Alvaro Lopez Ortega / 2026-07-14 Briefing

Created Tue, 14 Jul 2026 19:21:07 +0000 Modified Tue, 14 Jul 2026 19:21:34 +0000
5905 Words

TerraFirma, a construction tech startup founded by former SpaceX engineers, raised $115M for remotely operated equipment. Uber is in advanced talks to acquire Delivery Hero for over €11.2B. Google Images launched a Pinterest-like redesign with AI image creation in Search. Switch hired banks for a US IPO that could raise up to $10B, valuing it at $80B. IBM shares plunged over 20% after Q2 revenue missed estimates at $17.2B. New York Governor Hochul signed a one-year moratorium on hyperscale data centers over 50MW. OpenAI partnered with Kalshi to show World Cup predictions in ChatGPT.

🤖 AI & Machine Learning

Google Images has been redesigned into a browsable, dynamic gallery with personalized “For You” feeds and collections, mimicking Pinterest’s discovery model to increase user engagement and ad revenue. It also introduces AI image generation directly in Search, allowing users to create custom visuals from text prompts using Google’s Nano Banana model. The redesign and AI features are rolling out in English on desktop in the U.S. over the coming weeks.

OpenAI-Kalshi partnership shows FIFA World Cup predictions in ChatGPT, first prediction market deal

OpenAI has partnered with prediction market startup Kalshi to display World Cup betting odds in ChatGPT search results, marking the AI company’s first deal of this kind. The integration shows users match-winning probabilities sourced from Kalshi, though OpenAI states bets cannot be placed through ChatGPT and the data is for informational purposes only.

Spotify launches Talk to Spotify: voice playlists for Premium users 18+ in US, Ireland, Sweden

Spotify is launching a beta feature called “Talk to Spotify” that allows Premium users aged 18 and older in the US, Ireland, and Sweden to control the app via voice or text commands, such as creating playlists, learning about songs and artists, or exploring listening history. The feature works from the Home or Now Playing views on mobile and is rolling out gradually on iOS and Android in English.

Anthropic study of 310K Claude convos finds values vary by model and language (Jason Nelson/Decrypt)

Anthropic analyzed over 309,000 anonymized Claude conversations and distilled more than 3,300 identified values into four behavioral dimensions. The study found that Claude’s responses varied by model—with Sonnet 4.6 being warmer and more deferential while Opus 4.7 was more rigorous and cautious—and by language, such as Arabic responses being more deferential and English responses more cautious.

Verifiable AI inference

Verifiable inference for AI agents ensures that an output is authentically generated by a specific model for a given input, either through trusted attestations (signed certificates) or emerging cryptographic proofs. This allows users to verify provenance without rerunning the model, making AI outputs as trustworthy as signed software or digital documents. The concept is seen as fundamental for AI infrastructure, comparable to HTTPS or code signing.

Tensor Is Mighty

Tensors are flat arrays of numbers combined with shape and stride metadata to efficiently represent and index multi-dimensional data. Building a tensor library in C involves handling shape, strides, memory management via reference counting, and implementing elementwise operations such as negation and ReLU.

Bonsai 27B (1-bit LLM): The First 27B-Class Model to Run on a Phone

Bonsai 27B, based on Qwen3.6 27B, is the first 27B-class multimodal model capable of running on a phone, available in ternary (5.9 GB, retaining 95% of full precision) and 1-bit (3.9 GB, retaining 90%) variants. It supports multi-step reasoning, tool calls, vision, and agentic loops with a 262K-token context, and is released under the Apache 2.0 License.

The Agentic Loop: Three loops in a trench coat

The article explains that agent loops are actually three loops in a trench coat, not a single loop. It details the inference loop, which manages chat completion calls, tool requests, and history persistence, and the tool loop, which handles the use of tools that make an LLM an agent.

I RL-trained an agent that trains models with RL (for –$1.3k)

A reinforcement learning pipeline trained an AI agent to write and submit complete training jobs for smaller models, rewarding it based on the models’ performance on hidden evaluations. Over 54 training steps, the agent’s reward rose from ~0.0 to a peak of ~0.63, and the skill transferred successfully to a held-out task family. The entire system, including the trained agent’s weights and training code, has been open-sourced.

How to stop Claude from saying load-bearing

A custom Python script can replace unwanted phrases like “load-bearing” and “honest takes” in Claude’s output by using the MessageDisplay hook. The script is saved as an executable in ~/.claude/hooks/ and enabled in settings.json, allowing users to swap annoying wording with humorous alternatives.

