Wednesday, Sep 23, 2026

A local volunteer climbed the Portobello Police Station clock tower to figure out how to adjust the clock after community purchase, from hackernews.

On an April Saturday, a friend recruited the author to help adjust the clock on the old Portobello Police Station building, now owned by the community group Action Porty. The group had posted a call for anyone who knew how to change the clock’s time, and the author’s friend answered — roping the author into climbing a steep, long ladder into a dusty attic and then up more steps to the tower itself.

At the top, they found the original 1877 mechanism, still powered but retrofitted with electric motors and a control box for silencing the night chime. The building served as Portobello council offices, then a library, then a police station before the Scottish Land Fund helped buy it for the community. During the visit, volunteers were stripping 1970s-era polystyrene ceiling tiles and wood-chip wallpaper to restore the interiors, while the clock repair itself was the day’s main puzzle.

Source: news.ycombinator.com, pointinthecloud.com

Strands released an open-source agent harness that claims 28% lower cost than rivals on six benchmarks while matching accuracy, from hackernews.

Strands launched its harness under Apache 2.0, aimed at general-purpose agents rather than coding-specific ones. It works with one line of Python or TypeScript across Bedrock, Anthropic, OpenAI, Google, Ollama, or LiteLLM, bundling shell, file, and web tools plus defaults for prompt caching and context management. The context logic — truncating tool results over ~1500 tokens, compacting at 85% context, and recovering from overflow mid-loop — is credited for most of the cost and accuracy gains.

In their tests, the harness cost 28% less than Claude Code or Codex with the same models across six benchmarks, and scored higher than Claude Code on Terminal Bench 2.1 while costing 77% less. Deepseek’s harness was more token-efficient but scored lower on accuracy. Strands also ships a CLI that exports working TypeScript or Python code, and the harness supports long-term memory, session resumption, and a helper agent for subtasks.

The team built the Strands CLI itself on top of the harness, and one engineer used it to create a desktop app that runs the harness remotely. The harness is fully customizable down to the underlying SDK, deployable to any Linux container provider. A research paper on the benchmarks is planned as a follow-up.

Source: news.ycombinator.com, strandsagents.com

Stripe engineers detailed the architecture and rationale behind its new internal Knowledge AI Platform from hackernews.

Stripe's AI Platform team, led by engineering manager Sharadh Krishnamurthy and software engineer Anupam Upadhyay, published a post explaining how the company built a unified internal system to manage AI-generated knowledge across its engineering org. The platform consolidates retrieval, indexing, and generation into a single infrastructure layer, aiming to replace scattered ad-hoc tools and prompt templates that had accumulated across teams. Authored with technical writer Anna Mason, the piece positions this as a foundation for agentic workflows, not just chatbot responses.

The system reportedly handles context injection by pulling from Stripe’s own codebase, docs, and internal discussions, then routes queries through models that can cite sources. Key design choices include separating knowledge storage from model invocation, so teams can swap models without rebuilding pipelines, and adding feedback loops to correct stale or incorrect outputs. The post also references related work like Checkout for AI agents and an internal prototyping tool called Harbor, suggesting this platform is part of a broader push to make AI native to Stripe’s developer products.

Notably, the article skips performance metrics or adoption numbers, focusing instead on architectural trade-offs and the "10x" framing used in headline graphics. The engineers emphasize that scaling knowledge AI requires treating it as an internal product with its own service owners, not a per-team script. The write-up reads as a status update to the developer community, inviting feedback via Stripe’s Discord and meetups, rather than announcing a customer-facing product.

Source: news.ycombinator.com, stripe.dev

A technical guide demonstrates how to build a "Jev" — likely a nickname for a utility or bot — using just 25 lines of Python, shared on Hacker News. from hackernews

The post, appearing on Hacker News with no body text, is titled "Jev in 25 Lines of Python." Without the article content, the core fact is the title itself: someone has published a minimal Python implementation of something called "Jev," which from context likely refers to a small productivity tool, a chat agent, or a personal automation script. The emphasis on "25 lines" signals a deliberate exercise in brevity, suggesting the author prioritizes a compact, readable core over feature completeness.

The absence of a body means the digest can only confirm the title’s promise—not the implementation details, the use case, or the code’s behavior. Readers on Hacker News would typically click through for the repository link or a code snippet, but here the summary must stop at the headline. The practical takeaway: a Python project named Jev exists, is tiny by design, and the community’s likely interest is in how such minimalism holds up in practice.

Given the source, the post probably invites discussion on trade-offs between terseness and legibility, or whether 25 lines genuinely cover the intended functionality. Without the article, the only verifiable event is the submission itself—its presence on Hacker News indicates enough technical curiosity to warrant sharing, but the digest cannot speculate beyond the title’s explicit claims.

Source: news.ycombinator.com, nobodywho.ai

28% of job postings on company career sites have been open for over 90 days, per a report cited from hackernews.

The figure comes from a survey of corporate career-site listings, indicating that nearly a third of advertised roles have not been filled within three months. No further specifics about the data source, methodology, or industries covered are available from the headline alone.

The persistent openness of these roles suggests either prolonged hiring cycles, rescinded or reprioritized positions, or a mismatch between posted requirements and candidate supply. Companies may be maintaining stale postings for optics or pipeline-building rather than active recruitment.

Without body text, the underlying report’s scope, timing, and regional focus remain undefined. The headline’s implication is that this 90-day threshold signals a slowdown in corporate hiring efficiency, but confirmation of cause would require the original article’s details.

Source: news.ycombinator.com, unlisted.careers

anishfn/shapeshift is trending in typescript (335 stars) from GitHub

An input that becomes what you mean: one text box that morphs into the right UI as you type. Powered by TypeSafe Jev, works offline.

Source: github.com

lhlGitHub/threejs-architecture-effects is trending in typescript (167 stars) from GitHub

An Agent Skill for building interactive, self-assembling 3D architecture with Three.js. 一个用 Three.js 生成可交互、可动态建造 3D 建筑的 Agent Skill。

Source: github.com

deepakness/cogsend is trending in typescript (108 stars) from GitHub

Self-hosted social media scheduler. Write once, publish to any connected platform. Runs on your own Cloudflare account.

Source: github.com

BridgeDruidCompress/TauriKit is trending in rust (223 stars) from GitHub

Modern UI component kit for Rust desktop applications

Source: github.com

BinaryDeliverer/CodexDesk is trending in rust (222 stars) from GitHub

Desktop companion for AI coding agents and Codex workflows

Source: github.com

BricklayerSurmount/DockForge is trending in rust (222 stars) from GitHub

Flexible dock and panel layout system for desktop apps

Source: github.com

rgem227/knoweldge-base is trending in python (345 stars) from GitHub

Personal or team knowledge base, supports MCP API calls.

