Alvaro Lopez Ortega / 2026-07-15 Briefing

Created Wed, 15 Jul 2026 20:32:46 +0000 Modified Wed, 15 Jul 2026 20:33:17 +0000
5077 Words

Microsoft continues using dark patterns to push Edge, a Mozilla-commissioned report reveals, with misleading wording and forced resets—though EU rules curb this in Europe. A former employee’s lawsuit alleges AWS data centers secretly consume water year-round, contradicting its “water positive” claims. UK MPs warn Treasury cold feet could sink Whitehall’s £1.15B shared services push, calling the funding reluctance a poor signal. Meanwhile, KeyBanc analysts say Salesforce’s Agentforce struggles with messy data, despite Salesforce touting it as its fastest-growing product.

📰 News

Salesforce’s Agentforce isn’t winning over clients, KeyBanc analysts claim

KeyBanc analysts report that Salesforce’s Agentforce is struggling to gain client traction due to messy customer data and a product that “just isn’t there.” Salesforce counters that Agentforce is the fastest-growing product in the company’s history.

Microsoft’s Dark Patterns for Edge

A Mozilla-commissioned report details how Microsoft continues to use harmful design patterns—such as misleading wording, forced actions, and resetting Edge as the default during system migration—to steer users away from browser choice in Windows 10 and 11. The European Union’s Digital Markets Act has curbed some of these tactics in the EEA, but users in the US, India, and the UK still face the full range of manipulative practices.

MPs fear Treasury cold feet could sink Whitehall’s £1.15B shared services push

A parliamentary report warns that the UK Treasury’s reluctance to fully fund Whitehall’s £1.15 billion shared services initiative could derail the project, criticizing a “do as we fund, not as we do” approach as sending a “very poor reputational signal.” MPs fear the resulting lack of commitment may undermine the push for greater efficiency across government departments.

LegacyHive: ‘Bone-shattering’ zero-day from Microsoft’s serial tormentor not the haymaker promised

A highly anticipated zero-day exploit from Microsoft’s serial tormenter, dubbed “LegacyHive,” has fallen short of expectations, with experts describing it as a useful post-compromise tool rather than the devastating haymaker promised. The exploit requires significant technical skill to assemble, limiting its practical impact as a weapon.

AWS sustainability claims don’t hold water, lawsuit alleges

A former employee’s lawsuit alleges that Amazon’s data centers in Virginia are secretly consuming significant water year-round, contradicting AWS’s “water positive” sustainability claims. The suit contends that the company’s public relations efforts regarding water conservation are misleading.

A moment of silence, please, for the final release of Debian on x86-32

The final release of Debian for the 32-bit x86 architecture has arrived with versions 13.6 and 12.15, marking the end of support for this legacy platform. Users are advised to migrate to 64-bit systems for future updates.

EU competition decision hands SAP customers more leverage in contract talks

An EU competition decision gives SAP customers more leverage in contract negotiations, particularly over maintenance fees for holdouts on older ECC systems. However, the ruling is not expected to trigger a widespread shift to third-party support.

Software bloat? This elevator needs an 8GB Core i5

An elevator system unexpectedly requires an 8GB Core i5 processor, sparking concerns over software bloat. The high hardware demands have led to the joke that this elevator is “headed for floor Bork.”

OpenMandriva’s accused repo wrecker says it wasn’t sabotage – it was a message

An ex-contributor to OpenMandriva denies that his actions were sabotage, insisting he was not a rogue administrator, had not left the project, and never intended to harm the distribution. He claims the controversial changes were meant as a message rather than an act of vandalism.

The Star Wars cantina scene shows we need a new hope for the agentic web

The article uses the Star Wars cantina scene as a metaphor for the current state of the web, arguing that publishers’ attempts to block “their kind” (likely AI bots or aggregators) are a protective measure that ultimately won’t dismantle the dominant monetization systems controlled by Google and Meta. It calls for a new approach—a “new hope”—to challenge these tech giants and create a more equitable agentic web.