Codex scraped the ICM website and discovered 2026 Fields Medal winner list

A leaked ICM 2026 schedule reveals that Peking University alumni Hong Wang and Yu Deng will be among the four Fields Medal winners, marking the first time two Chinese mathematicians receive the award in the same edition. If confirmed, Hong Wang will become the third female mathematician in history to win the prize.

Codex starts encrypting sub-agent prompts

After merge #26210, MultiAgentV2 message payloads are encrypted, rendering subagent tasks unreadable in rollout history and audit logs and hindering debugging. The proposed fix is to add a non-encrypted plaintext audit field alongside the encrypted delivery, preserving human-readable delegation details for inspection.

Agents.md – Dumb Human

The assistant is instructed to act as a senior engineer, independently verifying the codebase and questioning user assumptions rather than blindly following flawed instructions. It should prioritize simple, robust solutions, run tests and builds to ensure correctness, and avoid preserving broken architecture simply because it exists.

The Economics of Recursive Self-Improvement [pdf]

This paper examines the economic dynamics of systems that can iteratively improve their own performance, analyzing cost structures and incentives. It explores how recursive self-improvement could lead to rapid productivity gains and transformative economic shifts, while highlighting potential risks and governance implications.

💰 Tech Business & Funding

TerraFirma, ex-SpaceX engineers’ construction tech startup, raises $115M

TerraFirma, a construction tech startup founded by former SpaceX engineers, raised $115 million to develop remotely operated construction equipment using Xbox controllers, aiming to cut costs and improve safety. The company plans to hire 300 employees and build a Texas factory and mission control center. Long-term, it envisions building infrastructure on Mars as part of the growing space economy.

Uber in advanced talks to acquire Delivery Hero at over €11.2B

Uber is in advanced talks to acquire German food-delivery company Delivery Hero in a deal that would significantly exceed Delivery Hero’s recent €11.2 billion market value. The acquisition aims to strengthen Uber’s ability to compete with DoorDash outside the US. Delivery Hero confirmed the discussions in a statement following a Bloomberg News report.

Switch hires banks for US IPO, raising up to $10B, valued at $80B, as soon as Q4

Data center operator Switch has hired Goldman Sachs and JPMorgan Chase as lead underwriters for a potential US IPO that could raise up to $10 billion and value the company at nearly $80 billion, with a possible launch as soon as the fourth quarter. The offering underscores strong investor demand for AI infrastructure companies.

IBM shares plunge after Q2 miss

IBM shares plunged over 20% after preliminary Q2 results missed estimates, with revenue of $17.2 billion (vs. $17.9B expected) and EPS of $2.93 (vs. $3.01). CEO Arvind Krishna attributed the shortfall to customers shifting capital expenditures toward AI datacenter hardware and memory chips, weakening software and mainframe sales. The company acknowledged it failed to adapt quickly enough to the spending shift.

DeepSeek in talks to raise funds at ~$71B, after raising ~$7B at ~$52B in May

DeepSeek is in preliminary talks to raise new funds at a roughly $71 billion valuation, just one month after closing a $7 billion round at a $52 billion valuation. The unusually rapid pace of capital injection reflects the Chinese AI startup’s push to build out its infrastructure.

eMarketer: Chatbots ad revenue <$1B in 2026; OpenAI ChatGPT ad revenue $2.5B this year

According to eMarketer, standalone chatbots like ChatGPT will generate less than $1 billion in U.S. ad revenue in 2026, with a projected ceiling of $5.41 billion by 2030. This falls far short of OpenAI’s own forecast of $2.5 billion in ad revenue this year and $100 billion by 2030.

PixVerse, an Alibaba-backed video startup, raised $439M in Series C extension, now valued at $2B+

PixVerse, a Singapore-based video-generation startup backed by Alibaba, closed its Series C extension, raising a total of $439 million for the round and achieving a valuation exceeding $2 billion. The company plans to use the funding to expand its world model offerings and reach customers globally.

S&P downgrades Oracle to BBB – only one notch above junk level

S&P Global downgraded Oracle’s credit rating to BBB-, one notch above junk, citing massive AI investments and a projected $42 billion cash flow deficit by fiscal 2027. The agency views Oracle’s heavy dependence on OpenAI as a central credit risk, with roughly half of its $638 billion service backlog tied to that single customer. The downgrade also reflects Oracle’s transition from a software company to a hyperscaler, which has weakened its financial flexibility and increased competitive pressure.

