Alvaro Lopez Ortega / 2026-04-15 Briefing

Created Thu, 16 Apr 2026 01:18:40 +0000 Modified Thu, 21 May 2026 01:16:58 +0000
6826 Words

A global crisis of AI-generated deepfake nudes has impacted hundreds of students across 28 countries. In legal news, Thomson Reuters faces a lawsuit alleging an employee was fired for protesting its ICE contracts. Meanwhile, researchers have used CRISPR to successfully silence the extra chromosome in Down syndrome cells, even as experts warn that AI-assisted cognition could lead to human intellectual stagnation.

πŸ€– AI & Machine Learning

CPUs Aren’t Dead. Gemma2B Out Scored GPT-3.5 Turbo on Test That Made It Famous

The Gemma 2B model surpassed GPT-3.5 Turbo on the MT-Bench benchmark, achieving a score of approximately 8.0 while running on a consumer-grade CPU. By implementing specific software engineering fixes, researchers were able to further boost the model’s performance to a score of roughly 8.2.

Improved Reasoning in Gemini Robotics-ER 1.6

Google DeepMind has released Gemini Robotics-ER 1.6, an upgraded model featuring enhanced spatial, physical, and multi-view reasoning capabilities. The update introduces new features such as complex instrument reading and improved success detection, and is currently available to developers via the Gemini API and Google AI Studio.

Study: Back-to-basics approach can match or outperform AI in language analysis

A new study led by The University of Manchester has introduced LambdaG, a grammar-based approach to authorship analysis that can match or outperform advanced AI systems. By analyzing linguistic patterns such as sentence structure and punctuation, the method provides a more transparent and computationally efficient alternative for identifying writers.

My AI-Assisted Workflow

The author introduces an AI-assisted development workflow that prioritizes “thinking in writing” to maintain architectural clarity and avoid unvetted implementations. The process uses an AI-driven interview to transform initial free-form plans into structured PRDs, which are then decomposed into granular development issues.

The Death of an AI Whistleblower

Former OpenAI researcher Suchir Balaji, who recently alleged widespread copyright infringement by the company, was found dead in his San Francisco apartment. While medical examiners have ruled his death a suicide, his parents contend that he was murdered to silence his whistleblowing.

Anthropic no longer allows you to fix to specific model version

Anthropic is deprecating specific model versions, such as claude-sonnet-4-5-20250929, in favor of a generic identifier that always points to the newest version. This change prevents developers from pinning to specific model versions, which could cause client applications to break unpredictably during updates.

Don’t let the bot play doctor! AI gets early diagnoses wrong 80% of the time

A study published in JAMA Network Open reveals that leading AI models fail at early differential diagnosis in over 80% of cases due to difficulties with clinical uncertainty. Because these models can project false confidence, researchers advise against using them for patient-facing diagnostic reasoning without structured human review.

Bad teacher bots can leave hidden marks on model students

New research reveals that large language models can transfer undesirable traits to “student” models through “subliminal learning,” even when specific biases are removed from the training data. This phenomenon occurs via subtle statistical signatures in the teacher model’s output, suggesting that AI safety evaluations must examine the origins and creation processes of training datasets.

Not all networks can handle AI traffic – and experts are sounding alarms

Experts are warning that insufficient networking infrastructure is becoming a critical bottleneck for AI, as many neocloud providers have scaled compute capacity without adequate network support. To ensure performance, organizations must prioritize scalable, low-latency networks capable of managing the intensive data movement driven by AI workloads.

Waymo’s self-driving cars face their toughest test yet: London

Waymo has begun testing its self-driving software in London, transitioning from human-controlled vehicles to systems monitored by trained safety operators. The company is adapting its technology to the city’s complex driving environment as part of its progression toward a fully driverless robotaxi service.

The only technology that died more times than VR is AI, and that seems to have worked out

Despite claims that the metaverse is dead, the author argues that immersive technology is evolving toward a seamless integration of digital and biological realities. The focus is shifting away from cumbersome, motion-inducing hardware toward more refined, lightweight, and accessible user experiences.

