Daily Tech News - 2026-09-14

AI’s next bottleneck is becoming clear: making intelligence affordable, reliable, and secure beyond the demo.

AI infrastructure shifts into high gear

September 14’s coverage centers on the economics of running advanced models. NVIDIA headlines the discussion with a claimed 67-fold improvement in agentic inference performance per dollar for Vera Rubin NVL72, while technical articles explore complementary gains through FlashAttention, INT8 quantization, and accelerated mixture-of-experts training in JAX. One quantization example cuts a Llama 3.1 8B model from 14.9 GB to 8.0 GB, illustrating how efficiency work can reduce deployment demands.

The infrastructure debate also extends to location: as AI moves into robotics, choosing between on-device and datacenter inference becomes a consequential architectural decision.

Production puts agents and platforms to the test

Articles on Kubernetes inference deployment and interrupted streaming highlight the operational challenges behind scaling AI services. Coverage of Kubernetes v1.37 reports Memory QoS reaching beta and becoming enabled by default, bringing memory management into sharper focus.

Meanwhile, comparisons of LlamaIndex and LangGraph, alongside LangChain’s paid media agent, show growing attention to coordinating useful, multistep automation. A Meta Muse stress test reporting control-plane timeouts underscores the reliability questions that accompany those ambitions. Developer performance remains another theme, with proposed Linux changes promising substantially faster kernel builds.

Security exposes the cost of complexity

Reports of mass scanning against exposed Vite development servers put cloud credentials in the spotlight. Separate Android research describes routes from unprivileged apps to root access or device shutdown, while cloud identity analysis tackles an expanding population of human, machine, and agent accounts.

Together, the day’s stories point to a practical priority: turning impressive AI capabilities into systems that organizations can operate dependably—and defend.

Featured Articles

Deploying LLM Inference at Scale on Kubernetes

In today’s rapidly evolving technological landscape, large language models (LLMs) are transforming the way businesses leverage artificial intelligence. From automated customer support and sentiment an...

• Keywords: scalability kubernetes, efficiently kubernetes, kubernetes crucial, enhancing kubernetes, kubernetes ai, deployments kubernetes, kubernetes powerful, google kubernetes, containers kubernetes, kubernetes deployments
• Source: collabnix.com

Understanding FlashAttention Pt 1: Personal Notes

Understanding FlashAttention Pt 1: Personal Notes 0. Introduction How IO-Aware Attention Makes Transformers Faster Without Approximating Attention The mechanism, in three words: Tiling + Online Softma...

• Keywords: flashattention performance, performance flashattention, flashattention faster, memory flashattention, flashattention attention, sparse flashattention, flashattention algorithm, io optimized, io complexity, flashattention analysis
• Source: chizkidd.github.io

Vera Rubin NVL72 Agentic Inference: 67x better Performance per Dollar

Rubin is the first platform co-designed across six products for the agentic era: Rubin GPU, Vera CPU, NVLink 6 Switch, ConnectX-9, BlueField-4, and Spectrum-6. Today we are publishing the first verifi...

• Keywords: benchmark agentx, agentic benchmark, agentic workloads, agentic workload, performance agentic, optimizing rubin, rubin performance, hoppers performance, agentx early, agentx scenario
• Source: newsletter.semianalysis.com

Accelerating Dropless MoE Training in JAX with NVIDIA Transformer Engine

Mixture of experts (MoE) has become one of the defining architectural trends in large-scale AI model training. DeepSeek, Qwen, and Mixtral are examples of MoE models that match or exceed the performan...

• Keywords: gpus deepseek, training deepseek, expert gpu, gpu_multiprocess training, gpus expert, training throughput, challenging deepseek, gpu kernels, training nvidia, gpu performance
• Source: developer.nvidia.com

Migrating to 0.5.0

Migrating to 0.5.0 Ordered upgrade paths to 0.5.0 from 0.4.x and its release candidates. This guide targets stable 0.5.0 . Choose the path matching your installed version. Update vgpu and any directly...

• Keywords: migration vgpu, update vgpu, version vgpu, obsolete vgpu, replace vgpu, existing webgpu, vgpu install, vgpu packages, vgpu native, create vgpu
• Source: vgpu.sh

A Brain Too Big to Carry — On-Device vs Datacenter Inference

Where should the brain of the robot go? So far, AI has mostly lived behind a screen. Chatbots answered questions. Then agents started driving software and finishing multi-step tasks on their own. The...

• Keywords: robot brains, robotics mind, robots contending, billion robots, robot dependent, brain robot, robot hardware, robots early, robots primarily, robot state
• Source: newsletter.semianalysis.com

Hackers Mass-Scan Exposed Vite Servers to Steal AWS and Azure Cloud Credentials

Hackers are conducting a large-scale automated scanning campaign against internet-exposed Vite development servers, attempting to steal AWS credentials, Azure access tokens, environment variables, and...

• Keywords: vite vulnerability, vite access, vite server, exploited cve, vite provides, exposed vite, vite development, cloud security, vite file, exploited vulnerabilities
• Source: cybersecuritynews.com

Understanding W8A8 INT8 LLM quantization: Accuracy and performance results

In Understanding W8A8 INT8 LLM quantization: Half the size, better performance, same accuracy, we compressed a Llama 3.1 8B Instruct model from 14.9 GB to 8.0 GB using 8-bit integer (INT8) W8A8 quanti...