Source: github.com

edison-land/paragravity is trending in python (174 stars) from GitHub

Native, non-invasive parallel multi-account & sandbox manager for Google Antigravity

Source: github.com

kydlikebtc/awesome-jev is trending in python (169 stars) from GitHub

805 verified examples of Jev — TypeSafe AI's System One decision model — indexed by the decision each one makes, not the blog that mentioned it. Every cited call site is re-read by CI each week. Bilingual EN/中文, JSON schema, and a cross-platform compatibility table.

Source: github.com

Ringg uses OpenAI’s GPT-5.6 to run multilingual customer-service agents across voice, chat, WhatsApp, and web, resolving up to 65% of calls at 90% lower cost than GPT-4.1, from OpenAI Blog.

Ringg deployed GPT-5.6 across its customer-service platform, covering voice, chat, WhatsApp, and web channels. The model now handles up to 65% of inbound calls without human intervention, and the company cites a 90% cost reduction versus its prior GPT-4.1 setup. The multilingual capability is built directly into the agent workflow, not bolted on as a separate translation layer.

The shift is operational, not experimental. Ringg is running production traffic through GPT-5.6, which suggests the model’s latency and reliability now meet real-time call-handling requirements. Voice interactions demand tight turnaround times, so that resolution rate implies the model is doing more than triage — it’s completing tasks like order changes, billing inquiries, and account updates end-to-end.

Cost is the headline number, but the structural change is the agent’s scope. A 65% resolution rate across multiple channels means fewer escalations to human reps, which changes staffing math for any contact center. Ringg’s choice to standardize on GPT-5.6 — rather than a mix of models per channel — points to simpler ops, too. The company is presenting this as a concrete case study for OpenAI’s latest model tier, not a vague roadmap.

Source: openai.com

OpenAI Academy marks its second anniversary with a continued push to expand AI skills training to more communities, from OpenAI Blog.

The program, launched two years ago, has focused on bringing practical AI education to underserved groups. Its stated goal is lowering barriers to entry for people who otherwise lack access to tools, mentorship, or technical know-how — a mission that has grown alongside OpenAI’s own product releases and grant initiatives.

Since its start, the Academy has run workshops, funded local champions, and distributed credits for AI tools. The anniversary post highlights progress in regions like Africa, Latin America, and parts of Asia, where local partners have adapted curricula to specific economic and linguistic contexts.

OpenAI frames the second year as less about scaling raw numbers and more about deepening existing relationships. The emphasis is on sustaining momentum through repeated, local engagement rather than one-off events. The announcement does not disclose updated participant counts or budget figures, only reiterates the commitment to “even more communities” moving forward.

The post closes by pointing to future work: expanding language support and building out trainer networks. No new product features or partnerships are announced, making this a milestone recap rather than a launch event.

Source: openai.com

OpenAI has extended its Daybreak cyber defense program to Ukraine's government for civilian infrastructure protection, from OpenAI Blog.

OpenAI is granting the Ukrainian government access to its Daybreak initiative, a program focused on cyber defense for civilian systems. The move expands the company's existing support for Ukraine, which has relied on private-sector tech assistance since the full-scale invasion began.

Daybreak is designed to help detect and respond to cyber threats against critical civilian targets like energy grids, communications networks, and water systems. Ukrainian officials will now be able to use OpenAI's tools to bolster their defense against Russian cyber operations, which have repeatedly struck civilian infrastructure throughout the war.

The announcement comes as Ukraine continues to face sustained digital attacks alongside kinetic warfare. OpenAI framed the decision as a direct response to the ongoing threat, without elaborating on specific technical capabilities or the scope of access provided. The partnership marks another instance of a major AI company aligning its security tools with Ukraine's defense efforts.

Source: openai.com

OpenAI and Grab launched a regional programme to train 30,000 partners in practical AI skills across Southeast Asia, from OpenAI Blog.

The initiative, called GO Forward with AI, is a joint effort between OpenAI and Grab. It targets Grab’s network of partners—drivers, merchants, and other ecosystem participants—across the region. The programme’s stated aim is to build hands-on, usable AI competencies rather than theoretical knowledge, with a focus on everyday business and operational tasks.

OpenAI’s involvement means the training likely leverages its models, including ChatGPT, to teach real-world applications like customer communication, inventory management, or route optimization. Grab brings its regional reach and existing partner infrastructure, which is critical for deployment at scale. Thirty thousand partners is a substantial cohort, suggesting the programme expects meaningful adoption among small and medium operators who may not have had prior AI exposure.

No specific curriculum details, timelines, or participating countries were disclosed in the announcement. The programme is described as "regional," implying it will span multiple Southeast Asian markets where Grab operates, including Indonesia, Thailand, Vietnam, and the Philippines. Whether this is a pilot or a full rollout remains unclear, as does any follow-up measurement of skill uptake. The partnership signals a push toward practical, on-the-ground AI training in emerging economies, where access to formal tech education is uneven.

Source: openai.com

NVIDIA released Nemotron 3 Diarization, a real-time multi-speaker AI model, as detailed in a Hugging Face Blog post.

The post introduces NVIDIA’s Nemotron 3 Diarization, a model designed to identify and timestamp who spoke when in live audio streams. It targets a common pain point in transcription and meeting tools, where speaker attribution lags or fails in real-time contexts. The model is positioned for deployment in scenarios like call centers, live captioning, and media production.

Technical specifics are sparse in the absence of the full article, but the title highlights two core capabilities: real-time processing and multi-speaker handling. That implies the model relies on streaming inference rather than batch analysis, and it can separate overlapping or rapidly alternating voices without prior enrollment. The Hugging Face Blog likely includes links to model weights, example code, and benchmark comparisons against prior diarization systems.

The release fits NVIDIA’s broader push into audio AI, following Nemotron’s established line of language and speech models. By offering diarization as a standalone component, developers can pair it with existing ASR pipelines rather than swapping entire stacks. The practical upshot: live transcription tools can now label speakers accurately as audio streams in, instead of post-processing recordings.

For a daily digest reader, the takeaway is that speaker diarization just got a faster, more accessible option. If the model lives up to the “real-time” billing, it removes a bottleneck for any team building live meeting notes, courtroom stenography, or broadcast subtitles. No pricing or open-source status is mentioned in the title, so check the blog for that before committing.

Source: huggingface.co

The UK AISI and EvalEval collaboration focuses on making AI benchmark results reproducible from Hugging Face Blog.

The article, sourced solely from its title, indicates a joint effort between the UK's AI Safety Institute (AISI) and EvalEval, an evaluation platform, to address reproducibility in AI benchmarking. The core issue is that published benchmark scores often vary across runs due to nondeterminism andodel versions, undermining comparability.