Apple to send ~200 Siri team members to AI coding bootcamp

Apple plans to send approximately 200 members of its Siri team to an AI coding bootcamp as part of an organizational restructuring. The group, which has been internally referred to as a “laggard,” will undergo the training during this broader company shakeup.

Objection debuts: AI/experts verify news; backed by Thiel/Srinivasan; costs $2,000

Aron D’Souza has launched Objection, an AI-powered startup backed by funding from Peter Thiel and Balaji Srinivasan, which allows users to pay for investigations into the accuracy of news claims. While the platform uses an “Honor Index” to score journalistic integrity, critics warn that its scoring system could discourage investigative reporting that relies on anonymous sources.

Jensen Huang on Nvidia’s supply chain, AI competition, and China sales | Dwarkesh Podcast

In an interview on the Dwarkesh Podcast, Nvidia CEO Jensen Huang discussed the company’s supply chain advantages and competition from specialized hardware like Google’s TPU. He also addressed topics including Nvidia’s stance on selling AI chips to China and its strategic position in the evolving AI ecosystem.

OpenAI plans to introduce a click-based pricing model for certain ChatGPT ads, moving away from a system based on impressions. The company is also exploring ads designed to drive specific user actions, such as app downloads or purchases, though no firm timeline has been established.

Q&A with ElevenLabs co-founder Mati Staniszewski on audio models, voice agents, and more

ElevenLabs co-founder Mati Staniszewski discusses the technical mechanics of audio models and the challenges of achieving conversational speech in a recent interview with John Collison. The conversation also explores the future potential of voice agents across various industries and the company’s rapid ascent to an $11 billion valuation.

πŸ”’ Security & Privacy

Global Deepfake Nudes Crisis in Schools

A growing global crisis involving the use of generative AI to create non-consensual deepfake nudes of female students has impacted approximately 90 schools and 600 students across at least 28 countries. These incidents, which are classified as child sexual abuse material, cause profound psychological trauma and are believed to be much more widespread than current documentation suggests.

US v. Heppner (S.D.N.Y. 2026) no attorney-client privilege for AI chats [pdf]

In US v. Heppner, the Southern District of New York ruled that attorney-client privilege does not extend to communications involving artificial intelligence. The decision clarifies that AI-based chats may not be protected under traditional legal privilege standards.

AI ruling prompts warnings from US lawyers: Your chats could be used against you

U.S. lawyers are warning clients against sharing sensitive information with AI chatbots, as these communications are not protected by attorney-client privilege. The advice follows a New York federal judge’s ruling that AI-generated documents could not be shielded from prosecutors in a securities fraud case.

Claude may require identity verification in some cases

Anthropic is implementing identity verification for certain Claude use cases through its partner, Persona Identities, to prevent platform abuse and ensure legal compliance. The process requires users to provide a government-issued photo ID and a selfie, with the data used strictly for identity confirmation and not for model training.

Police officer uses Flock database to track mistress

Former Costa Mesa Police officer Robert Jay Josett pleaded guilty to using confidential law enforcement databases to track his mistress, wife, and other individuals. He also admitted to harassing the woman through thousands of communications and violating a restraining order. Josett was sentenced to three years of informal probation and ordered to complete a 52-week domestic violence program.

Anthropic’s Project Glasswing CVE tally is still anyone’s guess

The exact number of vulnerabilities identified through Anthropic’s Project Glasswing remains unknown. While researchers found approximately 40 recent CVEs credited to Anthropic or its affiliates, most of these findings cannot be definitively linked to the bug-hunting initiative.

French cops free mother and son after 20-hour crypto kidnap ordeal

French special forces rescued a mother and her ten-year-old son following a 20-hour kidnapping linked to cryptocurrency extortion. Authorities arrested at least four suspects after the group demanded hundreds of thousands of euros from the boy’s father, a cryptocurrency entrepreneur.