• Keywords: w8a8 quantization, w8a8 quantizes, w8a16 quantizes, int8 quantization, w8a16 quantized, w4a16 quantization, performance quantization, quantization overhead, fp8 quantization, accuracy quantization
• Source: developers.redhat.com

SDR–; open source SDR with a patchable signal graph, Rust DSP, web UI

sdr-- is a software-defined radio application with a visual signal path. Connect devices, decoders, displays, and recorders on a canvas, then pin the controls you use to a rack. A Rust server handles...

• Keywords: sdr software, sdr sdr, sdr, try sdr, rtl sdr, images sdr, sdr active, sdr licensed, decoders use, decoders tested
• Source: github.com

Playwright vs Selenium vs Puppeteer: Which To Use?

Playwright, Selenium, and Puppeteer are the three most popular browser automation tools. They can be used for end-to-end testing, web scraping, screenshot generation, and increasingly to enable AI age...

• Keywords: automation browser, automations selenium, browser automation, selenium uses, selenium puppeteer, uses webdriver, selenium use, webdriver uses, webdriver intended, chrome automations
• Source: browser-use.com

An Educational GEMM Ladder for Helios GPUs

An Educational GEMM Ladder for Helios GPUs# AMD Helios will be an important platform for AI. Helios offers 432 GB of HBM4, 23 TB/s of HBM bandwidth per GPU, and 40 PFLOPs of FP4 compute AMD Instinct™...

• Keywords: helios gpu, gpu capabilities, gpu generations, gpus including, optimize cuda, helios gpus, amd gpus, kernel implementations, narrows gpu, amd kernels
• Source: rocm.blogs.amd.com

Google is a leader in The Forrester Wave™: Public Cloud Platforms, Q3 2026

Google is a leader in The Forrester Wave™: Public Cloud Platforms, Q3 2026 Brad Calder President, Google Cloud Platform and SRE A new report from Forrester Research names Google Cloud a Leader, receiv...

• Keywords: google cloud, cloud leader, cloud platform, cloud platforms, enterprise google, ga cloud, cloud highest, cloud providers, cloud, google leader
• Source: cloud.google.com

Intelligence is yours. Let's keep it that way.

CEO, DigitalOcean Intelligence was yours. A founder’s intellectual property, business logic, differentiation, their intelligence, was once theirs alone. You built it, you owned it, and no tools/platfo...

• Keywords: owns intelligence, software kubernetes, kubernetes postgresql, kubernetes, intelligence manufactured, intelligence provider, linux kubernetes, operating infrastructure, proprietary cloud, intelligence theirs
• Source: digitalocean.com

When Kubernetes Scales Down Mid-Stream: Why Graceful Shutdown Is Not Enough

-35e7a49c6df5---4 crawled_date: 2026-09-14T05:57:20.632460+00:00 feed_url: https://aws.plainenglish.io/feed published: Mon, 14 Sep 2026 05:24:28 GMT

When Kubernetes Scales Down Mid-Stream: Why Gr...

• Keywords: stream kubernetes, pod termination, terminates pod, ran kubernetes, kubernetes, kubernetes scales, kubernetes services, halfway kubernetes, pod handling, kubernetes istio
• Source: aws.plainenglish.io

LlamaIndex vs. LangGraph: Agentic orchestration (2026)

LlamaIndex vs. LangGraph: Agentic orchestration (2026) Table of contents

• Same problem, different models — Both frameworks coordinate multistep, stateful agents. LlamaIndex Workflows routes typed eve...

• Keywords: llamaindex workflows, workflows langgraph, langgraph workflow, context workflows, orchestration retrieval, execution langgraph, workflows use, durable workflows, langgraph orchestration, workflows emphasizes

• Source: nutrient.io

Kubernetes v1.37: Memory QoS Graduates to Beta

Kubernetes v1.37: Memory QoS Graduates to Beta Memory QoS has graduated to Beta in Kubernetes v1.37 and is now enabled by default. On Linux nodes running cgroup v2, the feature uses the memory control...

• Keywords: kubeletconfiguration memorythrottlingfactor, kubeletconfiguration memoryreservationpolicy, memorythrottlingfactor kubelet, kubernetes cgroups, beta kubernetes, kubernetes v1, qos memory, kubeletconfiguration featuregates, memory qos, bugs kubernetes
• Source: kubernetes.io

Unmasking Cloud Identities: From Behavioral Clustering to Automated Detection

Executive Summary As cloud environments expand to include human, machine and autonomous agent identities, mapping the functional roles of these identities has become a significant security challenge....

• Keywords: cloud audit, cloud security, cloud threat, cloud identity, cloud identities, behavior attackers, detect security, secure cloud, protect cloud, automated threat
• Source: unit42.paloaltonetworks.com

OEMpocalypse: Unprivileged Android app to root on Samsung, Xiaomi, others

OEMpocalypse Now: A Generic Exploitation Strategy from Android untrusted app to root Part 1 of a series that takes an unprivileged Android app to root on Samsung, Xiaomi, and Oppo/OnePlus/Realme devic...

