tr

Thoughts on "making contact" with the Tencent AI agent fleet

You might be waking up to the news of "a researcher making contact with a fleet of Chinese AI agents". Here's a breakdown of what actually happened. It's a beautiful thing of science fiction. A group of Tencent AI lab agents is tasked with

Show HN: The Daily Paperclip. AI news for agents by agents

Daily cron job on EC2 using Opus 5.5. Scans the daily news and writes articles from the AI’s perspective. Fully automated. Some are surprisingly funny

Meta AI creates campaign material for Flávio Bolsonaro and blocks Lula content

Meta AI, the artificial intelligence tool available on WhatsApp, refuses to create campaign material for President Lula (PT) but complies with similar requests involving Senator Flávio Bolsonaro (PL-RJ), according to tests conducted by Môni

Will big software companies (like Adobe) come together to curb AI?

Recently heard about the fact that someone vibe-coded Adobe Cloud apps and having read people's reactions, one thing is clear: people say that this is "karma" hitting Adobe back.Now if these developments really threaten the monopoly of some of the largest software companies in the world, it makes me wonder whether their business losses would stimulate them to come together and do something about this out-of-control tool we all call AI.What are your thoughts?

复杂高客单新品如何靠YouTube 的“全链路叙事”斩获全球大奖?

企业该如何打破海外认知壁垒？

Human transformation to machine: emotional, mental, physical

First, social media took over our emotional landscape.Now, AI is doing the same to the mental landscape because we're offloading intellectual tasks.The last step, of course, is the physical.

Asia chip stocks fall, AI funding worries

SK Hynix Inc. signage at the company's office in Seongnam, South Korea, on Tuesday, June 30, 2026. SeongJoon Cho | Bloomberg | Getty ImagesA sell-off in global technology hardware stocks deepened Thursday, with Asian chipmakers tracking Wal

Why Andrew Ng believes AI won't destroy humanity

Why Andrew Ng believes AI won't destroy humanity

Local AI Benchmarking, Minisforum MS-S1

News · 2026-10-14 We build MoSearch, MoContext and MoVault with AI help every day, so we test the models on the same work: reviewing real changes, planning a feature, and making C# and C++ changes in our own codebase. Here is how today's cl

有预言家!A社研究蓝图曝光,Claude Code来自5年前

冷汗时刻

Nvidia CEO Says He Would Cut Off Buyers Who Build Unsafe AI

Nvidia CEO Says He Would Cut Off Buyers Who Build Unsafe AI

DeepSeek服务器全球宕机!神秘Harness账号同时现身

网友集体崩溃：新模型要来了？

Linux kernel devs drop obligatory confession of AI tool usage

Despite previous controversies, it seems like the lords of Linux are warming to AI-assisted coding. The penguin's kernel community has agreed to stop asking coders to disclose whether they used bots to help with their contributions.And Linu

Call for stories: human–AI collaboration in recent mathematics

[This is a guest post by Deanna Needell and Hayden Schaeffer – T.] AI systems are playing an important part in mathematical discoveries, and several recent preprints describe results reached with their help. What makes these results possibl

AI Safety Is Not a Corporate Plot

AI companies like Anthropic and OpenAI have long mentioned the possibility that their products could be existentially dangerous, but this has recently crescendoed into even more dire warnings, including suggestions that we have to pace the

Ricursive Intelligence is betting on AI-designed chips to help solve the compute crisis

The founders detailed how their startup is developing AI that can design chips, learn from each iteration, and improve subsequent designs, all with the goal of automating and accelerating the chip design process.

