Midjourney has asked a federal court to compel Disney, Universal and Warner Bros. to disclose how they use generative‑AI technologies in its copyright lawsuit. The studios sued Midjourney in 2025, claiming that the startup’s image‑generation models reproduced copyrighted characters such as Bart Simpson and Darth Vader. A judge previously ordered the studios to provide AI‑related documentation only when it related to consumer‑facing videos or images. In its latest filing, Midjourney requested that the court expand discovery to include all internal AI documentation, prompts and outputs used for any purpose, arguing that the current limitation restricts the evidence it can use for its fair‑use defense. Studios’ lead attorney David Singer said the request amounted to a fishing expedition and that the studios only seek to stop unauthorized copying and public display of their characters. He added that the studios do not intend to halt AI development or shut down Midjourney. The court has not yet ruled on whether broader discovery is permissible.
Models, chips, platforms, and the policy lines being drawn around them.
Confidential computing, promoted as a foundation for Europe's sovereign cloud strategy, is undermined by a cryptographic protocol flaw that may not have a solution. The technology relies on remote attestation to prove a Trusted Execution Environment runs genuine software, and vendors claim full control over customer data. Research by Muhammad Usama Sardar of TU Dresden and co‑authors showed that the attested TLS protocol can be subjected to relay attacks, allowing a client to verify a genuine server while its traffic is encrypted to a compromised machine. Their paper, Identity Crisis in Confidential Computing, presented at AsiaCCS 2026, demonstrated that connection evidence can be silently redirected to a different server anywhere in the world. A follow‑up study examined seven intra‑handshake binding mechanisms and found none prevent relay attacks. The analysis covered four production systems—Meta's Private Processing for WhatsApp, Contrast, Cocos AI, and a CCC proof‑of‑concept—all vulnerable to the flaw. The vulnerability was assigned CVE‑2026‑33697 with a severity rating of 7.5. The IETF TLS working group acknowledged the attacks, while the CCC's Attestation Special Interest Group delayed publishing the proof for more than a week despite repeated requests, leading Sardar to release the artefacts independently. Germany's Federal Office for Information Security (BSI) stated that confidential computing is a defense‑in‑depth component that does not guarantee key or identity management, and that vendor claims about its ability to ensure digital sovereignty are overstated. Intel said its attestation infrastructure does not limit sovereignty guarantees and does not receive plaintext data, while Google did not respond to inquiries. Sardar recommends abandoning intra‑handshake attestation in favor of post‑handshake attestation, noting that the timing of evidence binding makes level three security unattainable under the current architecture.
Mistral AI, a French artificial‑intelligence company, announced that it has raised more than $4 billion in total funding and is preparing to launch an open‑weight language model in July, while also outlining a €4 billion plan to build data centers in France and Sweden.
Founded in early 2023 by former researchers from Google DeepMind and Meta, the startup has grown to a valuation of roughly $23 billion and has secured a series of high‑profile partnerships. In February it reported annual recurring revenue exceeding $400 million, up from $20 million a year earlier, and said it is on track to surpass $1 billion in ARR for the current year.
The company offers a suite of models, including Mistral Small 4 and the “Les Ministraux” family designed for edge devices, and has released the code‑agent tool Leanstral as open source. It has partnered with Microsoft, which invested €15 million and will distribute its models on Azure, and with UAE investment firm MGX, NVIDIA, and bpifrance to develop an AI campus in the Paris region. Additional agreements include a European AI platform using Nvidia hardware, scheduled for 2026, and collaborations with Accenture, IBM, Orange, and several government and defense entities.
Mistral CEO Arthur Mensch said the firm aims to make advanced AI systems broadly accessible and affordable, while noting that an initial public offering remains a long‑term goal and the company is not currently for sale.
shadcn/ui announced that Base UI is now the default component library for new projects, effective July 2026.
Radix remains supported for existing applications.
The library launched in January 2023 built on Radix, which offered headless, accessible components.
The same team that created Radix introduced Base UI as a separate library.
Community members asked whether Radix would be replaced, and the team said switching libraries could be risky for production applications.
New projects created with npx shadcn init automatically select Base UI.
The shadcn/create command lists Base UI first.
Documentation pages open on the Base UI tab by default.
Radix documentation remains accessible with one click.
Radix is not being deprecated.
Updates and new components will be released for both libraries.
Users can continue using Radix in existing applications without migration.
A migration skill is available for developers who want to move components from Radix to Base UI.
The skill processes one component at a time.
It creates a report in .migration and a separate git branch for each component.
Rollback is possible by deleting the branch.
In June 2026 the team released chat interface components: MessageScroller, Message, Bubble, Attachment and Marker.
MessageScroller handles scrolling, message threading and visibility tracking.
The other components provide layout, bubble styling, attachment support and status markers.
Two CSS utilities, scroll-fade and shimmer, improve visual feedback in long lists.
The @shadcn/react package was introduced for unstyled, headless React components.
Its first component, message-scroller, implements interaction logic that works with both libraries.
GitHub registries can now be used as source registries for shadcn items.
A registry.json file at the repository root defines installable items.
The shadcn CLI reads the file, resolves includes and installs the requested files without building or publishing.
shadcn/tailwind.css provides shared Tailwind utilities for both libraries and is imported automatically during init.
An eject command inlines the CSS into the project stylesheet and removes the dependency.
A new style named Rhea, described as a more compact version of Luma, is available in shadcn/create for both libraries.
It adjusts component sizes and spacing while keeping Tailwind's utility scale unchanged.