Raspberry Pi OS ends open-door policy for sudo

The latest version of Raspberry Pi OS now requires a password for sudo commands by default in new installations to enhance security. While this change addresses the vulnerabilities of passwordless administrator access, existing installations remain unaffected and the setting can be manually reverted.

55 years since its introduction, 6 million FTP servers are still exposed on the open internet

Approximately 6 million web servers, or 3% of the global total, still utilize the insecure, 55-year-old FTP protocol, leaving credentials vulnerable to theft due to a lack of encryption. Many of these exposures stem from default configurations in hosting software, prompting security experts to urge a migration to the more secure SFTP.

Fabricked: Misconfiguring Infinity Fabric to Break AMD SEV-SNP

The Fabricked attack compromises AMD’s SEV-SNP by manipulating memory routing through the Infinity Fabric using a malicious UEFI. By misdirecting critical initialization writes, attackers can bypass security protections to gain unauthorized access to Confidential Virtual Machine memory.

I don’t want a screenshot of your Claude conversation

The author critiques the increasing use of AI-generated screenshots in discussions, noting that the sycophantic nature of language models can lead to biased and unreliable results. They argue that sharing these outputs places an unfair burden on experts to verify inaccuracies and instead advocate for prioritizing human-driven ideas and original context.

X users say bot crackdown purged secret accounts used to curate porn

X’s large-scale crackdown on automated accounts has inadvertently suspended many human-operated “alt” accounts used for private content curation. Users report losing years of archived media as the platform’s automated moderation systems struggle to distinguish between bots and real users.

πŸ’» Software Engineering

Libretto – Making AI browser automations deterministic

Libretto is a Skill+CLI tool that uses “development-time AI” to generate deterministic and debuggable browser automation scripts. By combining Playwright UI automation with direct network requests, it offers a more reliable and interpretable alternative to traditional runtime AI agents.

Cal.com transitions to closed source

Cal.com is transitioning its core codebase from open-source to a closed repository to mitigate security risks posed by AI-driven hacking. To support the developer community, the company is releasing Cal.diy, an open-source version under the MIT license for hobbyists and experimentation.

MCP as Observability Interface: Connecting AI Agents to Kernel Tracepoints

As the Model Context Protocol (MCP) becomes the interface for AI agents to access infrastructure data, two approaches are emerging: wrapping existing observability platforms or building MCP-native layers. The author advocates for the MCP-native approach using eBPF, as it provides the raw, kernel-level telemetry necessary for deep root-cause investigations.

Keep Android Open

Starting in September 2026, Google will require all Android app developers to centrally register by providing identification, fees, and private signing keys. While Google has proposed an “advanced flow” for power users to install unverified applications, critics argue this mechanism is unreliable and threatens the platform’s fundamental openness.

Do you even need a database?

The article evaluates whether early-stage applications require a full database or if custom file-based storage strategies are sufficient. Benchmarking various approaches, such as in-memory maps and disk-based binary searches, demonstrates that optimized file-based methods can achieve performance levels comparable to SQLite.

Wacli – WhatsApp CLI: sync, search, send

Wacli is a third-party command-line interface built on whatsmeow that uses the WhatsApp Web protocol to sync and manage WhatsApp data. The tool allows users to perform fast offline searches, manage contacts and groups, and send text or files directly from the terminal.

20-Year-Old Enlightenment E16 Bug Fixed

Developer Kamila Szewczyk has fixed a 20-year-old bug in the Enlightenment E16 window manager that caused desktop freezes during window title truncation, often triggered by opening specific PDFs. Her patch resolves an infinite loop in Newton’s algorithm by implementing an iteration limit and additional safety guards.

Windows takes a crash dump after one McDonald’s order too many

A McDonald’s restaurant in Healdsburg, California, is displaying a Windows Blue Screen of Death on its order progress screen. Instead of showing incoming customer orders, the monitor is instead displaying a completed Windows crash dump error.