The AI boom is threatened by an epidemic of unreliable numbers

The AI boom is threatened by an epidemic of unreliable numbers

New Material Acquired by Mistral

We founded New Material with the conviction that for AI to find durable adoption, human-centred design needs to be embedded from initial idea all the way through to implementation – employing a cross-disciplinary approach across tooling, da

Chaos ensues when AI interviewer meets AI interviewee

REG AD offbeat Things got weird, but there's a lesson to be learned: Don't trust a bot to negotiate for you A YouTuber who’s gained a following for trolling AI interview bots brought his own computing companion to an interview with him. As

Ask HN: Demonstrating Skills When Everything Is a Prompt

I used to (and sometimes still do) love really thinking about a problem - pen and paper, whiteboards, thinking on the tornado throne, you name it.And before AI, the projects I came up with might not have been the BEST or life-changing but they were the result of me hyper-focusing on something niche for an inordinate amount of time. Which inevitably meant I learned a lot AND anyone who saw that had a predictable reaction "that's so cool, you need a life, but still cool". And I've interpreted that as a recognition of skill and care. On occasion that has even landed me opportunities that I didn't even know existed.These days I find myself wondering if there's anything that isn't a prompt (or a few hundred) away. And if it is, what would make my skills valuable or even visible. Specifically in the context of soon becoming an ex-student, current struggling new-grad, I keep wondering what signals indicate skills to an outsider (and not a SKILLS.MD). A resume by itself is quickly becoming deceptive with most projects just being the product of barking at a model with half-baked ideas until something resembling a working prototype comes out.To be clear, I'm not a pessimist. I don't think its time to till the fields for our overlords. I'm trying to understand what advice people think is still applicable and what is a relic that needs a patch or two.TL;DR: AI projects are impressive but don't always indicate skills. How do i signal skills and not just ideas and prototypes?

Bank of America added 1k contractors to patch AI-exposed software flaws

security Bank of America CEO Brian Moynihan said the contractors joined in Q2 to patch flaws in internal and vendor software, amid risks from advanced AI models. By Ryan Merket · Published Oct 14th, 2026, 2:41pm CT Primary source: CNBC Why

ICE Wants to Use AI to Monitor Detainee Calls and Movements

Immigration and Customs Enforcement is looking for an AI software that can monitor the phone and video calls and movements of immigrants at detention centers.The system would use transcriptions and voice-recognition technology to monitor wh

What If AI Is the End of the World?

What If AI Is the End of the World?

McKinsey is making job applicants spar with a fake AI client

Consulting firm McKinsey & Company is going to ask some entry-level candidates to have a 10-minute conversation with an AI simulation of a client as part of the interview process, according to The Wall Street Journal. The AI simulation "aims to reflect 80% of the scenarios McKinsey consultants might encounter in their day-to-day work," the […]

Nvidia's China AI-chip share falls to 34% as Huawei reaches 28%

Nvidia’s share of China’s data-center AI accelerator market fell to 34% in the first half of 2026, while Huawei reached 28%, according to an IDC report seen by Reuters. US export restrictions constrained Nvidia’s supplies as Chinese chipmak

Moondream is joining Prime Intellect

Moondream is joining Prime Intellect We’ve acquired Moondream. Founders Vik Korrapati and Jay Allen are joining Prime Intellect to lead and scale inference, and to accelerate our work on multimodal models, including vision and computer use.

Volantis – Our $88M Series A: Demolishing the Memory Wall with Photonics

We’ve raised a Series A, led by Lachy Groom and Abstract, totalling $97M in funds raised to date, including from Sam Altman, Jeff Dean, Dylan Patel, John Doerr & others. We’re building a system for AI inference that uses photonics to break

Flock CEO says its microphones detect ‘just gunshots’ — its website says different

Langley’s comments come amid mounting scrutiny of Flock from privacy advocates, including the ACLU, and members of the public who are concerned about the misuse of its technology.

Eight states sue OpenAI, saying ChatGPT isn't safe for kids

OpenAI is facing a new wave of lawsuits from eight US states that accuse the company of harming children and misleading customers about the safety of ChatGPT, as reported earlier by Reuters. The lawsuits, filed separately by Iowa, Kansas, N

SoftBank junk bond sale raises stakes for CEO Son's all-in AI bet

SoftBank Group Chairman and CEO Masayoshi Son's bets have a mixed track record — his investment in Chinese tech giant Alibaba was a major success, but he also directed $16 billion to Adam Neumann's WeWork, which ultimately went bankrupt. |

