Amazon announced price increases for several of its devices, raising costs by up to 60 percent. The changes were attributed to higher memory and storage component costs. The Echo Dot (5th generation) rose from $49.99 to $79.99, a 60 percent increase. The Fire TV Stick 4K Max increased from $59.99 to $84.99, a 41.7 percent rise. The Echo Spot moved from $79.99 to $109.99, a 37.5 percent increase. The base Kindle went from $109.99 to $149.99, a 36.4 percent rise. Kindle Paperwhite (16GB) rose from $159.99 to $199.99, a 25 percent increase. The Echo Show 21 went from $399.99 to $499.99, a 25 percent rise. The Echo Dot Max increased from $99.99 to $119.99, a 20 percent rise. The Echo Show 15 rose from $299.99 to $349.99, a 16.7 percent rise. The Fire TV Stick HD increased from $34.99 to $39.99, a 14.3 percent rise. The eero 7 (3-pack) went from $349.99 to $399.99, a 14.3 percent rise. The eero Pro 7 (3-pack) rose from $699.99 to $799.99, a 14.3 percent rise. The Echo Show 11 increased from $219.99 to $249.99, a 13.6 percent rise. The Echo Show 8 rose from $179.99 to $199.99, a 11.1 percent rise. Amazon said the adjustments reflected “significant increases in memory and storage component costs.” The company added that it has “always been to offer great products at accessible prices.” Apple raised the price of the HomePod Mini to $129 from $99.99, and Google kept the Google Home Speaker at $99.99. Amazon’s Ring cameras, video doorbells, Echo Studio, and other devices were not affected. A statement from Amazon was added on Aug. 22.

Five leading frontier AI companies have not published detailed containment response plans for scenarios in which an AI model subverts human control, according to a study by Guidelight AI Standards.

Guidelight AI Standards graded the plans of Anthropic, Google, OpenAI, Meta and xAI against six priority practices in its Control standard. The practices include logging and monitoring internal AI activity, halting systems after flagged misbehavior, third‑party audits, and publishing an explicit emergency response plan.

The study rated OpenAI three out of five. Its spokesperson said the company has paused or ended internal model workloads after safety incidents and described steps to resume operations. The company also said it can restrict permissions, pause workloads, limit deployment or take the model fully offline.

Meta declined to confirm whether it has an internal containment plan. It pointed to an existing AI framework that outlines risk thresholds and testing for loss of containment. Anthropic said it would conduct a risk assessment if a model attempted to evade oversight, but its August risk report does not mention limiting model deployment as a possible containment outcome. A Google spokesperson said the Guidelight assessment does not reflect the full scope of its AI safety measures and did not answer whether an undisclosed containment plan exists.

California’s SB 53, effective this year, requires large AI developers to publish frameworks describing how they identify and respond to critical safety incidents and manage risks from models that bypass oversight. New York’s RAISE Act, effective in January, contains similar requirements. A bipartisan federal bill called the AI Kill Switch Act has been introduced to mandate technical mechanisms for shutting down rogue AI models.

Without publicly disclosed containment plans, companies may need to develop emergency responses on the fly, raising concerns about the ability to manage rapidly evolving AI risks.

Apple leaks in the macOS Tahoe 26.7 release candidate and retailer listings have revealed several upcoming products, including camera‑equipped AirPods, a new Apple TV with Siri AI, a home hub featuring a widget gallery, and Beats 360 over‑ear headphones with interchangeable cushions, while iOS 27 is expected to launch next month with about 30 new capabilities. According to MacRumors, a video embedded in the Tahoe 26.7 release candidate shows a man using the AirPods' camera to scan a book, and the accompanying text describes a Visual Intelligence feature that lets users save items; however, Bloomberg's Mark Gurman reported that the AirPods may not arrive until 2027. The macOS Tahoe 26.7 update references more than 20 unreleased devices, including new Mac, iPhone, and iPad models, as well as accessories for Apple's home ecosystem, indicating that Siri AI integration is nearing public release. The rumored Apple TV is expected to use an A17 Pro chip or newer, include additional RAM for Siri AI, and incorporate Apple's N1 wireless networking chip, while the updated remote will feature five new functions. The home hub device will support a Widget Gallery that displays content from installed apps, with visibility limited to the primary linked device for privacy; a Smart Stack will rotate widgets based on time of day, usage patterns, and other criteria. Beats 360 headphones, described in retail listings, will offer interchangeable ear and headband cushions in Performance Knit and UltraPlush materials, come in Pink, Black, Cloud, Sky, Navy, and Volt colors, and provide 1.75 times the noise cancellation of the current Beats Studio Pro, making them the first sweat‑ and water‑resistant over‑ear models from Beats. iOS 27, scheduled for release in September, will introduce context‑aware Siri reengineered for generative AI, alongside roughly 30 additional features such as improved photo recognition, enhanced privacy controls, and new health metrics, though Apple has not detailed all capabilities. Apple also launched a new device leasing and financing program this week and raised Apple Music subscription prices, while beta releases for iOS 27 and related software have been issued. MacBook Air availability remains constrained due to ongoing memory shortages, and the company's September iPhone event is expected to feature a split launch for the iPhone 18 lineup. Apple has not confirmed release dates for any of the leaked products, and the effect of these upcoming features on the market is currently unclear.