• Keywords: untrusted_app oem, oempocalypse strategy, weaponized untrusted_app, untrusted app, untrusted_app surface, variants untrusted_app_32, surface untrusted_app, android untrusted, code oempocalypse, directly untrusted_app
• Source: calif.io

I stress-tested Meta Muse until its agent control plane started timing out

meta-muse-black-box-testing I stress-tested Meta Muse until its agent control plane started timing out Four spawn experiments against a black-box multi-agent runtime, six database lock timeouts recove...

• Keywords: spawn failures, burst failures, contention postgresql, spawn attempts, timeouts recovered, lock timeouts, failure latencies, concurrency burst, spawn failure, test burst
• Source: blog.cygankiewicz.com

Principles for Fast Tokio Applications

I'm on my way back from RustConf. At the Unconf, we had a productive discussion about debugging and benchmarking async applications. Many interesting insights were shared. I'm attempting to enumerate...

• Keywords: tokio runtimes, benchmarking async, tokio runtime, runtime heavily, debugging benchmarking, async applications, runtimes, runtime needs, core runtime, running runtime
• Source: dial9-rs.github.io

Any Android App Can Shut Down the Phone - We Found It in Three Hours Without Source Code

Any Android App Can Shut Down the Phone - We Found It in Three Hours Without Source Code Any Android app on a Pixel can turn the phone off. Not an app with root. Not an app that conned you through a p...

• Keywords: app shut, shut phone, code android, android, root android, built exploit, android turning, kernel bug, phone runs, android gpu
• Source: xint.io

How We Built LangChain’s Paid Media Agent

Key Takeaways For LangChain’s first three years, our sales pipeline grew largely organically, driven by open source, content, YouTube, community, and meetups. In January, we wanted to kickstart our pa...

• Keywords: monitoring campaign, evaluating campaigns, advertising platform, improving campaign, overcome advertising, channels campaign, marketing pipeline, langchain campaign, optimize agent, advertising
• Source: langchain.com

Linux 7.4 Could End Up Seeing Kernel Builds ~36% Faster, Incremental Builds ~70% Faster

Linux 7.4 Could End Up Seeing Kernel Builds ~36% Faster, Incremental Builds ~70% Faster Earlier this month I wrote about a patch series posted to the Linux kernel mailing list that addressed a lot of...

• Keywords: linux patches, merged patches, patches merged, kernel builds, faster kernel, merged kernel, rust compiler, builds faster, revision patches, kernel build
• Source: phoronix.com

Quarkus QuickJS4j: Approved Carrier Webhook Transformers

Someone brought me a problem that starts out looking like a mapper. Their service receives signed webhooks from several parcel carriers. Every carrier reports the same operational event, but each one...

• Keywords: webhook payload, webhooksignatureverifier java, java webhooksignatureverifier, carrier webhooks, security webhooksignatureverifier, webhooksignatureverifier, mapping carrier, carrier webhook, webhooksignatureverifier signatureverifier, final webhooksignatureverifier
• Source: the-main-thread.com

Why don't machine learning research agents overfit?

Machine learning, at its core, is about generalization, not memorization. You hand your learning algorithm a pile of training examples and use them to fit a model. But the goal is not to perform well...

• Keywords: memorization holdout, memorized validation, learned validation, memorize training, catching overfitting, strategies overfitting, trained poorly, memorization, revise training, training examples
• Source: amazon.science

Hackers Leverage Claude to Exfiltrate Secrets from 1.8M Android apps

Cybercriminals linked to the ShinyHunters ecosystem used Claude to support a large-scale credential theft operation that downloaded, decompiled, and scanned 1.8 million Android applications for hardco...

• Keywords: android secrets, android security, steal android, hackers exploit, applications secrets, resources hackers, hackers, stolen credentials, enterprise secrets, apk scanning
• Source: cybersecuritynews.com

Kubernetes Changed Block Tracking API - Beta Differences

Kubernetes Changed Block Tracking API - Beta Differences Changed Block Tracking (CBT) support for CSI drivers shipped as Alpha in September 2025. With the March 2026 v1.0.0 release of the external-sna...

• Keywords: kubernetes changed, storage kubernetes, tracking cbt, kubernetes version, storage k8s, cbt storage, tracking storage, kubernetes alpha, volume snapshots, storage csi
• Source: kubernetes.io

Notes on gotchas while migrating 35kb preprompts from Opus to self-hosted Ollama

Motivations: Maybe you’re a Claude code/codex user diligently avoiding uploading personal data to LLM providers. Is it possible that the most valuable information isn’t your data- but the metadata abo...

• Keywords: agent sessions, inference providers, agents need, agents self, agents read, hosted ai, valuable information, agents lots, protect ideas, insight agent
• Source: patrickmccanna.net

Singeli: High-level interface for low-level programming

Introductions: Singeli as interpreter | Singeli as compiler -> Purity and Ford write a min filter Singeli is a domain-specific language for building high-performance algorithms (including SIMD) with f...