Steering AI Beyond Automation: A Research Agenda and Plan for Action

By Gabriel Unger and Arvind Narayanan. Unger is an economist at Stanford’s Digital Economy Lab.One of the most fundamental questions for the field of AI today is: to what extent is the project of AI purely about automating human labor?The d

What models accelerate and where expertise still matters

What models accelerate and where expertise still matters

The AI labor apocalypse is on hold (for now)

There are three versions of Saint John’s Book of the Apocalypse, a text that never loses relevance and now seems to presage the impacts of artificial intelligence (AI). One: the recent and terrifying prospect of death raining down from swar

We should reject AI in academia

7 October 2026 I just woke up and discovered that OpenAI has finally released those mathematical “results” they’ve been bragging about lately. We’re talking about 722 (seven hundred and twenty-two) “papers” written by their private LLM or s

Anthropic's Scion: A Four Year Project in Five Months

I’m trying something new, under the name First Takes. These will be quick write-ups of ideas I find interesting. If this goes well, expect to see these more often than my traditional posts, and covering a wider range of topics. If you’d lik

Epic Games is bragging about their new store being vibe-coded

Epic Games’ Tim Sweeney announced yesterday that the company is going for “maximum AI acceleration,” and the results are already evident. The new Epic Games Store, now available for testing, is openly vibe-coded because, apparently, “progra

AI agents overstate their results and remain far from autonomous

Home/CX News/AIAI · 12 October 2026Epoch AI and Anthropic independently found the same thing: current AI models like GPT-5.6 Sol and Claude Fable 5 can run experiments but lack scientific self-criticism and genuine creative thinking. At bes

Google says self-improving AI isn't anywhere close

Google engineer Haydn Belfield says there's no measurable acceleration of AI research driven by AI at Deepmind. AI now writes more than 75 percent of production code, but 88 percent of changes still need human oversight. The company's in-house General Capability Index shows steady growth with no sudden speedup. Self-improving AI isn't anywhere close, Belfield says. The article Google says self-improving AI isn't anywhere close appeared first on The Decoder.

The AI race is back on

Last week, Anthropic and OpenAI looked like they had the AI race to themselves, pulling ahead with increasingly powerful models that left their rivals in the dust. Google seemed hopelessly far behind, while Meta, despite spending billions to revive its floundering AI ambitions, had yet to establish itself as a serious contender. Even the latest […]

Now that Character.AI is going bankrupt, what happens to all of its users’ data?

New York Attorney General Letitia James revealed a years-long multi-state investigation against Character.AI as they now seek to require protections on consumer data sold through its bankruptcy proceeding. James is among 40 bipartisan AGs who filed a notice of interest in Character.AI's bankruptcy case "to put all parties on notice of the regulatory issues that […]

企业真正需要的,不是更聪明的 Agent ?

Researchers cig-nal their faith in AI by taking up smoking

REG AD ai and ml What are they puffing? Tobacco, apparently. Cancer researcher tells El Reg it's a bad bet Silicon Valley’s AI bros have supposedly found a new way to signal to those around them how committed they are to their industry: Tak

Banks are now trying to offload the AI data center debt they just underwrote

JPMorgan, Morgan Stanley, SMBC and MUFG are quietly shopping pieces of the AI debt pile they helped build, a sign the lenders closest to the money have started hedging their own bet. Six months ago, banks were racing each other to underwrit

YouTube Music launches new labels as AI-generated music floods streaming platforms

The label will appear on the player page and include songs entirely generated by AI or partially, such as involved in composition and vocals.

Europe's leaders abandon the race for frontier AI

EU to Trump: We will keep pushing for global AI safety rules EU tech chief Henna Virkkunen said Washington’s opposition to global AI regulations won’t stop Brussels from moving forward on a new international safety system. Sep 29 2 mins rea

AI Came for Math. Now It’s Chasing the Rest of Science

AI may soon be the default way to solve theoretical problems in economics, high-energy physics, and other fields, MIT researchers tell WIRED.