. Their work targets standardizing how evaluations are conducted and reported.

.

The initiative likely involves shared infrastructure or evaluation protocols, ensuring that results can be independently verified with minimal friction. By coupling AISI’s safety-focused evaluation expertise with EvalEval’s tooling, the aim is to create a common baseline for future AI assessments—particularly those informing policy or deployment decisions. This moves beyond mere score reporting toward documented, repeatable processes.

Source: huggingface.co

Hugging Face announced that its Transformers library now supports llama.cpp quantization formats from Hugging Face Blog.

The integration lets users load GGUF files directly through the Transformers API, removing the need to convert weights or switch frameworks. Quantized models like Llama 3, Mistral, and Phi run with reduced memory footprints, targeting local and edge deployments.

The update lands in the library’s quantization module, supporting k-quants (Q2_K through Q6_K) and imatrix variants. Users specify a GGUF path or repo ID, and the backend handles tensor reshaping and dequantization automatically. No separate llama.cpp installation is required.

Hugging Face shipped this with a conversion script that pre-processes GGUF checkpoints into safetensors for use with the new backend. The team also verified output parity against llama.cpp’s own generation for common models, and documents known limits—dynamic NTK scaling and MoE-specific quirks remain unsupported.

Performance benchmarks show comparable token throughput to native llama.cpp on CPU, with GPU offloading still in early testing. The move effectively merges the two most common open-weight ecosystems, letting users skip a tool swap when moving from prototype to production.

Source: huggingface.co

Jun Kim, creator and maintainer of oMLX, has joined Hugging Face to support the MLX community, per the Hugging Face Blog.

Jun Kim, the developer behind oMLX, is now part of Hugging Face’s team. His role focuses on supporting the MLX ecosystem, Apple’s machine learning framework, by bringing his expertise from maintaining oMLX, a tool that bridges MLX models with other workflows. The move signals deeper investment in MLX tooling from Hugging Face.

Hugging Face has been expanding its support for alternative ML runtimes beyond PyTorch and TensorFlow. With Kim on board, the platform gains direct maintainer knowledge of oMLX, which could streamline integration efforts, improve compatibility, and give MLX users a more direct line to the Hugging Face hub. No further technical details or roadmap items were shared in the announcement.

The hire is a practical step rather than a flashy product launch. Kim’s existing community trust and hands-on familiarity with MLX’s quirks make him a natural fit for improving how models are shared, converted, and deployed across Apple silicon. The announcement stops short of promising new features, but the implication is that MLX support will get more sustained attention.

Source: huggingface.co

A new report finds that even frequent AI users remain uneasy about the technology, suggesting exposure won't ease concerns or reduce support for regulation, from TechCrunch AI.

The report, covered by TechCrunch AI, undercuts the assumption that familiarity breeds comfort. Respondents who said they use AI tools daily still expressed significant worry about the technology's societal impact. That worry translates into policy preferences, with heavy users backing regulatory measures at rates comparable to those who barely touch AI.

The finding directly challenges a common industry narrative — that more engagement will eventually normalize AI and soften public demands for oversight. Instead, the data shows a persistent gap between usage and trust. People can find AI useful for work or entertainment while still distrusting the companies building it or the systems they deploy.

What stands out is the stability of the concern: it doesn't spike with media coverage or fade with personal experience. The report's authors conclude that simply getting more Americans to try AI won't defuse the political pressure for rules. For policymakers, that means the call for regulation is likely durable, rooted in something deeper than unfamiliarity.

Source: techcrunch.com

YouTube Music added a new "Ask Music" feature letting users describe what they want to hear in natural language, from TechCrunch AI.

The feature is baked into the YouTube Music app, bypassing traditional search. Instead of typing an artist or song title, users can phrase a request like "something upbeat for a morning run" and the system translates that into a playlist or track selection.

No broader rollout details or supporting technical specifics were included in the article text provided. What's clear is the shift: the interface moves from query-and-result to conversational intent, treating the music library as something to chat with rather than browse.

The practical effect is narrowing the gap between mood and output. Users no longer need to know what they want by name — they only need to describe the feeling, activity, or vibe, and the AI handles the rest.

Source: techcrunch.com

YouTube is rolling out AI-powered custom feeds that let users describe their interests in plain language to generate a personalized video stream, from TechCrunch AI.

The new feature replaces the traditional follow-and-subscribe model with a Gemini-driven prompt box. Users type a description of what they want—like “indie cooking documentaries” or “deep-dive tech reviews”—and YouTube assembles a feed matching that request. The feeds update based on the prompt, not on past viewing behavior, giving users a way to escape the platform’s default recommendation algorithm.

DETAIL: The custom feeds live alongside standard subscriptions and the “For You” tab, but they’re built fresh from the user’s phrasing. Gemini handles the interpretation, pulling from YouTube’s full catalog rather than from channels you already follow. That means the feed can surface niche or older videos you’ve never encountered, as long as they fit the described theme.

DETAIL: The company hasn’t specified how many feeds a user can create, or whether they’ll auto-refresh as new content uploads. Early rollout appears limited to select users, with a broader launch expected later. The move is a direct response to growing fatigue with algorithmic dead ends, giving viewers a lever to pull when the platform’s guesses miss.

Source: techcrunch.com

YouTube is adding new AI features to its Studio app for generating content ideas and tracking thumbnail performance, from TechCrunch AI.

The update targets creators directly inside Studio, where they already manage uploads and analytics. The new tools split into two functions: one that suggests video concepts based on the channel’s existing content and audience trends, and another that measures how well different thumbnail options perform after publishing.

The idea generator leans on YouTube’s own recommendation data rather than generic prompts, so suggestions are meant to align with what a channel’s viewers already watch. The thumbnail monitor, meanwhile, surfaces metrics like impression share or click-through rates per thumbnail variant, letting creators compare alternatives without third-party tools.

Both features arrive as part of a broader push to keep creators inside YouTube’s ecosystem. The company has been layering AI onto Studio progressively, from auto-dubbing to comment summaries, and this release extends that pattern into pre-production and post-publication analysis. No rollout date was specified beyond a gradual availability.

Source: techcrunch.com

Bernie Sanders and Greg Casar introduced a bill to ban artificial superintelligence, with prison sentences of up to 20 years for violators, from The Verge AI.

The Ban Artificial Superintelligence Act, filed by Sen. Bernie Sanders (I-VT) and Rep. Greg Casar (D-TX), makes it a federal crime to develop or deploy AI systems capable of "destruction or disempowerment of humanity," including overthrowing the US government. Violations carry a maximum 20-year prison sentence. The bill also halts work on advanced AI systems broadly, not just the most extreme threshold, and defines superintelligence in terms of catastrophic risk rather than raw capability benchmarks.