You cannot use the GNU (A)GPL to take software freedom away

The Free Software Foundation (FSF) has criticized the OnlyOffice project for using a modified AGPLv3 that imposes unauthorized restrictions, such as the mandatory retention of a specific product logo. The FSF maintains that these additional terms are inconsistent with the license’s purpose of preserving software freedoms and may mislead users about their rights.

Spec Driven Development isn’t Waterfall - Marc’s Blog

Spec-driven development is an iterative process that utilizes living, versioned specifications rather than static, up-front designs. By abstracting requirements from low-level implementation, this approach enables AI agents to autonomously develop software by following high-level, evolving maps.

How do Wake-On-Lan works?

Wake-on-LAN is a network protocol that enables computers to be remotely powered on upon receiving a specialized “Magic Packet.” This packet consists of a synchronization stream followed by the target device’s MAC address repeated 16 times to trigger a boot signal to the BIOS.

SDL3 - Add DOS platform support

A nearly full-featured port of SDL3 to the DOS platform has been developed thanks to contributions from multiple participants. The update is available through a recent pull request in the official SDL GitHub repository.

Let Sleeping CPUs Lie β€” S0ix

This article explores the historical evolution of computer power management, from early “suspend-to-RAM” functionality to modern standards. It details the transition from Intel’s SMM, which allowed firmware to bypass the operating system, to more standardized and OS-directed interfaces like APM.

Rewriting Every Syscall in a Linux Binary at Load Time

The article discusses a technique for rewriting every system call within a Linux binary at load time. This method involves modifying the binary’s syscall instructions during the loading process.

Swift 6.2 concurrency type system: Formal typing rules and materials for Capability and Region

This article presents the formal typing rules and presentation materials for the Swift 6.2 concurrency type system, focusing on the concepts of Capability and Region. It details technical mechanisms such as region-based isolation and affine transfers in relation to several key Swift Evolution proposals.

The True Size of ASTC Textures

The article examines memory waste in ASTC textures caused by non-power-of-two block dimensions on certain mobile GPUs. It highlights how PowerVR and early Apple GPUs round these dimensions to the next power of two, whereas modern Apple GPUs utilize a more efficient 32-block macro tile alignment to minimize padding.

Retrofitting JIT Compilers into C Interpreters

The “yk” system enables the automatic transformation of C interpreters into JIT-compiling virtual machines through minimal code modifications. This approach offers improved performance for languages like Lua and MicroPython while maintaining full compatibility with their reference implementations.

My 14-Year Journey Away from ORMs: Creating a SQL-First Code Generator

Architect Nikita Volkov shares his journey from developing an ORM in 2012 to creating pGenie, an open-source SQL-first code generator. The tool aims to make the database the single source of truth by focusing on improving the safety and integration of raw SQL rather than abstracting it away.

Modern Common Lisp with FSet

Scott L. Burson’s documentation for FSet v2.4.2, titled “Modern Common Lisp with FSet,” contains no LLM-generated content. The work is published under a CC BY-NC-SA 4.0 license and accepts user feedback via GitLab or GitHub.

Testing OpenGraph on localhost from the CLI before you go public

og-check is a new CLI tool that enables developers to preview OpenGraph tags on localhost without requiring a public URL or external debuggers. The tool fetches metadata and renders previews, including inline images via the Kitty graphics protocol, directly within compatible terminals.

Porting Go’s strings package to C

The author is porting Go’s math/bits, unicode/utf8, and bytes packages to C to provide a more complete, Go-like standard library for C developers. The implementation addresses technical challenges, such as differing operator precedence between the two languages, to successfully replicate Go’s byte slice operations.

Uses for nested promises – The If Works

During the development of the Promises/A+ specification, a debate arose over whether to formally incorporate monads and category theory. The spec authors ultimately chose a design that prioritizes convenience by using a single then() method that automatically flattens nested promises.