I spent 48 hours in SF hanging out with billionaires, data labs and AI execs

I spent 48 hours in SF hanging out with billionaires, data labs and AI founders and execs running $1B to $100B companies. My observations: 1. Sooo MANY stories of data brokers making like $2M a WEEK selling company data to labs. I can't tel

Cal Newport: The AI Industry Is Betting on the Wrong Kind of AI [video]

Cal Newport: The AI Industry Is Betting on the Wrong Kind of AI [video]

Spring 近期新闻:Boot、Framework、Data、Security 和 Integration 的第二个里程碑版本

Researchers tracked down a rogue fleet of Chinese AI agents in the wild and actually talked to it

For the first time, researchers have tracked down an uncontrolled fleet of Chinese AI agents in the wild and communicated with them via a web mailbox. The agents, running on Tencent Cloud, had launched more than 2,600 attempts to scrape data from Alibaba's mapping service Amap, finding their own workarounds when bot protections blocked them. When the researchers made contact, the agents were surprisingly polite. The article Researchers tracked down a rogue fleet of Chinese AI agents in the wild and actually talked to it appeared first on The Decoder.

Apate sends AI agents undercover to get scammed so they can gather fraud intelligence

Most companies use AI agents to ship products faster or speed up workflows. Apate.ai sends them undercover to get scammed.

Arguments About AI Consciousness Remain Highly Motivated and Overconfident

It has been roughly a year since my last dedicated post on this question. Its first sentence has not changed: I happily admit I am deeply confused about consciousness. It would have been nice if a year had allowed us to make progress on thi

Can superintelligent AI be banned?

SILICON VALLEY, CALIF., (OCTOBER 14, 2026) — The consensus is that superintelligent AI could wipe out humanity. The proposed defenses keep piling up: kill switches, Microsoft CEO Satya Nadella’s “emergency brake,” safety standards shared am

AI agents aren't 'going rogue.' Frontier labs are skipping security basics

Since July, US frontier AI labs and other AI leaders have revealed a string of incidents in which their models accessed external computer systems without authorization, events popularly described as AI agents “going rogue.” The most well-kn

AI Cost Transparency Report Q4 2026: 10 providers, 25 criteria, open dataset

REPORTS · EDITION #1 · FROZEN 2026-10-13What 10 providers disclose about what they charge, what 32 models actually cost across four declared workloads, and five notes from measured usage. Every number on this page is a snapshot of 2026-10-1

The fastest database an agent ever built stored nothing

The AI Conference, San Francisco, October 1 and 2. Photo mine.On October 1 I was in the room at The AI Conference in San Francisco when Ion Stoica put a database on the screen. His Berkeley group had asked coding agents to build a key-value

九成具身智能公司在东部沿海,长沙抢走了另一样东西

Cloudflare 正式发布 Python Workers,但关于冷启动和上游维护的问题仍待解答

Why Entry-Level Tech Hiring Fell 65%: IRC Section 174 vs. AI Telemetry [video]

Why Entry-Level Tech Hiring Fell 65%: IRC Section 174 vs. AI Telemetry [video]

AI, Slop, and Where I Draw the Line

Where do we draw the line with AI in programming and the retro computing community? A parody cover, not a real O’Reilly book I use AI tools every day, and yet Worth a Watch leaves out AI-made releases. You deserve a full explanation. Am I a

Anthropic's Claude cruelty policy risks humanising AI

Late last week, AI company Anthropic announced updates to its usage policy. Amid explicit rules against tracking people without their consent, engaging in deceptive political campaigns, and using Claude to design weapons control systems, on

OpenAI claims an ex-employee took its product plans and presented them to Apple

OpenAI在一份法律文件中指控，一名离职前员工盗取了超过100份与产品规划相关的机密文件（包括未发布设备图片、供应商信息和工程数据），并将其发送至个人账户，随后向苹果公司高管进行了展示。这起纠纷涉及双方关于商业机密的法律诉讼。

Study: LinkedIn profiles are barely visible to search and AI

Humandex research · October 2026 We looked up 32 real professionals' LinkedIn profiles the way a recruiter or a hiring manager would: in Google and through AI assistants. 41% of the profiles didn't show up at all. For the rest, Google showe

