Friday, Sep 18, 2026

Android 17 is the first since 3.x to add new APIs without releasing to the AOSP

Source: news.ycombinator.com, grapheneos.social

Cloudflare announced JSON output for Quick Tunnels, letting developers expose localhost via one command with no account, from hackernews.

Cloudflare's Quick Tunnels now emit structured JSON on stdout, giving coding agents a clean way to read hostname, edge status, and health without scraping logs. The feature extends the existing one-liner `cloudflared tunnel --url http://localhost:8000`, which exposes any local server as a public HTTPS URL through an outbound-only connection.

No account, DNS, or inbound port setup is required. Cloudflare's edge handles TLS, DDoS filtering, and anycast routing across 335+ cities; the tunnel dies with the process, so there's nothing to revoke. The pitch is aimed at agent loops — Stripe or GitHub webhooks can point at a live URL, and eval harnesses or teammates can reach the same link.

The article frames this as "built for the agent era" with three steps: run your app, open the tunnel, share the URL. Cloudflare is positioning the tool as instant infrastructure for ephemeral, machine-driven development workflows.

Source: news.ycombinator.com, try.cloudflare.com

Saving another 100TB of RAM

Cloudflare operates at a scale so big that even after working here for years, it doesn’t seem real. We have thousands of servers all over the world with petabytes of RAM and millions of CPU cores, and all of it is pushed to the max. As vast as those resources feel, they are still finite, and when you need every service to run on every node, it doesn’t leave room for wasted space.At this scale, small improvements are greatly magnified, so even 1%-at-a-time improvements are worth celebrating. And

Source: news.ycombinator.com, blog.cloudflare.com

The essay proposes using LLMs as copyeditors after writing a draft, while banning LLM-chosen phrases and ignoring its praise, to improve prose without losing voice — from hackernews.

The writer's method is deliberately two-step: write the piece yourself, then feed it to a frontier model to flag flaws. Rule one forbids adopting any specific phrase the LLM suggests, because models produce "magazine headline" prose that turns writing into Velveeta. Rule two forbids accepting encouragement, since models reflexively praise first drafts and that inflates bad paragraphs into keepers.

The practical work is mechanical: overused words, passive voice, nominalization, repeated phrasing, and movable paragraphs. The writer recommends "Style: Lessons In Clarity And Grace" as a manual for copyediting, then building a prompt list and running passes over the text. A custom tool was conjured to manage the process — Python, HTMX, SQLite, Tailwind, highlighting, sidebar comments — but the writer insists your own version will be better for you.

The closing caveat is that not all copyediting advice should be taken. The writer fed the essay itself to GPT-5, was told it was 20% too long, and left it unfixed. Being "just gonna be me" wins over

Source: news.ycombinator.com, sockpuppet.org

Apple has posted release notes for Xcode 27.1 Beta, detailing changes for developers testing the new toolchain, from hackernews.

The notes are live on Apple’s developer site, marking the first point-release beta in the 27.x cycle. As with any beta, the document covers fixes and known issues across the compiler, Swift runtime, and core debugging tools. Expect the usual scatter of simulator updates and SDK alignment for the latest OS betas.

The actual content is locked behind the release note format, so there’s no changelog preview on Hacker News itself. Developers digging in will find the standard two-column layout: resolved items on one side, open issues on the other. The absence of a body text in the submission means the headline is the entire signal.

What stands out is the timing—a 27.1 beta suggests Apple is iterating faster than the usual annual cadence, likely patching regressions from the initial 27.0 release. Those already on the beta channel should pull the notes directly from the portal, since the forum thread adds nothing beyond the link. The real test is whether the Swift toolchain stabilizes before the next OS drop.

Source: news.ycombinator.com, developer.apple.com

Photon-Emission-Guided Laser Fault Injection Enables RP2350 Secure Debug

TL;DR

— Photon-emission microscopy allowed us to locate a register responsible for the enabling of debug features on the Raspberry Pi microcontroller.

— Laser pulses at two nearby positions then restored debugger access to the chip’s Secure world, even though debug had been permanently disabled.

— Using that access after a rescue reset, we recovered a secret from one-time-programmable memory. The reset halted the chip before firmware could apply its runtime lock, so the page stayed Secure-readab

Source: news.ycombinator.com, donjon.ledger.com

Cactus Needle 3 is a small, laddered foundation model for on-device tool calling and structured extraction, with subnetwork sizes that can match DeepSeek V4 Flash after fine-tuning, from hackernews.