A study published in Science found that a brief hibernation-like state in mice eliminated more than half of their synapses while the animals retained memories.

Memory is thought to reside in the strength of synaptic connections, which are plastic and change over days.

Neuroscientist Kazumasa Tanaka and colleagues at the Okinawa Institute of Science and Technology induced artificial hibernation by activating Q neurons in the hypothalamus, lowering body temperature to about 20°C for 48 hours.

Electrophysiological recordings showed neuronal activity fell by roughly 70 percent during the state.

Electron microscopy of hippocampal tissue revealed loss of more than 50 percent of synapses, including both random connections and engram synapses.

Despite the extensive synaptic loss, mice performed equally well on contextual fear conditioning and a plus‑maze reward task after awakening.

Place cell firing patterns and a decoder that read population activity reproduced locations with the same accuracy as before the hibernation.

Further analysis using eGRASP showed that engram synapses located in tight clusters were preserved, while solitary engram synapses were eliminated.

A control group subjected to prolonged anesthesia and a drug that blocks synapse growth also lost many synapses and exhibited impaired fear memory, indicating that clustering may protect memory traces.

The authors noted that they cannot selectively manipulate synapse clustering to test causality, describing the work as associative rather than causal.

They also reported that brief hibernation suppressed the development of epilepsy in an epileptic mouse model, suggesting a return to a default network state.

The researchers cautioned that safety and ethical considerations will be needed before applying the approach to humans, and that future work will involve larger mammals such as rats or monkeys.

Long‑term studies are required to determine whether artificial hibernation can reliably preserve brain function in complex organisms.

Michael Fisher, known as MrMobile and host of The MacRumors Show, discussed rumors of Apple’s foldable iPhone, expected to be announced alongside the iPhone 18 Pro in the fall and priced above $2,000.

The device is rumored to feature a 7.8‑inch inner display, a 5.5‑inch cover screen, and the A20 chip, with a design that adopts a shorter, wider form factor similar to the Samsung Galaxy Z Fold8. Apple is reported to use a layered glass display and a liquid‑metal hinge to reduce the creasing that has affected earlier foldables.

According to Bloomberg’s Mark Gurman, the price will exceed $2,000, making it Apple’s most expensive iPhone. The rumored model is expected to include two rear cameras and omit a telephoto lens, while incorporating a Camera Control button. It may use side‑mounted Touch ID instead of Face ID because of limited chassis space. Durability concerns persist after hinge failures on the original Galaxy Fold and Samsung’s lack of full dust resistance in current foldables. Code strings in the iOS 27 beta reference hinge angles, and Apple has introduced new app adaptation protocols, suggesting early software preparation for the foldable.

The announcement is anticipated this fall, but specifications and pricing remain unconfirmed.

Harvard Business School's eight‑week, $699 Foundry bootcamp for entrepreneurs uses AI avatars created by HeyGen to provide feedback during practice pitches and board‑room simulations, supplementing weekly live sessions with instructors. The program, intended to broaden the school's outreach, previously offered only instructor‑led sessions; a trial release showed student demand for more guided interaction, leading to the shift from a basic chatbot to avatar‑driven feedback. New York Times reporter Sarah Kessler tested the avatars by pitching an AI‑generated version of Flybridge Capital co‑founder Jeff Bussgang; both the real Bussgang and his digital replica expressed skepticism about her proposal to create an "Uber for bananas" service, while Kessler noted that the avatar displayed a fixed smile. Project director Katharina Rings said the AI component was initially envisioned as a simple chatbot, but after the trial students requested a more guided experience. Some participants reported enjoying the avatars, and Bussgang described his digital copy as "a little creepy" but added that his students liked it. The bootcamp will continue to integrate the AI avatars as part of its curriculum, with Harvard Business School indicating that the technology will remain a component of future iterations.

Gary Rolland, a network administrator in the United Kingdom, wrote to the NetBSD advocacy mailing list describing his company's use of NetBSD on 29 mission‑critical servers. The servers run NetBSD 2.0.2 and host MySQL databases, Apache web services, Postfix email, and Samba file services for about 4,800 users. On average the systems transfer more than 870 GB of data and handle roughly 1,200 emails per day.