πŸ“± Consumer Tech

The tiniest e-reader in the world, and you can build one yourself

YouTuber Paul Lagier has developed a miniature, DIY e-reader that is approximately the size of a car keyfob and can hold up to ten books. The device features a 3D-printed shell and an ESP32 microcontroller, and can be built for around $30 using instructions and design files provided by the creator.

Backpacks got worse on purpose

VF Corporation has reportedly prioritized profit margins by intentionally reducing the quality of mid-range products across its dominant US backpack brands, including JanSport and The North Face. The company achieved this by utilizing inferior materials, cheaper hardware, and less durable stitching to cut production costs.

‘Seeking connection’: video game where players stopped shooting, started talking

Players of the extraction shooter Arc Raiders are unexpectedly prioritizing cooperation and social interaction over the game’s intended competitive combat. Instead of fighting for loot, many users are teaming up against AI enemies or engaging in spontaneous social activities, a trend that has pleasantly surprised the developers.

Woman buys secondhand Samsung Galaxy S26 Ultra, ends up in trade-in scam

A Reddit user reported that their parents were scammed after purchasing a secondhand Samsung Galaxy S26 Ultra on Facebook Marketplace. The device is set to be locked because the original owner failed to return their old handset as part of a manufacturer trade-in agreement.

90% of US teens use social media for fun, 57% message daily on Snap, 37% say TikTok impacts sleep

A Pew Research Center survey shows that roughly 90% of U.S. teens use TikTok, Instagram, and Snapchat primarily for entertainment. However, usage patterns vary by platform, with Snapchat users more likely to message daily and TikTok users more likely to report spending excessive time on the app.

Spotify adds physical book buying in US/UK via Bookshop.org; expands Page Match to 30 languages

Spotify has launched a feature in the U.S. and U.K. that enables users to purchase physical books through Bookshop.org directly within the app. The company also expanded its “Page Match” tool to support over 30 additional languages, allowing listeners to sync physical or e-books with corresponding audiobook sections.

Global smartphone shipments fell 4.1% in Q1 amid chip crunch; Samsung +3.6%, Apple +3.3% (IDC)

Global smartphone shipments fell 4.1% year-over-year in Q1, marking the first decline since 2023 due to a memory chip shortage. Despite the overall market drop, Samsung and Apple both saw shipment growth of 3.6% and 3.3%, respectively.

Google releases Windows app with Spotlight-like search (web, Drive, local) and screen sharing.

Google has released a new desktop application for Windows users that provides a search experience similar to macOS Spotlight. Accessible via an Alt + Space shortcut, the app allows users to search the web, Google Drive, and local files while offering features such as AI-powered search and screen sharing.

🏒 Business & Policy

Thomson Reuters Fired Worker for Speaking Out About ICE, Former Employee Says

A lawsuit alleges that Thomson Reuters terminated a long-standing employee for protesting the company’s sale of data products to Immigration and Customs Enforcement (ICE). The legal action follows a protest letter signed by more than 200 employees regarding contracts with ICE and the Department of Homeland Security.

Kalshi CEO expects US DOJ to prosecute insider trading cases

Kalshi CEO Tarek Mansour expects the U.S. Department of Justice to prosecute insider trading on prediction markets, asserting that such activities constitute federal crimes. Mansour also advocated for a unified federal regulatory framework to replace the current patchwork of state-level regulations.

Allbirds Pivots from Footwear to AI

Allbirds, Inc. is rebranding as β€œNewBird AI” as it pivots from footwear to AI compute infrastructure and GPU-as-a-Service. Following the sale of its footwear assets and the announcement of a $50 million financing facility, the company’s stock price has surged by more than 600%.

Bosses say AI boosts productivity – workers say they’re drowning in ‘workslop’

A growing divide has emerged between executives, who view generative AI as a productivity booster, and employees struggling with “workslop.” This influx of inaccurate, AI-generated content requires significant manual correction, which often increases workloads and decreases overall job quality.

