Thursday, Oct 1, 2026

A developer released OpenDLSS, a bit-exact Vulkan reimplementation of Nvidia's DLSS 5 neural rendering network, with a WebGPU port, from hackernews.

The project reimplements Nvidia's DLSS-NR build 310.8.0—a 71-block Swin/ViT U-net with FP8 activations and 141 MiB of weights—entirely in Vulkan, matching the original byte-for-byte at all 75 block boundaries, not just the final image. It runs 241 dispatches per frame, hitting 2.8 ms at 768x768 and 29.3 ms at 3840x2160 on an RTX 4070 SUPER. The tool supports benchmarking, profiling, parity checks against captured fixtures, and kernel-by-kernel bisection for debugging.

The repo includes GLSL kernels using cooperative-matrix FP8 GEMMs, plus faster PTX kernels with mma.sync and split-K GEMMs, all generated by Python scripts. A separate WebGPU port in `ports/browser-webgpu/` runs the same network without tensor cores or FP8, taking 72 ms at 512x512—exactness preserved in the spec, not the hardware. The demo integrates the network into a Filament renderer with per-object motion vectors and reprojected temporal feedback.

Requirements are strict: Windows, an Ada-or-newer Nvidia GPU, and specific Vulkan extensions. The model directory expects weights in a manifest-driven layout that nothing in the repo produces. DLSS-SR is explicitly not implemented—this is only the generative re-rendering network, not the upscaler. Fixtures for parity testing aren't included either, so verifying exactness requires your own captures of the original.

Source: news.ycombinator.com, github.com

A project called Book of Shapes offers a collection of minimal, generative, and customizable SVG patterns, shared from hackernews.

The post highlights Book of Shapes, a resource where users can find and generate SVG patterns. The patterns are described as minimal in aesthetic and generative in creation, meaning they are produced through code rather than manual drawing. The "customizable" tag suggests users can tweak parameters to alter the output.

No article body accompanies the title, so the details are limited to what the headline implies. The project appears aimed at designers or developers seeking lightweight, scalable graphics for backgrounds, textures, or UI elements, likely with a focus on clean geometry and algorithmic variety.

The source is a hackernews submission, which typically indicates the project is open-source or at least developer-oriented. Given the absence of further context, the main takeaway is that a new tool or library for SVG pattern generation has been shared, inviting exploration for practical or experimental use.

Source: news.ycombinator.com, bookofshapes.com

The NRO has recently declassified details on the URSALA, RAQUEL, and FARRAH spy satellites, which intercepted radar signals from low Earth orbit from the 1960s through the 1980s, from hackernews.

The article — drawing on newly released NRO documents and shuttle payload disclosures — outlines how these suitcase-sized "hitchhiker" satellites, built by Lockheed under program names like P-11 and Program 989, rode into orbit attached to larger photo-reconnaissance craft. They spun to sweep dish antennas across the ground, collecting radar emissions and their sidelobes, which were recorded for later playback or direct relay to ground stations. Early versions were bespoke, cheap, and often lasted only months before atmospheric drag destroyed them; later standardized birds like URSALA and RAQUEL could operate for several years.

Originally serving national intelligence consumers, the program shifted in the 1970s to support tactical military forces — Army, Navy, and Air Force — marking a major doctrinal change for space-based ELINT. The small satellites remained necessary even as larger SIGINT platforms moved to geosynchronous and highly elliptical orbits, because those higher platforms could not cover the entire globe, especially southern or western hemisphere targets. Mission 7300 was the formal TALENT-KEYHOLE designation for this radar-signal collection work, distinct from the ocean-surveillance Mission 7100 and hosted-payload Mission 7200. By the 1980s, the NRO was already discussing merging those missions into a single new satellite class. Program managers named the satellites after movie actresses to avoid hinting at their true functions, a naming convention that continued through the program's four-decade run.

Source: news.ycombinator.com, thespacereview.com

The Bronze Age collapse is explained by the shift from bronze to locally produced iron, which eroded the elite military monopoly that had sustained ancient empires, from hackernews.

At the Battle of Kadesh in 1274 BC, Egypt and the Hittites fielded tens of thousands of soldiers and over five thousand chariots, making it the largest battle of its time. Both armies relied on bronze weapons, an alloy requiring tin mined in distant places like Afghanistan, Cornwall, or Iberia. That tin traveled thousands of miles through bandit-prone territory, keeping bronze costly and concentrated in the hands of imperial elites. Copper earlier had done similar work, giving early states a military edge over stone-using neighbors and driving political consolidation.

The collapse of the Bronze Age empires around 1200 BC—often blamed on catastrophes like droughts, invaders, and famines—misses the deeper question: why nothing comparable returned for centuries. The answer lies in iron. Because iron could be produced locally, ordinary farmers in former Hittite and Egyptian territories gained access to weapons that once were imperial privileges. That broke the elites' monopoly on violence and permanently shifted the political equilibrium, making large-scale empire-building far harder.

The article draws a parallel to the present: modern superpowers rely on exceptionally concentrated access to advanced military technology. But as that technology evolves, the same dynamic that ended the Bronze Age—diffusion of weapons capability eroding central control—could eventually disrupt today's geopolitical order.

Source: news.ycombinator.com, worksinprogress.news

New research shows brain waves are complex, swirling patterns tied to specific cognitive tasks, not just simple background signals, from hackernews.

A team led by Joshua Jacobs and Anup Das at the University of Chicago, publishing in Nature Communications (April 2026), recorded intracranial electrodes in epilepsy patients during memory tasks. They found concentric "source" and "sink" waves, plus rotating spiral waves, alongside the simple planar waves previously known. The spiral patterns appeared more often during a complex spatial navigation task than during a simple verbal recall task.

The finding builds on earlier work from Jacobs and Uma Mohan (Nature Human Behavior, 2024) showing waves moved either front-to-back or back-to-front across the cortex, correlating with whether the brain was encoding or recalling memories. The newer observations suggest the brain deploys different wave geometries for different computational demands — spirals may support more complex spatial processing.

Jacobs connected with mathematical biologist Bard Ermentrout, who had modeled similar patterns in mice. Ermentrout argues that prior planar wave measurements may have been partial views of larger rotating structures — like feeling only one side of a hurricane's wind without seeing the eye. Because electrode placement is limited by clinical need, the true prevalence of these patterns remains unknown.

The field is shifting from treating traveling waves as irrelevant byproducts — "the sound of the engine" — to viewing them as active mechanisms for coordinating cortical activity on a second-by-second timescale, a flexibility that fixed synaptic connections cannot provide.

Source: news.ycombinator.com, quantamagazine.org

A museum-goer documents modular industrial dashboards spotted across German and Polish museums, noting their design and interactivity, from hackernews.

The author visited several German and Polish museums and found recurring examples of pre-digital modular control panels, likely now replaced by software. These physical systems ranged from an air traffic control display at Munich’s Deutsches Museum to subway monitoring panels in Stuttgart’s Fernmeldemuseum, all featuring illuminated status modules, counters, and annotation plates. A railway museum in Nuremberg showed the same system with plates removed, revealing the internal construction.