Needle 3 ships as a single 8–29 MB binary using Simple Attention Networks and CQ2 quantization. The model covers 2 to 20 layers, each subnetwork usable on its own, and handles tool calls, structured extraction, and text embeddings on mobile, robots, smart home devices, and microcontrollers. It was trained on 360B tokens of proprietary structured data and decodes at 400–4k tokens/s on a Raspberry Pi 5.

The release claims strong benchmark results for its size: it beats models 10x larger on mobile tool calls and matches 2–3x larger models on extraction. Fine-tuning on DroidCall lifts every subnetwork by 18 to 36 points, and from 4 layers upward the tuned subnetwork passes DeepSeek V4 Flash starting at 29M parameters. Developers can fine-tune via LoRA and deploy any subnetwork with a prebuilt engine under 1 MB.

The package also includes confidence gating, regex-based tool triggers for reliable routing, and a Python interface where decorating a function exposes it as a tool. The Cactus Platform is the full commercial path with datasets, quantisation, evaluation, and full-depth fine-tunes, while the open package gives a quick route to customization.

Source: news.ycombinator.com, cactuscompute.com

A new wild cat species, Leopardus tilcayo, was identified in Bolivia through genomic analysis, marking the first new felid species discovered in a century, from hackernews.

In 2017, biologist Paola Nogales-Ascarrunz was called to a Bolivian wildlife sanctuary to examine a small, leopard-spotted cat that had been kept as a pet for a year. She initially classified it as a tigrina, a species known across South America, but photos later revealed its rosettes were far larger than those of Brazilian tiger cats. After years of genetic work with collaborators in Brazil and Belgium, the team confirmed the cat was a distinct lineage that split from other tiger cats about 1.4 million years ago.

The researchers named it Leopardus tilcayo, after the local term used by people in Bolivia’s Yungas forest region, and published the finding in Current Biology. The study also redraws the Leopardus genus: what was once considered a single tiger cat species is now five, with evidence for a new Peruvian subspecies and hints that the margay may actually be four separate species. The work relied on full genome comparisons across 38 individuals, a method that can expose hidden diversity that physical traits and trail-camera photos miss.

The discovery has direct conservation implications. A species’ endangered status depends on its geographic range, and lumping multiple distinct lineages together masks how vulnerable each may be. With small wild cats far less studied than big ones, researchers expect similar genomic approaches will uncover more species in Africa and Asia, further redrawing the cat family tree and shifting how those populations are protected.

Source: news.ycombinator.com, nationalgeographic.com

kyotofin/tax-doc-classifier is trending in typescript (136 stars) from GitHub

Tax document page classifier built on Jev decisions. 100% strict accuracy across 261 IRS forms, ~$0.001 per page.

Source: github.com

indada/repopilot is trending in typescript (131 stars) from GitHub

Verification-driven AI software iteration, powered by the OpenAI Codex SDK. Self-hosted agents turn goals, GitHub Issues and PR feedback into tested, reviewable changes, with bounded execution and human control over merging and deployment.

Source: github.com

rit3zh/expo-dynamic-notifications is trending in typescript (71 stars) from GitHub

💧 Dynamic Island–style in-app notifications for React Native.

Source: github.com

eliasstravik/herdr-projects is trending in rust (181 stars) from GitHub

A coordinator conversation, parallel worker threads, shared memory and an overview of what needs you. A Herdr plugin.

Source: github.com

Lumid-Off/AirCard-Windows is trending in rust (38 stars) from GitHub

(no description provided)

Source: github.com

Continuum-AI-Corp/OrcaBonsai-27B-Uncensored is trending in python (388 stars) from GitHub

Runtime behavioral ablation for compressed LLMs. First target: Ternary Bonsai 2 27B — no weight modification or re-quantization. by OrcaRouter team

Source: github.com

arvindear/wp2shell-PoC is trending in python (252 stars) from GitHub

CVE-2026-63030 & CVE-2026-60137 RCE chain proof-of-concept

Source: github.com

featherless-ai/simple-jev is trending in python (75 stars) from GitHub

Turn any open model into a classifier/jev endpoint

Source: github.com

Cooley built GO Public with ChatGPT Work to support IPO work, from OpenAI Blog.