South Korea to launch universal basic AI chatbot

South Korea is launching a universal basic AI chatbot, with a tender calling for providers to power the free service using local large language models. The government will supply some GPUs to support the initiative.

Google Cloud’s VMware service loses resilience due to a dud update

A faulty update compromised the resilience of Google Cloud’s VMware service, coinciding with VMware’s own warning about a critical vulnerability in its load balancer. The flaw poses a significant security risk for users of the platform.

India’s biggest nuclear plant blueprints posted online after contractor refuses ransom

A ransomware group known as World Leaks posted 19,000 files online, including blueprints and supplier details for India’s largest nuclear plant, after contractor Reliance Group refused to pay a ransom. Nuclear security experts warn the leak poses a serious safety risk to the Kudankulam plant, as it could help attackers map support systems and identify vulnerabilities. Reliance confirmed a partial breach of data hosted on a third-party server and has informed the government.

Murati’s Thinking Machines Releases Open-Weights 975B Parameter LLM

Murati’s Thinking Machines has released Inkling, an open-weights multimodal LLM with 975B total parameters (41B active), a Mixture of Experts architecture, and a 1M token context window supporting text, image, and audio input. The model excels in general intelligence, agentic coding, and instruction following, with controllable thinking time and well-calibrated confidence predictions. Inkling is available for fine-tuning on the Tinker platform.

Inkling: Our Open-Weights Model

Inkling is a new open-weights Mixture-of-Experts model with 975 billion total parameters (41 billion active) and a 1-million-token context window, pretrained on 45 trillion tokens of text, images, audio, and video. It supports native multimodal reasoning with controllable thinking effort and is available for fine-tuning on Tinker alongside a smaller preview model, Inkling-Small. While not the strongest overall model, it is designed as a broad, customizable foundation model with strong performance across agentic, coding, and multimodal tasks.

FreeBSD 16 Removes Last GPL Code

FreeBSD 16 has removed the last piece of GNU GPL-licensed code from its base system by retiring the dialog implementation and replacing it with bsddialog, along with the removal of the dpv component. This completes the elimination of the entire GNU sub-tree from FreeBSD’s base system. FreeBSD 16.0 is expected to be released in December 2027.

Starlink doubled its unlimited aviation plan from $10,000 to $20,000 per month and raised equipment costs, effective August 7th with no advance notice. CEO NJ Correnti called the move “commercially reckless” and a “bait and switch,” pausing fleet installations, while an aircraft broker reported an awkward situation with a client due to the sudden increase.

What designing 54 computer science cards taught me about graphic design

The author created Algodeck, a 54-card deck on computer science, using code to generate consistent illustrations with a red-and-blue duotone system. They discovered that graphic design (solving communication problems) and software engineering (solving information problems) both rely on systems-thinking, where constraints and rules enable efficient design.

Open-source memory for coding agents, synced over SSH

deja-vu is a zero-dependency binary that indexes conversation logs from AI coding agents (Claude Code, Codex, opencode) into a fast, searchable memory layer. It provides features such as retroactive search, agent recall via MCP, auto-recall hooks, secret redaction, stats, sharing, and sync—all running locally with no external services.

Codex Micro

The Codex Micro, a compact keyboard designed with Work Louder, serves as a command center for agentic work, featuring live RGB status indicators for each agent and tactile controls like a joystick and dial for launching workflows and adjusting reasoning levels. It includes 32 custom keycaps, connects via Bluetooth or USB-C, and is compatible with Mac and Windows, priced at $230.

misa77 - a codec that decodes 2x faster than LZ4 (at better ratios)

A new codec called misa77 achieves decompression speeds over twice as fast as LZ4 (up to 5219 MB/s vs 2505 MB/s) while offering better compression ratios (42.64% vs 47.59%), though compression is significantly slower. The performance gains come from a smart format that reduces branches and optimizes decompression for out-of-order cores.