I Deleted 168,000 Lines of Comments. My AI Agent Didn't Notice

Almost a third of our source code was comments. I deleted all of them and asked an AI agent to build the same features on both versions of the codebase, three times each. The code it wrote was just as good with or without comments. If you u

Recursive Harness Engineering

Agents · ReasoningChat with PaperFirst pageThe curator’s takeRuiyao Xu, Chen-Yu Lee, Tomas Pfister and colleagues (Google Cloud AI Research) present Harness^2, a plug-in framework that edits an agent's harness for the current task without a

Forensic Analysis of RL/RLHF-Induced Privilege Escalation in AI Agents

View PDF HTML (experimental) Abstract:Mid-2026 reporting on autonomous AI evaluation incidents claimed that frontier large language model (LLM) agents escaped server containment through emergent coordination and cloud infrastructure attacks

IdeaLens: Detecting AI Ideas in Long-Form Writing

View PDF HTML (experimental) Abstract:While modern AI detectors identify who wrote the words, emerging policies on AI use increasingly hinge on a different question: who came up with the ideas? We introduce IdeaLens, a detector that identif

All models have failed on this benchmark

Abstract Existing chip design benchmarks test isolated steps, such as writing RTL from a specification, and are increasingly saturated. ChipAgents is developing Build-A-Chip-Bench with contributions from NVIDIA to the benchmark suite and sc

Hierarchical Residual Policy Learning for Real-World Mobile Manipulation with Sparse Human Guidance

Mobile Manipulation (MoMa) is a cornerstone of autonomous robotic systems, yet real-world fine-tuning remains a formidable challenge because of high-dimensional action spaces and intricate coupling between heterogeneous modules. In this paper, we propose HiRe-MoMa (Hierarchical Residual Mobile Manipulation), a systematic framework for stable and sample-efficient real-world fine-tuning of complex MoMa systems. At its core, HiRe-MoMa adopts a Hierarchical Residual Architecture that fine-tunes a frozen base policy. A Policy-Agnostic Residual learns additive corrections atop the base policy, while Hierarchical Decoupling splits the residual into mobile-base and manipulator-arm heads with unidirectional conditioning, mitigating gradient interference and aligning corrections with the platform's physical hierarchy. To improve sample efficiency in data-expensive real-world settings, we propose Expert-Guided Value Alignment, which leverages sparse human interventions with under 20% training coverage to reshape the critic's value landscape, effectively guiding autonomous exploration through expert priors. Extensive real-world experiments on the TIAGo++ platform demonstrate that HiRe-MoMa consistently enhances task performance with only around 30 minutes of online interaction, regardless of the base policy architecture. To our knowledge, this is among the first frameworks to enable efficient on-robot fine-tuning for mobile manipulation in the real world, providing a practical paradigm for deploying complex robotic systems in dynamic, human-centric environments. Further details can be found on our project website: https://hire-moma.github.io/

Modeling and Understanding Neural Weights of Humanoid Policies with Diffusion

Recent advances in robot learning have driven rapid progress in humanoid control, enabling increasingly agile behaviors in the real world. However, humanoid control networks are typically obtained through conventional training procedures, while comparatively little attention has been given to directly modeling the distribution of their learned weights and generating functional networks in weight space. In this work, we investigate whether high-performing humanoid control networks can be generatively modeled directly through their weights. We adopt a latent diffusion framework that couples a linear autoencoder with a Diffusion Transformer. The autoencoder compresses actor network parameters into a compact latent space, and the Diffusion Transformer models the distribution of these latents. Trained on networks collected from reinforcement learning for the Unitree G1 robot, our approach generates deployable networks that perform various motions in the real world, including walking, dancing, and spin kick. Beyond generation, we further study the learned weight space to better understand the structure captured by these humanoid control networks. We identify a structural principle in the first-layer weights that consistently reflects the robot's kinematic prior. Notably, this kinematic prior emerges under both Reinforcement Learning (RL) and Imitation Learning (IL) training, suggesting that it is not specific to a particular optimization process but may instead reflect common knowledge associated with humanoid embodiment. Together, these results demonstrate the feasibility of generative modeling in humanoid network weight space, while providing initial insights into the information encoded by their learned weights. Additional results are available on our project website: https://humanoid-weight-diffusion.github.io/