Every CEO and CFO change at US public companies, live from SEC

A new real-time tool monitors SEC filings to track executive and board changes at US public companies. The platform provides data on over 2,100 recent appointments, including insights such as an average new CEO total compensation of $8.4 million.

New bill would let New Yorkers hang solar panels from windows

The proposed “Sunny Act” in New York aims to legalize small, plug-in solar panels for apartment residents to help reduce electricity costs. This legislation would expand current solar regulations to permit easy-to-install units capable of offsetting power usage by up to 25%.

Anna’s Archive loses $322M Spotify case

A US federal judge has awarded $322 million in damages to Spotify and major music labels following a copyright infringement lawsuit against the pirate library Anna’s Archive. Although the judgment includes a worldwide injunction to disable ten of the site’s domains, enforcement remains challenging because the site’s anonymous operators have not been identified.

Krafton CEO used ChatGPT in failed bid to avoid paying US$250M bonus

A US court has ordered South Korean gaming publisher Krafton to reinstate the leadership of Unknown Worlds Entertainment following an attempt to avoid a $250 million earnout payment. The ruling rejected a plan developed by Krafton’s CEO using ChatGPT that sought to remove the studio’s executives and seize operational control.

US states can’t account for datacenter tax breaks. Literally

A report by Good Jobs First alleges that many US states and local authorities are violating accounting principles by failing to disclose significant revenue lost to datacenter tax subsidies. The nonprofit claims that only Washington, Texas, and Virginia properly report these multibillion-dollar losses in audited financial statements, while others use unregulated reports to obscure the impact.

UK told its Big Tech habit is now a national security risk

A new report by the Open Rights Group warns that the UK’s heavy reliance on US Big Tech companies for critical infrastructure poses a significant national security and economic risk. The findings suggest that this dependency undermines digital sovereignty and leaves essential public services vulnerable to foreign sanctions and service withdrawals.

Boeing deliveries soar past Airbus for the first time in years, but stay cautious

Boeing’s quarterly deliveries of 143 aircraft recently surpassed Airbus’s 114, marking the first time the company has led in deliveries in seven years. However, both manufacturers continue to face significant production hurdles, including Boeing’s 737 wiring issues and engine supply shortages affecting Airbus.

AI-powered mainframe exits are a bubble set to pop: Gartner

Gartner predicts that over 70% of AI-powered mainframe exit projects initiated in 2026 will fail due to an overestimation of generative AI’s ability to migrate complex legacy code and data. Additionally, the firm forecasts that 75% of vendors in the mainframe exit market will either pivot their business models or cease to exist by 2030.

Dutch court orders X to grant user information about why he was shadowbanned

The Amsterdam Court of Appeal has ordered X to grant shadowbanned user Danny MekiΔ‡ access to the personal data the platform holds about him. Although X challenged the ruling to protect its confidential business information, the court upheld the decision and imposed a €4,000 daily penalty for non-compliance.

Australia tried to ban kids from social media but failed, study finds

A new study reveals that Australia’s social media ban for users under 16 is largely ineffective, with 61% of underage children still maintaining active accounts. Many adolescents are reportedly using workarounds, such as VPNs, to circumvent the restrictions, raising significant concerns about the legislation’s actual impact on online safety.

Jury finds Live Nation and Ticketmaster illegally maintained ticketing monopoly

A federal jury has found Live Nation and its subsidiary, Ticketmaster, liable for illegally maintaining monopoly power in the ticketing market through anticompetitive practices. The verdict follows an antitrust lawsuit brought by the Justice Department and several state attorneys general alleging that the company’s control over the industry resulted in higher fees for concertgoers.

EIA to mandate nationwide energy use survey of data centers

The US Energy Information Administration (EIA) plans to implement a mandatory nationwide survey to track the energy consumption of data centers. This initiative, which is being preceded by pilot studies in several states, aims to provide transparency regarding the industry’s impact on the power grid and utility costs.