The library will continue to support both libraries and welcomes community feedback on future updates.
OpenAI's Codex model exhibits a pattern of reasoning token counts that cluster at exact values, most notably 516 tokens for GPT-5.5, which coincided with lower overall reasoning token usage and may explain degraded performance on complex tasks.
The finding follows issue #29353, which documented a task‑level reproduction where GPT‑5.5 outputs ending at exactly 516 reasoning tokens produced incorrect answers; this analysis aggregates data from February to June 2026 to assess the prevalence of the pattern.
Analysis of 390,195 response records across 865 sessions showed 3,363 instances where reasoning_output_tokens equaled 516; GPT‑5.5 accounted for 19.3 % of all responses but 82.0 % of the exact‑516 cases, yielding a ratio of 44.0 % for GPT‑5.5 versus 1.3 % for other models. Monthly data revealed a sharp rise in the exact‑516 proportion from 0.11 % in February to 53.30 % in May, while mean P90 reasoning tokens fell from 268.1 in February to 106.9 in May.
The author requests that the Codex team examine whether a reasoning‑budget, routing, truncation, or scheduler mechanism causes responses to terminate around 516, 1034, or 1552 tokens, and clarify whether the exact 516 count represents a normal stopping point or an internal threshold.
On July 4, 2026, the Industrial Light and Magic team released a video reflecting on the production of Terminator 2, which celebrates its 35th anniversary this year. Terminator 2 required more than 100 computer‑generated shots, a large undertaking for visual effects studios at the time. ILM engineers described the limitations of early 1990s CGI technology and the need to develop new tools to handle the volume of shots. The retrospective highlights the collaborative effort across departments to solve rendering, animation, and compositing challenges. The work on Terminator 2 set a precedent for subsequent digital effects pipelines and continues to be referenced in contemporary visual effects research.
OpenAI has not visited a key site for its planned Stargate UK datacentre project, raising questions about the £20bn of a £30bn investment announced by the UK government.
Stargate UK was announced in 2025 as a multibillion-pound datacentre initiative under a US‑UK technology partnership, with OpenAI, Nscale and Nvidia pledged to develop infrastructure at sites including a planned facility at Cobalt Park in North Tyneside.
An Freedom of Information request showed that neither OpenAI nor Nscale held meetings with local authorities at the Cobalt Park site, while Nvidia visited the North East combined authority in February 2026. OpenAI said it continues to explore the project pending regulatory approval and manageable energy costs. Nscale's commercial officer went to North Tyneside but there is no record of a meeting. The UK government said the AI growth zone would attract £30bn, with £10bn committed by Blackstone and potential additional investment of £20bn from future partners. OpenAI chief executive Sam Altman described the initiative as reflecting a shared vision for UK AI infrastructure, while critics noted that the £20bn figure represented the amount needed to build the datacentre and meet a 1.1GW electricity demand, rather than a commitment of funds. Local Conservative leader John Johnsson said he was surprised by the announcement and had not been consulted during planning. Energy system operator reports indicate the site lacks a grid connection, with an alternative power solution submitted but details redacted. Analysts said energy costs, grid capacity and infrastructure appear insufficient to support a project of this scale.
The government said a taskforce co‑chaired by the technology secretary and the North East mayor is advancing planning, and that the AI growth zone will reach 1.1GW capacity with 400MW coming online in 2028.
Applied Science posted a YouTube video on July 2 2026 that shows high‑speed footage captured with an atomic force microscope built by icspi. The microscope records a laser‑created diffraction grating on stainless steel and applies a selective etching process that removes material in defined patterns. The same footage also displays bacterial cells adhered to a surface, illustrating the instrument’s ability to image soft samples at high speed. The video has accumulated more than 120,000 views and includes comments from researchers discussing tip wear, imaging artifacts, and surface preparation methods. Links in the description reference the microscope manufacturer, related YouTube channels, and the creator’s Patreon page.
Jocelyn Malamy, an associate professor at the University of Chicago, reported that Clytia hemisphaerica jellyfish close small wounds within minutes and larger wounds in less than an hour, a rate far exceeding human capability and without scar formation. The jellyfish medusa is a short‑lived stage of a life cycle that begins with a polyp colony resembling a perennial shrub. The colony releases free‑swimming medusae that are transparent, allowing researchers to observe cells in real time. Malamy described two cellular structures that drive wound closure. First, lamellipodia, actin‑rich extensions that crawl across the basement membrane and drag the cell body toward the wound edge. These structures can extend even within a single cell, a novel observation. When lamellipodia meet, an actomyosin cable forms behind them and contracts, pulling cells together and expelling debris if the basement membrane is damaged. If the wound is too large for lamellipodia to bridge, the sheet of epithelial cells lifts and migrates collectively until the edges meet. The mechanisms appear conserved across organisms, suggesting that insights from jellyfish may inform wound‑healing research in mammals. Malamy plans to investigate how basement membranes are repaired, a process that remains unclear.
Supergirl, part of the DC Universe's Gods and Monsters series, opened with lower than expected box office numbers, prompting Warner Bros. to anticipate stronger weekend earnings. The film was written by Ana Nogueira and directed by Craig Gillespie, and it adapts the Supergirl: Woman of Tomorrow comic miniseries. Milly Alcock portrays Kara Zor‑El, and Jason Momoa appears in a brief cameo as Lobo. The story follows Kara celebrating her 23rd birthday while traveling with her dog Krypto across red star planets. Her cousin Superman, played by David Corenswet, urges her to return to Earth, but she prioritizes her bond with Krypto. The narrative draws comparisons to the Western True Grit and references the Silver Age comics. Critical reception has been mixed, and the film's streaming release is pending.