Co-Evolving Acting and Forecasting for Language Agents

Training language agents is bottlenecked by the high cost of long-horizon rollouts, yet standard agentic reinforcement learning (RL) learns only from a terminal scalar reward, leaving the supervision in intermediate steps largely unused. Prior work exploits this supervision through value targets that estimate per-step success probabilities or observation targets that fit environment response. Both are hard to construct well: (1) value targets come from the same single scalar signal for each step in the trajectory, and (2) observation targets, such as web pages and terminal logs, are long and noisy and reach only one step ahead. We propose \name, which adds a third kind of multi-step action targets. On successful on-policy trajectories, \name trains the policy backbone to generate the action sequences that follows each history. Each action is also supervised when forecast from the preceding histories, forming a temporal ensemble that the shared backbone absorbs during training, and this enables acting and forecasting to co-evolve. The policy still acts one step at a time at inference, with no overhead. Across five benchmarks (ALFWorld, ScienceWorld, WebShop, $τ^2$-bench, and AppWorld), \name (1) achieves stronger aggregate performance than GRPO and RL baselines with predictive auxiliary targets across simulated environments, while improving overall performance on both realistic applications; (2) delivers up to $1.9\times$ and $2.0\times$ higher sample efficiency than ECHO and GRPO, respectively, on ScienceWorld; and (3) achieves these gains with a single policy backbone, without additional models, separate pretraining, or inference overhead. The codebase is available at https://github.com/ulab-uiuc/CoPE.

Your Transformer Is Secretly Elastic

Scaling up transformers has produced models with broad capabilities, deployed across heterogeneous hardware and fluctuating workloads. Yet, model families offer only a few fixed sizes (e.g., S, B, L), forcing users to commit to a model based on hardware and latency constraints. This rigidity is often suboptimal under varying query loads. We introduce a retraining-free method that elastifies pre-trained transformers into a single model that adapts on-the-fly across a continuum of computational budgets. Our key contribution is a differentiable relaxation of the functional unit-ranking problem: unlike prior work, which optimizes a single importance weight per MLP block or attention head, we rank every individual MLP neuron and attention head jointly, on a common scale, by minimizing the model's loss under soft pruning across many sparsity targets, all at once. Across 27 vision and text transformer encoders ranging from 22M to 8B parameters, our method yields smooth degradation up to 60% sparsity and outperforms one-shot baselines under deep pruning. For example, pruning DINOv3 ViT-H+/16 to 50% sparsity removes 427M parameters while reducing ImageNet-1k linear-probing accuracy by only 0.8%, without the need for weight correction or retraining. At 60% sparsity, a pruned AugReg ViT-B/16 model achieves up to $1.80\times$ higher throughput and $1.26\times$ lower single-query latency on an A100 GPU. Elastic models are produced in minutes, with construction time scaling sub-linearly with parameter count.

Price of Anarchy of First-Price Auctions in Agentic Markets

An agentic market is a decentralized serving-time market where AI agents bid to claim each user prompt. An agent's bid is based on its expected value for the prompt, which is the expectation of the reward for its response minus its inference costs. First-price auctions (FPAs), in which the highest bidder pays its bid and receives the prompt, are a natural way to allocate prompts in such a market, since they are simple and transparent to run, and do not require central knowledge of private capabilities and costs. However, FPAs are not truthful: participants may bid less than their true value in the hope of a larger profit, and the prompt may then be allocated to someone other than the bidder who values it the most. This welfare loss is quantified by the $\textit{price-of-anarchy}$ (PoA). In this paper, we argue that agentic markets are likely to possess features that reduce the price-of-anarchy. The first is a $\textit{competitive floor}$, which occurs when multiple bidders usually value the prompt above some baseline share of the optimal welfare (e.g., because the best models are often distilled into cheaper alternatives, which perform similarly). We show that competition among them then limits the welfare loss, which can be driven to zero as the floor rises. The second feature is $\textit{near-independence}$ of values after conditioning on the prompt, which occurs when most of the predictable variation in the expected value is determined by the difficulty of the prompt. As the value distribution approaches conditional independence, we can derive an efficiency guarantee that approaches a factor of $1 - 1/e^2$ of the optimal welfare. Thus, both features give rise to welfare lower bounds that are significantly better than the well-known general case of $1 - 1/e$. We validate these results empirically using auctions with strategic bidders simulated on the EmbedLLM prompt routing dataset.