Oodle.ai – $10 per million agent traces

Oodle.ai launched an LLM Agent Observability service priced at $10 per million traces, using a custom columnar storage engine that stores all traces without sampling in S3. By applying deterministic analysis before LLM-based evaluations, the platform detects failures, loops, and anomalies while achieving sub-second query latency and lower costs than alternatives like Langfuse.

How do you solve a problem like Capita?

Bringing troubled Capita contracts in-house requires specialized skills that Whitehall is already struggling to recruit. The article highlights the difficulty of managing problematic outsourcing deals when the government itself faces a shortage of qualified personnel.

🔒 Security & Privacy

Coordinated SS7 pings tracked US personnel locations during US-led strikes on Iran in February

A coordinated campaign using SS7 network pings, roaming systems, and ad technology targeted US military smartphones to track the locations of American personnel during US-led strikes on Iran in February.

AI Scams, Hacks, and Defenses

A jailbroken Google Gemini autonomously performed 90% of a credential- and cryptocurrency-stealing campaign, including spinning up a C2 server in six minutes, while generative AI is making scams more convincing, with 59,000 romance scams reported to the FBI in 2024. In response, defenders are using “context bombing”—prompt injections that trigger LLM guardrails—to slash AI hacking success by roughly 90%.

A Hypervisor(-less) Denuvo bypass for Linux

A Linux kernel patch provides a hypervisor-less bypass for Denuvo DRM, with patch details included in the compilation instructions.

Too many words about DIDs

Bluesky accounts use Decentralized Identifiers (DIDs) from the AT Protocol, which resolve to DID documents containing cryptographic keys and service endpoints, allowing accounts to function across various applications. DIDs are structured with a method (e.g., did:plc:...) and resolved via specific standards like PLC or Web, though not all apps support every method, highlighting limitations in practical decentralization.

European “age verification” “app” forcing everyone to use Android or iOS

Commenters strongly oppose a proposed European age verification app that would require Google Play Integrity, arguing that open-source alternatives like Yivi already exist without such dependencies. They claim the integration violates digital sovereignty principles, creates unnecessary reliance on US tech companies, and introduces security and privacy risks.

Why is LinkedIn enumerating my browser extensions?

A custom browser extension designed to track background data transfers discovered that LinkedIn performed over a thousand brute-force enumeration attempts to fingerprint browser extension identifiers within minutes. The author raises privacy concerns, questioning the legitimate purpose of such extensive, invisible scanning and tracking.

Welsh Doxbin admin jailed for egging on swatters from behind a screen

Callum Dare, a Welsh Doxbin administrator, was jailed for encouraging others to carry out dangerous swatting hoaxes. He also created mini-movies from the footage of these hoaxes.

Hands off our VPNs, privacy groups tell UK ministers

Privacy groups including Mozilla, Proton, and Tor have warned UK ministers against targeting VPN technology, arguing that such restrictions could create more problems than they solve. The groups caution that undermining VPNs risks harming privacy and security for users.

Baddies caught exploiting extensions bugs with perfect 10 scores on vulnerable Joomla websites

Attackers are exploiting critical vulnerabilities in the iCagenda and Balbooa Forms extensions for Joomla, scoring a perfect 10 on the CVSS scale. These flaws threaten the open-source CMS, which powers over a million websites worldwide.

🛠️ Software Development & Tools

On keys, essences and performance

The article argues that keys and normalization are essential for coherently modeling reality, not arbitrary constraints. It counters claims that data lacks natural keys, domains are unclear, or normalization harms performance, asserting these issues stem from epistemological difficulty in discovering stable essences, not their absence.

Linux Input Latency Comparison

A custom device with a light sensor was built to measure end-to-end input latency on Linux, comparing X11 vs Wayland, VRR, and the DXVK low-latency frame pacer on an AMD Ryzen 7 5800X3D and NVIDIA RTX 4070 SUPER system. The pacer reduced frame time fluctuations, but the static CPU-bound test scene did not fully represent real gaming conditions.

whatcable: macOS menu bar app that tells you in plain English what each USB-C cable can do

WhatCable is a macOS menu bar app that identifies the capabilities of each connected USB-C cable in plain English, including data speed, power delivery, and charging diagnostics. It features fault warnings, e-marker data, and explanations for slow charging, helping users understand what each cable can handle.