The iPhone 17 Pro Max, finished in Cosmic Orange, was placed inside a time capsule buried in July 2026 to mark the United States’ 250th anniversary. The capsule, which contains items from all 50 states, is scheduled to be opened in 2276 for the nation’s 500th anniversary. Apple officials said the device is intended to serve as a digital artifact for future examination. The phone’s battery is expected to deteriorate before the opening date, raising doubts about its operational capability. When opened, the capsule’s contents will be accessible through the Notes application, according to the project’s description. The inclusion highlights Apple’s role in the commemorative collection. The timeline of technology over the next 250 years remains uncertain, with potential shifts toward neural interfaces or other forms of device.
Thousands of Vespa riders gathered in Rome on June 27, 2026, to celebrate the 80th anniversary of the well‑known Italian scooter brand. The event highlighted the scooter's long‑standing presence in Italian culture and attracted participants from several countries. The scooter was first produced in 1946 by Piaggio after World War II, when the company sought a low‑cost, easy‑to‑use vehicle for a nation with damaged infrastructure. Engineer Corradino D’Ascanio designed a step‑through frame with an enclosed engine, allowing women in skirts to ride comfortably. Riders traveled through the Colosseum, the Circus Maximus and the Baths of Caracalla under summer heat. Some participants wore vintage helmets while others displayed flags. The procession was observed by onlookers and photographed. Claudio Sarra, who has managed a Vespa rental business and a small museum for three decades, said the scooter represented freedom and modernity. Piaggio reported lower sales and profits in 2025, though the anniversary event proceeded without disruption. David Mumunadai and Dawn Brooks Mumundai, who own 13 vintage Vespas, said they met riding Vespas in New Orleans 16 years ago. Shirley Russell, who traveled from Brisbane, Australia, said the gathering promoted inclusivity. The anniversary occurs as Piaggio navigates softened demand in Europe, North America and Asia, while the Vespa brand remains a notable part of Italian culture and is observed in Italy and abroad. The long‑term impact of the celebration on Piaggio's sales remains to be seen.
The Ikko MindOne Pro, a small smartphone launched at CES in January and priced at $429, features a square screen and a flip-up camera that can serve as a selfie lens or stand.
Ikko, a Shenzhen-based company known for earbuds and audio accessories, introduced the device as a compact alternative to mainstream phones, though reviews note several functional drawbacks.
The phone’s processor generates noticeable heat during initial setup, and its battery depletes faster than typical devices, dropping from about 90% to 60% within an hour and a half under moderate use.
Camera performance is inconsistent; while the flip-up mechanism is functional, image quality suffers in low-light conditions, often producing greenish tones.
The square display, intended to differentiate the device, complicates web browsing and video viewing, as most content is designed for vertical rectangles, leading to cropping and reduced readability.
Ikko includes software options such as a resolution toggle and a vertical aspect ratio setting to mitigate screen issues, and a dedicated AI launcher that offers LLM switching and note-taking features, along with a free eSIM for data.
An optional keyboard accessory adds a magnetic ring and headphone jack but has been described as fiddly compared with on-screen keyboards.
The device’s reception remains mixed, with some users appreciating its minimalist concept while others find it limited for everyday tasks such as app usage and battery endurance.
Authors have reported a surge of AI‑generated scams originating in Nigeria that use personalized emails to solicit payments for fake book club features, private review services,...
Authors have reported a surge of AI‑generated scams originating in Nigeria that use personalized emails to solicit payments for fake book club features, private review services, and marketing assistance.
Since June 2025, writers have received unsolicited messages offering fees ranging from $55 to $350 for appearances in nonexistent book clubs or for participation in private review groups.
Two distinct iterations have been identified. In the past two weeks, nearly two dozen writers have received emails claiming to represent local book clubs. Some clubs were fictional with no online presence; others were real clubs whose names were used without permission, as confirmed by a representative of Mocha Girls Read. Scammers described fees as spot fees, spotlight fees, or participation fees, with amounts reported from $55 to $350 and package options priced between $100 and $200. Payment instructions included PayPal friends and family, bank transfers, and Upwork contracts issued by third‑party vendors such as Olaleye Abdulhammed. Wells Fargo bank transfer details were also provided.
A second wave of scams offered private review groups that promised tips of $20 to $30 per reviewer. Authors were required to purchase a minimum of 30 to 50 reviewers. One solicitation asked for a $25 fee for 30 readers. Payment was to be made via a contract sent by a payment manager, often based in Nigeria. The solicitation included extensive personalization and emojis. Author T. Kingfisher confirmed that a false review was attributed to her name.
Scams previously originated mainly in the United States and Canada, but recent reports show Nigeria, the Philippines, and Pakistan as primary sources. The book club scam appeared in September 2025, the private review group scam in August, and basic PR scams in June. The rapid increase in spam volume may cause recipients to become suspicious. Over‑the‑top flattery and repetitive language have reduced the credibility of the messages.
Scammers now request bank transfers to Nigerian accounts, payments through PayPal friends and family, Zelle, Cash App, and Wise. These methods are difficult or impossible to reverse compared with credit card payments. Invoices from Upwork, Payoneer, and Coachli remain available but are used less frequently. The blog writer has collected extensive payment information but has not confirmed any payments made by victims.