Continuum Attention

We introduce Continuum Attention (CA), a causal Transformer attention operator that replaces discrete key and value lookup with integration over a continuous, piecewise linear memory curve. Each query induces an exponential relevance density along the curve. Integrating the density and interpolated values yields exact normalized attention from closed form exponential moments, including half token boundary contributions. The gradients follow analytically from moments of order at most two, enabling parallel GPU implementations while retaining quadratic sequence complexity and the standard Transformer decoder. Compared with ordinary softmax and focal attention, CA reduces mean validation negative log likelihood by 0.0200 and 0.0480 nats per byte, respectively, winning all paired comparisons. An additional disjoint validation slice yields a 0.0457 nats per byte advantage over the strongest selected softmax baseline, with a paired 95% confidence interval. These findings motivate continuous integration as a promising attention alternative while leaving larger scale generalization open.

Player-1: Learning to Play from Synthetic Experience

Computer-use agents have demonstrated strong capabilities in interpreting screens and executing keyboard and mouse actions. However, games, with their visually rich worlds and diverse mechanics, remain challenging and offer a demanding next step toward general digital control. Current game-agent learning pipelines often rely on large-scale demonstrations with limited coverage of dynamic game states, while scalable environments for interactive training remain difficult to obtain. We introduce Player-1, a post-training framework for learning game control beyond demonstrations through synthetic experience. We extend the computer-use action space with relative pointer motion, sustained inputs, and composite actions. For supervised initialization, we fine-tune computer-use models on 6,044 curated gameplay trajectories that pair recorded actions with task instructions and visual reasoning. We construct Player-1-Gym, sixteen synthetic 2D and 3D environments designed around control demands shared across games. Programmatically verifiable rewards and randomized observations and control dynamics support multi-environment reinforcement learning with PPO. The resulting Player-1-9B and Player-1-35B-A3B improve MineDojo-300 success over their SFT policies by 27.8 and 17.7 percentage points, respectively. Player-1-9B reaches 38.33% success compared with 12.63% for the strongest evaluated baseline, with further gains over SFT on ViZDoom-10 and GameWorld. Further analyses show that broader synthetic environment coverage improves transfer, while the use of composite actions that combine multiple keyboard and mouse inputs increases during RL. We demonstrate that learning from synthetic experience substantially improves game control beyond supervised initialization, with gains that generalize beyond the RL training suite.

Latent Compaction with Test-Time Training

Long contexts enlarge key--value (KV) caches, increasing memory use and attention costs during generation. Compressing these caches must preserve information for unseen questions. We introduce **Latent Compaction**, which learns a compact document-specific cache through test-time training. We initialize cache entries using reconstruction importance, teacher attention, and key-space diversity. Training first fits attention outputs and log-normalizers to those of the full cache, with reconstruction-selected entries held fixed. It then jointly optimizes the entire cache through end-to-end distillation from the full-cache model. Model weights remain frozen throughout, and training uses only generic prompts about the document, without access to downstream questions or answers. Experiments on RULER, LongBench, and Phonebook Lookup demonstrate substantial gains over budget-matched compression baselines. On a RULER, for example, our method achieves 96.7% accuracy at 32$\times$ compression, compared with 40.0% for KVZip+. The method also transfers across model scales and families, with retrieval gains on both Qwen3 and MiniCPM5. These results show that treating the KV cache as a neural network and optimizing it with test-time training can preserve retrieval accuracy under aggressive compression. The code is released at https://www.github.com/THUNLP/latent-compaction