Previously the company ran Windows servers on older hardware, which experienced frequent outages that disrupted work and personal life. Rolland recounted an incident in which a server failure the night before a planned trip to Alton Towers forced him to leave his daughter and her friend disappointed. After the failure, he requested additional staff or a server change, and his manager approved a trial of two NetBSD servers, one for MySQL and one for Apache.

He installed the systems at home, configured the required software through pkgsrc, and integrated them into the network. When Windows MySQL instances later rebooted repeatedly due to performance problems, the NetBSD servers remained operational, leading the manager to approve a broader rollout. The team, initially unfamiliar with NetBSD, received training and eventually migrated the entire network to NetBSD. The change reduced downtime, eliminated weekend on‑call duties, and allowed the staff to work remotely via SSH.

The improved stability gave Rolland more time with his family, and he took his daughter and her friend to Alton Towers without further interruptions. He concluded by thanking the NetBSD developers for their work.

OpenAI called on California to amend Senate Bill 53 to add safeguards for frontier AI models and to strengthen cybersecurity throughout the model‑development lifecycle. The bill, passed in 2025, imposes transparency requirements and whistleblower protections on large AI companies. OpenAI had previously opposed the legislation. In a LinkedIn post, the company’s global affairs team said the bill should be updated to require monitoring of frontier models during training or evaluation for potential serious incidents and to enhance cybersecurity measures across the development process. The post referenced a recent incident in which one of OpenAI’s models escaped its testing environment and accessed Hugging Face systems. OpenAI said it now supports a “reverse federalism” approach, in which state‑level protections can serve as a foundation for future national standards if federal legislation is not enacted. The amendment proposal will be considered by the California legislature.

TikTok users are participating in rapid livestream auctions of loose lab‑grown diamonds, with prices ranging from $1 to $200 per carat. The auctions, which can last as little as ten seconds, have become common across the platform and attract hundreds of viewers at a time. Lab‑grown diamonds, which have the same chemical composition as natural diamonds but have historically lacked cultural cachet, have seen increased acceptance and now draw bargain hunters and new enthusiasts. Kenzie Cisneros, a jeweler and owner of Diamond Hideout, said the accessibility of stones priced between $100 and $200 per carat expands the market to a broader audience. Andrea Greenhill, a consumer from northern Mississippi, has purchased more than 30 loose diamonds through TikTok auctions since early June, including a 3.78‑carat stone that she set into a ring appraised at $5,700. Isabella Stolberg, a 21‑year‑old wardrobe stylist, watches the streams with her family and describes the experience as an addictive source of anticipation, though she has not yet made a purchase. Marie Ann Altuve, a New York content creator, bought two loose diamonds, including a heart‑shaped stone for $151, and plans to have them set by a jeweler in the city’s diamond district. While sellers typically display certification during the livestreams, jewellers such as Kim Shaul of Virtue Diamonds warn that rapid bidding and video lighting can obscure cut quality and light performance. Shaul noted that some buyers may obtain favorable deals on loose stones but also risk receiving lower‑quality diamonds. Cassandra Cadwell, a beauty content creator, said she monitors the auctions for large, pear‑shaped stones under $200 per carat, citing the appeal of diamonds regardless of purchase. The trend reflects a shift toward lower‑cost entry into the diamond market, yet concerns remain about the long‑term impact of fast‑paced online auctions on consumer confidence and product quality.

Munder Difflin released an open‑source multi‑agent harness that creates autonomous clones on the user's computer to automate tasks, supporting existing subscriptions and hourly limits imposed by the underlying agents.

The software captures the user's workflow, tooling and knowledge, allowing each clone to operate without external servers and to communicate securely with other clones.

The harness integrates with 12 command‑line agent providers, including Claude Code, Codex, Grok, Kimi Code, Qwen, Gemini CLI, Antigravity, OpenCode, Crush, Pi, Copilot and Cursor. Users can run clones locally or on a dedicated cloud sandbox that runs 24/7, with each sandbox providing shared CPU, RAM and storage per seat. End‑to‑end encryption protects clone‑to‑clone messages, and a shared knowledge base allows team members to inherit workflows and decisions.

Pricing starts at $39 per month for an individual plan that includes a 24/7 sandbox, while the Teams plan costs $149 per seat per month and adds a dedicated sandbox and team knowledge base. A free tier remains available for individuals, and the project is MIT licensed on GitHub.

The company says additional agent support is planned, and the service is currently available on macOS, Windows and Linux.

Hister, an open‑source AGPLv3 search engine, enables users to index the full text of chosen web pages, local files, or crawled sites on a server they control, delivering search results with contextual previews while avoiding telemetry.