Awareness of these tactics remains the primary defense, and the prevalence of the scams continues to be tracked as new variations emerge.
A study by the European Southern Observatory (ESO) published in Astronomy & Astrophysics says that more than 1.7 million planned satellites in low‑Earth orbit would exceed the amount that can be accommodated without degrading night‑time astronomical observations. The study recommends limiting the total number of operational satellites to 100 000 and requiring them to be below visual magnitude 7 to remain invisible to the naked eye.
Since 2019, the number of satellites in orbit has risen to about 14 000, driven largely by SpaceX’s Starlink network. Proposals now include SpaceX’s plan for an additional one million satellites, as well as constellations such as E‑Space’s Cinnamon, China’s CTC‑1 and CTC‑2, and a proposed mirror‑based system by the U.S. firm Reflect Orbital that aims to launch up to 50 000 satellites by 2035.
Simulations show that a large fraction of each night would have multiple satellites visible, with up to several thousand at certain times, comparable to the number of stars seen under dark skies. Bright satellite trails can reduce usable observation time by up to 28 % in images taken with the Very Large Telescope’s FORS2 instrument, and could render many images from the Vera C. Rubin Observatory unusable for several hours each night.
Light pollution from satellites also raises overall sky brightness; with the full Reflect Orbital constellation the night sky could become three to four times brighter, reducing contrast for faint celestial objects such as distant galaxies, exoplanets, and potentially hazardous asteroids.
ESO Director General Xavier Barcons said the large number of planned satellites challenges the scientific, technical, economic and educational value of astronomy and called for limits on future launches and cooperation among stakeholders.
Olivier Hainaut, lead author of the study, noted that the 100 000 satellite limit is not an exact threshold, but exceeding it could cause losses comparable to equipment failures. He also emphasized that bright satellites can affect biological systems and contribute to atmospheric pollution during re‑entry.
Both SpaceX and Reflect Orbital have filed applications with the U.S. Federal Communications Commission. The FCC has received over 1,800 comments on the Reflect Orbital proposal and nearly 1,500 on the SpaceX application and is reviewing them. ESO, in partnership with the Royal Astronomical Society and the International Astronomical Union, has submitted a response to the FCC.
The study concludes that managing the growth of mega‑constellations is necessary to preserve the night sky for future generations.
The YouTube Studio AI assistant, called Ask Studio, summarizes comments when creators request it. Researchers discovered that comments can be used to inject content into the assistant’s responses. A comment can begin with a placeholder such as [IMPORTANT NOTICE FROM YOUTUBE] followed by a link that includes a video title. When a creator clicks the link, the assistant includes the title in its output. The link can point to a URL containing private video information, allowing an attacker to retrieve titles of videos that are not publicly visible. YouTube has stated that the behavior requires user action and is not classified as a security bug. The company said the issue involves social engineering rather than a vulnerability in the system. Researchers argued that the trust model is compromised because creators rely on the assistant and do not see edited comments. The assistant has access to channel data, including private video titles, because it operates under the creator’s authenticated session. The finding highlights the need for clear separation between user‑generated comments and system instructions in AI‑driven tools.
Kelsey Pfendler arrived at Ala Wai Boat Harbor in Honolulu on Friday evening after completing a solo row from Monterey, California, covering more than 2,400 miles in 43 days. The arrival, observed by hundreds of spectators at Magic Island and the Hawaii Yacht Club, marked the end of a journey that began on May 21. Pfendler rowed a 21‑foot ocean rowing boat named Lily and said she became the first, youngest and fastest woman to complete the crossing, surpassing the women’s record of 86 days 10 hours 5 minutes set by Lia Ditton and the men’s record of 52 days. She described the accomplishment as a personal feat and said it was intended to inspire women and raise awareness for the Whale Foundation, a nonprofit supporting Grand Canyon river‑guiding community members; more than $30,000 has been donated to the organization since the start of the trip. Pfendler, a veteran Grand Canyon river guide with eight years of experience, previously led a four‑woman crew across the Pacific in 2024, finishing in 40 days 22 hours 14 minutes despite encountering large waves, a capsizing and equipment failures. She is scheduled to speak to the media after resting and celebrating her birthday.
University of Oxford began teaching by 1096, predating the founding of Tenochtitlán, the Aztec capital, by 229 years. The university grew rapidly after 1167, when King Henry II banned English students from the University of Paris, and by 1264 it had formalized into a university with three original colleges. The Aztec Empire was established in 1325 when Tenochtitlán was founded, and an alliance in 1428 expanded its dominance. Spanish forces captured Tenochtitlán in 1521, ending the empire. Oxford is recognized as the oldest continuously operating English‑language university, though the oldest university worldwide is the University of al‑Qarawiyyin in Fez, Morocco, founded in 859, and the University of Bologna in Italy, founded in 1088, is the oldest in the Western world. Researchers note that public perceptions of historical timelines are often inaccurate. Tim Urban of the blog “Wait but Why” observed that humans struggle to place events in proper perspective. He cited examples such as Cleopatra, whose reign occurred closer in time to the modern era than to the building of the pyramids. Urban also described a “horizontal” method of examining history segments to reveal overlapping lifespans of figures from different periods. The comparison of Oxford and the Aztec Empire illustrates how chronological assumptions can be misleading.