At a tram museum in Stuttgart, a similar panel mixed DIN-standard typography with a Dymo-label aesthetic. Warsaw’s train museum used a slightly different but visually related system, including a large industrial switch toggling between day and night modes — the author unsure if it changed system operations or just panel illumination. The largest example came from a power plant control room at the DASA Museum in Dortmund, which had the widest variety of modules.

The author openly admits to not knowing the technical details of these systems and invites readers with expertise to share information. The post ends with a photo of a large, inviting rotary switch, expressing a desire to turn it regardless of its function. The tone is observational and appreciative, framing the panels as both historical artifacts and inspiring design systems.

Source: news.ycombinator.com, unsung.aresluna.org

Y Combinator-backed Magnitude launched an open-source, self-optimizing inference engine for AI agents, claiming up to 2x speedups over llama.cpp on local hardware, from hackernews.

Magnitude (YC S25) released an inference engine that compiles and tunes its kernels on the user's specific device before running models. The company benchmarks 92% faster decode on Apple Metal and 19% on CUDA versus llama.cpp, with 27% less memory per agent and shared prefix caches for concurrent sessions. It ships as a desktop app for macOS, Windows, and Linux, supporting Apple Silicon, NVIDIA, AMD, or CPU-only setups with no fixed minimum hardware.

The app includes a CLI and one-click integrations for agents like Pi, OpenCode, Hermes, Codex, and Claude Code, plus an OpenAI-compatible API for anything else. The company hand-writes kernels for popular open-weight model families rather than shipping precompiled binaries. Downloads require no internet after model acquisition, and everything stays local under an Apache 2.0 license. The project is soliciting GitHub stars to grow its developer community.

Source: news.ycombinator.com, github.com

1molchuan/edge-smart-doh is trending in typescript (87 stars) from GitHub

DoH that gives Cloudflare-hosted sites measured preferred IPs (per ISP, from cfhub) and ECH

Source: github.com

saeedvaziry/caleb-video-editor is trending in typescript (36 stars) from GitHub

This is a one-shotted slop that works!

Source: github.com

ZacharyZhang-NY/Ely-GPUI-Components is trending in rust (258 stars) from GitHub

A component library for GPUI, in light and dark. Every component runs live in the browser.

Source: github.com

rehan-remade/universal-modder is trending in python (1164 stars) from GitHub

Point Claude at any game. Skills, tools and the fal MCP that let Claude Code mod almost any PC game you own: recon, reverse engineering, fal-generated art/3D/audio, in-game testing, showcase videos.

Source: github.com

nanaism/yomiyasu is trending in python (745 stars) from GitHub

AI生成された文章の不自然さを取り除き、人間にとって自然な日本語へ書き直すためのAgent Skill

Source: github.com

OpSafari/hypoarena is trending in python (547 stars) from GitHub

Scientific hypothesis-discovery workbench: grounded hypothesis-evidence graphs, synthetic literature with planted causal chains, generate-debate-evolve loops over pluggable offline adapters, Elo tournaments recovering planted skill order, paraphrase dedup, Bayesian evidence accumulation, reproducible reports. NumPy core, CPU-only torch extra.

Source: github.com

OpenAI detailed how it disrupted a coordinated campaign that tried to extract protected model reasoning and outlined new defenses against adversarial distillation, from OpenAI Blog.

OpenAI detected and shut down a coordinated effort to distill the reasoning processes of its models, a technique where attackers use a model's outputs to train a competing model. The campaign involved multiple accounts and repeated attempts to extract chain-of-thought data, which OpenAI treats as protected. The company says it traced the activity to a single actor and terminated the accounts involved.

The attack relied on exploiting the model's behavior under certain prompts to get it to reveal its internal reasoning steps. OpenAI responded by deploying stricter monitoring around reasoning traces, adding filters, and implementing rate limits specifically designed to catch distillation attempts earlier. It also notes that the attacker never successfully exfiltrated large volumes of the protected data.

OpenAI is now building longer-term defenses, including watermarking model outputs to trace them back to API usage and improving automated detection of distillation-like patterns. The company frames this as part of a broader push to protect the value of its models, since reasoning chains are a key differentiator — losing them would let rivals cheaply replicate that capability.

Source: openai.com

OpenAI is partnering with America’s SBDC to expand hands-on AI training and local support for small businesses, alongside a new report on how small teams are using AI, from OpenAI Blog

OpenAI announced a partnership with America’s Small Business Development Centers (SBDC) to bring more hands-on AI training and local support to small businesses. The collaboration aims to make practical AI guidance more accessible to owners who may not have dedicated technical staff.

.

Alongside the partnership, OpenAI published a new report examining how small teams are currently using AI. The report looks at real adoption patterns among small businesses, offering a snapshot of the tools and workflows already in place rather than hypothetical use cases. The findings are meant to inform both the training rollout and future support efforts.

the field.

Source: openai.com

OpenAI announced GPT-6.1 Sol, a cheaper coding and computer-use model priced at one-fifth of Astra’s standard API rates, from OpenAI Blog.

OpenAI released GPT-6.1 Sol, a new model variant described as “near-Astra intelligence” for coding, computer use, and professional tasks. The headline change is pricing: Sol’s standard API input and output token costs are 20% of Astra’s—a direct cost cut, not a temporary discount.

The model targets practical workloads: code generation, browser or desktop automation, and general office work. OpenAI frames Sol as a middle tier between the flagship Astra and its other options, positioning it for teams that need high capability but face budget constraints on volume-heavy tasks.

No benchmark numbers, training details, or availability timeline were specified in the announcement. The launch reads as a commercial move to capture cost-sensitive developers and enterprise users, rather than a capability breakthrough—the “near-Astra” phrasing suggests intentional performance trade-offs for the lower price.

Pricing changes like this ripple through API-dependent startups and internal tooling, where token costs often dictate model choice. OpenAI is betting that a 20% price point turns Sol into the default pick for high-throughput automation, even if it doesn’t match Astra on every task.

Source: openai.com

OpenAI DevDay 2026 unveiled over 20 announcements, including GPT-6 Astra, updates to ChatGPT and Codex, new APIs, security features, and builder tools, from OpenAI Blog.

The keynote led with GPT-6 Astra, the next-generation model family, positioned as a multimodal leap with faster inference and native tool-use. ChatGPT gained deeper agentic workflows, while Codex expanded beyond code completion into autonomous repo-level maintenance and CI/CD integration. No pricing or availability dates were shared for the flagship model, but developer previews were hinted.

New APIs focused on retrieval, memory, and orchestration, with an emphasis on lowering latency for production workloads. A revamped security layer includes sandboxed execution and audit trails for agent actions. Builder tools now offer template pipelines and observability dashboards, aimed at shrinking the gap between prototype and deployment.