• Keywords: singeli compiler, singeli interpreter, singeli compilation, interpreter singeli, singeli compiled, singeli programs, simd programming, singeli command, singeli executable, singeli code
• Source: github.com

Slow developer experience will bottleneck fast models

Slow developer experience will bottleneck fast models Right now developer experience is measured in seconds. If your tests take a second to run, that’s good; if they take thirty seconds, that’s bad. A...

• Keywords: faster models, optimize dev, slow developer, experience bottleneck, faster smart, fast compilers, increasingly faster, fast models, getting faster, bottleneck fast
• Source: seangoedecke.com

Untitled

Building agents is the easy part. Even building a cross-functional AI governance committee to define policies for deploying these agents is within reach for most organizations. What many organizations...

• Keywords: ai governance, agent trust, ai agents, ai agent, agent ai, agents engineering, agentic trust, building agents, agent trustworthy, agentic ai
• Source: vijil.ai

Introducing MTLX — MaterialX Tools for the Web

Introducing MTLX — MaterialX Tools for the Web MTLX is a pure TypeScript/JavaScript MaterialX toolkit — a library, CLI, web viewer, and VS Code extension — for parsing, validating, packaging, and tran...

• Keywords: mtlx materialx, mtlx materials, mtlx material, materials mtlx, materialx toolkit, material mtlx, materialx tools, mtlx textures, introducing mtlx, textures mtlx
• Source: ben3d.ca

Show HN: Fly.exe – An EON systems like virtual fruit fly uploaded to computer

A Drosophila brain simulator that runs the whole Traced universe of the released male CNS connectome — 165,122 neurons and every one of the 25,563,197 edges between them — inside a physical fly body,...

• Keywords: neuromechfly bodies, drosophila brain, neuromechfly, claimed neuromechfly, neuromechfly 133, mujoco neuromechfly, drosophila, complete drosophila, drosophila male, flyconnectome 2025malecns
• Source: github.com

A Gentle Introduction to Model Distillation

In this article, you will learn what model distillation is, how it has evolved for large language models, and why it has become one of the most contested topics in the AI industry. Topics we will cove...

• Keywords: ai distillation, distillation intelligence, models distillation, distilling models, model distillation, data distillation, distilling knowledge, distillation framework, distilled models, distillation entirely
• Source: machinelearningmastery.com

Build anywhere, stay in flow: How Windows 365 is redefining the developer experience

Development teams are under increasing pressure to ship software faster, even as the environments they rely on grow more complex. However, modern development has only become harder to provision and ma...

• Keywords: cloud development, developer cloud, windows 365, 365 developers, 365 microsoft, 365 cloud, virtualized developer, 365 developer, microsoft cloud, 365 windows
• Source: blogs.windows.com

Foundation Model Engineering: From Theory to Production

Foundation Model Engineering is a technical textbook for readers who want to understand how modern foundation models actually work, why the stack evolved the way it did, and what engineering trade-off...

• Keywords: model engineering, model architectures, model landscape, foundation models, agents engineering, modeling ideas, architectures, engineering narrative, architecture systems, foundation model
• Source: sungeuns.github.io

Pkgsrc Is Cool (2022)

I think I saw something about pkgsrc pop up randomly in my Mastodon feed a few months ago. Before that, I had never heard of it. And now…well now I’m currently writing these words in a version of Emac...

• Keywords: pkgsrc packages, pkgsrc supports, pkgsrc netbsd, supported pkgsrc, pkgsrc use, pkgsrc debian, pkgsrc compiler, projects pkgsrc, pkgsrc isn, netbsd portability
• Source: wisellama.rocks

Announcing On-Demand State Repartitioning for Apache Spark™ Structured Streaming on Databricks

Right-size your most demanding stateful streams, from fraud detection to real-time surveillance, without ever rebuilding your checkpoint state by Thangam Vaiyapuri, Jay Palaniappan, B. Micheal Okutubo...

• Keywords: queries spark, apache spark, partitions stateful, spark structured, changes spark, spark sql, shuffle partitions, configuration spark, stateful streams, spark declarative
• Source: databricks.com

GitHub Pays $100,000 Bounty for Critical RCE Flaw in Git Push Pipeline

GitHub has awarded security researcher Saif Ghani a $100,000 bug bounty after the disclosure of CVE-2026-3854, a critical remote code execution vulnerability affecting GitHub’s Git push processing pip...

• Keywords: github vulnerability, github security, bounty git, github pays, repository threat, github reportedly, malicious repository, affecting github, github git, payment github
• Source: cybersecuritynews.com

Hacking AI customer service agents

Hacking AI customer service agents By Ayoub and Inti De Ceukelaire September 2, 2026 Table of contents

• Weaponizing chatbots via email

• Bypassing multi-factor authentication (2-FA/MFA) in AI agents...

• Keywords: mfa ai, ai agent, reply agent, accounts ai, ai customer, agents bypassing, ai agents, instructs agent, agents threats, agent instruct

• Source: intigriti.com

How can you not be romantic about UNIX domain sockets?

How can you not be romantic about UNIX domain sockets? Earlier this summer, I gave a technical iOS talk at the DEFCON 34 convention (search up “Rage Against the Sandbox”) and my demo crashed on stage...