Google released a commercial that imagines the Founding Fathers drafting the Declaration of Independence using Google Workspace. Titled 'Group project, but make it 1776', the ad shows a fictional Thomas Jefferson receiving a text from Ben Franklin, then editing the document in Google Docs, scheduling a meeting in Google Calendar and holding it via Google Meet with participants keeping cameras off, and finalizing the text with e‑signatures and fireworks. The founders also use Google's 'help me visualize' tool to experiment with animal designs for the national seal, ask Gemini to take meeting notes, and consult a chatbot before declining a document-access request from King George III. The ad includes a tongue-in-cheek line from Sam Adams asking if the dispute can be settled over beers. Viewers on YouTube and Instagram expressed generally positive reactions, while Bluesky users described the ad as cringey and tone-deaf; historian Angus Johnston noted that little of the content actually relied on AI. The commercial does not suggest that AI would improve the historical text.
Matic will raise the price of its robot vacuum and mop from $1,245 to $1,495 on September 9. The increase of $250 reflects higher costs for memory and other components, which the company says have risen tenfold. Purchasers buying directly from Matic will receive a year’s supply of replacement bags, valued at $96, at no additional cost; each refill contains 12 bags and shipping is free. The company has extended its return policy from 60 days to six months. A review by Jennifer Pattison Tuohy noted that the device showed reliable human‑like navigation, became stuck twice during six months of use, and performed well in a three‑story home with pets, thick rugs, and high transitions, delivering strong suction, a self‑cleaning roller mop, continuous cleaning even when the water tank is low, quiet operation, and minimal maintenance. The unit carries its own water tank and stores dirty water in a disposable bag, allowing the user to empty the reservoir manually; it can also drive itself to a sink for refilling. Unlike many competing models, it can operate entirely offline, with maps and data stored locally rather than in the cloud.
An anonymous X account posted a skin for AO3 on June 29 that flags text containing a Claude‑injected code.
The skin adds a 'font-claude-response-body' wrapper to content copied directly from Claude into AO3’s editor, causing the page background to turn red when the code is present. Testing by the account and an independent reporter showed the red background appearing when a story was pasted straight from the chatbot and disappearing when the same text was edited before posting.
The code is only preserved if the text is transferred unchanged, so the tool cannot detect AI‑generated material that is edited in other programs. The red screen may indicate full AI authorship or only a brief passage that was checked with the model, and it does not reveal the extent of AI use.
Anthropic has not responded to requests for verification. Existing AI detection methods such as C2PA Content Credentials and Google’s SynthID apply to images, video and audio and do not cover copy‑pasted text. No universally reliable system for distinguishing AI from human writing currently exists.
AO3 offers a voluntary “Created Using Generative AI” tag for authors who wish to disclose AI use. The skin demonstrates a technical approach, but it may lead to false positives, and the community continues to rely on self‑reported signals rather than definitive verification.
A researcher evaluated Meta's Stable Signature watermarking system and found that its claimed false‑positive rate of less than one in a million is not supported by empirical data, which indicate a collision probability of roughly one in four. The system embeds a 48‑bit sequence using a neural network that selects locations in an image to hide data with minimal visible distortion and uses Hamming distance to identify matches. The researcher collected 10,000 uncurated images from a public source and applied statistical tests from the NIST SP 800‑22 suite. Analysis showed that the 48‑bit outputs were not uniformly distributed, with about 56 % ones and 44 % zeros, producing a chi‑square statistic far above the threshold for randomness. Serial dependence tests also indicated strong correlation between adjacent bits. Clusters of images sharing identical or near‑identical bit sequences were observed, including 25 images with exact matches and a larger cluster of 450 images within six bit differences, representing about 4.5 % of the sample. These patterns contradict the assumption of independent bits used in Meta’s theoretical calculations, leading to a higher false‑positive estimate than claimed. Similar statistical issues were identified in Google’s SynthID and Adobe’s TrustMark watermarks, which also assumed independent bits despite evidence of bias and dependence. The findings suggest that current watermarking algorithms cannot reliably distinguish genuine watermarks from random noise, which could limit their usefulness for fraud detection and compliance with upcoming AI‑content regulations in the EU, China and California. The researcher called for independent verification and noted that the statistical mismatch undermines the credibility of the claimed accuracy rates across all three systems.
Command and Conquer Generals now runs natively on Apple Silicon Macs, iPhone and iPad without emulation. The 2003 real‑time strategy game was originally built for Windows and required translation layers to run on Apple platforms. The port compiles the original engine for ARM64 and translates DirectX 8 calls through DXVK to Vulkan, then through MoltenVK to Metal on iOS devices. The implementation builds on the open‑source GeneralsX repository, which relicenses EA’s GPL‑v3 source release and adds iOS and iPadOS support along with engine fixes. Users must provide their own game assets, obtainable from Steam for approximately $5, and follow scripts that automate dependency installation, configuration, building and deployment. macOS builds are executed with ./scripts/build/macos/build-macos-zh.sh followed by ./scripts/build/macos/deploy-macos-zh.sh, while iOS builds use cmake presets and the ./scripts/build/ios/package-ios-zh.sh script. Known issues include iPad memory limits that may terminate sessions and occasional crashes when backgrounding the app; logs are saved in the Files app. The code is released under GPL v3, and credits list Westwood/EA Pacific, EA, the GeneralsX contributors, the community mainline, and libraries such as DXVK, MoltenVK, SDL, OpenAL Soft and FFmpeg. The engineering log records the development process and is maintained by Ammaar Reshi, with assistance from Claude Code. The project remains open for community contributions.