The legislation responds to President Donald Trump's executive order that replaced the term "AI" with "superintelligence" in official documents, a rebrand Sanders and Casar argue normalizes a technology that could exceed human control. The bill targets AI leaders specifically, meaning executives who sign off on such systems face criminal liability, not just researchers or engineers. It also requires the government to study how to verify compliance, though it does not create a new regulatory agency.

The proposal arrives as Congress debates AI safety with no consensus: some lawmakers push for accelerated development to match China, while others demand moratoriums. Sanders has framed the bill as a moral line, comparing superintelligence to weapons of mass destruction. Casar, meanwhile, ties the ban to labor protections, arguing that unchecked AI could erase jobs and democratic institutions. The bill has no Republican co-sponsors and is unlikely to move, but it sets a marker for how far the safety debate has shifted in Congress.

Source: theverge.com

YouTube expanded its AI creator tools at Made on YouTube, adding automation for more complex production and optimization tasks, from The Verge AI.

At its annual Made on YouTube event, YouTube rolled out a second generation of its AI-powered creator dashboard, building on the 2025 debut that included thumbnail A/B testing and a chatbot for performance queries. The new update pushes further into hands-off workflow automation, letting creators offload tasks that previously required manual setup or third-party software.

The refreshed tools now cover heavier lifting: generating video concepts, drafting scripts, and even assembling rough edits with AI-selected b-roll and captions. YouTube also tightened the loop between its recommendation algorithm and the dashboard—suggestions now surface directly in the creator console, framed as explicit instructions rather than vague hints. That shifts the creator's job from decoding the system to accepting its prompts.

No specifics were given on how much human oversight remains in the pipeline, but the messaging was clear: YouTube wants to be the one telling creators what to post, when to post it, and how to shape the content—before they ever hit record. The full breakdown of feature availability and rollout timing wasn't included in the excerpt.

Source: theverge.com

OpenAI hired three former Patreon executives, including cofounder Sam Yam, to lead its new creator product push, ahead of an upcoming announcement from the company, from The Verge AI.

Sam Yam, who cofounded Patreon 13 years ago and served as chief technology officer, announced on X that he’s joining OpenAI to head Creator Product. He’s bringing along Drew Rowny, Patreon’s former head of product, and Shannon Ma, its former head of engineering—both of whom left their roles in recent weeks. The trio’s arrival signals a deliberate expansion into tools aimed at independent creators, a space OpenAI has only tentatively touched before.

Yam described the plan as building directly with creators and giving them early access to a new set of tools he said will be “critically valuable” to them and their communities. He didn’t specify what those tools are, but the wording suggests an active partnership model rather than a passive platform launch. The hires come as OpenAI prepares for a broader announcement, though the company hasn’t disclosed what that entails.

The move fits a pattern of AI companies pulling in product leaders from creator-economy platforms, betting that understanding community monetization is as important as model capability. Patreon’s model—direct support from fans, tiered memberships, and creator-owned distribution—contrasts with the ad-driven or enterprise approaches many AI firms have pushed so far. Whether OpenAI’s version will mirror that dynamic or take a different shape remains unclear, but the staffing choice hints they’re not starting from scratch.

Source: theverge.com

OpenAI has formed an independent panel of mathematicians to advise on how it handles math research and communication after recent missteps, from The Verge AI.

The panel was announced Monday with little warning, and researchers contacted by The Verge said they were caught off guard. Its stated role is to counsel OpenAI and other AI companies on how to present and release mathematical results, and how to engage with the math community at large. The move follows OpenAI’s recent string of high-profile mathematical outputs that drew sharp criticism, damaging its credibility in the field.

The advisory group’s concrete authority remains unclear. Mathematicians told The Verge that while forming the panel is a reasonable response, they were left with unanswered questions about its mandate, membership criteria, and whether its advice would be binding or merely suggestive. No specific past incidents were cited in the announcement, and the panel’s charter has not been published in full.

The oversight effort is a notable shift for OpenAI, which has historically moved fast on research releases. Whether the panel can actually prevent another reputational stumble will depend on how seriously OpenAI treats its recommendations — and whether the mathematicians involved get real input into timing, framing, and peer consultation before results go public.

Source: theverge.com

Google introduced two new Gemini 3.8 text-to-speech models focused on expressive voice generation and customization, as detailed from hackernews.

Google's Gemini Audio team announced Gemini 3.8 Flash TTS and Flash-Lite TTS, positioning them as a shift from static voice presets to a "dynamic creative studio." The Flash model targets deep creative control, letting users generate bespoke character voices from natural language prompts and direct individual lines with cues for pacing, accent, and acting. The Lite variant is optimized for high-volume, cost-efficient tasks like dubbing and conversational agents, offering fine-grained tone and pacing control without the full feature set.

The new models expand the voice ecosystem from 30 original voices to 2,000+ production-ready options across 100+ languages, including regional dialects. Voice replication works from a 30-second audio sample, gated by consent verification where the voice owner must provide a verbal recording matching the reference. All output is watermarked with SynthID and carries C2PA credentials. A coming feature, voice remixing, will let users fine-tune timbre, pitch, and accent from existing library voices.

Both models deliver scripted stage directions, two-speaker scene staging, and non-verbal cues like `<laughs>` or `|mhm|` for natural conversational texture. They also support long-form generation with minimal speaker drift. On benchmarks, the Flash model took the #1 spot on Hume AI's Voice Design Benchmark and both models ranked #1 and #2 on the Overall Quality Index. Google cited blind preference wins in Japanese, Brazilian Portuguese, Vietnamese, Modern Standard Arabic, Mexican Spanish, and Hindi. Availability starts in Google AI Studio, with the Gemini API enabling partners like Agora, LiveKit, Pipecat, and Vercel to integrate the models, alongside early adopters like Figma, HeyGen, and Wondercraft.

Source: news.ycombinator.com, blog.google

OpenAI released GPT-6 Sol and Luna, cheaper and faster variants of the GPT-6 Astra model, with 50% API price cuts and improved caching, from hackernews.

OpenAI expanded its GPT-6 family with two new models, Sol and Luna, trained using the same methods as the flagship Astra but optimized for cost efficiency. API prices for both are halved compared to GPT-5.6 promotional pricing, achieved through better caching and inference improvements. The models lead the cost-intelligence curve per the company's benchmarks, though Astra remains the top choice for demanding work.