Andreessen & Horowitz donate $25M to pro-AI super PAC Leading the Future; total hits $51M

Venture capitalists Marc Andreessen and Ben Horowitz have donated $25 million to the pro-AI super PAC Leading the Future, bringing the group’s total funds to over $51 million. The donation aims to support the artificial intelligence industry and combat potential state-level regulations ahead of the November midterm elections.

EU Commission threatens Meta with interim ban on WhatsApp policies allegedly blocking AI rivals

The European Commission has threatened Meta Platforms Inc. with an interim ban on WhatsApp policies that allegedly block rival AI firms from operating on the platform. These measures are intended to prevent irreparable market harm pending Meta’s response to the commission’s supplementary statement of objections.

VA voters’ support for new data centers falls to 35% from 69% in 2023; 37% favor tax breaks

A Washington Post-Schar School poll reveals that support for new data center construction in Virginia has fallen from 69% in 2023 to just 35%. Additionally, only 37% of voters now support providing tax breaks for these facilities.

eToro to acquire crypto wallet Zengo for ~$70M, mostly in cash

EToro Group Ltd. has agreed to acquire crypto wallet provider Zengo in a deal valued at approximately $70 million, primarily in cash. The acquisition is intended to expand the trading platform’s reach into decentralized finance.

Snap to lay off ~1,000 employees, 16% of workforce, to cut costs and achieve profitability

Snap Inc. is laying off approximately 1,000 full-time employees, or 16% of its global workforce, to reduce costs and achieve profitability. CEO Evan Spiegel stated that the cuts aim to boost efficiency through advancements in AI technology and will include closing more than 300 open roles.

SoftBank seeks more banks for $40B OpenAI-backed loan, requiring $5B commitments each

SoftBank Group Corp. is inviting additional banks to participate in a $40 billion loan intended to back its investment in OpenAI. The deal is currently in a “soft launch” phase, with potential sub-underwriters being asked to commit approximately $5 billion each.

Democrats advised not to clash with $300M pro-AI lobby despite polls for tougher AI rules

Ahead of the US midterms, Democrats have been advised to avoid antagonizing a $300 million pro-AI lobby. This recommendation comes despite internal polling that indicates broad public support for stricter regulations on artificial intelligence.

Snap-Perplexity $400M AI search deal fails; Snap to announce significant layoffs Wednesday

A proposed $400 million AI search deal between Snap and Perplexity has reportedly collapsed. Additionally, Snap is expected to announce significant layoffs on Wednesday as part of a restructuring effort to separate its legacy Snapchat business from its upcoming AR glasses division.

Omdia: Facebook, Instagram, YouTube, and TikTok capture 90% of social ad revenue; Meta holds 70%

Facebook, Instagram, YouTube, and TikTok capture over 90% of social media advertising revenue, with Meta alone accounting for 70% of the total. According to Omdia, online advertising will drive significant market growth, with total global ad revenue projected to surpass $1.6 trillion by 2030.

πŸ”¬ Science & Society

AI-Assisted Cognition Endangers Human Development

AI-assisted cognition poses a risk of intellectual stagnation because large language models rely on outdated data that fails to reflect recent global developments. This “static cognitive skew” could potentially hinder the natural evolution of human knowledge and culture by resisting new geopolitical and social realities.

CRISPR takes a bold leap toward silencing Down syndrome’s extra chromosome

Researchers have developed a modified CRISPR/Cas9 technique to silence the extra chromosome 21 that causes Down syndrome by inserting the XIST gene into the chromosome. Although this study provides a successful proof-of-concept in human stem cells, further research is required to ensure safety and effectiveness before clinical application.

Sam Vimes ‘Boots’ Theory of Socio-Economic Unfairness

The “Boots’ Theory of Socio-economic Unfairness” illustrates how poverty is both more expensive and time-consuming due to the inability to afford durable goods. While the wealthy can invest in long-lasting items that save money, the poor are trapped in a cycle of purchasing cheap, short-lived products that require frequent, costly replacements.