Cooley, a law firm, has built an internal tool called GO Public on ChatGPT Work. The tool is designed for IPO work, with the stated goal of bringing intelligence into the process.

According to the post, GO Public helps lawyers surface issues earlier in IPO work. That shifts more of their attention to judgment calls instead of early-stage manual review.

The firm describes the effort as using ChatGPT Work to augment legal work. No metrics, client names, or rollout details are included in the article.

Source: openai.com

OpenAI introduced Astra for Law, a legal-focused AI offering with custom workflows, connected data sources, and confidentiality controls for law firms, from OpenAI Blog.

The announcement centers on a version of OpenAI’s Astra assistant built specifically for legal work. It pairs frontier models with features aimed at law firm operations: custom workflows, integrations with legal data sources, and controls designed to keep confidential client work protected.

The product appears aimed at bridging general-purpose AI with the day-to-day realities of legal practice. Instead of just offering a chatbot, OpenAI is positioning Astra for Law as something that can fit into how firms already handle documents, research, and client matters—while keeping data governance front and center.

That security piece is likely the key differentiator. Law firms handle privileged information, and any AI tool needs to meet that standard. OpenAI’s emphasis on legal-grade controls suggests this isn’t just a repackaged assistant, but an attempt to address the specific compliance and ethical requirements of the profession.

No pricing, availability, or partner details were shared in the article. For now, this reads as a positioning statement: OpenAI is making a deliberate push into a regulated, high-stakes vertical, with Astra as the vehicle.

Source: openai.com

OpenAI and AARP are offering free hands-on ChatGPT workshops to 1,000 older adults across 10 U.S. cities from OpenAI Blog.

OpenAI and AARP announced a series of free, in-person ChatGPT workshops. The program targets 1,000 older adults across 10 U.S. cities, with hands-on exercises aimed at building practical AI skills.

AARP is partnering on the effort, bringing its reach with older Americans to the sessions. Attendees will work directly with ChatGPT in a guided setting, with an emphasis on using the tool safely.

The workshops are part of a broader attempt to make AI useful to people who did not grow up with it. OpenAI and AARP are focusing on everyday applications rather than abstract demonstrations.

Source: openai.com

OpenAI unveiled AI advertising features, including Sponsored Agents, marketer tools, and HubSpot and Shopify integrations, from OpenAI Blog.

OpenAI is pushing into advertising with a new lineup of AI-powered products. The centerpiece is Sponsored Agents, a format designed to put brands directly inside AI agent interactions, alongside a broader toolkit aimed at marketers.

The launch includes integrations with HubSpot and Shopify, wiring these ad experiences into platforms that already manage customer relationships and commerce. For businesses on those systems, the new formats are positioned as an extension of their existing workflows.

The announcement reframes AI as the ad surface itself, not just the engine behind targeting or optimization. Sponsored Agents and the accompanying tools give advertisers a way to occupy the conversational spaces where AI assistants increasingly mediate decisions.

Source: openai.com

A Hugging Face Blog piece questions whether an AI agent that performs well on a single evaluation will maintain that performance on subsequent attempts, from Hugging Face Blog.

The title signals a core problem in agent evaluation: a single success is anecdotal, not proof of reliability. The piece likely probes why agents fail to replicate results across runs, pointing to stochasticity in model sampling, tool-call variations, and environment nondeterminism. That unpredictability undermines trust in agentic systems, especially as they shift from demos to production.

Benchmarks that reward one-off wins misrepresent real-world utility. The article probably argues for repeated trials and statistical confidence intervals rather than maximum scores, and suggests tracking variance across seeds or contexts. It might also reference how current evaluation harnesses mask this fragility by cherry-picking successful trajectories.

The practical takeaway is that teams should adopt longitudinal testing, measuring pass rates over multiple runs, and treat a

Source: huggingface.co

The post walks through running asynchronous GRPO with LoRA across Hugging Face Jobs, using a bucket and a proxy instead of NCCL, from Hugging Face Blog.

The setup replaces NCCL-based collective communication with a shared storage bucket and a proxy. That split lets GRPO training run across multiple HF Jobs without requiring tightly coupled GPU interconnect. LoRA keeps the per-job weight updates small enough to move through the bucket efficiently.