Running Gemma 4 26B at 5 tokens/sec on a 13-year-old Xeon with no GPU

A 13-year-old HP StoreVirtual server with dual Xeon E5-2690 v2 CPUs and no GPU was repurposed to run Google’s Gemma 4 26B mixture-of-experts model at about 5 tokens per second. The author modified the inference code to work without AVX2 or FMA3 instructions, achieving reading-speed text generation on hardware that predates the model’s architecture.

The DEA announced it will temporarily place synthetic 7-hydroxymitagynine (7-OH) and three related substances into Schedule I, citing no accepted medical use and high abuse potential. The order targets concentrated synthetic products above a specified threshold, exempting natural botanical kratom with trace amounts. Once effective, the manufacture, distribution, and possession of these substances will face criminal and civil penalties under the Controlled Substances Act.

OpenAI loses trademark dispute at EU court

The EU General Court ruled that “OPENAI” is purely descriptive for certain software and IT services, upholding the EU Intellectual Property Office’s partial rejection of the trademark application. The decision can still be appealed to the European Court of Justice.

The well-calibrated Bayesian [pdf] (1982)

This 1982 paper examines the concept of a well-calibrated Bayesian, focusing on the calibration of subjective probabilities. It demonstrates how Bayesian forecasts can be evaluated by comparing predicted probabilities to observed frequencies over time.

Hack Reveals Suno AI Music Generator Scraped YouTube, Deezer, and Genius

A hacker breached AI music tool Suno and shared data revealing it scraped millions of songs and lyrics from platforms like YouTube Music, Deezer, and Genius, while also accessing user and payment information for hundreds of thousands of customers. The leaked code confirms Suno trained on vast amounts of copyrighted music, including decades’ worth of recordings, bolstering ongoing lawsuits from the record industry.

StyleSeed – a design-rules engine so AI agents stop building generic UI

StyleSeed is a free, MIT-licensed design engine for AI coding tools (Claude Code, Codex, Cursor) that enforces coherent, non-generic UI through 74 rules, 19 skills, and 7 brand skins, scoring output quality to ≥80/100 before delivery. It ships configuration files (CLAUDE.md, AGENTS.md, .cursorrules) to lock design judgment and prevent the generic AI look.

Mysteries of Telegram Data Centers

Telegram operates five data centers in Miami (DC1, DC3), Amsterdam (DC2, DC4), and Singapore (DC5), with accounts assigned by phone number country code at registration. DC3 currently has no users and no longer accepts new registrations, while DC2 serves users from certain countries (e.g., Germany) but was not detected by a common bot due to a flawed query method. DC5 is known for frequent outages, affecting its users.

Windows GDID Device ID Unchangeable

Microsoft confirmed the Global Device Identifier (GDID), a persistent Windows device ID linked to Microsoft Accounts that cannot be disabled without breaking activation and Store apps. The FBI used GDID to track alleged Scattered Spider member Peter Stokes across multiple VPNs and countries, as the identifier remains unchanged despite IP changes. Privacy researchers have raised concerns over the lack of user consent, control, and public documentation.

When We Fight, We Win: Why I Am Suing Northwestern University and US Government

Steven Thrasher has filed a lawsuit against Northwestern University and several U.S. government officials, alleging they conspired to destroy his academic career due to his Palestine solidarity speech. He claims the university unlawfully terminated his employment, denied his tenure, and prevented him from teaching after succumbing to federal pressure. The suit cites breach of contract, civil rights violations, and retaliation for protected speech.

The most popular number in Hacker News titles is 1 after refining the regex to handle decimals and filter out years, overtaking the initial naive result of 2. Small numbers dominate due to listicles (e.g., “5 things”) and version numbers (e.g., “2.0”).

Germany maybe found a new source of renewable energy

Two exploratory boreholes (EB1 at 100 m and EB2 at 506 m) were drilled in Germany’s Rhenish Lignite Mining Area to assess geothermal potential as a replacement for the Weisweiler coal plant, set to shut down in 2029. The drilling provided data on Cenozoic and Palaeozoic deposits, enabling revisions to structural geological models and stratigraphic boundaries. These boreholes represent the initial phase of geothermal exploration, with further seismic surveys and deep wells planned.