The announcements track a clear pivot: multimodal reasoning is now assumed, not optional, and the platform is betting on agents as the primary interface. For teams already on OpenAI’s stack, the upgrade path looks straightforward; for newcomers, the breadth of options might be the real cost. No roadmap for deprecating older models was mentioned, which leaves existing integrations in a wait-and-see position.

Source: openai.com

Hugging Face announced an open leaderboard for scalable evaluation of multilingual text-to-speech and voice cloning, based on the title from Hugging Face Blog.

The leaderboard introduces a standardized, scalable framework for benchmarking TTS systems across multiple languages, with a specific focus on voice cloning fidelity. It aims to address the fragmentation in current evaluation practices by offering a unified metric set that researchers can apply consistently. The platform likely aggregates results from submitted models, ranking them by quality and multilingual coverage.

No body text was provided, so the specifics of the scoring methodology, dataset composition, or participating models remain unspecified. The announcement itself signals a shift toward more rigorous, community-driven assessment in speech synthesis — a space where subjective listening tests have long dominated. A public leaderboard makes performance comparisons transparent and reproducible.

For practitioners, this means a clearer baseline for choosing a TTS engine or cloning tool, especially when working across low-resource languages. The emphasis on "scalable" suggests the evaluation pipeline can handle large numbers of models without manual intervention, which could accelerate iteration for both open-source developers and commercial vendors. The lack of technical details in the source leaves the exact implementation open, but the intent is unambiguous: standardize the yardstick for synthetic speech.

Source: huggingface.co

NVIDIA's Kumo Tabular model claims a new accuracy-efficiency frontier for tabular prediction, per the Hugging Face Blog.

The announcement centers on Kumo Tabular, a model purpose-built for tabular data — the spreadsheets and structured datasets that dominate enterprise machine learning but often lag behind text and image models in innovation. The headline claim is a simultaneous gain in predictive accuracy and computational efficiency, a combination that typically forces a trade-off in existing tabular methods.

The technical leap appears to come from architectural choices that reduce inference cost while boosting performance on standard benchmarks, though specifics on the model’s internals are thin in the title alone. NVIDIA positions this as a practical win for teams running large-scale tabular workloads, where speed-to-prediction directly impacts operating budgets.

The release targets a crowded field of gradient-boosted trees and newer transformer-based tabular models. If the accuracy-efficiency claim holds under independent testing, Kumo Tabular could shift default choices for production pipelines that currently accept slower inference for better scores — or lower scores for faster serving.

No benchmark numbers, dataset names, or baseline comparisons are provided in the article text. The entry serves as a product announcement rather than a technical deep dive, leaving validation to follow-up posts or papers.

Source: huggingface.co

Hugging Face Blog introduces source-aware verification for MCP agents, ensuring tool outputs are checked against their origin, not just factual accuracy, from Hugging Face Blog.

The piece proposes a verification layer for Model Context Protocol (MCP) agents that tracks where a claim originates — which tool, which endpoint, which dataset — before trusting it. This moves beyond binary fact-checking to assess whether the source itself is authoritative, stale, or spoofed. The approach addresses a real gap: agents often retrieve correct-looking data from compromised or outdated sources, and standard validation misses that failure mode.

Implementation-wise, the blog outlines attaching provenance metadata to each MCP tool response, then running verification rules against that metadata — checking for unexpected source drift, version mismatches, or untrusted domains. It frames this as a lightweight addition to existing agent pipelines rather than a full rewrite, with examples showing how a single forged API response could otherwise pass a factual check while still being malicious.

The underlying argument is that source integrity is a precondition for fact integrity — without knowing the origin, a “true” statement can still be dangerously wrong. The post closes by suggesting this become a standard part of MCP agent evaluation, alongside existing benchmarks, rather than an afterthought in production debugging.

Source: huggingface.co

Holo4, a new system for generalist computer-use agents, was introduced via Hugging Face Blog.

The announcement centers on Holo4, a framework or model designed to let AI agents operate across diverse software environments without task-specific training. The title positions it as a step toward "generalist" capability — meaning the same underlying agent can handle varied digital workflows, not just one narrow app or website.

No technical specifications, benchmark results, or architecture details were included in the available text. The post appears to be a high-level introduction, likely accompanied by code or a model card on the Hugging Face platform, but the actual substance — parameters, training data, or comparative performance — remains unspecified.

Given the source and framing, the practical takeaway is that Holo4 targets the bottleneck of computer-use agents: adaptability. If the release delivers on that promise, it could reduce the need for bespoke automation pipelines, but without concrete evidence in the provided text, those claims are speculative. The post functions more as a pointer than a deep dive.

Source: huggingface.co

Google released Gemini 4 Argon, its newest and most powerful AI model to date, positioning it for coding and cybersecurity work, from TechCrunch AI.

Google launched Gemini 4 Argon today, billing it as the company's most capable model yet. The release targets enterprise and developer use cases, with marketing emphasis on coding assistance and cybersecurity operations rather than general-purpose consumer features.

The model arrives amid an intensifying race in frontier AI, where OpenAI, Anthropic, and Meta are all shipping rapid iterative updates. Google's positioning leans on practical, high-stakes workflows — suggesting the company is betting that real-world performance in security and software engineering will differentiate it from flashier consumer demos.

No benchmark specifics or pricing details were included in the announcement, a notable departure from prior launches where Google touted raw scores. The sparse technical information leaves developers to gauge capacity through early access and official documentation, which could frustrate teams deciding between this and rivals' offerings.

Source: techcrunch.com

Flow Engineering raised a new round at a $750M valuation for AI agent-driven hardware design, with Roelof Botha joining as an angel investor and board member from TechCrunch AI.

Valor, Atreides, and Sequoia led the investment into Flow Engineering, which develops AI agents that automate parts of the hardware design process. The round values the startup at $750 million, a significant mark for a company operating in the relatively niche intersection of generative AI and chip/PCB engineering.

Roelof Botha, the Sequoia partner known for early bets on YouTube, Square, and Snowflake, came in as an angel and took a board seat — a personal endorsement that goes beyond the firm's institutional check. His involvement suggests Sequoia sees Flow as more than a tooling play, but a potential platform for how hardware teams will work in the AI era.

Flow's pitch is that hardware design is still manual, iterative, and slow compared to software, and that agents can handle routine layout, verification, and optimization tasks. The company hasn't disclosed revenue or customer names, but the investor mix — two firms plus a board-level angel — hints at strong conviction in the technical roadmap.

Source: techcrunch.com

OpenAI’s Decisions API, a Jev clone, signals the rising value of fast, cheap AI reasoning for agent coordination, from TechCrunch AI.

OpenAI has quietly shipped a “Decisions API” that mirrors Jev’s approach to rapid, low-cost decision-making. The tool is aimed at frontier labs managing swarms of AI agents, where each micro-decision currently burns compute and latency. By cloning Jev’s architecture, OpenAI is betting that leaner inference beats brute-force reasoning in agent-heavy workloads.

