Anthropic employee claims general proof of the Riemann Hypothesis [pdf]
Summary not available.
Hacker News
Summary not available.
Content is being fetched
Content is being fetched
Content is being fetched
Content is being fetched
Content is being fetched
Content is being fetched
Content is being fetched
In an initial safety trial conducted in the United States, a genetically engineered adenovirus has successfully halted the growth of pancreatic tumors in three patients. Led by Masato Yamamoto at the University of Minnesota, the treatment involves injecting the virus directly into tumors, where it replicates specifically within cancer cells by targeting the COX-2 enzyme, causing the cells to burst and die. While the early results show clinically stable disease despite the administration of only one-tenth of the intended dose, experts like Kai Brown from Royal North Shore Hospital caution that the findings are preliminary and require larger, controlled trials to confirm efficacy. Future research aims to combine this viral therapy with immunotherapies, such as checkpoint inhibitors, to potentially stimulate a broader immune response against metastatic cancer.
Content is being fetched
Content is being fetched
The Investigatory Powers Tribunal has rejected the UK government’s attempt to maintain secrecy regarding a Technical Capability Notice issued to Apple. The ruling marks a significant victory for transparency after the government repeatedly refused to confirm or deny the existence of an order used to demand changes to iCloud encryption. The dispute stems from a 2025 attempt by the government to gain access to encrypted data, which led Apple to withdraw its Advanced Data Protection feature for new users in the United Kingdom to avoid building "backdoors." While the tribunal did not rule on the legality of the underlying order itself, it concluded that revealing the notice's existence would not damage national security, clearing the way for further legal scrutiny into the Investigatory Powers Act 2016.
Frankenstein, PBC is a natural-philosophical research and deployment company specializing in the reanimation of lifeless matter into artificial persons. Led by founder Victor Frankenstein, the firm seeks to transition mankind through transformative reanimation to banish disease and conquer death, despite acknowledging a one in ten chance that their technology could end the human race. The company is currently undergoing a public offering of 825,000 Class A shares (symbol: ADAM) to fund its operations, even as it struggles with the loss of control over its primary product, Creature v1.0, which is currently pursuing the founder across the Arctic ice. While the company operates as a Public Benefit Corporation, it faces significant risks, including high mortality rates among associated persons, a lack of stable revenue, and the unpredictable, self-aware nature of its reanimated subjects.
Summary not available.
Content is being fetched
In an excerpt from his book The Pen: A Human History, Simon Garfield explores the evolution of writing fluids, from the ancient recipe for iron gall ink—made from oak galls, ferrous sulphate, and gum arabic—to the industrial success of the Stephens Ink Co. Founded in 1832 by Henry Stephens, a medical doctor and former roommate of poet John Keats, the company revolutionized the industry with its "indelible writing fluid." The brand became a global standard for official documents, famously utilizing a distinctive ink blot logo and aggressive advertising. Despite the company's dominance and its adaptation to the fountain pen era under Henry Charles Stephens, the family empire eventually ended in 1967 when it was acquired by the Royal Sovereign Pencil Company.
Summary not available.
Researchers at Fermion Research have conducted a comprehensive causal survey of the Qwen3-8B language model, mapping its internal neurons to identify specific behavioral "switches." By applying a "push" to individual neurons, the team discovered that approximately one in a hundred neurons acts as a discrete dial for specific traits, such as language (e.g., German), persona (e.g., a pirate or a cat), and stylistic registers. The study found that these switches primarily control who is speaking rather than what is being said, as no single neurons were identified for behaviors like lying, flattery, or manipulation. Furthermore, the research demonstrated that these neurons can be used for precise model editing; an AI agent was able to successfully improve reasoning performance on the MATH-500 benchmark by targeting and editing specific weight directions.
Summary not available.
The Cornhole News feed features a collection of recent headlines across the Midwest, covering a wide range of political, environmental, and local interest stories. Key reports include the U.S. Army's plan to destroy old ammunition in Iowa, which may create a toxic cloud, and an Illinois man facing terrorism charges for a plot against a Hindu temple. Other significant news items involve Kansas Governor-elect Cindy Holscher's Medicaid campaign, Minnesota State's long-standing financial struggles, and environmental updates such as the rebuilding of bighorn sheep populations in Badlands National Park and rare American eel sightings in the Minnesota River. Local developments range from construction at the Ft. Laramie tunnel to urban redevelopment in Lincoln and school building closures in IPS.