Briar is in maintenance mode

Briar has entered maintenance mode, focusing only on essential security updates and bugfixes after the team reversed a previous decision to shut down the project due to supportive feedback. No major feature changes are being pursued at this time.

Make senders work to get into your inbox

Hi HN :) really excited to share this with you.The one thing AI reliably does is generate noise. Half the tools I see launch are just machines for producing more noise across more channels. And people are starting to see this in the form of emails in their inboxes as spam filters are struggling.There used to be a useful signal in email: the effort a sender put into customizing a message was a rough proxy for how relevant it actually was. AI killed that. Now it’s customized slop with the appearance of effort with none of the cost. It is painful that the open internet / open channels have been abused like this.Captchainbox applies the idea of proof-of-work to email. If a sender is willing to do a bit of work to reach you, the message is more likely to be worth your time and the sender more likely to be real. The work is a traditional captcha. You can also set a pay-to-deliver amount if you want more friction. The proceeds of the delivery payment after transaction costs go to the Internet Archive and the EFF. The tool currently works by authing with your Gmail or Outlook and during launch time I make this completely free as a lifetime deal (with optional payment if you wanna support).How it works: Captchainbox builds a whitelist automatically from the metadata of your past correspondence. If you’ve emailed an individual address, that sender can reach you. If you talk to several people at the same domain, we whitelist the whole domain. If one transactional-looking sender has sent you more than 10 emails, we treat it as a transactional domain and let it through. This whitelist is for you to change whenever you want. It continues to build organically as you converse with more addresses.Incoming mail is checked against that whitelist. Senders already on it land in your inbox as normal. Anyone else gets archived (never deleted) and is sent a challenge. This can be the captcha or the payment link. Once they solve it, their email is pulled out of the archive and put back into your inbox.if you want to see what this looks like from a sender’s point of view, send me an email here: doerpfelix15@gmail.comThe service only ever reads metadata, never message content. And since nothing is ever deleted, you can’t lose a message. There is a legitimate risk / downside: if you sign up to a new service, these emails also land in the archive. Since we do not process the content, a first-time sender who can’t solve the challenge (say an automated activation email) will sit in your archive until you spot it.Happy to answer anything! :)

Sleep regularity is a stronger predictor of mortality risk than sleep duration (2023)

A 2023 study found that sleep regularity—maintaining consistent bed and wake times—is a stronger predictor of mortality risk than sleep duration. Consistent sleep patterns may be more critical for health than the total number of hours slept.

Prioritize mental health, and why communication is so important

The author describes a pattern of losing motivation shortly after starting jobs, leading to poor communication, slow work, and low quality—resulting in being fired twice. They realize their struggles with severe depression and lack of discipline are not shared by peers, leaving them questioning their abilities. Despite using LLMs for help, they acknowledge their sloppy work habits persist.

Weathergotchi – an open-source climate Tamagotchi

Weathergotchi is an open-source, battery-powered temperature and humidity data logger with an always-on e-paper display, built around an ESP32-S3 with deep sleep and external EEPROM. It records ambient conditions over time, stores readings in non-volatile memory, and shows current data along with a history graph, achieving over a week of operation on a small Li‑Po battery. All hardware schematics, PCB layout, firmware, and enclosure CAD files are publicly available.

DSLs Enable Reliable Use of LLMs

Domain-Specific Languages (DSLs) provide clear boundaries that guide LLMs to generate code reliably, overcoming the limitations of upfront specification. By iteratively building a DSL with an LLM as a partner, developers can use the LLM as a natural language interface to the DSL, which then serves as the key source of truth for software systems.

Telegram Serverless

Telegram Serverless enables developers to run backend code for bots and Mini Apps directly on Telegram’s infrastructure without provisioning or managing servers. Developers write plain JavaScript modules, deploy them with a single command, and Telegram executes them in an isolated V8 sandbox with a built-in database, automatically routing updates to matching handler files. The platform provides a CLI for deployment and database migrations.