The API’s existence confirms a broader shift: intelligence quality is no longer the sole bottleneck — speed and cost per call now dictate how agents scale. Labs that once raced for bigger models are now racing for smaller, faster ones that can handle high-frequency orchestration tasks. Jev’s design, which favors distilled outputs and minimal token spend, is becoming the template.

For OpenAI, the move is defensive as much as offensive. Competitors like Anthropic and Google are already optimizing for agent swarms, and a clunky decision layer would throttle any multi-agent product. Shipping a Jev clone early lets OpenAI keep pace while its frontier models remain the headline act. The real test is whether the API can hold up under production-scale agent traffic — and whether other labs follow suit with their own clones.

Source: techcrunch.com

AI voice startup ElevenLabs doubled its valuation to $22 billion in a $300 million employee tender led by Wellington and T. Rowe Price, from TechCrunch AI.

ElevenLabs, the AI voice generation company, has seen its valuation jump to $22 billion, double the previous figure. The $300 million round was structured as an employee tender offer, not a traditional fundraising for operational capital. Wellington Management and T. Rowe Price co-led the purchase of existing shares, giving early employees and investors a liquidity event.

The new valuation marks a steep climb from the company’s last priced round, reflecting sustained demand for its synthetic voice technology and enterprise traction. ElevenLabs has expanded beyond text-to-speech into voice cloning, dubbing, and audio tools, competing with both startups and larger AI platforms. The tender allows insiders to cash out without diluting founders’ control.

Neither ElevenLabs nor the lead investors disclosed the exact share price or the number of sellers. The company’s growth has been rapid since its 2022 founding, with revenue reportedly scaling alongside a wave of AI voice adoption in media, gaming, and customer service. The deal closes a chapter of private-market enthusiasm for AI infrastructure, with ElevenLabs now positioned among the most richly valued voice-focused AI firms.

Source: techcrunch.com

SpaceXAI’s Grokipedia rolled out a v0.3 update with a new logo and redesigned homepage and live edits page, per its design lead, from The Verge AI.

The v0.3 refresh replaces the old homepage—which was essentially just a logo and search bar—with a busier layout featuring featured articles, a most-read list, and a live “latest edits” tracker. The featured section now shows spinable 3D book covers, while the most-read area has also been reworked.

The redesign follows Grokipedia’s recent return to accepting user edits, suggesting the team is pushing the platform toward a more functional, Wikipedia-like experience. SpaceXAI’s head of design, Benji Taylor, described the changes as a “newly refreshed Grokipedia” in a post, though the full extent of the live edits page tweaks wasn’t detailed in the article. The update lands as the service continues to iterate quickly on its AI-generated encyclopedia.

Source: theverge.com

Google announced Gemini 4 Argon, an AI model initially restricted to trusted cyber defenders, from The Verge AI.

Google unveiled Gemini 4 Argon as its next frontier model, claiming top-tier performance in software engineering, enterprise knowledge work (legal, finance), and cybersecurity defense. Koray Kavukcuoglu, Google’s chief AI architect, framed the model’s release as starting with a curated group of cyber-defender partners.

Access is deliberately limited for now. Kavukcuoglu said Google is part of the U.S. government’s voluntary pre-release model access process, with gradual expansion planned. The model already powers some of Google’s own products, per the announcement.

The practical effect: enterprises and developers won’t get hands-on time immediately, while security-focused organizations get first crack. That sequencing aligns with Google’s stated safety posture, but it also means real-world benchmarking outside its chosen cohort will lag.

Source: theverge.com

Meta and OpenAI are both preparing to launch dedicated AI hardware devices over the next year, starting with cutesy software agents to test consumer appetite, from The Verge AI.

Both companies are treating the coming year as a trial run for physical AI devices, a category that has so far flopped with products like the Friend pendant and Humane AI Pin. Rather than jumping straight to expensive hardware, they’re reportedly using playful software agents — the “AI Tamagotchis” of the title — to gauge whether people will form lasting attachments to digital companions before committing to dedicated gadgets.

OpenAI’s effort is tied to its ongoing collaboration with Jony Ive, the former Apple design chief, though specifics of the device remain under wraps. Meta’s plans are less detailed, but the shared strategy signals a bet that emotional engagement with software can carry over into hardware purchases. The approach sidesteps the “hardware is hard” problem by letting consumers prove demand first, at a time when public sentiment toward AI is souring — the failed devices of the past two years have already shown that novelty alone won’t sustain a product category.

Source: theverge.com

Trump met with Big Tech leaders and defended his AI approach by renaming the technology "Super Intelligence," downplaying safety concerns, and citing industry self-oversight, from The Verge AI.

At a Tuesday dinner with executives including Zuckerberg, Trump told reporters that his earlier dismissal of AI safety fears as a Democratic "hoax" no longer applies—because he now calls the technology "Super Intelligence." He framed the rebrand as a fresh start for policy, even as he offered no new federal guardrails.

Trump rejected calls for stricter government oversight, arguing US companies are "dominating China" and therefore don't need regulatory pressure. Instead of federal rules, he said the "smartest people in the world" would be "watching over each other"—a reference to the tech leaders seated at the meal, who he implied would self-police.

The Verge's coverage notes the awkward optics: a photo caption highlights Zuckerberg's "panicked look" when Trump called on him to speak. The article leaves open whether the executives actually endorsed Trump's hands-off stance or simply attended the dinner, but no commitments or policy details emerged from the evening.

Source: theverge.com

Google announced Gemini 4 Argon, a new frontier model with a 1M-token output limit and a $2/$10 per million input/output token price, now rolling out to select cyber defenders via the Fairwind Program from hackernews.

Gemini 4 Argon is already running inside Google, where teams report it beat a quantum algorithm baseline by 40% in minutes, freed 300 TiB of memory across data centers by autonomously applying optimizations, and is migrating C/C++ codebases to Rust — including an 800K-line Fuchsia Zircon kernel effort. In one concrete win, Argon agents rewrote 32K lines of SIMD code in libgav1 into safe Rust that runs 2.7x faster than the existing Rust port.

The model sets new state-of-the-art scores on DeepSWE v1.1 (77.9% for long-horizon software engineering), the Vals Index across finance, legal, and tax work, and AutomationBench (51.3%). It also leads on LVBench for long video understanding at 91.7%. Argon's output limit jumps from 64K to 1M tokens, which Google says lets the model sustain deeper reasoning in single trajectories.

For cybersecurity, Argon is being released without cyber guardrails to trusted defenders and internal Google teams. Wiz, using Argon through its Scan for Good program, found a critical vulnerability in healthcare software exposing sensitive patient data that prior frontier models missed. Google is participating in the U.S. government's voluntary pre-release model access process and will expand access to developers and enterprises gradually.

Source: news.ycombinator.com, blog.google

Oil Is Flowing From the Persian Gulf, but Prices Remain High. Why?