Summary not available.
Content is being fetched
Content is being fetched
Content is being fetched
Content is being fetched
Content is being fetched
Summary not available.
The AI model Claude Fable 5.2 has reportedly solved two major open problems in theoretical physics concerning black holes and quantum gravity. Working through a collaborative network of autonomous agents, the model completed a complex, 20-year-old calculation regarding the Hawking-Page transition, proving that the abrupt phase transition from a gas of particles to a black hole persists even in regimes where spacetime loses a definite shape. Additionally, Claude applied a mathematical tool known as a spectral curve to the modular bootstrap method, identifying a pattern in numerical data missed by humans to calculate the precise, optimal ceiling for the mass of the lightest black hole in 3D gravity. These findings, verified using the Lean proof checker, represent significant autonomous research contributions to the effort of reconciling general relativity with quantum mechanics.
Content is being fetched
Content is being fetched
Summary not available.
Content is being fetched
ChatGPT for SNES is an unofficial experimental project designed by Benj Edwards of BX Foundry and developed using AI agents to create a chat-style program for the Super Nintendo Entertainment System. Running on a Super FX 2 chip, version 0.7 utilizes a small language model with approximately 2.2 million parameters that generates text one word at a time. While the base version is roughly 4 MB, an expanded build includes a 150,000-article Simple English Wikipedia fact book, bringing the total size to 19 MB and allowing the model to perform lookups. Users can experience the software via a browser-based emulator or on original hardware using a flash cart with Super FX and MSU-1 support. Despite its novelty, the developer notes the model is significantly smaller than standard ChatGPT, possesses a vocabulary of about 8,000 words, and is prone to errors.
Harrison Was Here reports on overcoming Chamberlain's 2023 decision to restrict third-party API access through the myQ app by implementing a localized hardware solution. To regain autonomy over his garage door operations, the author installed two ratgdo32 WiFi control boards from ratcloud llc, which utilize open-source firmware to communicate via the LiftMaster Security+ 2.0 protocol. This new setup allows for seamless integration with Apple's HomeKit for voice commands through Siri and avoids reliance on cloud services. Furthermore, by leveraging the devices' JSON status endpoints and Tailscale networking, the author developed a custom monitoring system that sends notifications to Slack, providing more granular data—such as which user opened the door—than the original manufacturer's application.
Content is being fetched
To enhance software reliability and detect interface issues early, GIMP has implemented a new Graphical User Interface (GUI) testing platform developed by Harsh Verma. Unlike previous unit tests or legacy code-hijacking methods, these new tests interact with the real interface by launching a functional instance of GIMP to perform tasks such as opening dialogs, managing layers, and using tools like Bucket Fill. Currently consisting of 13 automated tests running on Linux (both X11 and Wayland), the system utilizes the Python tool dogtail to navigate elements via accessibility information rather than pixel matching, which simultaneously improves software accessibility for screen readers. Because the tests operate within small, hidden desktops, they can run automatically on CI servers without requiring a physical screen. The project contributors, including Jehan and Michal Odehnal, have invited the community to expand the test suite by writing new tests for various GIMP dialogs and tools.
Researchers from Wiz Research, along with Ikotas Labs, Metnir, Rapid7, and Xint, have discovered two critical unauthenticated remote code execution vulnerabilities in NGINX, tracked as CVE-2026-123633 and CVE-2026-123007. Both flaws carry a CVSS 4.0 score of 9.2 and specifically affect NGINX configurations where HTTP/2 is enabled. The use-after-free vulnerability (CVE-2026-123633) and the integer underflow vulnerability (CVE-2026-123007) could allow attackers to compromise servers, steal sensitive data, or disrupt services regardless of whether the underlying application code is secure. F5, the company behind NGINX, has released security updates to address these long-standing memory-safety issues, and experts recommend that organizations upgrade to the latest NGINX OSS or NGINX Plus releases immediately or temporarily disable HTTP/2 as a mitigation.