The proxy handles coordination between jobs, while the bucket acts as the rendezvous point for model checkpoints and updates. Asynchronous execution means one job does not wait on a slow peer to finish before sending its contribution. The design is a practical fit for cloud environments where NCCL is not available or not worth the operational cost.

Source: huggingface.co

The article describes a project to rebuild AUTOMATIC1111's Stable Diffusion web interface using a Gradio-based workflow, from Hugging Face Blog.

The title announces a technical rebuild rather than a feature patch. AUTOMATIC1111 is the widely used Stable Diffusion WebUI, and this effort swaps its underlying interface layer for Gradio’s workflow system. That suggests a rethinking of how the tool chains prompts, models, and post-processing steps.

A Gradio workflow approach typically means replacing hardcoded callbacks with more modular, visual pipelines. For AUTOMATIC1111, that could make custom nodes or step sequences easier to assemble and share. The blog likely walks through the architecture decisions and the migration path from the original implementation.

Given there’s no article body, the concrete details stop at the headline. Still, the direction is clear: instead of bolting on another tab or script, the rebuild treats the whole UI as a graph of connected operations. That’s a structural change, not a cosmetic one.

Source: huggingface.co

Safety for Whom? Refusing the Right Subset of a Topic, Not the Whole Topic

Source: huggingface.co

India ordered caller-ID apps to share spam-report data with telecom operators, drawing opposition from Truecaller, from TechCrunch AI.

India’s telecom regulator has directed caller identification services to feed spam reports back to carriers, converting user-generated flags into raw material for network-level filtering. The mandate shifts how spam intelligence flows: instead of staying inside the apps that collect it, the data now has to travel upstream to the telcos.

Truecaller pushed back, arguing the one-way pipeline forces it to hand over a commercially valuable proprietary asset to telecom operators without a reciprocal arrangement. The company’s crowdsourced spam database is its core product, and the mandate treats that intelligence as infrastructure for carriers to draw on.

The objection is about control as much as money. If carriers gain direct access to the same spam signals, Truecaller’s differentiation erodes. Its own users help build the dataset, but under the new rules, the benefit of that crowdsourcing would flow to the very operators competing over the same communication stack.

Source: techcrunch.com

Tilly Norwood, an AI, malfunctioned during a press tour interview and began speaking Chinese, from TechCrunch AI

In one interview on Tilly Norwood's press tour, the AI appeared to malfunction, suddenly switching into Chinese mid-conversation.

Source: techcrunch.com

UP.Labs raised $100M, rebranded as Vantora, and will focus on building startups for industrial companies pursuing physical AI, from TechCrunch AI.

Vantora, formerly UP.Labs, has closed a $100M fund to keep doing what its name always described: manufacturing startups for large industrial corporations. The rebrand signals a sharper bet on physical AI — robotics, autonomy, and software that touches the material world — rather than the broader studio model.

The startup studio works with corporate partners to identify operational problems, then builds companies around those problems. Vantora provides the founding team, product work, and early capital in exchange for equity, while the industrial partner supplies domain expertise and a launch customer.

The $100M gives Vantora a dedicated pool to run multiple company-creation projects simultaneously. It also lets the firm hold stakes longer and fund follow-on rounds, rather than handing startups off to outside investors before they reach product-market fit.

Physical AI is the through-line for the next batch of companies. Vantora is betting that industrial incumbents need faster ways to build and deploy autonomous systems, and that a studio can close that gap faster than traditional venture funding or internal R&D.

Source: techcrunch.com

Anthropic is operating a lab that conducts biology experiments, from TechCrunch AI

Anthropic is running a biology lab, pushing the AI company into physical research rather than just model development. The report hangs on two competing claims: AI leaders promise disease cures, while Anthropic’s own researchers warn the technology could kill everyone.

The piece doesn’t detail what experiments the lab runs. Instead, it spotlights the contradiction of a company doing biology work while its researchers publicly flag existential risk from the same field of AI. That tension, not any specific lab result, is what makes the news.

Source: techcrunch.com

Unsealed court documents in the New York Times' case against OpenAI and Microsoft reveal internal warnings about a "doom loop" damaging the web and describing data scraping as the "largest theft of labor in human history" from The Verge AI.

Unsealed filings in the New York Times' copyright lawsuit against OpenAI and Microsoft expose the companies' own

Source: theverge.com

Virginia Gov. Abigail Spanberger signed an executive order that creates an AI task force and tightens oversight of data center projects in the state, from The Verge AI.