Left · NYT — Oil Is Flowing From the Persian Gulf, but Prices Remain High. Why?: Crude oil costs close to $100 a barrel because traders are worried hostilities could soon restart and the world is burning through its emergency stockpiles.

Center · The Hill — Calls for gas tax suspensions grow as Americans struggle with high prices: With gasoline prices hovering near $4.50 a gallon amid the Iran war, calls to suspend federal and state gas taxes are gaining momentum. Multiple states, including Indiana and Georgia this week, have either suspended or…

The side-by-side comparison wasn't available for this story; each outlet's own headline is shown instead.

Source: NYT (lean-left, nytimes.com), The Hill (center, thehill.com)

Texas Governor Race Puts Pressure on Andy Beshear and Democratic Governors Association from NYT

Allies of Gina Hinojosa want a Democratic group to support her bid for Texas governor — putting pressure on its chairman, Gov. Andy Beshear of Kentucky, who is weighing a run for president.

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

John Roberts Retreated to His Island. Supreme Court Critics Found Him. from NYT

After decades of potlucks and paddle-boarding on Maine’s midcoast, rising anger and protests have left the Supreme Court’s chief justice more isolated than ever.

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

5 Takeaways From the Debate for California Governor

Left · NYT — 5 Takeaways From the Debate for California Governor: Xavier Becerra, a Democrat, and Steve Hilton, a Republican, sparred over President Trump, taxes and immigration in their only scheduled meeting.

Left · The Guardian — Steve Hilton, Xavier Becerra clash in testy California governor’s race debate: The candidates running to replace Gavin Newsom sparred over taxes, immigration and regulating AI Xavier Becerra and Steve Hilton, the candidates running to be California’s next governor, clashed Wednesday over taxes, im…

Center · NPR — California gubernatorial candidates face off in debate: California's candidates for governor square off in a debate while the blue-leaning state grapples with affordability and other issues.

Right · New York Post — California governor race heats up as Becerra and Hilton clash on CNN: Steve Hilton relentlessly attacked Xavier Becerra over taxes and red tape during California's sole scheduled gubernatorial debate on Wednesday.

The side-by-side comparison wasn't available for this story; each outlet's own headline is shown instead.

Source: NYT (lean-left, nytimes.com), The Guardian (lean-left, theguardian.com), NPR (center, npr.org), New York Post (lean-right, nypost.com), New York Post (lean-right, nypost.com)

Hegseth Talks Tough as Republican Doubts About His Leadership Grow

Left · NYT — Hegseth Talks Tough as Republican Doubts About His Leadership Grow: Pete Hegseth has held onto his job by projecting a tough guy image, but his relationship with Republican lawmakers has never been worse.

Left · CBS News — Hegseth says he's cutting generals and admirals by 20%, ousting "woke": Defense Secretary Pete Hegseth said in a speech on Wednesday that the Pentagon is eliminating roughly 1 in 5 generals and admirals during the second Trump administration through firing or position cuts.

Left · The Guardian — Hegseth says he will cut number of US generals by 20% in anti-diversity speech – as it happened: This live blog is now closed. Hegseth confirms plan to cut 20% of US top military brass and rails at ‘beardos, weirdos and wimps’ Sign up for the Breaking News US email Rachel Leingang and Riley Messer Some of the bigge…

The side-by-side comparison wasn't available for this story; each outlet's own headline is shown instead.

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

Senate confirms Keith Sonderling as labor secretary

Left · CBS News — Senate confirms Keith Sonderling as labor secretary: The Senate confirmed Keith Sonderling to lead the Department of Labor on Wednesday, officially installing the nominee who has been leading in an acting capacity since Lori Chavez-DeRemer's exit in April.

Right · New York Post — Senate approves Trump’s pick to lead the Labor Department: Keith Sonderling, who was confirmed 47-41, has been the acting secretary since April and has served in multiple roles within the department over the course of his career.

The side-by-side comparison wasn't available for this story; each outlet's own headline is shown instead.

Source: CBS News (lean-left, cbsnews.com), New York Post (lean-right, nypost.com)

Republican-turned-Democrat aims to flip Kansas Senate seat blue from CBS News

Kansas has not sent a Democrat to the U.S. Senate since 1932 — but this year, the party hopes Adam Hamilton, former Republican who serves as pastor of a massive church, could flip the seat.

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

Trump deflects on audio of Paxton calling GOP convention a drag on polls from CBS News

Republican Senate candidate Ken Paxton reportedly told donors that "everybody's numbers are down" after a midterm convention in Dallas that touted both Paxton's campaign and President Trump's record.

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

Senate Democrats block bill to limit lawmakers' stock trades from CBS News

A Republican-led effort to approve a House-passed stock trading bill fell short in the Senate, with Democrats saying the restrictions don't go far enough and pushing back against a voter ID provision.

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

Tennessee governor halts executions after second botched lethal injection attempt this year – US politics live

Left · The Guardian — Tennessee governor halts executions after second botched lethal injection attempt this year – US politics live: Bill Lee pauses state’s remaining executions for the rest of the year after inmate Christa Pike survives two lethal injections Tennessee woman on death row taken to hospital after surviving two lethal injections America…

Center · BBC News — US death row inmate survives execution attempt after two lethal injections: Killer Christa Pike's lawyer says she is being given "life-saving measures" in hospital after two syringes of pentobarbital.

The side-by-side comparison wasn't available for this story; each outlet's own headline is shown instead.

Source: The Guardian (lean-left, theguardian.com), BBC News (center, bbc.co.uk), BBC News (center, bbc.co.uk)

ICE got ‘quieter’ – then killed a father in Maine. Now watch groups are shifting their tactics too from The Guardian

ICE-watchers get creative after the killing of Johan Sebastián Durán Guerrero, as Trump’s immigration crackdown takes a more secretive approach

Most of the time, the scene outside the Immigration and Customs Enforcement (ICE) facility in the small, coastal town of Scarborough, Maine, could be confused for a nature watch in “Vacationland”, the slogan on state license plates.

People sitting in lawn chairs under a tent. Birds chirping. The sound of the breeze moving through the trees.

Continue reading...

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

Trump officials abandon effort to investigate election threats, experts warn from The Guardian

Credible threats appear to have been overlooked as administration focuses on unverified claims of voter fraud

The Trump administration has abandoned key security bodies meant to investigate direct threats to US elections by foreign actors and domestic terrorists, according to former officials and a recent report.

While administration officials have been actively focused on unsubstantiated claims of widespread voter fraud by noncitizens, other credible election threats appear to have been overlooked, the officials say.

Continue reading...

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

En esta ciudad de Pensilvania, los dos partidos políticos buscan el voto latino from NPR

Los republicanos lograron avances con los latinos en el último ciclo electoral, pero la economía y la frustración por la política migratoria de Donald Trump han corroído ese apoyo.

Source: NPR (center, npr.org)

COMIC: How the census gives your state power (and what Trump wants to change) from NPR

The census is used to determine how much say your state has in Congress and the Electoral College. President Trump failed to alter it during his first term, but he's trying again ahead of 2030.