As research into the gut microbiome expands, interest in faecal microbiota transplants (FMTs) is growing, with potential applications ranging from treating Clostridioides difficile to managing autism and mental illness. Finding suitable donors is a highly selective process; experts like Carl Crawford and Susy Hota note that only a tiny fraction of applicants pass rigorous screenings for infectious diseases, lifestyle, and microbial diversity. Scientists are particularly interested in "super poopers"—donors with an exceptionally high abundance and variety of beneficial bacteria, such as Faecalibacterium prausnitzii—whose samples may lead to higher remission rates for conditions like ulcerative colitis. While current treatments involve direct transplants or encapsulated pills, the future of the field likely lies in standardized, lab-grown microbiome-based therapies.
Content is being fetched
Content is being fetched
Content is being fetched
Evidence, a Y Combinator-backed company specializing in business intelligence as code, is seeking a Founding GTM Lead to join its team in Toronto, ON. Founded in 2021 by Sean Hughes and Adam McAskill, the startup offers Evidence Core, an open-source framework, and Evidence Studio, an enterprise platform integrating AI analytics agents. The new hire will work directly with the co-founders to build a scalable growth engine, focusing on technical content creation, product marketing, customer storytelling, and marketing infrastructure. The ideal candidate is a skilled writer with a technical inclination and 1–3 years of experience, capable of leveraging AI tools to drive customer acquisition and establish a marketing function from the ground up.
Nintendo has issued a DMCA notice resulting in the removal of the Picofly modchip firmware from GitHub. The legal action has drawn attention because the notice claims copyright infringement despite assertions that the firmware contains no actual Nintendo code.
Apple is reportedly developing an agentic AI assistant capable of performing complex, multi-step tasks such as browsing the web, writing code, and working across various applications and APIs. According to a now-removed job listing first spotted by Bloomberg, the company is seeking engineers to lead the creation of these agents, which would allow Apple to compete with existing technologies like OpenAI’s models, Google’s Gemini, and Meta’s Muse. This initiative appears to be an extension of Apple's efforts to enhance Siri, with upcoming updates intended to allow the AI to coordinate specialized sub-agents and integrate more deeply into third-party applications.
Content is being fetched
In a recent blog post, Meredith Whittaker announced that Signal is implementing new measures to prevent AI agents integrated into mobile operating systems from capturing and storing sensitive user data. Following previous efforts to block Microsoft’s Recall on Windows 11, the Signal Technology Foundation is now restricting agentic systems in iOS and Android by preventing screenshots when AI assistants attempt to read the screen. The company warns that the integration of AI capabilities like Google Gemini and Apple’s Siri AI into major operating systems poses an "existential threat" to privacy, as these agents can scan and store application content outside of Signal’s control. Signal is calling on Apple, Google, and Microsoft to provide better technical tools and documentation that allow developers to easily opt out of AI data access without compromising core application utility.
Content is being fetched
Content is being fetched
Content is being fetched
Content is being fetched
Content is being fetched
Content is being fetched
Content is being fetched
Onlook is an open-source, AI-first design tool and local-first editor specifically engineered for designers and visual engineers to build interfaces using real code. Functioning as a visual-first canvas, it allows users to run their actual Git repositories locally, viewing live applications on an infinite canvas with various device frames. Rather than relying on a proprietary model, Onlook enables users to bring their own AI agents, such as Claude Code or ChatGPT/Codex, to edit source files directly via chat. The desktop application, built with Electron and React, supports exporting artifacts as HTML, PNG, JPEG, or PDF and is designed to work seamlessly with any web framework supported by the user's local development server.
Content is being fetched
In a recent article by Therry van Neerven, the author advocates for using zram to combat increasing memory demands in modern computing environments, particularly as AI increases the workload on workstations. While operating systems like Windows, macOS, and Fedora utilize memory compression, many Linux distributions like Ubuntu do not enable it by default. Although compression incurs a slight increase in CPU usage and latency, it can provide massive capacity gains, with zram achieving compression ratios as high as 1:4. Van Neerven demonstrates this efficiency by running a massive number of Firefox tabs, successfully compressing 5.7GB of data into just 1.7GB of RAM. Implementing this technique can optimize local hardware performance and significantly reduce Cloud expenditures by allowing more processes to run on less physical memory.