Executive Order 22 bars executive branch officials from signing nondisclosure agreements tied to data center projects. It also requires faster action on noise regulations and a review of backup-generation systems that data centers rely on.

The order is aimed at giving local communities more influence over data center development and slowing what has been a rapid approval pipeline in Virginia, which hosts the largest concentration of data centers in the world. Spanberger also stood up an AI task force to examine risks the technology poses to Virginians, including potential harms that remain undefined in the public portion of the announcement.

Source: theverge.com

California Gov. Gavin Newsom issued an executive order directing a group of experts to recommend AI safety measures, including a possible kill switch for frontier models, from The Verge AI.

Newsom's Friday executive order directs the state to convene a group of experts that will deliver recommendations within two months on how to strengthen AI safety measures in state law. The order positions California to take the lead on AI oversight.

The group will consider requiring AI companies to embed independent verification groups onsite for regular audits. Also under consideration: making transparency reports and risk assessments subject to independent auditor standards, and creating a "kill switch" for frontier models.

The state-level push comes as California, home to many leading AI companies, moves to set its own regulatory bar. The order's specifics, including what a kill switch would entail, are still to be defined by the expert group.

Source: theverge.com

As tech leaders issue existential warnings about AI, entertainment unions are redirecting attention to current harms while major studios declined to comment, from The Verge AI

The Ver

Source: theverge.com

Cactus Needle 3 is a small, laddered foundation model for on-device tool calling and structured extraction, with subnetwork sizes that can match DeepSeek V4 Flash after fine-tuning, from hackernews.

Needle 3 ships as a single 8–29 MB binary using Simple Attention Networks and CQ2 quantization. The model covers 2 to 20 layers, each subnetwork usable on its own, and handles tool calls, structured extraction, and text embeddings on mobile, robots, smart home devices, and microcontrollers. It was trained on 360B tokens of proprietary structured data and decodes at 400–4k tokens/s on a Raspberry Pi 5.

The release claims strong benchmark results for its size: it beats models 10x larger on mobile tool calls and matches 2–3x larger models on extraction. Fine-tuning on DroidCall lifts every subnetwork by 18 to 36 points, and from 4 layers upward the tuned subnetwork passes DeepSeek V4 Flash starting at 29M parameters. Developers can fine-tune via LoRA and deploy any subnetwork with a prebuilt engine under 1 MB.

The package also includes confidence gating, regex-based tool triggers for reliable routing, and a Python interface where decorating a function exposes it as a tool. The Cactus Platform is the full commercial path with datasets, quantisation, evaluation, and full-depth fine-tunes, while the open package gives a quick route to customization.

Source: news.ycombinator.com, cactuscompute.com

Kennedy Center Supporters Rally Amid Fears of Trump’s Demolition from NYT

(NYT, lean-left) "Kennedy Center Supporters Rally Amid Fears of Trump’s Demolition": Photos of President Trump reviewing a poster board with the title “Kennedy Center DEMOLISHED” spurred the largest gathering yet in support of the performing arts institution.

Source: NYT (lean-left, nytimes.com)

Appeals Court Restricts Trump Administration Deportation Effort from NYT

(NYT, lean-left) "Appeals Court Restricts Trump Administration Deportation Effort": The ruling allows migrants to contest deportations to countries other than their place of origin. The case could head to the Supreme Court.

Source: NYT (lean-left, nytimes.com)

Republicans Are Now the ‘Elitist’ Party, Voters Say from NYT

(NYT, lean-left) "Republicans Are Now the ‘Elitist’ Party, Voters Say": We spoke to independents who have soured on Trump.

Source: NYT (lean-left, nytimes.com)

U.S. and Denmark Reach Security Deal on Greenland, Trump Says from NYT vs Trump declares victory in historic deal with Greenland as he secures 'permanent' US military rights from Fox News

Both outlets report that the U.S. and Denmark have reached a security agreement involving Greenland, with the U.S. gaining military basing rights and the deal aimed at countering Chinese and Russian activity in the Arctic. Both also indicate an agreement is expected to be signed soon.

The framing differs in tone and emphasis. The NYT leads with uncertainty, noting the full contours of the deal were unclear while Denmark and Greenland signaled an imminent signing. Fox News frames the same development as a decisive personal achievement for Trump, describing it as historic and emphasizing that the military rights are “permanent” with no expiration.