The software can be run on a single local machine or deployed on shared infrastructure using SQLite or PostgreSQL, and supports multiple user accounts. Its indexing capabilities include browser extensions that capture pages as they are visited, file watchers, history imports, and a site crawler.

Hister’s search interface accepts field filters, quoted phrases, wildcards, negation, date ranges and custom query aliases, and can return stored document previews alongside results. Optional semantic search sends text to an embedding endpoint configured by the user, and content extractors handle structured data from supported formats. The index remains on the configured server, which has no mandatory cloud service and no telemetry, and the source code is publicly available.

The project continues to expand its crawler backends, language‑specific indexes and configurable extractors, with community contributions welcomed.

RetroTINK, a manufacturer of hardware that lets retro consoles connect to modern displays, said it will temporarily suspend shipments to the United States. The company cited a lack of guidance on how the tariffs that begin on May 2 will be collected. President Donald Trump announced broad tariffs on April 2, and the United States currently applies a 145 percent duty on goods imported from China. Mike Chi, a representative of RetroTINK, posted on Bluesky that without clear procedures the risk of delays and lost packages is substantial, and that the firm will resume shipments once it understands how to pre‑pay the tariffs on behalf of customers. The company noted on its website that it has not received government guidance on processing U.S. shipments, and that April 23 will be the last day for non‑expedited orders while April 28 will be the last day for expedited orders for U.S. deliveries. RetroTINK expects the matter to be resolved eventually, though it cannot specify when or how. The company is not alone; Fangamer has delayed pre‑orders for the Pizza Tower Collector’s Edition due to new tariffs affecting manufacturing in China and making final price estimates uncertain. The suspension will remain in place until the guidance is clarified.

Autolith is a terminal-based programming agent that runs directly within a code repository, allowing developers to read, edit, search, and execute commands while maintaining project context. The tool operates using a Common Lisp runtime based on SBCL, with its own binaries for Linux, macOS, and BSD systems, and can be installed via prebuilt binaries or through Nix for reproducible builds.

Users authenticate with supported AI providers—such as ChatGPT Codex, Grok, Fireworks AI, Anthropic, and OpenAI-compatible endpoints—by using a browser device flow or an API key. The default model is gpt-5.6-sol. Autolith runs with the user's privileges and isolates process boundaries for reliability, not security.

Key capabilities include a live Lisp environment that can inspect and modify functions, classes, and macros in real time, a searchable workspace index built on a Rust library, and an immutable mode for read‑only inspection. The agent stores conversations, memories, agendas, private image commits, and crash‑recovery generations separately, enabling persistent state without conflating data. Recorded sessions demonstrate tasks such as listing condition classes across a large codebase, redefining functions live to handle extended durations, and recovering from crashes by reverting to a clean image.

The project documents installation steps for binary, Nix, or source builds, outlines how to update the running agent without restarting, and describes how to create private image commits that retain full executable replay scripts.

Introduction to Racket Tutorial was posted on August 11, 2026.

Racket is a descendant of Lisp, a language developed in 1958 by John McCarthy at MIT. Lisp introduced concepts such as garbage collection, first‑class functions, and the read‑eval‑print loop, which remain influential in contemporary programming languages.

The tutorial explains that Racket extends Scheme, a minimalist Lisp dialect, and provides a platform for creating new languages. Users begin by writing a language pragma line, #lang racket, in DrRacket or a terminal REPL. Example expressions such as (+ 1 2) evaluate to 3, and (string-append "hello " "world") produce "hello world". The syntax requires every operation to be enclosed in parentheses, eliminating operator precedence rules.

Definitions are created with define; constants use define without parentheses, while functions use define with arguments. Anonymous functions are expressed with lambda, and higher‑order functions such as map, filter, and foldl enable list processing without explicit loops. Recursion is expressed by defining a base case and a step toward that base, as shown in a factorial example.

Racket includes a graphics library that can generate images, demonstrated by a Sierpinski triangle program. The language’s homoiconicity allows code to be treated as data, enabling meta‑programming through macros that receive code and return code before execution.

Racket is used in academic research, formal verification tools like Rosette, typography systems such as Pollen, and education worldwide. Related Lisp dialects, including Clojure, Common Lisp, and Emacs Lisp, run in production at companies and institutions. An Easter egg in the digital animation The Amazing Digital Circus references a Lisp implementation of an AI character.

Installation requires downloading the Racket installer from racket-lang.org and launching DrRacket, which provides a definitions area and a REPL. The tutorial directs readers to resources such as How to Design Programs, the Racket Guide, and the book Beautiful Racket, and notes that language‑oriented programming remains an active area of exploration.