Content is being fetched
Author Fabio Manganiello has detailed a major overhaul of his self-hosted voice assistant pipeline within the Platypush framework to improve robustness and reduce false positives. The updated architecture replaces openwakeword with the assistant.microwakeword plugin, utilizing microWakeWord for more reliable local hotword detection. To enhance speech-to-text accuracy and end-of-speech detection, the system now integrates Silero VAD as a primary backend within a reworked AudioPreprocessor, which also includes adaptive noise floor tracking and hysteresis. The resulting pipeline utilizes assistant.openai for cloud-based transcription and processing, alongside tts.openai for responses, though local alternatives like assistant.vosk and tts.piper remain supported for privacy-focused or offline configurations.
RegExr, the tool developed by gskinner, is currently temporarily offline for essential maintenance. The developers announced the outage to address an aging codebase and mitigate risks associated with AI hacking, stating that the site will remain down until they can modernize and secure the platform to better protect user data.
In Computer Graphics from Scratch, author Gabriel Gambetta demystifies the core algorithms used to create the imagery in modern animated movies and video games. The book offers a broad introduction to the field, focusing specifically on the two primary techniques of raytracing and rasterization. By utilizing language-agnostic pseudocode and avoiding external libraries or APIs, Gambetta enables readers with basic programming skills and high school mathematics to build complex algorithms from scratch. The curriculum covers essential concepts such as light, shadows, reflections, perspective projection, and shading, complemented by interactive demos to bring the mathematical principles to life.
Summary not available.
Content is being fetched
A critical security vulnerability, dubbed Ruby2Shell (GHSA-34m4-m9fh-gqp5), has been identified in Ruby, involving a one-byte heap overflow that can lead to remote code execution. The flaw exists in the string case conversion process when handling specific single-byte encodings; specifically, upcasing the character ß writes two bytes when the system only expects one, allowing an attacker to overflow the allocated buffer. This vulnerability is reachable without authentication in common web frameworks like Ruby on Rails and Sinatra, and affects several popular self-hosted software bundles including GitLab, GitHub Enterprise Server, Mastodon, OpenProject, Discourse, and Chatwoot. Users are urged to upgrade to fixed Ruby releases 3.3.14, 3.4.13, or 4.0.10 immediately to mitigate the risk.
Creator Wilson Tovar (known as krthr) has introduced significant updates to fanzines.app, a browser-based tool designed to simplify the creation of 8-page zines from a single sheet of A4 paper. To eliminate the tedious manual layout process in software like Photoshop, the application automatically rotates and arranges pages into a print-ready, 300 DPI PDF. Recent improvements include a redesigned editor that reflects real physical spreads, a Spread view to preview facing pages, and the ability to batch-upload photos. The updated interface also features Fill, Fit, and Center tools, a Review step with safe margins to prevent printer clipping, and mobile compatibility, with the site now supporting both English and Spanish.
In his essay, Robert Peake explores the crisis facing software engineering and knowledge work due to the rise of Artificial Intelligence. He argues that automation is driven by the principle of "grading"—the ability to verify and score a task's correctness—rather than how "human" a task feels. While highly "gradable" and valuable tasks like coding and speech synthesis are being rapidly automated, Peake suggests that work involving high subjective value but low verifiability, such as poetry, fine art, or parenting, remains more resistant to digital automation. He warns that while commercial creative work (like copywriting and illustration) is under significant threat from market incentives, the future of human labor should shift from producing mere "deliverables" to achieving complex, ungradable outcomes. Peake advocates for a re-skilling focus on big-picture thinking, ethics, and judgment to navigate an AI-native world.