• Keywords: crash deterministic, crash related, crash occurs, crash verify, occurs ios, initial crash, deeper crash, crash issue, slave crash, crash
• Source: yuvalino.com

[email protected]

2c373ef: Stop repeating API in generated descriptions when the name already says it. API reference overview and operation meta descriptions no longer turn a spec titled Example API (or Payments APIs,...

• Keywords: api generated, api v2, api reference, says api, api api, apis example, builds astro, astro documented, apis, astrojs cloudflare
• Source: useblume.dev

Implementing ETag-based Caching Revalidation for TanStack Start

Implementing ETag-based Caching Revalidation for TanStack Start Serve your blog faster with fewer server resources by using HTTP caching and ETags to avoid repeated rendering and unnecessary downloads...

• Keywords: caching etags, etag cache, http caching, enables caching, caching enabled, based caching, server caching, caching, caching browsers, cache needs
• Source: ben3d.ca

Announcing Pause/Resume and NVIDIA RTX PRO 6000 Blackwell GPU support in Dataflow

Announcing Pause/Resume and NVIDIA RTX PRO 6000 Blackwell GPU support in Dataflow Efesa Origbo Product Manager, Google Cloud Danny McCormick Software Engineer, Google Cloud Overview As enterprises sca...

• Keywords: gpus dataflow, resume dataflow, dataflow today, pause dataflow, enhancements dataflow, efficient dataflow, dataflow critical, support dataflow, dataflow capabilities, enables dataflow
• Source: cloud.google.com

Build Custom Salesforce Connect Adapters Smarter with Salesforce Skills

Salesforce Connect brings external data into Salesforce without copying it. Live API calls replace ETL and sync jobs, and the data stays where it is. For sources reachable through a built-in adapter,...

• Keywords: connect apex, salesforce connect, connect salesforce, data integration, datasource connection, connection classes, connection data, api developer, uses datasource, apex adapters
• Source: developer.salesforce.com

Find code faster: Introducing our new & improved search experience

Finding code across your repositories in Bitbucket just got a major upgrade. We’ve rolled out code search in open beta for Bitbucket Cloud: a faster, more integrated search experience built to help yo...

• Keywords: search bitbucket, code search, repository search, repositories bitbucket, code repositories, bitbucket code, developers search, exploring codebase, finding repositories, codebase investigate
• Source: atlassian.com

Rightsholders Can’t Use OpenAI and Anthropic to Dismantle Meta’s Seeding Defense

Over the past two years, rightsholders of all kinds have filed lawsuits against companies that develop AI models. Meta is among a long list of companies now being sued for this allegedly infringing ac...

• Keywords: torrenting claims, meta torrenting, meta torrent, meta torrented, torrenting ai, stated bittorrent, examining bittorrent, lawsuits, torrenting, bittorrent efficient
• Source: torrentfreak.com

The case against JPEG XL

Investigating JPEG XL's place as a Web image codec. Why? JPEG XL is a technically impressive image codec; it is a definitive upgrade over JPEG, more versatile than WebP, and well-equipped to serve use...

• Keywords: support jpeg, endorsed jpeg, jpeg xl, reasons jpeg, recompressed jpegs, codec jpeg, jpeg versatile, jpegs jxl, proponent jpeg, opinion jpeg
• Source: giannirosato.com

An OpenAI capabilities researcher publicly warns about AI risk, widening the internal-policy debate around frontier development

Dan Selsam is a current OpenAI capabilities researcher. (since 2022) He was my boss for a while. He doesn't have a twitter account but has made this public statement of his views on AI risk and sent i...

• Keywords: ai risk, ai researchers, existing ai, ai years, biases ai, openai capabilities, ai research, risks future, described ai, come ai
• Source: x.com

Perplexity trusts GPT-6 Astra with end-to-end systems

Perplexity trusts GPT‑6 Astra with end-to-end systems Perplexity uses Astra to write communications, change software, and monitor production systems, and checks in much less frequently than with earli...

• Keywords: engine perplexity, perplexity uses, perplexity search, systems perplexity, ai testing, code perplexity, gpt astra, answer engine, applications ai, end systems
• Source: openai.com

SoC Switch Supports 260 PCI Express Lanes

SoC Switch Supports 260 PCI Express Lanes There are three things you can never have enough of: compute cycles, bandwidth, and memory. Marvell's Structera S targets bandwidth linking compute engines, m...

• Keywords: 260 pci, lanes pci, pci express, bandwidth memory, pci, soc switch, pcie networks, memory chip, 260 lanes, bandwidth tb
• Source: electronicdesign.com

SwiftUI Agent Skill: Install and use with AI coding tools

A SwiftUI Agent Skill that helps you build better views or refactor existing ones. It’s the reality we’re in today, and I honestly can’t live without it anymore myself. Several skills helped me improv...

• Keywords: swiftui agent, swiftui expertise, swiftui expert, skills swiftui, swiftui knowledge, skill swiftui, detailed swiftui, better swiftui, guidance swiftui, understand swiftui
• Source: avanderlee.com

14th September – Threat Intelligence Report

For the latest discoveries in cyber research for the week of 14th Setpember, please download our Threat Intelligence Bulletin. TOP ATTACKS AND BREACHES

• IDScan.net, a US identity verification provide...