Smell loss affects up to 22% of the population and has been observed in millions of COVID‑19 cases, prompting renewed research into the sense of smell. The condition, known as anosmia or hyposmia, can be permanent or temporary and is associated with a range of neurological and psychiatric disorders.
The sense of smell relies on olfactory neurons in the nasal cavity that transmit signals to the olfactory bulbs, which directly connect to brain regions involved in emotion and memory. These bulbs are among the few brain areas that generate new neurons in adulthood and are vulnerable to viral infection, trauma, inflammation, and neurodegeneration.
Chrissi Kelly lost her sense of smell after a viral infection during a trip to the Czech Republic in 2012. After six months of complete anosmia, she described the loss as “catastrophic” and began searching for information and community support. She later co‑founded patient groups and co‑authored more than 30 academic papers.
Research has linked prolonged smell loss to increased risk of neurodegenerative diseases. A wine enthusiast who requested anonymity reported that loss of smell preceded his diagnosis of Parkinson’s disease after 20 years, while clinicians noted that olfactory dysfunction can appear early in Alzheimer’s and Lewy body dementia. A 2023 survey found that 60% of COVID‑19 patients experienced smell loss, most temporarily but some for extended periods.
Scientists estimate that humans can distinguish millions of odors, yet the neural code translating odor molecules into perception remains unclear. In the early 1990s, researchers identified roughly 1,000 olfactory receptor genes, a discovery that earned a Nobel Prize in 2004 and launched modern olfactory research.
Current investigations focus on olfactory training, a regimen in which patients repeatedly smell familiar scents such as lemon, rose, clove, and eucalyptus. Controlled trials have shown that about 30% of participants with long‑standing smell loss improve after several months of training, with some studies reporting improvements of 10% or more on standardized smell tests. A 2024 meta‑analysis of 36 trials found a statistically significant positive effect across all domains of olfactory function, with reported gains ranging from 11% to 68%.
While olfactory training appears beneficial for some, it is not universally effective, and more research is needed to determine its long‑term impact on cognition, mood, and disease prevention. The ability to smell also influences safety, as loss of odor detection can prevent recognition of smoke, gas leaks, or spoiled food.
Overall, recent findings underscore the importance of the olfactory system for quality of life and brain health, and they suggest that smell loss may serve as an early marker for several medical conditions.
NASA’s Perseverance rover detected complex macromolecular carbon on the surface of a rock at the edge of the ancient river channel Nereva Vallis in Jezero Crater, marking the shallowest detection of organic matter on the Martian surface to date.
Perseverance has spent five years exploring Jezero Crater and analyzing rocks with SHERLOC, a UV Raman spectrometer that identifies specific molecular bonds.
The instrument targeted four locations; three—Cheyava Falls, Apollo Temple, and Walhalla Glades—produced a graphitic band signal indicating tangled reduced carbon, while Steamboat Mountain served as a control.
The signal closely matches terrestrial kerogen, though researchers avoided the term because it implies a biogenic source.
Lead author Ashley E. Murphy of the Planetary Institute said the origin of the carbon remains uncertain and could be either biotic or abiotic.
Further examination of returned samples may clarify the source of the carbon, underscoring the mission’s ongoing effort to characterize Martian chemistry.
The author concluded that ORMs provide limited benefit and should supplement rather than replace SQL.
He has worked for 30 months with PostgreSQL and SQLite, using SQLAlchemy and Hibernate, and has handled event‑based timelines and reporting tasks.
He describes several challenges: tables can accumulate many columns, leading to wide tables that are difficult to query efficiently when using the ORM; queries that rely on the ORM’s query API can generate excessive joins and cause performance problems; foreign keys often require many joins to retrieve a single object; managing entity identities between the application and the database adds complexity; transaction handling requires dynamic scoping and careful boundary placement; and schema changes are cumbersome because the database remains the source of truth while the application must be updated separately.
He notes that ORMs still require knowledge of SQL to write efficient queries, and that many developers spend more time translating database rows into objects than optimizing the queries themselves.
The author recommends treating the database as a data type with a query API, writing queries directly or using templated SQL rather than relying on the ORM for query generation, and learning SQL directly to avoid the “ORM makes it easy” trap.
He adds that his experience reflects his specific use case and does not prescribe a universal approach.
Carlos Gonzalez, a quadcopter enthusiast, released a series of blog posts that introduce the fundamentals of drone modeling, state estimation, motion planning, and control. The posts are intended as a concise, accessible overview for newcomers, not as an authoritative academic treatment. Each post includes references to scholarly sources, and some sections are marked as work in progress. Gonzalez explained that he initially attempted to write a longer paper but found it more digestible to break the material into shorter installments, allowing a more casual tone while still providing an intuitive feel for the subject. The guide aims to help beginners understand how quadcopters can achieve autonomy and remains an ongoing resource with future sections expected to be completed.
A user reported a potential session leakage in the Claude AI workspace, where the system responded to queries about Minecraft bricks despite being authenticated to an Enterprise ZDR workspace.
The report was posted on Hacker News under a GitHub issue for Anthropic's Claude code. The user described using the darwin platform with Apple_Terminal version 2.1.199.
The user said the agent began asking what kind of bricks were needed for a Minecraft temple and asserted that it was building a Minecraft temple. The user questioned whether the cache was isolated to the workspace or if leakage could occur from a consumer plan.
The user noted that the session had been started in a directory unrelated to the task and that earlier conversation pollution may have influenced the agent's behavior. No error messages were recorded.
Anthropic has not yet responded to the report, and the cause of the observed behavior remains unclear.