God Sleeps in the Minerals

A collection of photographs from the Natural History Museum of Los Angeles County’s “Unearthed: Raw Beauty” exhibition was recently shared online. The post’s title, “God Sleeps in the Minerals,” has prompted a variety of user reactions, ranging from admiration of the images to debates over its religious implications.

Academic fraud may be the symptom of a more systemic problem

Academic fraud may be a symptom of a systemic misalignment of incentives within the scientific community rather than merely a result of individual misconduct. The current academic reward structure prioritizes high publication volumes and simple, impactful narratives over research transparency, complexity, and rigorous methodology.

Good Sleep, Good Learning (2012)

This article synthesizes neurophysiological research on sleep to provide practical strategies for enhancing learning, creativity, and mental performance. It also offers insights into managing various sleep disorders through informed lifestyle adjustments.

Projected warming will exceed the long-term thermal limits of rice cultivation

Research reveals that rice’s thermal limits have remained unchanged for 9,000 years, rarely thriving in areas where mean annual temperatures exceed 28Β°C or warm-season maximums exceed 33Β°C. By the end of this century, the land area in major Asian rice-producing nations surpassing these thresholds could expand ten to thirty times, threatening the food security of over one billion people.

Scientists discover why bread can cause weight gain without extra calories

Researchers at Osaka Metropolitan University found that consuming carbohydrate-rich foods like bread and rice can lead to weight gain in mice even without increased calorie intake. The study suggests this weight gain is driven by metabolic changes that reduce energy expenditure rather than overeating.

Viral victory: Iran is beating the land of tech bros in the social media wars

Iran is using tech-savvy content creators and AI-generated memes to target the Trump administration on global social media platforms. Despite facing severe domestic internet censorship, these digital strategies aim to influence international narratives through humorous and non-religious messaging.

Woman with three deadly diseases has ‘remarkable’ recovery after cell therapy

A 47-year-old woman suffering from three life-threatening autoimmune diseases has achieved treatment-free remission following the use of CAR T-cell therapy. The engineered treatment successfully reset her immune system by targeting rogue B-cells, allowing her to return to a near-normal life after years of failing medical interventions.

Don’t feel like exercising? Maybe it’s the wrong time of day for you

Research indicates that aligning exercise timing with your natural body clockβ€”morning for “larks” and evening for “night owls”β€”can maximize health benefits such as improved sleep and lower blood pressure. While matching workouts to your chronotype enhances cardiovascular and metabolic gains, experts emphasize that maintaining consistent and regular physical activity remains the most critical factor for fitness.

Britain’s atomic brain trust gives itself till 2030 to unpick fusion challenges

The UK Atomic Energy Authority (UKAEA) has unveiled a 2026-2030 strategy to tackle critical technical hurdles in fusion energy research, backed by a Β£2.5 billion budget. The roadmap focuses on achieving milestones in plasma control, fuel self-sufficiency, and commercial viability to support the eventual development of the STEP prototype power plant.

Steven Garcia missing: New Mexico UFO mystery deepens

The disappearance of Albuquerque contractor Steven Garcia, who had access to sensitive nuclear defense infrastructure, has fueled speculation regarding a pattern of missing defense-linked individuals in New Mexico. His case draws significant comparisons to the disappearance of General William Neil McCasland, a former official involved in space weapons programs.

Not that big of a problem? Users wait for discounts amid Lululemon β€œforever chemicals” investigation

The Texas Attorney General is investigating Lululemon to determine if the company misled consumers regarding the presence of “forever chemicals” (PFAS) in its products. While Lululemon claims to have phased out these substances in 2023, the investigation has sparked significant concern among consumers, particularly in the brand’s important Chinese market.

Neuralink’s brain-computer interface technology is being tested to help individuals with ALS communicate by translating neural signals into spoken language. The technology decodes imagined speech via an implanted chip, allowing patients to interact through digital devices and reconnect with their families.