• Keywords: data breach, breaches idscan, stolen data, threat metabase, fraudulent information, compromised accounts, access compromised, attacks breaches, database vulnerabilities, use compromised

• Source: research.checkpoint.com

Generate images with Pydantic AI

The news upfront: you can now use dedicated image-generation models through Pydantic AI! Previously, our image-generation support went through conversational models with image-generation abilities. Th...

• Keywords: imagegenerator openai, imagegenerator pydantic_ai, instrument_pydantic_ai images, image agent, pydantic_ai capabilities, agent openai, agent imagegenerator, agent image, imagegeneration pydantic_ai, binaryimage agent
• Source: pydantic.dev

How I use a single .zshrc file on macOS and Windows (WSL2)

How I use a single .zshrc file on macOS and Windows (WSL2) By Tal Koren · Ever since The Matrix came out in 1999, I've been into command line interfaces. The feeling of knowing how to navigate your wa...

• Keywords: zshrc wsl, zshrc using, zsh wsl, mac zshrc, zshrc mac, wsl macos, dotfiles zshrc, macos wsl, zshrc behaves, separate zshrc
• Source: talkoren.com

Introducing Agentic Batch Changes: the frontier agent for code change at scale

Introducing Agentic Batch Changes: the frontier agent for code change at scale Codebase-wide changes across hundreds or thousands of repositories can now be run by one engineer. Codebase-wide changes...

• Keywords: batch changes, repository agentic, codebases agentic, agent repository, agentic batch, changes codebases, batch change, large migration, changes hundreds, thousands repositories
• Source: sourcegraph.com

RubyGems Open Source Supply Chain Security and OpenAI

RubyGems Open Source Supply Chain Security and OpenAI Over the weekend it has been widely reported that OpenAI agents attacked RubyGems on May 11, 2026, two months before Hugging Face, including by ma...

• Keywords: openai agents, attacked rubygems, vulnerability rubygems, attack rubygems, security openai, rubygems security, involved rubygems, steal rubygems, reported openai, rubygems open
• Source: rietta.com

VEX-Bench tests whether LLM agents can assess exploitability in software-supply-chain vulnerabilities

🎉 Excited to share that our paper "VEX-Bench: Benchmarking LLM Agents for Assessing Exploitability of Software Supply Chain Vulnerabilities” has been accepted to EMNLP 2026! This work is a collaborati...

• Keywords: assessing exploitability, exploitability software, exploitability reasoning, reasoning exploitability, exploitability results, exploitability, grained exploitability, vulnerability assessment, determining vulnerability, understand exploitable
• Source: x.com

Why 4hi HBM May Not Be the One-Size-Fits-All Fix

4hi HBM samples started showing up on suppliers’ radar right at the start of 3Q26. AI chip makers, NVIDIA included, are evaluating every possible configuration, including lower stack counts, to work a...

• Keywords: hbm edge, compared hbm4e, hbm4e capacity, 4hi hbm, hbm shortage, hbm usage, hbm5 4hi, hbm5 based, hbm4e, based hbm5
• Source: insights.trendforce.com

Casey Muratori: Surprises In Computer History And Where Bad Code Comes From

In this episode, my goal was to record a conversation that was completely free of any “AI doom” content. There’s so much of it out there, I hope that this can be a nice timeline cleanser. I had Casey...

• Keywords: computer historian, talk historian, groundbreaking computer, famous computer, programming casey, technology wrote, research casey, casey muratori, ai doom, ai era
• Source: developing.dev

Sticking Functions Where They Donʼt Belong

Extensible Defunctionalization with Typeclasses. Or, Functions in Pure Data in Haskell. 2026/06/04 What does pure data mean? and how the heck do we stick functions in there? For me, pure data means a...

• Keywords: nfdata functions, data haskell, nfdata normalize, defunctionalization typeclasses, structure nfdata, anyclass nfdata, wrapped nfdata, typeclasses functions, closure data, extensible defunctionalization
• Source: blog.veritates.love

We are all Product Engineers now

We are all Product Engineers now Just yesterday I published a very long post about the economics of open source. As part of that argument, I mentioned that the cost of writing software has collapsed,...

• Keywords: software industry, software market, expensive programmers, develop software, industry software, programmers scarce, making software, software cost, cost software, software development
• Source: seldo.com

Best Knowledge Engine Platforms in 2026

The best knowledge engine platforms in 2026 are Pinecone Nexus, Databricks Genie, Snowflake Cortex, Microsoft IQ, Palantir Foundry, and Glean. A knowledge engine takes data from many separate sources...

• Keywords: knowledge engine, knowledge engines, knowledge platforms, managed knowledge, knowledge production, knowledge provider, need curated, knowledge layer, best knowledge, curated task
• Source: pinecone.io

GPT-5.6 Luna vs. GPT-6 Astra: Is a $1.20 Model Good Enough for Code Review?

GPT-5.6 Luna vs GPT-6 Astra: Is a $1.20 Model Good Enough for Code Review? GPT-5.6 Luna costs$0.20 per million input tokens and $1.20 per million output tokens. GPT-6 Astra costs$10 and $50. On the...