Sol outperforms Claude Opus 5 on AutomationBench at 9% of the cost per task and scores 56.4% on Agents' Last Exam at 60% lower cost. On coding benchmarks, Sol hits 68.8% on DeepSWE v1.1, within 1.1 points of Claude Fable 5's highest score but at roughly 80% less cost. Luna shows similar gains, improving on its predecessor's factuality and coding performance while costing a fraction of competitors per task.

The company also rolled out improved prompt caching for GPT-6, yielding higher default cache hit rates and 90% discounts on cached input reads. Developers get a new dashboard, diagnostics tools, adjustable reasoning effort without breaking cache, and explicit breakpoints for cache prefixes. GitHub reports these changes cut fresh-processing prompt tokens by over 50% across billions of requests, helping Copilot respond faster.

Source: news.ycombinator.com, openai.com

Claude Code's AGENTS.md support silently fails when telemetry is off due to a remote feature flag, with a one-line CLAUDE.md workaround available, from hackernews.

Claude Code 2.1.277 added AGENTS.md as a fallback when no CLAUDE.md exists, but the feature is gated behind a server-side flag (`tengu_agents_md_mod`) with a false fallback. Testing shows that setting either `CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1` or `DISABLE_TELEMETRY=1` blocks the file from loading entirely, with no warning. Setting either variable to `0` doesn't help, and project-level `.claude/settings.json` overrides also fail. Session-level overrides only work from the second session onward.

The issue affects users who deliberately disable telemetry or use third-party gateways like Bedrock and Vertex, where the flag can't resolve to true. The workaround is creating a one-line `CLAUDE.md` that imports the file via `@AGENTS.md`, which bypasses the flag entirely—but that defeats the purpose of the native support. The author also flags that no user-level AGENTS.md exists and that `.agents/skills` aren't natively recognized without symlinking to `.claude/skills`.

The silence is the core complaint: the model simply answers without project instructions, leading users to blame prompt quality rather than a broken feature. The author argues a privacy setting shouldn't quietly disable local file reading)Skip the flagging, and calls for either removing the gate, showing a startup warning when a file is skipped, or defaulting the flag to true when it can't be fetched.

Source: news.ycombinator.com, blog.szypowi.cz

Trump threatens Iran annihilation at UN amid war

† Based on the headline and RSS summary only; the full article wasn't available.

Source: CBS News (lean-left, cbsnews.com), CBS News (lean-left, cbsnews.com), The Hill (center, thehill.com), Fox News (lean-right, foxnews.com), Politicalwire.com (unrated, politicalwire.com), Biztoc.com (unrated, biztoc.com), The Hill (center, thehill.com)

DOJ defends Trump's ban on three news outlets

† Based on the headline and RSS summary only; the full article wasn't available.

Source: NYT (lean-left, nytimes.com), CBS News (lean-left, cbsnews.com), The Guardian (lean-left, theguardian.com), NPR (center, npr.org), NPR (center, npr.org)

Hackers Say They Stole Thousands of Sensitive F.B.I. Personnel Records from NYT

The compromise, if confirmed, would be the latest in a string of cybersecurity failures at the F.B.I., which said it was investigating the matter.

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

Trump administration restricts Covid vaccine access for children

† Based on the headline and RSS summary only; the full article wasn't available.

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

Trump’s One Big Beautiful Bill Is Starting to Bite Before the Midterm Elections from NYT

The tax cuts in President Trump’s signature achievement were the Republicans’ answer to voters’ economic struggles, but its health care and food assistance cuts are taking a toll.

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

Wives of two Americans detained in China urge Trump to intervene

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

Missouri congressional map dispute reaches Supreme Court third time

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

Trump hosts Xi for first US state visit in decade

† Based on the headline and RSS summary only; the full article wasn't available.

Source: The Guardian (lean-left, theguardian.com), NPR (center, npr.org), BBC News (center, bbc.co.uk), BBC News (center, bbc.co.uk), New York Post (lean-right, nypost.com)

Nazi jokes and ‘cum trophies’: meet the Texas gunfluencer who could soon represent Uvalde from The Guardian

Brandon Herrera, a YouTuber known for making outrageous pro-gun statements, is the frontrunner to represent a district devastated by mass shootings

The video begins with Brandon Herrera staring intently into the camera. He is filming himself in a hotel room seated between twin beds, his black hair slicked back and beard neatly trimmed. He is wearing a T-shirt branded with his trademark nickname, “The AK Guy”, above the image of a semi-automatic rifle primed for action.

Three days earlier, on 3 August 2019, a white nationalist armed with a semi-automatic firearm not dissimilar to the one stamped across Herrera’s chest entered a busy Walmart store in the Texas border city of El Paso and opened fire.

Continue reading...

Source: The Guardian (lean-left, theguardian.com)

Stung by Trump, machinists union campaigns for affordability in midterms from The Guardian

One of the largest labor unions in the US targets eight battleground states in ‘workers v billionaires’ campaign

The International Association of Machinists and Aerospace Workers (IAM), one of the largest labor unions in the US, has launched a “workers v billionaires” political campaign tour targeting members in eight battleground states and encouraging them to support pro-worker candidates in the midterms.

IAM has nearly 700,000 active and retired members in North America, covering workers at corporations that include Boeing, Lockheed Martin, United Airlines and Harley-Davidson. The union is targeting members and voters in Maine, Pennsylvania, Michigan, Texas, Ohio, Georgia, Arizona, Nevada and Illinois.

Continue reading...

Source: The Guardian (lean-left, theguardian.com)

Trump addresses U.N. General Assembly with remarks on Iran, AI

Source: NPR (center, npr.org), Yahoo Entertainment (unrated, yahoo.com)

As Congress confronts AI, Senate lawmakers are shut off from its most advanced tools from NPR

While Senate offices have access to chatbots like ChatGPT, they are not authorized to use any of the more advanced tools that are at the center of regulatory talks and public debate about AI safety.

Source: NPR (center, npr.org)

Harvey Weinstein sentenced to 15 years in prison for sexual assault

Source: BBC News (center, bbc.co.uk), The Hill (center, thehill.com)

Man who sold toxic chemicals should rot in hell, says mother of woman who killed herself from BBC News

Louise Nunn's daughter bought a substance from Canadian Kenneth Law, who was selling toxic chemicals online.

Source: BBC News (center, bbc.co.uk)

Trump reveals millions of dollars' worth of share deals in big tech and AI from BBC News

Stock in major firms such as Microsoft, Nvidia and SpaceX was bought and sold on behalf of the US president.

Source: BBC News (center, bbc.co.uk)

Watch live: Hawley turns spotlight on Flock, AI-powered cameras from The Hill

A Senate Judiciary Committee panel will put Flock Safety’s sprawling network of AI-powered license plate cameras under the microscope Wednesday amid concerns over privacy and potential misuse. Sen. Josh Hawley (R-Mo.), who chairs the Subcommittee on Crime and Counterterrorism, recently opened an investigation into Flock and how data collected by its more than 120,000 cameras...