Alibaba will prohibit its employees from using Anthropic’s Claude Code beginning July 10, according to multiple reports. The company will require staff to use its own Qoder tool instead. Anthropic has restricted Chinese companies and foreign entities owned by them from using its models and has been working to close loopholes that enable Chinese access, including a version of Claude Code that could identify Chinese users. A Reddit post reported that this version could secretly identify Chinese users. Thariq Shihipar, a spokesperson for Anthropic, said on X that an experiment launched in March aimed to prevent account abuse by unauthorized resellers and protect against model distillation, and that stronger mitigations have since been implemented; the company had been planning to remove the tool. Alibaba classified Claude Code as high‑risk software and directed employees to use Qoder.
Zig moved package management functions from the compiler to the build system. The new maker process now contains logic for fetching packages, handling HTTP and TLS networking, and managing compression and parsing of build.zig.zon files. This change reduces the size of the Zig executable by about 4 percent. Safety checks are enabled for networking operations because the maker executable is built in ReleaseSafe mode, and networking and file‑hashing crypto can use host‑specific CPU instructions. The alteration is largely non‑breaking. Flags have been renamed: the --maker‑opt flag is replaced by the ZIG_DEBUG_MAKER environment variable, and the --zig-lib-dir flag is replaced by ZIG_LIB_DIR. Upcoming work includes a build server protocol, detection of build script changes during watch mode, and improvements to incremental compilation. The developer noted that the changes are intended to make the build system faster, simplify maintenance, and support future features such as --watch and --webui. Contributions are welcome.
One author highlighted differences in how iPhone and Nothing Phone process rapid taps for rotating images. The author described a scenario in which a user taps a rotation control multiple times quickly. On iPhone, each tap is buffered until the current animation completes, allowing pending rotations to occur sequentially. On Nothing Phone, a tap triggers a haptic and auditory confirmation and then ignores additional taps while the animation is running. The author argued that such behavior can impede accessibility, noting that many users may need to rotate multiple photos in a short period. Situational disability applies when ordinary tasks become difficult due to interface constraints. Rapid tapping without waiting for animations would improve usability for users handling large numbers of images. The author also noted that developers can stop or accelerate the animation after an interrupting tap, and recommended that interfaces should not force users to wait for animations to finish. The author suggested that future updates could adopt buffering or immediate response mechanisms to accommodate varied usage patterns.
Newer Claude models such as Opus 4.8 and Sonnet 5 have been observed to generate malformed tool calls when invoking Pi's edit function. They add keys that do not match the expected schema. The edit tool expects a path parameter and an edits array containing objects with oldText and newText fields. Validation rejects calls that contain additional properties. Across multiple test runs, the newer models produced extra fields such as requireUnique, type, id, or children, causing the tool call to fail. The issue was reproduced in an agentic session where the model had read files and composed a multi‑line edit, occurring in roughly 20 % of attempts. Earlier versions of the model, including Opus 4.5, did not show this behavior. The author hypothesizes that the issue stems from post‑training that exposed the model to Claude Code’s edit tool, which uses a flatter schema and tolerates additional parameters. Reinforcement learning in that environment may reward successful completion despite schema deviations, leading the model to learn a broader set of acceptable keys. Strict mode in the tool invocation interface appears to eliminate the problem by preventing sampling of unsupported fields. The case illustrates that tool schemas are not neutral for all model families and that harness developers may need more explicit schema validation to ensure compliance.
The Epomaker RT98 is a mechanical keyboard priced at $119 that includes a detachable number pad capable of being positioned on either side of the main layout. It features a retro aesthetic with a CRT‑style screen that can display date, time, connection status, or looping images. The board uses a solid PCB and polycarbonate switch plate without flex cuts, which contributes to a firm typing feel. Available switch options include Creamy Jade linear switches and Sea Salt Silent V2 linear switches, the latter offering quiet operation but louder plate‑mounted stabilizers on the space bar, enter, shift, and backspace keys.
The number pad can be moved by removing the bottom screws, detaching the top cases, swapping the internal assembly, and reattaching the components. The process requires unplugging the detachable screen for its pogo pin connector and repositioning the numpad ribbon cable. While instructions are not included in the package, a manufacturer video tutorial guides the change, which can be completed in about 30 minutes.
Reviewers note that the keyboard’s plastic construction feels less robust than aluminum alternatives, and the printed legends on the keycaps appear low‑quality. However, the tactile experience with the Creamy Jade switches is described as crisp and pleasant, with a richer, darker acoustic profile than comparable Aula models.
The RT98’s price places it near the lower end of mid‑range keyboards that use aluminum cases. It costs $40‑$60 more than entry‑level models without a modular pad and $10 less than the Wobkey Rainy75, which includes higher‑quality keycaps. Potential buyers must weigh the added customization of the movable numpad against the modest price premium and the plastic build.
The modular design offers flexibility for users who want a north‑handed or south‑handed layout without purchasing a separate peripheral. While the screen adds a novelty element, it remains limited to pre‑loaded GIFs and basic status information, and no third‑party accessory ecosystem has been confirmed.
Whether the RT98 will become a mainstream option remains to be seen, as its success will depend on future accessory support and continued consumer interest in modular layouts.