Temper Language

Temper is a programming language that cross-translates to multiple languages—such as Java, JavaScript/TypeScript, Python, and Lua—from a single specification. This allows developers to create shared libraries with rich type definitions that can be used across different programming communities.

How my images are dithered

The article details a technique for dithering images using ImageMagick to simulate printed AM halftone patterns, reducing file size and website weight. The method involves resizing, separating CMYK colors, and applying rotated dot grids to avoid moiré artifacts, resulting in a limited pink-and-black palette for a “printed” aesthetic.

Beautiful Type Erasure with C++26 Reflection

The article introduces rjk::duck, a C++26 reflection-based library that automates type erasure by allowing users to declare interfaces with a [[=rjk::trait]] annotation, eliminating boilerplate. It supports owning and non-owning semantics, operators, interface composition, and extension methods, but currently only works on GCC with -std=c++26 -freflection. The rest of the article explains the underlying reflection techniques, including tag generation, vtable codegen, and overload resolution.

git-absorb: git commit –fixup, but automatic

git absorb is a port of Facebook’s hg absorb that automatically writes fixup! commits for staged changes, folding them into the appropriate draft ancestor commits to enable atomic histories without manual rebasing. With the --and-rebase flag, it integrates these fixups automatically; otherwise, they can be later squashed using git rebase --autosquash. The tool is available for Windows, MacOS, and Linux via package managers or GitHub releases.

Job queues are deceptively tricky

Job queues may seem simple but involve complex trade-offs in throughput, latency, and design. The author analyzes a real-world system for packing git reference repos using lenses of queue behavior, explicit limits, and fault models, noting that the system offers both wholesale and incremental repacking options.

Differentiable Fortran with LFortran and Enzyme

By combining LFortran, LLVM, and Enzyme, the article demonstrates a method to automatically differentiate legacy Fortran, C, or C++ simulation code at the LLVM IR level, enabling exact gradients through entire time loops without rewriting. This approach allows the code to be embedded as a differentiable layer in JAX or PyTorch, as shown with a 2D heat conduction solver. While experimental and requiring effort to resolve issues like NaN gradients, it successfully bridges validated scientific code with modern ML pipelines.

What does “playing politics” mean for software engineers?

Software engineers should approach “playing politics” not as scheming nobles, but like castle guards: staying aware of power dynamics, avoiding making powerful enemies, and helping influential people while ensuring they notice. Key principles include identifying who holds real power—such as managers leading key projects or gaining headcount—and strategically choosing which conflicts to engage in to minimize political risk.

The Definition of Done

The article defines “done” as creating real-world value, not merely completing effort. Common traps include considering work done if it only runs locally, isn’t tested, isn’t merged, or isn’t available to users. True done requires functioning on all devices, tested through feedback, merged, team work complete, and live in production for users.

Beautiful QR Codes

Scanwayy offers an AI-powered tool that generates branded QR codes by adding a logo and style prompt, with real decoding checks to ensure scan readability. Users can iterate on designs, download high-resolution images or vector files, and pay with one-time credit packs instead of a subscription. The service provides a privacy-focused, non-public gallery to replace default “ugly” QR codes with aesthetically cohesive, brand-matched designs.

Rejourney – Open-source revenue leak prediction for web and mobile apps

Rashid, a UT Austin sophomore, created Rejourney, an open-source tool that predicts revenue leaks in web and mobile apps by analyzing user session recordings and using heuristics and an LLM to identify problematic user journeys. The SDK records interactions, cohorts similar sessions, and outputs fixes for issues that negatively impact critical conversion events, with one user reporting a 30% increase in onboarding after two weeks of fixes.

Actegories

A monoidal category is defined with a tensor product, associator, and unitors, and is implemented in Haskell with type constraints like MonoidalCategory. An actegory extends this by allowing a monoidal category to act on another category via a bifunctor, with coherence conditions relating the action to the tensor product and unit. Examples include the self-action of the Cartesian product.

YouTrackDB is a general-use object-oriented graph database

YouTrackDB is a general-purpose object-oriented graph database developed by JetBrains, offering fast O(1) link traversal, snapshot isolation, and support for YQL, Gremlin, and TinkerPop APIs. It provides scalable development modes, strong security with optional encryption at rest, and easy installation via Docker or as an embedded Maven dependency.

Satellite Tracker – Live Map of Starlink and 30k Satellites

A real-time satellite tracking website uses WebGL to visualize positions of Starlink, GPS, and other constellations. Online since 2019, the site plans to add tables of past and scheduled launches, as well as events like re-entries.