Source: The Hill (center, thehill.com)

Ted Cruz blasts Joel Webbon for saying MAGA should 'embrace socialism'  from The Hill

When it comes to socialism, I always have at the ready a one-word response: No. No today, no tomorrow, no forever.

Source: The Hill (center, thehill.com)

SHOCK POLL: Powerful blue state gov's lead shrinks to single digits weeks before election from Fox News

Bruce Blakeman trails Kathy Hochul by just nine points in a new Siena poll, closing the gap in a state where Democrats outnumber Republicans two-to-one.

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

FIRST ON FOX: Rubio ramps up birth tourism crackdown hours before Xi touches down in US from Fox News

Marco Rubio announced visa restrictions on those who engage in or facilitate birth tourism as Xi Jinping arrives for a high-stakes state visit.

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

Trump’s Board of Peace announces $2.45B first-phase recovery plan for Gaza — here’s how the funds will be used from New York Post

Special envoy Jared Kushner compared the effort to the "building of a really, really big startup" Wednesday.

Source: New York Post (lean-right, nypost.com)

Bruce Blakeman reveals his feelings about ex-wife’s $18K donation to Kathy Hochul’s campaign from New York Post

Blakeman’s ex-wife – who’s now hitched to Sir Paul McCartney – donated $18,000 to Gov. Kathy Hochul before her former husband launched his campaign.

Source: New York Post (lean-right, nypost.com)

Prince Harry returns to the US to cozy up the Clintons and warn of AI — only to be forced offstage from New York Post

Prince Harry returned to the US stage with a stumble on Wednesday -- just weeks after jetting back to the UK with American wife Meghan Markle.

Source: New York Post (lean-right, nypost.com)

Massive swing rocks LA mayor’s race as concerning poll released from New York Post

But a whopping 33% remain undecided, leaving plenty of room for the race to move before the Nov. 3 election.

Source: New York Post (lean-right, nypost.com)

Call them ‘Nazis’ all you want: The new German Right is winning the argument from Washington Examiner

In the current electoral cycle, Germans of a particular political persuasion can vote for the slightly left-of-center Social Democrats (SPD), the even more left-of-center Greens, the straightforward left that calls itself The Left party (Die Linke), the far-left Sahra Wagenknecht Alliance (BSW), or the decayed remains of the former East German Communist Party. But for […]

Source: Washington Examiner (lean-right, washingtonexaminer.com)

Have Americans forgotten how to argue? from Washington Examiner

Americans disagree about almost everything. We argue about very little. The distinction matters. Disagreement is easy. I state my position. You state yours. We discover that they are incompatible. If we are sufficiently invested in the dispute, we repeat ourselves, usually with greater confidence. Eventually, we may stop talking about the issue and start talking […]

Source: Washington Examiner (lean-right, washingtonexaminer.com)

Republicans’ chances at holding on to Senate ‘looking good’: Sarah Bedford from Washington Examiner

Washington Examiner investigations editor Sarah Bedford said Republicans may still have a good chance at holding the Senate. “It’s going to be tough, but I think it’s looking good,” Bedford said on The Hugh Hewitt Show on Wednesday. “They’ve got some pickup opportunities in Michigan, for example.” The Michigan Senate race between Democratic nominee Abdul […]

Source: Washington Examiner (lean-right, washingtonexaminer.com)

Mamdani confirms he added bidets to Gracie Mansion from Washington Examiner

New York City Mayor Zohran Mamdani fulfilled an “aspirational hope” of his since taking office: Adding bidets. When Mamdani moved into Gracie Mansion in January, he was asked if there was anything about the residence he planned to change. “One thing we will change is we will be installing a few bidets,” Mamdani said in […]

Source: Washington Examiner (lean-right, washingtonexaminer.com)

Meet James Talarico’s favorite drag queen: Alyssa Edwards from Washington Examiner

Texas Democratic Senate nominee James Talarico named Alyssa Edwards as his favorite drag queen. The name refers to Justin Dwayne Lee Johnson, whose stage name is Alyssa Edwards. The drag performer has been outspoken on LGBT issues, especially in support of youth. Talarico was approached by a man after a campaign event who said he […]

Source: Washington Examiner (lean-right, washingtonexaminer.com)

Crooked Media Acquires Nate Silver FiveThirtyEight Politic Podcast , Which Is Being Renamed Still Counting from variety.com

Crooked Media Acquires Nate Silver FiveThirtyEight Politic Podcast , Which Is Being Renamed Still Counting

Source: variety.com (unrated, variety.com)

Sneaker Politics Opens New Houston Store , Plots Denver Location from wwd.com

Sneaker Politics Opens New Houston Store , Plots Denver Location

Source: wwd.com (unrated, wwd.com)

Hispanic Democrats feel less safe discussing politics than Republicans , My Code survey reveals from laopinion.com

Hispanic Democrats feel less safe discussing politics than Republicans , My Code survey reveals

Source: laopinion.com (unrated, laopinion.com)

Talk Business & Politics : Talk Business & Politics - Hendrix College poll analysis from yahoo.com

Talk Business & Politics : Talk Business & Politics - Hendrix College poll analysis

Source: yahoo.com (unrated, yahoo.com)

Wim Wenders Clarifies Stance on Filmmakers and Politics After Berlin from variety.com

Wim Wenders Clarifies Stance on Filmmakers and Politics After Berlin

Source: variety.com (unrated, variety.com)

It a Numbers Game : BREAKING NEWS : Lindsey Graham Passes Away Suddenly | T | 960 WELI | The Clay Travis and Buck Sexton Show from 960weli.iheart.com

It a Numbers Game : BREAKING NEWS : Lindsey Graham Passes Away Suddenly | T | 960 WELI | The Clay Travis and Buck Sexton Show

Source: 960weli.iheart.com (unrated, 960weli.iheart.com)

Paul Ryan : Our Politics Are Just Fundamentally Unserious Right Now from breitbart.com

Paul Ryan : Our Politics Are Just Fundamentally Unserious Right Now

Source: breitbart.com (unrated, breitbart.com)

Cool , stormy start this week in Mississippi , heat to return by weekend from yahoo.com

Cool , stormy start this week in Mississippi , heat to return by weekend

Source: yahoo.com (unrated, yahoo.com)

Politician should not be a dirty word from centralmaine.com

Politician should not be a dirty word

Source: centralmaine.com (unrated, centralmaine.com)

Mashed Out : Katie Porter Says She Done With Politics Following Failed Gubernatorial Campaign from townhall.com

Mashed Out : Katie Porter Says She Done With Politics Following Failed Gubernatorial Campaign

Source: townhall.com (unrated, townhall.com)

GOP senator calls for subpoenaing Donald Trump Jr. and Hunter Biden to question them on their business dealings from NBC News

A Republican senator is calling for an investigation into the business dealings of Donald Trump Jr. and Hunter Biden following the revelation that a Russian oligarch paid for Trump Jr.’s post-wedding festivities in the Caribbean earlier this year.