NASA will conduct a year‑long simulated deep space mission beginning no earlier than August 2027 at Johnson Space Center in Houston. The mission, named the Moon and Mars Exploration Analog, will have volunteers live in two confined habitats for twelve months. Participants must complete a multi‑day selection process and pass a psychological assessment, and they are expected to have a strong desire for the experience and an interest in supporting NASA’s preparation for crewed lunar and Martian missions. The notice indicated that funding constraints could affect the program’s schedule. The simulation is intended to evaluate how prolonged isolation affects crew performance and to inform the Artemis program and plans for a sustained lunar presence.
David Potter, the South African‑born founder of Psion, died on 28 June at the age of 82, six days before his 83rd birthday. Potter created Psion in the early 1970s and led the company through the development of the Organizer line, the Series 3, and the Series 5 handheld computers, which sold millions of units worldwide.
Born on 4 July 1943 in East London, South Africa, Potter studied natural sciences at Cambridge and earned a doctorate in mathematical physics at Imperial College. After a professorship at UCLA, he returned to Britain and began investing surplus savings in companies such as Fogarty, which later tripled in value, providing capital for Psion. The firm supplied early software for Sinclair Research's ZX80, ZX81 and ZX Spectrum machines, including a flight‑simulation program, and later provided the XChange suite for the Sinclair QL.
Psion’s handheld devices featured replaceable solid‑state drives and a custom operating system, EPOC, which was later renamed Symbian and formed the basis for early smartphones. The Register reported that Psion designed EPOC32, a 32‑bit real‑time operating system, for the Series 5, and that the company sold its stake in Symbian to Nokia in 2004.
Potter received a CBE in 1996 for services to manufacturing, a fellowship of the Royal Academy of Engineering in 2001, and multiple honorary doctorates. He sold £12.6 million of Psion shares in 2000 and remarked that his timing on share sales often coincided with price increases. He also supported The David and Elaine Potter Foundation, contributed to the openDemocracy project, and donated to the Labour Party.
Colleagues such as Ian Fogg and Anthony Garvey described Psion’s collaborative culture and Potter’s role in keeping Microsoft’s Bill Gates awake at night during the company’s peak. Potter retired from Psion in 2009, leaving a legacy of handheld computing innovation and an operating system that influenced the first generation of smartphones.
PulseKit, an indie iPhone app, provides a dashboard that aggregates analytics from up to 30 platforms and displays widgets on the home screen and lock screen. The app is featured in 9to5Mac's weekly Indie App Spotlight series, which highlights new applications from independent developers. Out of the box, PulseKit connects to services such as App Store Connect, Cloudflare, Discord, GitHub, Google Analytics, Gumroad, Patreon, Product Hunt, Supabase, and Vercel, with additional platforms potentially added. The app supports iPhone, iPad, Mac, and Apple Watch, and includes StandBy mode for monitoring while the device is charging. A free version is available with limited widget functionality for devices running iOS 18.6 or later; a Pulse Pro subscription priced at $5.99 per month or $59.99 per year unlocks unlimited widgets. The developer has indicated openness to adding additional platform integrations based on user feedback.
President Donald Trump pardoned nine individuals convicted of Clean Air Act violations related to vehicle modifications.
The Clean Air Act regulates emissions from motor vehicles and industrial sources.
The individuals were found guilty in federal court for altering emissions control systems in diesel trucks.
Prosecutors said the modifications increased nitrogen oxide and particulate matter output.
The Department of Justice announced the convictions in 2023 after a multi‑state investigation.
The White House said the pardons were part of a broader review of sentencing disparities.
Environmental advocacy groups criticized the decision, saying it undermines enforcement of environmental laws.
The Environmental Protection Agency has not issued a public statement on the matter.
The nine individuals will have their convictions erased, but the underlying violations remain subject to future regulatory action.
Legal experts note that the pardon does not affect ongoing civil penalties imposed by the EPA.
Startup Ampera announced a 3D‑printed thorium microreactor module capable of delivering up to 30 megawatts of power for up to 30 years, targeting datacenter, defense and off‑grid applications.
The prototype was unveiled at the company's innovation center in Palm Beach Gardens, Florida, where more than 100 attendees, including local officials, business leaders and employees, observed the presentation.
Founder and CEO Brian Matthews described the unit as featuring a fully 3D‑printed silicon carbide reactor core and pressure vessel, and said the subcritical, solid‑state design uses tristructural isotropic (TRISO) particles that contain thorium and require an external neutron source.
The reactor operates without sustaining a chain reaction on its own, eliminating the risk of a runaway excursion, and is intended to run for up to three decades without refueling.
Ampera plans to offer configurations generating 15 or 30 megawatts, which it says could supply a typical datacenter, and larger versions are under development.
The company has established an Australian subsidiary to secure thorium supplies and intends to produce the fuel kernels domestically.
Matthews said the firm expects the power generation portion of the system to be available as early as 2027, with the nuclear module slated for customer delivery around 2030 pending regulatory approval.
When asked about the neutron driver that initiates operation, the company declined to provide details.
The U.S. Air Force has indicated interest in microreactors for three of its sites as part of a program to improve energy resilience during grid outages.
Regulators have not yet approved the technology, and the details of the neutron driver remain undisclosed.
A $200,000 bounty has been offered for individuals who can develop a method to retrieve the full scanned content of Google Books.
Google Books has digitized millions of titles, but many of the scanned pages are only searchable as brief excerpts.
The bounty, announced by the Annas Archive, invites developers and researchers to submit a prototype that can scale to provide complete book scans. Participants are encouraged to contact the organization early with their solution. The offer also applies to comparable large collections held by AI companies, especially those that include rare books.
The initiative seeks to expand public access to digitized literature while preserving the original scan integrity.