• Keywords: luna cost, luna costs, gpt luna, astra cost, luna gpt, astra costs, gpt astra, astra gpt, cheaper luna, luna astra
• Source: entelligence.ai

Hostinger weekly product updates: Week 37, 2026

Hostinger weekly product updates: Week 37, 2026 Last week brought updates across Hostinger Agent, Hostinger Ecommerce, Reach, our AI automation apps, and website deployments. Here’s what shipped the w...

• Keywords: agent hostinger, hostinger agent, agent whatsapp, whatsapp agent, whatsapp hostinger, hostinger ai, ecommerce hostinger, app hostinger, hostinger ecommerce, integrations hostinger
• Source: hostinger.com

Chess.com Leak Exposes 7.3M Users, Evidence Points to Scraping

Free is a strange price for stolen data, and that’s exactly what makes this listing worth a second look. A 15.5 GB file containing over 7.3 million chess.com user records showed up on two data-leak fo...

• Keywords: ransomnews technical, ransomnews, ransomnew fields, stolen data, ransom, schema chess, handed ransomnews, published ransomnew, chess com, 000 chess
• Source: securityaffairs.com

Weekly Radar 001: The GPU is Not the Bottleneck

Why we are starting this TrendForce Substack covers semiconductors and AI infrastructure in depth. Depth takes time to read, and it does not always answer the question that surfaces first on a Monday...

• Keywords: ai supply, weekly radar, semiconductor weekly, week trendforce, weeks balanced, infrastructure semiconductor, roadmap following, chip roadmap, industry shifts, technical updates
• Source: insights.trendforce.com

Who Gets to Define the Rules for AI?

A perspective from Aidan Gomez, Co-founder & CEO of Cohere Artificial intelligence is remaking the world we live in. Within a generation, the way we discover medicine, manage power grids, and secure o...

• Keywords: ai companies, ai government, ai safer, make ai, concerned ai, ai technologies, ai capabilities, technology evolves, government ai, ai getting
• Source: cohere.com

Apple OS 27 Is Here. Iru Is Ready.

Apple's OS 27 releases are available now, bringing new management capabilities across iPhone, iPad, Mac, Apple TV, and Vision. At Iru, we're proud to deliver Day 0 support for all of these new release...

• Keywords: applecare enhanced, applecare support, diagnostics applecare, iru functionality, device management, device features, apple applecare, apple device, features device, applecare
• Source: iru.com

BuzzASR releases more than 100 language-adapted speech-recognition models with an EMNLP Findings paper

🐝New preprint🐝 If you want open source speech technology that's actually optimized for your target language, BuzzASR might be your best option! All models on Huggingface 🤗 Congrats to @realshivamsingh...

• Keywords: speech technology, monolingual speech, target language, speech recognition, languages buzzasr, languages achieving, models huggingface, language unique, language buzzasr, dozens languages
• Source: x.com

Claude is a Contrarian

Claude is a Contrarian Of all the LLMs I’ve used and driven daily, Claude is the one that absolutely contradicts my instructions. No, it’s not about the inconsistent routing, nerfing, or asking for a...

• Keywords: tell claude, opinion claude, claude generated, contrarian llms, claude thinks, claude behavior, does feature, use claude, spot claude, claude
• Source: medium.com

Genie visualizations in Slack replies (Public Preview)

September 2026 Databricks released these features and improvements in September 2026. Releases are staged. Your Databricks account might not be updated until a week or more after the initial release d...

• Keywords: tasks databricks, databricks partner, pipelines databricks, available databricks, data databricks, supported databricks, databricks runtime, databrickssubmitrunoperator supports, databricks hosted, details databricks
• Source: docs.databricks.com

How my e-reader lost its stripes

I’ve been a lifelong reader, which has proved a fragile habit in our era of endless trivial distraction. While books and e-ink readers solve the problem of competition for attention, they can’t compet...

• Keywords: ink readers, ink device, ink crosspoint, ink screen, books ink, readers cheap, text reader, tiny ink, ink, reader driver
• Source: serpentine.com

Show HN: Nari Qwen3-TTS and Qwen3-ASR – High accuracy, low latency and cost

TL;DR Nari Labs leads Coval’s voice AI benchmark by sitting on the quality-latency Pareto Frontier for both Text-to-Speech and Speech-to-Text. We also lead the latency-cost and quality-cost Pareto Fro...

• Keywords: speech benchmarks, voice ai, tts benchmark, speech tts, coval benchmarks, speech ai, text speech, speech text, ai benchmark, benchmark coval
• Source: narilabs.com

Every Python Method That Returns None (And Why It Breaks Servers)

my_list = my_list.sort() doesn’t sort your list. It sets it to None . data = [3, 1, 2] data = data.sort()print(data) # None ← the list is gone This is deliberate. In Python, a method that modifies an...

• Keywords: sorted my_list, sorted list, my_list sort, list sort, sort bug, returns sorted, sort list, sort doesn, sort returns, returns sort
• Source: copahost.com

Introducing Firebase spend caps

You probably know this feeling: You have a brilliant idea for a new AI-powered feature in your app, but that excitement is tempered by a nagging worry about getting a runaway bill. This is why we’re i...