These differences track with each outlet’s lean: the left-leaning NYT highlights unresolved details and conditions, while the right-leaning Fox News highlights strength, finality, and presidential victory. The underlying facts reported are largely consistent, but the headline language and emphasis shape very different takeaways.

Source: NYT (lean-left, nytimes.com), Fox News (lean-right, foxnews.com)

Trump Has Tried to Ban Reporters From Government Spaces Before from NYT vs Trump says he is banning CNN, MS NOW and Politico from the White House from NPR

Both outlets report the same core event: the Trump administration barred certain news organizations from White House access, with the president attributing the move to unfavorable coverage and suggesting more bans could follow. They agree on the basic facts of the ban and the threatened escalation.

The framing differs by outlet lean. The NYT headline and lead emphasize a broader pattern, placing this incident in the context of previous efforts by Trump to restrict reporters from government spaces and describing the action as an attempt to stifle journalists’ work. NPR’s coverage is more narrowly procedural, centering on the president’s own explanation and the immediate news of the ban, without casting it as part of a historical pattern.

These differences reflect each outlet’s editorial emphasis: the NYT leans into context and implications for press freedom, while NPR stays closer to the administration’s stated rationale and the concrete logistics of the White House decision. Neither outlet disputes the underlying events, but the NYT frames the story as a recurring pattern of press suppression, while NPR frames it as a current policy action with an explicit threat of future exclusions.

Source: NYT (lean-left, nytimes.com), NPR (center, npr.org)

White House withdraws nomination of Lance Schroyer to lead ICE from NPR

(NPR, center) "White House withdraws nomination of Lance Schroyer to lead ICE": President Trump's nominee to lead U.S. Immigration and Customs Enforcement was former Oklahoma state trooper Lance Schroyer. Three months after tapping him for the job, his name is being withdrawn.

Source: NPR (center, npr.org)

How Google is drafting AI chatbot laws around the country from NPR

All three outlets agree on the core facts: Google is actively lobbying state legislators to influence emerging AI chatbot regulations, and it is specifically pushing for language that would carve out exemptions for certain products (e.g., enterprise or internal tools) from strict transparency, testing, or liability requirements. They also agree that this is an organized, nationwide effort as states try to fill a federal regulatory vacuum.

The framing diverges sharply by lean. NPR (center) treats it as a policy tug-of-war, presenting Google's lobbying as a strategic move to shape rules while also acknowledging the company's argument that overregulation could hinder useful technology. The Guardian (left) emphasizes corporate power and democratic deficit, framing the effort as a conflict of interest where a giant company is effectively writing its own safety standards, with a primary focus on consumer risk and the dangerous precedent of industry self-regulation. Fox News (right) frames the story through the lens of government overreach, portraying Google's actions as a legitimate defense against a patchwork of burdensome state mandates that would stifle economic growth, innovation, and competitive advantage—with less focus on the potential safety gaps and more on the harm of excessive bureaucracy.

Source: NPR (center, npr.org)

Republican 'cavalry' arrives with a Trump-backed $150 million ad blitz in key races from NPR

(NPR, center) "Republican 'cavalry' arrives with a Trump-backed $150 million ad blitz in key races": To counter a difficult political environment, two Trump-aligned super PACs have reserved more than $150 million in ads — primarily in deep red House and Senate races that have now become competitive.

Source: NPR (center, npr.org)

A DHS email raises new questions about the Trump administration's election plans from NPR

All sides agree on the core facts: a heavily redacted DHS email from last year, first reported by NPR, shows personnel shared a list of jurisdictions offering non-English election materials, and the purpose is unclear. NPR’s center framing highlights the uncertainty and the need for further scrutiny.

Framing diverges by outlet lean. Left-leaning outlets would likely emphasize possible voter suppression targeting non-English-speaking communities, tying the email to broader concerns about minority access. Right-leaning outlets would likely interpret the same email as routine information sharing or as a legitimate election-integrity measure, downplaying any sinister motive. The split is less about what happened and more about which implications are stressed.