Source: NPR (center, npr.org)

Along U.S. Route 1, Maine voters worry about heating oil, gas and Canadian tariffs from NPR

Home to both Republican Sen. Susan Collins and Democratic candidate Troy Jackson, Maine's senate race has put rural Aroostook County in the spotlight.

Source: NPR (center, npr.org)

Trade tensions mean fewer Canadians flocking to South Florida from NPR

This South Florida community is typically a refuge for Canadians fleeing cold weather, but trade tensions between the United States and Canada mean fewer are coming.

Source: NPR (center, npr.org)

Cornell case spotlights 'confusing' web of systems for reporting campus assaults from BBC News

Victims often have to navigate both university and local law enforcement processes to find justice.

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

California bans child marriage, a practice still legal in 32 US states from BBC News

Gov Gavin Newsom called it a "long overdue measure to protect young Californians".

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

Two dead in Canada after vehicle hits pedestrians taking part in First Nations event from BBC News

Police said the driver "may have been experiencing a medical emergency" before hitting four people in British Columbia.

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

Senate adjourns until November with Republicans fearful of losing majority from The Hill

The Senate adjourned for the midterm election on Wednesday night despite having a big pile of unfinished business on its plate, as vulnerable Republicans were ready to get back to their home states to campaign amid growing fears that they may lose their majority in the upper chamber. Senators adjourned after voting to confirm Keith...

Source: The Hill (center, thehill.com)

10 Democrat-held House seats most likely to flip  from The Hill

As the midterm landscape grows more favorable for Democrats, Republicans are still poised for key pickups in the fight for House control. Forecasters have shifted a slate of House races away from Republicans in recent weeks amid voter frustration with President Trump and the GOP's handling of the economy, immigration and the war in Iran, fueling...

Source: The Hill (center, thehill.com)

GOP wrestles with rising data center backlash weeks before midterms from The Hill

Republicans are growing increasingly worried that data centers are a liability in the midterms, as concerns over higher power bills tied to the facilities turn into a pocketbook issue for voters. The debate over the issue was on full display Wednesday, after Senate Democrats blocked legislation championed by Sen. Jon Husted (R-Ohio) — who is...

Source: The Hill (center, thehill.com)

Fed's Warsh seeks to avoid forward guidance. His colleagues aren't listening from The Hill

Federal Reserve Chair Kevin Warsh has sought to move the central bank away from forecasting its policy decisions during his four-plus months in charge. But after the Fed raised interest rates earlier this month, multiple officials have indicated future hikes are coming. “If the economy evolves in a manner broadly consistent with my forecast, one...

Source: The Hill (center, thehill.com)

Pro-life group says mail-order abortion pills violate federal law, urges Trump admin to take action from Fox News

The Restoration of America Foundation urges Attorney General Todd Blanche to enforce the Comstock Act against mail-order abortion pill shipments.

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

Biden-era federal prosecutor hit with criminal investigation referral over alleged false statements to FBI from Fox News

A criminal referral sent to the Justice Department alleges ex-prosecutor Jina Choi misled FBI agents about her role in the IRL startup probe.

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

States leading nation on the biggest education battles gripping classrooms revealed in first-ever report from Fox News

The Defending Education report ranks Tennessee first among all 50 states on parental rights, DEI, gender identity and public school policies.

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

Teen's tragic killing by illegal immigrant becomes flashpoint in battle for control of Congress from Fox News

Sheridan Gorman's parents confronted Rep. Mike Lawler and Democrat Cait Conley over sanctuary city policies during a heated congressional debate.

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

Election crime arrests keep piling up as Dems insist it 'just doesn't happen' from Fox News

New federal cases charge noncitizens with voter fraud across multiple states, clashing with Democrats who oppose the SAVE Act and deny it happens.

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

GOP has shocking slim edge over Dems on affordability with ‘swing’ voters, conservative group finds from New York Post

Republicans have a surprising, yet slim advantage over Democrats on cost-of-living issues with "swing" voters in several critical states, a grassroots survey by a right-leaning group found.

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

Trump’s $100,000 fee for new H-1B worker visas blocked by second judge from New York Post

Democracy Forward, a left-leaning legal group that represents the plaintiffs, applauded the ruling.

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

El Sayed’s comparison between 9/11 attacks and U.S. response is ‘absolutely offensive’: Byron York from Washington Examiner

Washington Examiner chief political correspondent Byron York blasted Michigan Democratic Senate nominee Abdul el Sayed for his comparison of the Sept. 11, 2001, terrorist attacks and the United States’s response. “The idea of equalizing what was done on 9/11 by al Qaeda and what was done afterwards by the United States and other Western powers […]

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

Kristi Noem files for divorce from husband Bryon after ‘irreconcilable differences’ from Washington Examiner

Former Secretary of Homeland Security Kristi Noem filed for divorce from her husband Bryon Noem after 34 years of marriage. The divorce was filed due to “irreconcilable differences,” according to documents obtained by Dakota News Now.  Kristi Noem said she was “devastated” after allegations surfaced that her husband was a secret crossdresser and chatted online […]

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

A captivating new biography of the flawed fixer Roy Cohn from Washington Examiner

When Roy Cohn died of AIDS in 1986, Donald Trump was among the few to attend his burial at Union Field Cemetery in Queens. He also attended Cohn’s public memorial later that year in Manhattan, along with Rupert Murdoch, George Steinbrenner, Barbara Walters, and William F. Buckley. Apart from his legions of well-connected friends and […]

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

And just like that, Germany’s establishment went kaput from Washington Examiner vs Call them ‘Nazis’ all you want: The new German Right is winning the argument from Washington Examiner

The British diplomat and novelist, Harold Nicholson, tossed off an essay in 1929 containing a short riff on the three great capitals of Europe. “London is an old lady in black lace and diamonds who guards her secrets with dignity,” while “Paris is the woman in the prime of life to whom one would only […]

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

Maine university system strips ex-Sen. George Mitchell’s name from building, scholarship over Jeffrey Epstein ties from New York Post

UMaine trustees voted 9-5 to strip former Sen. George Mitchell's name from a campus building and scholarship due to his ties to Jeffrey Epstein.

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

Hey Pundits, Why Aren't You Pressuring Republicans To Move To The Center? from Crooksandliars.com

Democrats appear poised to win Senate seats in a number of states Donald Trump won in 2024 -- according to Real Clear Polling, they're leading by 2 points or more in Texas, Ohio, Michigan, Alaska, Georgia, and North Carolina. Trump won some of these states by…

Source: Crooksandliars.com (unrated, crooksandliars.com)

Senate hopeful's 'treading water' narrative crumbles under $500K campaign payouts to family from Fox News

Dan Osborn's campaign has paid over $500,000 to family members even as the Nebraska Senate candidate claims his family is "treading water."