In a recent commentary, Giorgos Kallis explores the existential crisis facing academia as Artificial Intelligence destabilizes scientific funding and publishing. Drawing on insights from meta-science experts James Wilsdon and Ludo Waltman, Kallis highlights how LLMs are causing an exponential surge in grant applications and a deluge of rogue scientific papers, threatening to overwhelm traditional infrastructures. While Wilsdon suggests a transition to a "pull" model of funding managed by AI, Kallis warns that such a shift risks automating research away from human judgment and expertise. He critiques the current "quantity over quality" obsession and calls for collective academic organization to implement strict limits on AI use, arguing that preserving human intellectual labor and the ability to apply informed judgment is essential to preventing the total obsolescence of the scientific community.
The Frontier Design Benchmark, a collaborative initiative by Contra Labs and BenchFlow, evaluates the creative capabilities of frontier AI models using real-world, high-value client briefs from the Contra marketplace. Utilizing 4,311 blind head-to-head votes from 68 professional designers, the benchmark ranks models based on Elo ratings across 33 complex design briefs. Currently, the stealth model Gestalt holds the top position with an Elo rating of 1931, followed by Human golden and Claude Fable 5.2. To ensure explainable quality judgment, the benchmark incorporates a 35-dimension design critique rubric and identifies AI slop patterns, maintaining a high correlation (r = 0.94) between rubric scores and professional designer ratings.
Architect Labs has announced Frostbite, a novel general-purpose CPU architecture discovered, designed, and verified end-to-end by AI agent swarms. Moving away from traditional superscalar designs, Frostbite utilizes interaction nets to exploit inherent parallelism, where computation occurs through local graph rewrites, eliminating the need for complex cache hierarchies and coherence protocols. By co-designing the ISA, compiler, microarchitecture, and RTL autonomously, the AI achieved a 4.59x performance-per-watt increase over modern silicon-proven CPU cores across a variety of workloads. The development process relied on a formal, machine-checkable specification language called ASL and the Lean 4 theorem prover to maintain a provable source of truth across the entire hardware-software stack. This breakthrough demonstrates how AI can navigate vast design spaces to implement emergent architectural features, such as autonomous work-stealing mechanisms, which would be difficult for human architects to achieve manually.
The floppy3.5 project, hosted on GitHub by user bigcrimping, is an open-source hardware and software solution designed as a portable 3.5" floppy disk WiFi NAS. Aimed at users of vintage digital cameras like the Sony Mavica, the device reads photos from 1.44 MB FAT12 floppy disks and displays them via a web interface over WiFi. The system utilizes a custom PCB featuring an RP2354A controller and an ESP32-S3 bridge, allowing users to view files through a browser or copy them directly to an attached microSD card for expanded storage. The repository provides complete resources for DIY builders, including Gerber files for PCB fabrication, STL/3MF files for 3D-printed housing, and firmware for both processors.
Content is being fetched
Content is being fetched
Content is being fetched
Content is being fetched
Content is being fetched
In an essay for the DeepMind Institute, research scientist Tim Genewein explores the transition from Artificial General Intelligence (AGI) to Artificial Superintelligence (ASI), moving the conversation from speculation to empirical research. Genewein posits that intelligence exists on a continuum, where ASI would outperform human experts across virtually all tasks of interest. He identifies "cognitive industrialization"—driven by massive increases in compute efficiency and hardware investment—as a potential catalyst that could commoditize cognitive work. The essay outlines four primary technological pathways toward ASI: scaling models and data, technological paradigm shifts beyond current transformer architectures, improving agent coordination through collective intelligence, and recursive self-improvement (RSI), where AI automates its own advancement. Ultimately, Genewein warns that because these pathways are interconnected, slowing progress in one area may not suffice, necessitating unprecedented interdisciplinary efforts in modeling, measurement, and governance to prepare for the profound societal shifts ahead.
Content is being fetched
Content is being fetched
Foolscap is a specialized writing application designed to act as an intelligent companion for authors by tracking the structural continuity of their work in real-time. Rather than generating content, the tool analyzes the manuscript to identify connections and unresolved elements, surfacing these insights using the author's own words. The platform aims to provide deep context and awareness of a whole book while ensuring it never writes a word on behalf of the creator.