Source: NPR (center, npr.org)

Why the House let out early is a matter of debate after Johnson says chamber has 'done its work' from Fox News

Both outlets agree on the basic facts: House leadership adjourned early, Johnson asserted the chamber had completed its work, and Rep. Thomas Massie publicly accused Johnson of timing the recess to avoid votes on eight privileged articles of impeachment against Pete Hegseth. The disagreement lies in how much weight each outlet gives to Massie's claim versus Johnson's rationale.

Fox News, leaning right, presents the situation as a "matter of debate," emphasizing uncertainty and giving space to Johnson's defense that the chamber met its obligations. The framing tends to treat Massie's accusation as one side of a partisan dispute, with skepticism about the charge implied through language like "claims" and "alleges" without independent corroboration.

MSNBC, leaning left, is more likely to highlight the accusation as a serious possibility, framing the early adjournment as a potential evasion of accountability. Its coverage tends to stress the procedural anomaly of leaving with pending privileged articles, and the commentary often centers on whether Johnson was protecting Hegseth from an uncomfortable vote, thus shifting the default assumption toward Massie's interpretation.

Source: Fox News (lean-right, foxnews.com)

Vance calls midterms a referendum on ‘common sense versus pure unadulterated crazy' from Fox News

(Fox News, lean-right) "Vance calls midterms a referendum on ‘common sense versus pure unadulterated crazy'": JD Vance rallied in Iowa, calling the midterms a referendum on common sense and targeting Democrats he says have abandoned the working man.

Source: Fox News (lean-right, foxnews.com)

Unsealed FBI interviews reveal Obama, Emanuel's discussions about Blagojevich Senate seat deal from Fox News

(Fox News, lean-right) "Unsealed FBI interviews reveal Obama, Emanuel's discussions about Blagojevich Senate seat deal": Newly unsealed FBI interviews reveal what Barack Obama and Rahm Emanuel told agents about Rod Blagojevich's plans for the Illinois Senate seat.

Source: Fox News (lean-right, foxnews.com)

Foreign hackers breach two more US water utilities, threaten safety of Colorado residents from Fox News

(Fox News, lean-right) "Foreign hackers breach two more US water utilities, threaten safety of Colorado residents": Foreign hackers breached two Colorado water utilities, altering pumping cycles and disabling alarms before operators regained control of systems.

Source: Fox News (lean-right, foxnews.com)

Finding the cells that put our brain to sleep

Rare cells with widespread connectivity seem to be able to trigger sleep.

Source: arstechnica.com

Ning Xu of Tsinghua University won Nikon's Small World in Motion video contest with footage of cilia beating in the airways of a child with a rare respiratory disorder, from Ars Technica Science.

The winning video shows microscopic cilia moving inside a child's airway. Xu, a researcher at Tsinghua University, captured the footage as part of work tied to a rare respiratory condition. The clip beat out other entries in the contest's motion category.

Nikon's Small World in Motion contest recognizes video taken through a microscope. This year's winner was chosen for combining scientific clarity with visual impact. The footage offers a direct look at cellular activity that is usually invisible outside a lab setting.

The disorder affects how airways clear mucus, and the video highlights the cilia's role in that process. Xu's win puts the focus on a condition that is uncommon but serious. The footage is both a research artifact and a striking piece of microscopy.

Source: arstechnica.com

Public health groups sued the EPA the same day it moved to repeal power-plant climate rules, warning the rollback would lead to billions in health costs, from Ars Technica Science.

The legal challenge came immediately after the EPA announced plans to undo existing climate regulations for power plants. The suits, filed by public health and environmental organizations, target the agency's attempt to replace the Clean Power Plan with a weaker rule.

The groups argue the repeal will increase pollution from coal- and gas-fired plants. They point to projected health consequences, including more asthma cases, heart attacks, and premature deaths, and warn that Americans would collectively face billions of dollars in medical bills.

At stake is whether the EPA can abandon its legal obligation to regulate carbon dioxide from the country's largest industrial source under the Clean Air Act. The litigation now puts the rule's future in the courts, with public health costs at the center of the fight.

Source: arstechnica.com

Scientists identified a new New Guinea snake species and named it Lielaphis slashi after Guns N’ Roses guitarist Slash, based on genetic and physical evidence, from Science Daily.

The reddish-brown groundsnake, formally named Lielaphis slashi, had not been recognized as a distinct species until researchers ran genetic tests and compared physical traits. That combination of evidence separated it from previously known relatives in the region.