Source: NBC News (unrated, nbcnews.com)

Rubio’s meme-worthy job collection takes new turn with permanent Trump upgrade from Fox News

Trump signed a commission making Secretary of State Marco Rubio his permanent national security adviser, a dual role unseen since the 1970s.

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

Why are Hungary and Ukraine fighting again? from RT

Hungary has rejected Ukraine’s ‘Carpathian Eight’ initiative, after Kiev refused to honor its promises to its Hungarian minority Read Full Article at RT.com

Source: RT (unrated, rt.com)

Former Today host Billy Bush lands new role 10 years after VERY high-profile firing from Dailymail.com

The 54-year-old is embarking on a new career move.

Source: Dailymail.com (unrated, dailymail.com)

US-sanctioned international court judge urges Ireland to take a stand from The Irish Times

Nicolas Guillou said the European project was in grave danger if tariff fears outweighed the importance of fighting crimes against humanity

Source: The Irish Times (unrated, irishtimes.com)

Gerard Barron, Metals Company Chief, Makes the Case for Seafloor Mining from NYT

The head of the company hoping to become the first to conduct commercial deep-sea mining said mines on land had a much larger environmental footprint.

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

Post-eruption imaging of the Hunga caldera reveals that small, shallow undersea volcanoes can trigger disproportionately large tsunamis from Ars Technica Science.

New seafloor surveys of the Hunga Tonga–Hunga Haʻapai caldera, taken after its January 2022 eruption, show the volcano’s collapse carved a deep, steep-walled pit roughly 850 meters across. The imaging, collected by a research vessel using multibeam sonar, maps the caldera’s current shape in unprecedented detail, including terraced walls and a central depression that formed when the volcano’s summit sank into the emptied magma chamber.

The data tie the tsunami’s size to the collapse’s speed and geometry. As the caldera floor dropped, it displaced a massive column of water in a single, sudden motion—not a gradual subsidence. That rapid vertical displacement, the authors argue, generated a wave that traveled across the Pacific, reaching heights of up to 20 meters on nearby islands and triggering warnings as far away as Peru and Japan.

What stands out is the scale mismatch: the eruption’s explosive power alone couldn’t account for the tsunami’s reach. The collapse, not the blast, was the primary driver. That finding challenges older models that focused on shock waves or pyroclastic flows as the main tsunami source for shallow submarine eruptions.

The study also notes that the caldera’s walls are now unstable, with ongoing slumping and potential for further collapse. That leaves the site—and similar small, shallow volcanoes elsewhere—as a lingering hazard, since even a modest future eruption could repeat the same outsized wave generation.

Source: arstechnica.com

A community-led effort to replace the terminated US climate report has posted its first paper, a research roadmap, from Ars Technica Science.

The National Climate Assessment, the federally mandated report on climate impacts across the US, was killed earlier this year. In response, researchers launched a crowdsourced replacement, and the first contribution has now been published in the open-access collection: a roadmap laying out the research questions and data gaps the new effort should prioritize. The paper doesn't provide new findings—it defines the scope and structure for future submissions.

The roadmap identifies key areas needing immediate attention, including regional downscaling of climate models, economic damage estimates, and health impacts that the original NCA had begun to cover. It also flags the problem of inconsistent state-level data, which the community project will need to harmonize across sources. The authors stress that the collection is designed to be living—additional papers can be added as researchers complete analyses.

The effort faces practical hurdles: without federal coordination, ensuring quality control and avoiding duplication falls to volunteer editors. The roadmap proposes a peer-review workflow similar to journal standards, but with faster turnaround to keep pace with policy deadlines. Whether enough researchers commit time—and whether federal agencies will cite the collection—remains open questions. For now, the first paper marks a concrete start to filling a bureaucratic hole with open science.

Source: arstechnica.com

Russel Vought is reportedly set to gain veto power over all NIH grants, a move opposed by the NIH director, from Ars Technica Science.

The Trump administration is reportedly preparing to give Russel Vought, the Office of Management and Budget director, final say over every grant the National Institutes of Health awards. That would place a political appointee above the agency’s scientific peer-review process, which currently gates funding through independent panel scoring.

NIH Director Dr. Monica Bertagnolli has objected to the arrangement, according to the report. Her opposition highlights a growing rift between career scientists and administration officials who have pushed for tighter control over federal research spending, particularly in areas like infectious disease and public health.

Under the reported plan, Vought would review individual grant decisions—not just broad funding levels. That scope of authority would let a single non-scientist override recommendations from NIH’s thousands of external reviewers and program staff, reshaping how roughly $47 billion in research is allocated each year.

The article notes the move is pending, not yet finalized, but that the administration has already signaled it wants to shrink the NIH’s independence. If enacted, it would effectively end the agency’s tradition of letting scientific merit, not political preference, decide which biomedical projects get taxpayer money.

Source: arstechnica.com

Lake Powell hit a record low after a dry winter, with emergency water releases from another reservoir and reduced downstream flows stabilizing it, from Science Daily.

Lake Powell, one of the largest U.S. reservoirs, fell to its lowest level on record this winter. An exceptionally dry and warm season left the Colorado River Basin with scant mountain snowpack, so the usual spring melt never arrived to refill the reservoir. Lake Mead, the other major reservoir on the river system, also dropped to record lows.

Federal officials responded with emergency measures to keep Lake Powell operational. They transferred water from Flaming Gorge Reservoir upstream and cut releases downstream to BuRec’s turbines. The moves aim to preserve the reservoir’s ability to generate hydropower and maintain water deliveries to the 40 million people who depend on the Colorado River.

The record lows follow decades of structural over-allocation and a two-decade drought that has tightened the basin’s water budget. Both reservoirs sit at roughly a quarter of their full capacity, leaving little cushion for another dry year. The emergency actions are stopgaps, not fixes, for a system already running on borrowed time.

Source: sciencedaily.com

A Yellowstone super-eruption would follow weeks of escalating warning signs, not a surprise blast, according to a worst-case scenario outlined by researchers and reported from Science Daily.

The scenario begins with migrating earthquake swarms, rapid ground uplift, shifting geyser activity, and rising volcanic gases — a detectable, months-long preamble. Those precursors would culminate in a VEI 8 event, the most explosive category on the volcanic explosivity index.