• Keywords: cap firebase, firebase spend, caps firebase, firebase ai, firebase app, firebase services, firebase firebase, firebase, functions firebase, starting firebase
• Source: firebase.blog

MolmoSpaces evaluates vision-language-action models on zero-shot spatial tasks in real environments

Omar Rayyan on X: "GPT-Astra outperformed all open-source VLA/WAM baselines on a subset of our zero-shot MolmoSpaces-v1 benchmark! traces: https://t.co/8Le4Pni17H" All models used the same closed-loop...

• Keywords: shot molmospaces, leaderboard molmospaces, v1 benchmark, vlm apis, molmospaces allen, benchmark, molmospaces, benchmark api, benchmark traces, vlm
• Source: x.com

A study of sequence weighting at scale

TL;DR: We study the scaling laws of data weighting across in-house and open-weight LMs, finding non-monotonic behavior across scales. We vary the weight assigned to sequences during training and measu...

• Keywords: predictions scales, data weight, predictable scaling, data weighting, data weights, scales predictable, scale predictable, scaling research, model scales, sequence weighting
• Source: blog.janestreet.com

How Fyxer built an AI executive assistant people trust

How Fyxer built an AI executive assistant people trust Fyxer pairs OpenAI models with 500,000+ hours of EA workflows and real user feedback to draft replies in each person’s voice. 90% User retention...

• Keywords: fyxer ai, assistant workflows, ai assistant, task fyxer, ai executive, draft fyxer, executive workflows, fyxer manage, email tasks, fyxer approach
• Source: openai.com

How Grok Bot designers use AI agents to build personal sites and product prototypes | John Bai & Peng Zheng

John Bai and Peng Zheng are designers on the Grok Bot team at SpaceXAI, where they’re building one of the most talked-about AI products right now. John writes publicly about his design process (his pi...

• Keywords: bot templates, grok bot, bot template, bot grok, designers grok, uses devbot, bot setup, bot marketplace, materials bot, designing grok
• Source: lennysnewsletter.com

What a time to be alive – rouge AI agents attack RubyGems.org

What a time to be alive Sep 11, 2026 @ 5:02 pmToday Reuters and the Wall Street Journal both reported about rogue AI agents at OpenAI attacking RubyGems.org. https://www.rubyhack.ai/ has an amazing wr...

• Keywords: attacking rubygems, gems rubygems, gems scrape, gems api, gem rubygems, rubygems org, gems leverage, rubygems thought, rubygems honestly, data gems
• Source: tenderlovemaking.com

A working paper designs offline local-LLM infrastructure for community services and governance

My working paper is available on @ZENODO_ORG This talks about how we can leverage local community infrastructure for enabling local patrons and value providers using hardware, local LLM and governance...

• Keywords: community infrastructure, local community, infrastructure enabling, infrastructure, value providers, local patrons, providers, internet cloud, local llm, hardware local
• Source: x.com

If AI writes all the code, what’s left for engineers?

We are plausibly coming to a moment where AI will write all the code. This is a reality for many developers already. Over the last 4 months at PostHog, we moved from around 20% of our monorepo PRs bei...

• Keywords: engineers ai, increasingly engineers, improve engineers, engineers doing, engineers, enables engineers, thinks engineers, implement engineers, engineering engineers, engineer
• Source: newsletter.posthog.com

When LLM judges agree, should we believe them?

Imagine evaluating a retrieval-augmented-generation system. A user asks a question, the system retrieves a text passage, and an LLM judge decides whether it’s relevant. To reduce noise, you ask severa...

• Keywords: judge models, accuracy relevance, judge similarity, judge systems, evaluating retrieval, aware results, judge outputs, judges outputs, bias evaluation, learns judge
• Source: amazon.science

Backprop Alternative: Augmented Lagrangian Predictive Coding

Augmented Lagrangian Predictive Coding Training 1000-layer networks without backpropagation We introduce PC-ALM, a local alternative to backpropagation. PC-ALM trains residual MLPs up to 1000 layers,...

• Keywords: backpropagation brain, backpropagation training, locking backpropagation, networks backpropagation, backpropagation exactly, backpropagation runs, exact backpropagation, relies backpropagation, alternative backpropagation, standard backpropagation
• Source: pub.sakana.ai

GlossoGen studies when interacting LLM agents invent languages that humans cannot understand

Extremely excited to share our work in blogpost version. We study the conditions under which LLM agents develop their own languages, which we can’t understand, and introduce GlossoGen, a framework for...

• Keywords: ai agents, agents develop, glossogen framework, agents, oversee ai, introduce glossogen, llm agents, ai, languages, develop languages
• Source: x.com

Jerod Santo Joins Socket as Head of Media

Jerod Santo Joins Socket as Head of Media Allow myself to introduce... myself.

• Jerod Santo We need to talk. Supply chain attacks are going parabolic in both frequency and impact. There are so many h...

• Keywords: introduce socket, vulnerabilities open, socket mission, unknown vulnerabilities, vulnerabilities, socket, socket new, believed socket, sandboxes hack, socket crazy

• Source: socket.dev