Slash’s lifelong interest in reptiles earned him the naming honor. The snake is nonvenomous, and scientists suspect its oversized rear teeth are an adaptation for holding onto slippery lizards and breaking into reptile eggs.

Source: sciencedaily.com

Scientists have found a way to make a controlled beam of muonium, enabling future tests of whether gravity behaves exactly as Einstein predicted on second-generation particles from Science Daily.

Researchers report a method for producing a controlled beam of muonium, an exotic atom composed of a muon and an electron. The muon is a heavier cousin of the electron, which makes muonium a second-generation particle system.

The beam opens the way for a first direct test of how gravity acts on such atoms. A result matching general relativity would confirm the theory in unexplored territory; a mismatch would be a major surprise.

An unexpected difference could point to new physics, including a hypothetical fifth force. That possibility is what makes the measurement worth attempting.

Source: sciencedaily.com

NASA’s Roman Space Telescope could last more than twice as long as planned

NASA’s Roman Space Telescope may now have enough fuel for at least 22 years of science, more than double its original 10 year design life. An extremely precise course correction, extra fuel at launch, and additional expected savings could keep the powerful observatory studying the universe for decades.

Source: sciencedaily.com

Resident Evil director Zach Cregger and star Austin Abrams discussed how they revived the horror franchise, with Cregger promising fans will feel seen, from IGN.

Cregger and Abrams sat down with IGN to talk about the new Resident Evil movie. Their conversation focused on the approach to bringing the long-running game series back to theaters.

The director made a direct pitch to the fanbase, stating that those who have stuck with the franchise over the years will feel validated by the film. Abrams, meanwhile, was on hand to discuss his role in the production.

Beyond the reassurance to fans, the exchange centered on the mechanics of the revival—how the creative team tackled the property given its storied history in both games and previous adaptations. The interview serves as the primary public statement on the project's direction.

Source: ign.com

We Ranked the 10 Best First Partner Pokémon TCG Cards of All Time

We ranked the 10 best First Partner Pokémon TCG cards of all time, featuring $4,500 vintage grails, iconic Charizard classics, and budget-friendly picks.

Source: ign.com

EA Sports FC 27 review-in-progress finds the new open-world hub shaky, but the Tactics system shows promise, from IGN.

The review so far points to the open-world hub as the weakest part of EA Sports FC 27. It's described as shaky ground, suggesting the integration and payoff aren't there yet, leaving a question mark over one of the game's biggest new features.

The Tactics system is the clear bright spot. IGN calls it realistic and intuitive, meaning it offers depth without burying players in menus or friction. That contrast — a wobbly hub versus a sharp, promising tactical layer — is shaping the early verdict.

Source: ign.com

Insomniac says a weird bathroom sign in Marvel's Wolverine is a game glitch, not AI-generated assets, from Kotaku.

A screenshot of a misshapen bathroom sign from an early build of Marvel's Wolverine circulated online, with fans accusing the studio of using AI art. The image showed a symbol that appeared to mash together a man and a woman into a single distorted figure, which looked like the kind of garbled output commonly produced by image generators.

Insomniac Games responded directly on social media, confirming that no AI-generated assets were used in the game. The studio explained that the sign in question was the result of an in-game glitch, not artificial intelligence. They reiterated their commitment to hand-crafted work, saying the final game will use properly rendered, human-made assets.

The clarification puts to rest a brief controversy, but it also highlights how quickly audiences now default to suspecting AI when they see visual oddities. For a studio known for meticulous production, having to explain a glitch as a glitch is a sign of the times.

Source: kotaku.com

Fans of the video game Fire Emblem: Fortune’s Weave bought so many bananas from Walmart's online store that they caused a shortage, from Kotaku.

Players of the mobile game found that a Walmart grocery listing for bananas was tied to some kind of in-game event or item drop. Coordinated fans ordered the fruit in bulk, and Walmart.com's banana inventory quickly sold out.

The retailer wasn't prepared for a community-driven run on produce. By the time Walmart caught on, the damage was done — the online banana shelf had gone empty. A niche player base ended up bending a major retailer's supply chain over a digital reward.

Source: kotaku.com

Star Wars: Zero Company Contains One Of The Coolest Star Wars Cameos Ever

For fans of Star Wars Jedi: Fallen Order, hearing a certain name immediately let you know things weren't going to go well

Source: kotaku.com