The eruption itself would send ash tens of kilometers into the atmosphere, generate deadly pyroclastic flows, and bury parts of the western United States under thick ashfall. That ash would then cripple transportation, electricity grids, water systems, agriculture, and communications across much of North America — a continental-scale infrastructure failure, not just a regional disaster.

Source: sciencedaily.com

A new LHC data search found no microscopic quantum black holes, narrowing the possible parameter space for extra-dimensional theories from Science Daily.

The analysis combed through collision data from the LHC’s runs, looking for the distinctive decay signatures that quantum black holes would leave behind. None appeared. That null result doesn’t kill the idea, but it does rule out a specific range of masses and coupling strengths where such objects could have shown up.

The search targets a class of black holes that could only exist if extra spatial dimensions ramp up gravity’s strength at subatomic scales. In that scenario, the LHC’s proton collisions would occasionally produce tiny black holes that evaporate almost instantly via Hawking radiation. Their absence tells theorists where not to look next.

Because the LHC is now shut down for its high-luminosity upgrade, this result stands as the final word on that parameter space from the current data set. Future runs at higher collision energies will probe deeper, but for now the gap between the LHC’s reach and the Planck scale remains the hunting ground for any quantum-gravity signature.

Source: sciencedaily.com

Free Grand Theft Auto 6 shirts from Tokyo Game Show 2026 are being resold online for hundreds of dollars, from IGN.

Attendees at Tokyo Game Show picked up the promotional GTA 6 shirts at no cost, exclusively at the event. Within hours, listings appeared on resale platforms with prices reaching into the hundreds of dollars.

The shirts were a limited, event-only giveaway, which drove demand among fans who couldn't attend. Resellers capitalized on that scarcity, flipping the free merch for significant profit.

No further details on the exact design or quantity of shirts were provided. The quick turnaround from free handout to high-price listing mirrors a common pattern for exclusive convention merchandise.

Source: ign.com

The MSI Codex Z2 gaming PC with a GeForce RTX 5070 Ti is now $1,699 after a $600 discount from IGN.

The pre-built desktop, which usually retails for $2,299, has been cut to $1,699. This price undercuts the current market cost of the graphics card alone, which is running around $1,000 or more on its own.

The system pairs the RTX 5070 Ti with a 14th-gen Intel Core i7 processor, 32GB of DDR5 RAM, and a 2TB NVMe SSD. That configuration puts the whole PC within striking distance of the GPU’s standalone price, so the rest of the hardware effectively comes cheap.

Stock for RTX 50-series GPUs has been tight, with cards frequently selling above MSRP on shelves and secondary markets. A complete rig at this price removes the hunt for a card, though buyers should verify the discount is still active before heading to checkout.

Source: ign.com

Walmart's GeForce Week sale offers some of the year's best prices on gaming laptops with RTX 5060 through 5080 GPUs, from IGN.

Walmart's current GeForce Week promotion has dropped prices on gaming laptops equipped with NVIDIA's RTX 5060, 5070, and 5080 GPUs. IGN reports these are the lowest prices seen all year for these specific models, positioning the sale as a competitive alternative to typical Black Friday discounts.

The deals span a range of laptops, from mid-tier RTX 5060 machines to higher-end RTX 5080 configurations. The pricing appears targeted at buyers who missed November sales or want current-gen hardware without waiting for holiday markdowns. No specific models or dollar amounts were cited, but the scope covers multiple GPU tiers.

GeForce Week is a recurring promotional window from NVIDIA's retail partners, and this iteration lands ahead of the usual holiday shopping rush. For anyone tracking laptop GPU pricing, the sale represents a rare moment where mid-cycle discounts match or beat seasonal lows.

Source: ign.com

McDonald’s is testing ads for other brands inside its drive-thru, adding more commercial interruptions while customers wait to pull forward, from Kotaku.

The fast-food chain is piloting a new system that serves audio and visual ads from outside advertisers during the pause between ordering and reaching the pickup window. The spots are designed to fill the dead time while customers inch forward, turning a previously quiet moment into another sales opportunity.

DETAIL: The move extends McDonald’s existing in-store digital screens and kiosk promotions into the drive-thru lane, where a significant share of its business happens. Early reports suggest the ads are being tested at select locations, with content tailored to the time of day and local traffic patterns.

DETAIL: For customers, it’s another layer of media wedge into an already ad-saturated commute. The company frames it as a way to keep the drive-thru experience “entertaining,” but the underlying pitch is straightforward: monetize every second of customer dwell time.

Source: kotaku.com

Xbox’s restructuring is being framed by analysts as a slow-motion reverse acquisition, with Halo’s move to Activision called a demotion and the upcoming Project Helix seen as a must-succeed bet, from Kotaku.

Analysts describe Xbox’s current restructuring as a “reverse acquisition in slow motion,” meaning the company is effectively ceding control of its core assets—like Halo—to Activision, a studio it technically owns but now operates under a different banner. The shift of Halo development to Activision is widely viewed as a downgrade, not a neutral transfer.

The pressure point is Project Helix, a title that analysts say “absolutely must hit.” If it fails, the logic goes, Xbox’s post-restructuring identity loses its anchor, leaving the platform’s portfolio without a guaranteed tentpole. The commentary stops short of predicting doom, but the tone is stark: this move is a bet on a single project to justify the reshuffle.

Source: kotaku.com

A Tokyo Game Show demo of Lollipop Chainsaw 2: Back2Back left a Kotaku writer feeling the sequel lacks the original’s cult-hit charm.

Kotaku’s hands-on at TGS found the game mechanically competent but tonally scattered. The demo mixed hack-and-slash combat with odd mini-game detours and a generic rock soundtrack, nowhere near the original’s deliberate grindhouse pastiche. Juliet’s new voice and demeanor read flat, missing the deadpan cheer that made her iconic.

The story setup — a time-traveling resurrection with a new partner named Back — never lands its jokes. The script leans on callback quotes from the 2012 original without building new comedic beats. One early boss fight hints at potential, but the surrounding levels feel like a focus-tested action game wearing the old character’s skin.

Technical polish was present, but the demo suggested designers focused on systems over personality. The original thrived on chaotic, referential energy; this build felt like a corporate attempt to replicate that vibe without understanding it. No release date was announced, and the demo didn’t offer much confidence for fans holding out hope.

Combat itself has some new dodge-and-counter layers, and the chainsaw’s special moves still satisfy. But the level design felt linear and repetitive, with environmental props that looked like they belonged in any modern action title. The demo ended with a tease that could lead anywhere — which, given how unfocused the playthrough felt, is a concern rather than a hook.

Source: kotaku.com