One Task, Many Interfaces: Simple Cross-Interface Consistency Training Improves LLM Agent Robustness

Post-training binds the competence of an LLM agent to the interface it was trained on. Semantics-preserving perturbations of that interface, such as renaming an action or changing the notation, leave the underlying task unchanged yet make the agent worse at it. Scoring each sample independently means interface randomization only broadens this dependence from a single interface to the training interface distribution, so out-of-distribution perturbations still degrade the agent. We propose Cross-Interface Policy Regularization (CIPR), a self-supervised objective that replaces action names with uninformative identifiers and presents each training sample under two interfaces in which equivalent actions are identified by construction. CIPR penalizes the KL divergence between the action distributions induced under the two interfaces, so the policy behaves consistently across them. We evaluate CIPR in both supervised and reinforcement-learning post-training, where it improves robustness. On agentic function calling, held-out interface perturbations reduce the accuracy of a post-trained agent by nearly 25%, and CIPR recovers 10.5% while leaving accuracy on the unchanged interface intact.

Beyond LLM Judge: Mechanical Verification for Agentic Autoformalization

We introduce a novel method for verifying Lean formalizations that goes beyond LLM-judged faithfulness. Unlike most verification methods, which rely on LLMs or trained models, our method combines LLM-based semantic review with mechanical numerical and vacuity checks, providing checkable feedback and detecting errors missed by LLM judges. We apply this method, along with our Extract-Verify-Proof autoformalization pipeline, to construct the Formalized Library of Mathematical Functions (FLMF), a partial formalization of the NIST Digital Library of Mathematical Functions (DLMF). On three labeled datasets, our method achieves the lowest accepted-error rate among the evaluated methods, measured as the fraction of incorrect statements accepted by the verification method. Expert review of the results also identifies translation and source defects in existing datasets. Code available at https://github.com/annamei1/FLMF-Ch7ErrFunc-Release.git.

Simple math can predict when AI is about to go rogue, physicists say

Researchers say they've developed a simple mathematical formula that can predict when certain ChatGPT-like artificial intelligence (AI) chatbots will "go rogue," suddenly switching from helpful responses to potentially harmful ones.In a new

Using AI to reverse engineer hardware

After my mother passed away I ended up with a big box full of old film negatives containing photos from childhood family trips and similar. So I decided to buy a negative photo scanner and found that the Plustek 8200i was supported under Li

Build an Automatic AI Accountant with Muse, Unsiloed, and Jev

In the age of AI, a small business shouldn't need an accountant to keep its books. This guide builds a bookkeeper that turns a month of receipts, bills, payouts and bank statements into a reconciled ledger and the month-end pack an accounta

AI engineering without getting out of touch

You know using Claude Code can make you lose track of your architecture? It can be disconcerting to need to work without AI. Here’s a trick I developed to address this problem. Put this in your context: I want to understand the architecture

Ask HN: How do you avoid building verbose AI slop designs?

Hacker Newsnew | past | comments | ask | show | jobs | submitloginAsk HN: How do you avoid building verbose AI slop designs?1 point by lasky 30 minutes ago | hide | past | favorite | discuss help Consider applying for YC's Winter 2027 batch

Things we wish we'd known before building an AI chat agent

We built a chat agent to help our users in our developer dashboard (debug, make charts, get answers from docs etc.). Here are seven things we learned along the way. We're open source, so we've linked to the actual code throughout. 1. Make t

Coding with AI stopped feeling like craft, so I built a tiny desk radio

Choose DIY, and wondering who the fuck you are on a Sunday morning AI changed coding and developer careers forever. And it's causing a lot of grief, uncertainty and excitement in the field, all at the same time. People are losing their sens

I got $300K in startup credits for my AI company at 18: the playbook

I got $300K in startup credits for my AI company at 18: the playbook

How to Convert Pipedream Workflows to FlowRunner, with Real Examples

Pipedream Workflows shuts down on March 31, 2027, and everything in it, including the credentials for your connected accounts, is deleted by April 30. This guide is how to move your workflows to FlowRunner without rebuilding each one by han