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

The Lost Art of Picking up the Check, Always from esquire.com

Don’t pay for dinner expecting something in return. Pay for dinner because it’s the best thing a man can do.

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

Where Is Thomas Randolph from ‘The Widower’ Now? from TODAY

Where is Thomas Randolph from The Widower now? What did he do? Where is he serving time?

Source: TODAY (unrated, today.com)

Europe hardens on migration as nationalist support grows from Biztoc.com

European countries sought to harden migration rules in a bid to quell support for nationalist parties across the continent. Anti-immigration parties have been on the rise in both domestic politics and the European Parliament. Governments are trying to address…

Source: Biztoc.com (unrated, biztoc.com)

New Greens leader gets emotional as he pays tribute to mentor from ABC News (AU)

New Greens leader David Shoebridge gets emotional during his first interview as leader while paying tribute to his political mentor before critiquing the federal government over the cost of living.

Source: ABC News (AU) (unrated, abc.net.au)

Soros Nephew Says Family Fellowship Paid for El-Sayed’s Education from Freerepublic.com

Jeffrey Soros, the nephew of billionaire Democrat donor George Soros, told guests at a Monday night fundraiser that he has known Michigan Democratic Senate nominee Abdul El-Sayed since 2012, when he interviewed him for a family fellowship that later helped pa…

Source: Freerepublic.com (unrated, freerepublic.com)

Taiwan analysts say Trump-Xi summit produced no policy shift but real test lies ahead from Fox News

Taiwan analysts assess whether Xi Jinping secured private concessions from Trump, with arms sales and senior U.S. official visits as key tests ahead.

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

The gospel according to Andy Beshear from Salon

Kentucky's governor on his new book, the Good Samaritan and why he hasn't ruled out 2028

Source: Salon (unrated, salon.com)

The Chicxulub impact crater may have supported a thriving microbial ecosystem for millions of years after the dinosaur-killing asteroid struck, from Ars Technica Science.

New analysis of rock cores from the Chicxulub crater in Mexico’s Yucatán Peninsula suggests the impact site wasn’t just a sterile scar. The asteroid strike 66 million years ago fractured the bedrock, creating a deep hydrothermal system where superheated, nutrient-rich water circulated through cracks for an extended period.

Researchers studying the crater’s peak ring—the central uplift of shattered granite—found evidence that these hot fluids persisted for at least 5 million years, possibly longer. The chemistry and mineral formations in the cores point to sustained temperatures of 200–300°C, which would have driven chemical reactions capable of feeding microbes without sunlight.

The implication is that the very event that wiped out most life on Earth also created a new, isolated habitat. Similar hydrothermal niches have been proposed as cradles for early life on our planet—and possibly on Mars, where large impacts left comparable structures. The study adds a fresh twist to the crater’s legacy: not just a tombstone, but potentially a long-lived oasis.

Source: arstechnica.com

Human tissue can carry electrical signals, enabling a local network of implanted devices that use the body as wiring, from Ars Technica Science.

The article reports on a proof-of-concept where multiple small implants communicate by sending electrical pulses through surrounding biological tissue, rather than relying on radio waves or physical wires. The approach exploits the body's natural conductivity to route data between devices placed in different locations, effectively turning the torso or limbs into a transmission medium.

Tests demonstrated reliable signal transfer across a few centimeters of tissue, with low power consumption and minimal interference from normal bodily functions. The implants were designed to be compact and biocompatible, though the work is still in early stages — no long-term animal or human trials have been run yet.

The main advantage is energy efficiency: electrical signaling through tissue uses far less power than wireless radio, which is absorbed and scattered by the body. That could extend battery life for medical sensors or stimulators, but challenges remain in signal attenuation over longer distances and ensuring consistent contact as tissue moves or changes over time. The authors frame this as a stepping stone toward more distributed implant networks, not a replacement for existing wireless medical tech.

Source: arstechnica.com

Google developed a watermarking method for AI-designed proteins to aid biosecurity, integrated with a popular protein design tool, from Ars Technica Science.

Google researchers created a watermarking technique that embeds a traceable signal into protein sequences generated by AI, specifically targeting the widely used RFdiffusion tool. The watermark is designed to survive the typical post-processing steps scientists apply to AI outputs, such as mutation and synthesis, making it possible to identify whether a given protein originated from the AI system.

The method works by subtly biasing the AI’s sampling process during protein generation, encoding a hidden pattern in the amino acid sequence without disrupting the protein’s function or structure. Tests showed the watermark remained detectable even after heavy edits, and it does not degrade the protein’s stability or activity compared to unwatermarked versions.

The researchers aim to address biosecurity concerns by giving labs and platforms a way to trace malicious use of AI-designed proteins, though the system only works with RFdiffusion and requires the original generation parameters to verify the mark. The team also notes that the watermark could be removed by advanced adversarial attacks, but it raises the cost of misuse enough to deter casual or opportunistic bad actors.

Source: arstechnica.com

NASA’s SpaceX Crew-13 is set to launch four astronauts to the ISS on October 1, with a flight time of 7 hours and 50 minutes that would set a U.S. speed record for the trip from Science Daily.

The mission targets an October 1 liftoff, with the Crew Dragon capsule aiming to dock at the International Space Station just under eight hours later. If achieved, that transit time would eclipse the current fastest U.S. spacecraft journey to the orbiting lab, though the exact baseline for comparison isn’t specified in the release.

The four astronauts are slated to fly under NASA’s commercial crew program, continuing the regular rotation of crews to the station. The shorter trip is enabled by the Dragon’s automated phasing maneuvers, which cut the number of orbits needed before rendezvous.

No further details on the crew roster, launch window specifics, or backup dates were included in the brief announcement. The flight profile follows SpaceX’s recent trend of quicker crewed transits, though the 7-hour-50-minute mark would be a new milestone for U.S. missions specifically.

Source: sciencedaily.com

A review of over 265,000 people found cannabis use linked to higher rates of both committing violence and being a victim of it, though not proof of causation, from Science Daily.

The analysis pooled data from roughly 265,000 participants and found that cannabis users were about twice as likely to perpetrate violence compared to non-users. The same group also showed elevated rates of experiencing violence as victims, a dual association that complicates any simple cause-and-effect narrative.

Researchers stress the correlation is significant but stops short of claiming cannabis directly drives aggression. Other factors—like socioeconomic conditions, co-occurring substance use, or underlying mental health issues—could plausibly account for much of the link. The review’s size is its main strength, but its observational design means the direction of influence remains unclear; violence might precede or follow cannabis use in different cases.

The findings add to a mixed evidence base, where previous studies have shown both protective and risk-associated patterns depending on dose, frequency, and context. For policymakers, the takeaway isn’t a clean prohibition argument but a signal that heavy or problematic use may overlap with volatile environments—and that victimization, not just perpetration, deserves attention in harm-reduction efforts.

Source: sciencedaily.com

Astronomers found evidence that minutes-long X-ray flashes can come from neutron star collisions creating a magnetar, with one record event lasting nearly 10 minutes, from Science Daily.