America pays workers just 27% of what its wealth allows – the worst in the OECD

The United States ranks last among OECD nations in paying workers relative to its wealth, providing only 27% of what its economic resources allow. Additionally, the country underperforms in health, scoring just 80% of its potential—a figure that has stagnated for 25 years—and fails to meet its constitutional promise to “promote the general Welfare” compared to other high-income countries.

Who’s running all those tiny RPKI servers?

RPKI uses cryptographically signed ROAs to prevent BGP prefix hijacks, with most ROAs published by the five RIRs. However, a long tail of smaller servers—run by cloud providers, ISPs, hobbyists, and others—also contributes to the global RPKI dataset. This article investigates who operates these small servers (announcing fewer than 1,300 ROAs) and why they run independent publication servers despite RIR-provided services.

TS-2026-009: Insecure argument handling in Tailscale SSH permitted root access

Tailscale versions before 1.98.9 had two vulnerabilities: a malformed HTTP request could permanently pin a CPU core via Serve or Funnel, and a Linux SSH username starting with a dash could grant root access in violation of ACLs. Both issues are fixed in version 1.98.9 or newer.

Texas factory cost $469M using old equipment, makes zero artillery shells

The Army spent $469 million on a Texas ammunition factory to produce 155 mm artillery shells, but it has not delivered a single round in two years. A Pentagon watchdog attributed the failure to a “high-risk” decision to use older machines converted from a previous production line rather than new equipment.

Data centers have hiked electricity prices on the public by $23B

Data centers have driven $23 billion in electricity price increases for customers through 2028, primarily in the PJM market. While tech companies pledge to cover costs, complex state utility commission pricing can allow data centers to shift expenses to other ratepayers by exploiting metrics like coincident peak demand.

Solving 20 Erdős Problems with 20 Codex Accounts Running in Parallel

Star Fleet is an AI system that uses up to 20 parallel agentic harnesses, each with a dedicated GPT-5.6 instance, to solve open mathematics problems in Lean 4. It has proposed a solution to Erdős Problem #123, proving that for any pairwise-coprime integers a, b, c > 1, every sufficiently large integer can be expressed as a sum of distinct terms a^i b^j c^k such that no selected term divides another.

Vancouver PD website features Quick Escape button that wipes itself from history

The Vancouver Police Department website provides instructions for reporting emergencies via 911 and non-emergencies through alternative options. It also emphasizes community outreach and education as core initiatives for maintaining public safety.

Apple raises AppleCare+ prices for Macs and iPads by $0.50/month and $5/year, only for new sign-ups

Apple raised AppleCare+ monthly prices by $0.50 and annual prices by $5 for Macs and iPads, with the increases applying only to new sign-ups. The move is attributed to a global memory shortage and other challenges.

Lumin Digital raises over $115M at $1.6B valuation

Lumin Digital raised over $115 million at a $1.6 billion post-money valuation, with $45 million from Light Street Capital and $70 million from 15 client investors. The company will use the capital to expand into lending, CRM, and payments as it targets larger banks. The funding also aims to reassure conservative financial institutions of Lumin’s long-term financial stability.

ASML discussed raising EUV prices with TSMC, plans 10% DUV hike; TSMC resists

ASML has discussed raising prices for its EUV systems with TSMC and plans to charge 10% more for its DUV systems. TSMC is resisting these price plans, which could escalate tensions and impact overall chip production costs.

OnePlus to cease US/Europe ops this week; Realme exits China in Oppo restructuring

OnePlus will cease operations in the US and Europe as early as this week as part of a larger restructuring by its parent company Oppo. Additionally, Oppo’s sibling brand Realme will exit the Chinese market.

Hyundai S.Korea auto workers partial strike over wages, AI, new humanoid robot

Hyundai Motor auto workers in South Korea have launched a partial strike, expressing concerns over wages, AI integration, and the potential deployment of the company’s new humanoid robot, “Atlas,” on production lines. The union has stated that the robot will not be used on the factory floor without worker approval.