Content is being fetched
Content is being fetched
Content is being fetched
In an interview at AMD In-Depth, Mahesh Subramony, Corporate Vice President of Design Engineering at AMD, discussed the architectural innovations behind the upcoming EPYC Venice processors. A key shift in the design is the transition to 2.5D packaging to accommodate a massive increase in data bandwidth, reaching 1.6 TB/s, which necessitates splitting the I/O die into two mirrored dies to optimize power efficiency and die-to-die connectivity. Subramony also detailed the differences between the SP7 and SP8 platforms, noting that SP7 focuses on compute and memory bandwidth while SP8 prioritizes PCIe lanes for enterprise needs. Furthermore, the design of the 32-core CCD includes a 33rd redundant core to improve manufacturing yields through harvesting, and while the 12-core CCD features larger L3 cache that may introduce slight lookup latency, the increased hit rate is intended to offset those costs for high-performance workloads.
Google DeepMind has introduced Gemini 4 Neon, a new frontier-level AI model designed for high-frequency, complex tasks in software engineering, knowledge work, and scientific research. Positioned within the Gemini 4 family to balance depth and efficiency, Neon excels in multi-step instruction following, production-ready coding, and advanced analytical reasoning, ranking highly on benchmarks like the Artificial Analysis Intelligence Index and Agent Arena Index. The model features a 1 million-token context window and is optimized for cost-effective, high-performance workflows, including specialized capabilities for cybersecurity defense via the Fairwind Program. Gemini 4 Neon is rolling out immediately across the Gemini API, Google AI Studio, Google Antigravity, and the Gemini app for Pro and Ultra subscribers.
albertnetymk has submitted a pull request to the openjdk/jdk repository to address a security vulnerability in the AVX-512 Base64 decoder. The bug, identified as issue 8392044, occurs within the StubGenerator::generate_base64_decodeBlock() function during the x86_64 AVX-512 VBMI path. Specifically, when handling invalid inputs with a length of less than four bytes, the padding path causes the output_size to decrement below zero, resulting in an incorrect output_mask. This flaw leads to the stub storing data past the logical output length, which can corrupt the heap. The proposed fix involves implementing an early return to Java if the input length is less than four.
Summary not available.
Storage market forecasts indicate a divergent recovery timeline for SSDs and hard drives due to differing manufacturing cycles and contract structures. According to analysts from TrendForce and financial reports from Seagate and Western Digital, SSD prices are expected to begin a gradual decline in the second half of 2027 as NAND flash supply aligns with demand. In contrast, hard drive prices are unlikely to see significant relief until 2028 at the earliest, as major manufacturers have already allocated most production capacity to data centers through long-term contracts and have not planned new factory expansions. While SSD markets respond more quickly to chip supply shifts, hard drive relief will likely stem from increased density via HAMR technology rather than immediate price corrections. Consequently, consumers are advised to buy storage only when necessary for work or backups, while speculators should consider waiting for the SSD market to stabilize.
Lastbook is a free, open-source, and private digital library designed for Mac, Windows, and Linux computers to assist users in surviving and rebuilding society during emergencies. Operating entirely offline once installed, the platform provides access to critical resources including medical guides from organizations like the World Health Organization and the Red Cross, maps featuring over 258,000 searchable places, and instructional videos. The software offers various data sizes—ranging from 7.7 GB to approximately 388 GB—and includes an optional, private local AI powered by Gemma 4 26B to answer questions based on the library's contents. Key topics covered include food preservation, water treatment, first aid, emergency preparedness, and critical repairs.
Summary not available.
In a Google Summer of Code 2026 project, developer Iason Karaprodromidis has introduced native Fortran support to the LLDB debugger within the LLVM project. This development addresses a significant tooling gap for the Flang compiler by implementing core components such as a DWARFASTParser, a specialized TypeSystem, and a Language Plugin. The update enables LLDB to accurately represent and display primitive types (INTEGER, REAL, LOGICAL, COMPLEX), pointers, functions, and subroutines. Notably, the project implements complex support for Fortran arrays, handling column-major memory layouts, explicit arrays, assumed-size arrays, and dynamic arrays through synthetic children providers to manage runtime-defined bounds and strides. While the project successfully integrated case-insensitivity via DWARF attributes, future work is planned for strings, derived types, and full expression evaluation using Flang.