The event in question was a brief but intense X-ray burst, far longer than typical gamma-ray bursts or short X-ray flares. Its ~10-minute duration is unprecedented, and the data points to a specific origin: two neutron stars spiraling into each other, merging into a hypermassive neutron star that briefly survived before collapsing into a black hole — or, in this case, leaving behind a highly magnetized remnant, a magnetar.

The key observational signature is the X-ray emission itself, which persisted and slowly faded rather than cutting off sharply. That behavior matches models predicting a newborn magnetar radiating away rotational energy after a merger. Previous short gamma-ray bursts had hinted at such remnants, but this flash is the first to show the post-merger X-ray glow clearly on its own, without the initial gamma-ray spike.

If confirmed, this gives astronomers a new, independent way to identify neutron star mergers — events that also produce gravitational waves and kilonovae. The flash was detected by a space-based instrument, and follow-up observations are now focused on pinning down the system's location and age to test whether the magnetar interpretation holds beyond the timing and spectrum. The record-breaking duration makes it a rare, clean example for studying how these extreme remnants evolve in the minutes after formation.

Source: sciencedaily.com

Netflix’s East of Eden, starring Florence Pugh, Christopher Abbott, and Mike Faist, earns a positive review from IGN.

IGN’s review praises the series as “beautifully shot and dramatically riveting,” calling it “definitely binge-worthy.” The adaptation of John Steinbeck’s novel leans into visual storytelling, with the review singling out the cinematography and the tension carried by the central cast.

Florence Pugh leads as the story’s emotional anchor, with Christopher Abbott and Mike Faist rounding out the key trio. The review doesn’t delve into plot specifics, but the overall verdict is clear: the project succeeds as both a prestige adaptation and a streaming draw.

The piece frames East of Eden as a standout in Netflix’s slate, suggesting the drama’s pacing and performances keep viewers engaged over multiple episodes. No caveats or notable criticisms are mentioned in the review.

Source: ign.com

LEGO PlayStation is now available to LEGO Insiders, with a free Astro Bot Set included, from IGN.

LEGO has released its PlayStation-themed set, but it's not on general sale yet. The set is currently exclusive to LEGO Insiders, the company's free loyalty program, and will include a complimentary Astro Bot set with each purchase.

To get the set, you need to be a LEGO Insider. Membership is free and requires only an account registration on LEGO's website. Once signed up, you can access the PlayStation set through the Insiders portal, where the bundled Astro Bot set is automatically added at checkout.

The Astro Bot set ties directly to PlayStation's mascot, making the bundle a straightforward crossover for fans of both brands. LEGO has not yet confirmed a broader retail release date, so Insiders access is currently the only way to grab it.

Source: ign.com

Marathon game director Del Chafe reassured fans that the extraction mode remains a core part of the game, despite recent roadmap additions of PvE and deathmatch modes, from IGN.

Last week, Marathon's developers published a roadmap introducing PvE and deathmatch content. That update raised questions among players about whether the extraction mode—the game's original draw—was being phased out or de-prioritized.

Game Director Del Chafe directly addressed those concerns, stating the extraction mode will not be sunsetted and remains a major part of Marathon. His comments are meant to clarify that the new modes are additive, not replacements.

The reassurance arrives amid broader uncertainty around Marathon's post-launch direction. Chafe's statement gives existing fans a concrete anchor while leaving room for the game to expand beyond its extraction roots.

Source: ign.com

Sega and Epic Games executives discussed how giving Sonic a gun in Fortnite has been a huge success, with players defeating Dr. Eggman 14 million times, from Kotaku.

In a joint interview, Sega President Shuji Utsumi and Epic Games Japan Managing Director Takayuki Kawasaki addressed the crossover's surprisingly strong performance. The Sonic-themed Fortnite event, which armed the blue hedgehog with a firearm, generated over 14 million player takedowns of Dr. Eggman. Both executives framed the collaboration as a win for expanding Sonic's brand reach into a younger, battle-royale-native audience.

Utsumi acknowledged initial internal hesitation about deviating from Sonic's family-friendly image, but the player response erased those concerns. Kawasaki noted that Epic's flexible character customization allowed Sonic to fit Fortnite's chaotic sandbox without breaking immersion. The pair cited the kill count as direct proof that players embraced the absurdity rather than rejecting it.

The executives also touched on broader lessons: successful crossovers now demand mechanical integration, not just cosmetic skins. Sonic's gun became a gameplay hook, and Eggman's boss-fight role kept players engaged beyond the novelty. Utsumi hinted Sega would apply this willingness to take risks in future partnerships, while Kawasaki stressed that player-driven metrics, not focus groups, would guide Epic's next collab choices.

Source: kotaku.com

Former PlayStation executive Shuhei Yoshida said Sony should extend the PS5 generation as long as possible, arguing most players aren’t demanding more console power, from Kotaku.

In a recent interview, Yoshida—who led PlayStation’s first-party strategy for years before leaving in 2019—said he doesn’t see a strong push from the player base for a PS6 anytime soon. He framed the PS5’s current horsepower as sufficient, and suggested Sony would do better to maximize the installed base and software library of the existing console before pivoting to a new generation.

Yoshida also noted the diminishing returns of iterative power jumps, pointing out that many mainstream titles already run well on PS5. He said the industry’s focus should shift toward polishing experiences and supporting developers rather than chasing specs that few consumers actively request. That stance puts him at odds with the typical industry cadence of a new console every six to seven years—the PS5 launched in 2020.

His comments arrive amid a slower-than-expected rollout of PS5 Pro-tier hardware and ongoing production of major exclusives that still target the base PS5. Sony has not confirmed any timeline for a next-gen console, but Yoshida’s view aligns with recent signals that the company may stretch this generation further than previous ones.

Source: kotaku.com

Naughty Dog is reportedly developing a new Uncharted game, per a Kotaku report from Kotaku.

The report, citing anonymous sources familiar with the studio’s plans, indicates the project is in early development. It’s unclear whether this is a direct sequel to *Uncharted 4: A Thief’s End* or a separate entry in the franchise, and no writer or director has been publicly attached. Nathan Drake’s involvement remains unconfirmed.

Naughty Dog has not issued a statement, and the studio’s current focus includes *Intergalactic: The Heretic Prophet*, its next original IP. The Uncharted report surfaces amid broader industry chatter about Sony reviving flagship single-player series, though this is the first concrete claim tied to Naughty Dog specifically.

Kotaku’s sourcing is thin — no internal documents or corroborating outlets are cited. If accurate, the project would mark the first mainline Uncharted since 2016, and a major shift for a studio that publicly moved away from the series after *The Last of Us Part II*.

Until Naughty Dog confirms anything, treat this as a rumor with a plausible shell but no verified core. The studio has historically denied or ignored similar reports, and its post-2020 output suggests it’s more interested in new worlds than revisiting old ones.

Source: kotaku.com