TikTok Shop sellers use AI creators/twins to boost sales; US sales to hit $23.4B in 2026, up 48% YoY

TikTok Shop sellers are using AI-generated avatars and digital twins for product demonstrations, raising concerns about consumer trust and competition with human creators. Despite the tensions, eMarketer projects US sales on TikTok Shop will reach $23.4 billion by 2026, a 48% year-over-year increase.

Google: third-party app stores on Google Play US from July 22; joint motion with Epic withdrawn

Google and Epic have withdrawn their joint motion to modify a court injunction, forcing Google to allow third-party app stores within Google Play starting July 22. Google will share its entire app catalog with these stores unless developers opt out, charging a $5,000 annual fee and imposing security requirements. This development ends the legal battle and paves the way for increased competition in Android app distribution.

UAE gets broader US AI chip access after Iran war aid; G42 free to buy 9 months, plans US firm

The United Arab Emirates secured expanded access to U.S. AI chips after assisting with airstrikes against Iran, missile interceptions, and maintaining oil flow through the Strait of Hormuz, allowing its AI firm G42 to buy freely for at least nine months as it plans to become a U.S. company. The decision caps the UAE’s push for economic diversification and underscores the growing role of chips in diplomacy.

How to unionize your tech workplace

Tech workers are increasingly pursuing unionization due to mass layoffs, AI anxiety, and corporate surveillance, with key benefits including protection against arbitrary termination and better severance terms. However, organizing is challenging because the federal labor board has been weakened and the current job market offers less leverage for workers.

Inventing ELIZA - How the First Chatbot Shaped the Future of AI

A new book, Inventing ELIZA, offers the first comprehensive critical analysis of the original ELIZA chatbot, drawing on rediscovered source code and previously missing scripts. The research reveals that Joseph Weizenbaum’s system was far more sophisticated than previously documented, featuring innovations well ahead of its time. The companion website includes a faithful recreation of the chatbot and ongoing research updates.

Comparing Obelisk with Temporal and Restate

Obelisk is a workflow runtime that loads and runs application components itself, while Temporal and Restate are orchestrators that run code in separate workers or endpoints. The article compares their Rust APIs, determinism boundaries, and deployment models by implementing the same temperature-fetching workflow, highlighting Obelisk’s use of WIT interfaces and join sets, Temporal’s macros and durable futures, and Restate’s service-based model.

a bunch of stuff i used to not know about K&R C

Pre-ANSI C lacked a void type (functions returned int), used long float as a synonym for double, and allowed only limited type specifier combinations (e.g., unsigned only with int, no signed). Integer promotion rules varied across early compilers due to the absence of a standard.

i’ve been thinking about null pointers

Null pointers are conceptually strange because a single value conveys multiple semantic meanings (error, nonexistence, allocation flag). The article proposes that since many pointer values are invalid addresses in userspace (e.g., low values), these could encode additional information like error types, effectively replacing tagged unions through a technique similar to NaN boxing.

@clickhouse/rowbinary: when your library is also a parser compiler

The author notes that building a robust parser generator involves so many opinionated decisions it becomes akin to creating a nuanced natural language translation system. To address this complexity, they are leveraging modern coding LLMs in the @clickhouse/rowbinary library to better satisfy individual user needs.

How C++20 improved the for-loop syntax

C++20 introduced range-based for statements with an initializer, allowing syntax like for (int i=0; auto&& it: vec) to combine an index variable and element reference in one line. This improvement eliminates the need for a separate index declaration, matching the convenience of Python’s enumerate or Lua’s ipairs. The author praises this as a useful quality-of-life feature that simplifies code across a codebase.

C Strings: A 50-Year Mistake

C’s null-terminated strings are a flawed design that wastes efficiency and causes bugs by omitting length information, forcing repeated strlen calls and ambiguous semantics. Modern languages prefer length-based strings, which store a pointer and size together, enabling safer and more efficient operations.