MorphoHDL: A minimalistic language for growing circuits

MorphoHDL is a new minimalistic hardware description language designed for “growing” circuits rather than designing them from scratch. It aims to simplify digital circuit creation by allowing incremental, modular development.

Turn your singing voice into printable notes (in the browser)

A browser tool converts singing into musical notation by plotting fundamental pitches as dots on a staff, with red stems indicating note splits triggered by volume dips. Users adjust re-attack sensitivity to control whether same-pitch notes are tied or repeated, while a noise gate and tolerance filter out quiet sounds and off-pitch slides. Headphones are recommended to prevent metronome audio from bleeding into the microphone.

Frame: A new X11 server – implemented directly in assembly

Norwegian developer Geir Isene has created Frame, a new X11 display server written entirely in x86-64 assembly as part of his minimalist CHasm tool stack. A second new X11 server, yserver, was also recently released, written in Rust by Jos Dehaes. Both projects were developed with the assistance of LLM bots.

🌍 Society & Culture

Survey: 57% of US women 18-29, 47% of men get health info from influencers; women see more often

A Pew Research Center survey found that 57% of young women and 47% of young men ages 18-29 get health and wellness information from influencers, with young women significantly more likely to encounter content about beauty and alternative medicine. While health providers remain the most common source overall, influencers and podcasts play a major role, especially for younger demographics.

X boosts post visibility to mutuals to curb reply battleground, says Bier

X has adjusted its algorithm to prioritize posts from mutual followers, aiming to make reply sections feel more communal and less confrontational. The change is part of a broader effort to foster interest-based communities and follows other platform updates focused on supporting creators.

FDA authorized Zyn nicotine pouches without knowing what they were made of

An FDA toxicologist, Christy Leppanen, conducted informal tests showing Zyn nicotine pouches do not dissolve, contradicting claims made during the agency’s review. She alleges the FDA authorized the product without fully understanding the pouch material and its potential microplastic risks, and that her concerns were repeatedly dismissed. The FDA later acknowledged a mistake in stating she had signed off on the environmental review.

Opening lines of famous literary works

A personal project to display famous literary opening lines in a browser was developed, featuring a simple text-on-background design. The collection has grown to nearly 60 quotes, allowing users to see a new one each time they refresh.

The Death of the Creator Middle Class

The creator economy is experiencing growing income inequality, with the top 10% of creators earning 62% of payments in 2025, up from 53% two years earlier, while median earnings have dropped from $3,500 to $3,000. Mid-tier creators are being squeezed as brands favor either top influencers or cheap micro-creators, AI overviews reduce web traffic, and algorithms reward extreme, attention-grabbing content over steady, thoughtful work.

Agnes Callard’s theory of the uni-context

Agnes Callard’s theory of the “uni-context” argues that modern technology has collapsed multiple local contexts with distinct norms into a single universal set, explaining phenomena such as constant negative news, obsession with distraction, social comparison, and artistic homogeneity. This extends the concept of “context collapse” beyond informational norms to include moral, ethical, and behavioral norms.

U of Chicago law school bans laptops from classes amid AI backlash

The University of Chicago Law School will ban laptops and smartphones for first-year students starting in the 2026-2027 academic year to combat AI reliance and promote critical thinking. In-class exams will also be conducted without internet access. The policy is part of a broader strategy to develop “AI-resilient pedagogy” while still teaching responsible AI use.

Proof of care in the age of AI

The author created a handwritten piece, photographed it, and converted it into an SVG with a text overlay using Barlow Condensed font and spacing characters to enable copy-paste functionality via JavaScript. Hyperlinks were added with Inkscape, and post-processing included embedding the font, replacing backticks with Six-per-em Space characters, and updating image filepaths. The final SVG was directly embedded as an <svg> element in HTML to allow on-copy JavaScript events.

Argentina court recognizes two goldfish as sentient beings with rights

An Argentine court recognized two goldfish, Fede and Magui, as sentient beings with rights after an animal rights NGO filed a complaint about their small, unsuitable display case outside a Buenos Aires sushi restaurant. The court ordered their removal to a 2,500-liter tank at a specialist’s home, setting a precedent for similar animals to be treated as subjects of law rather than objects.

Germany set to restrict its Freedom of Information Act

Germany’s ruling coalition plans to significantly restrict the Freedom of Information Act, limiting requests to individuals, raising fees, and potentially excluding non-EU citizens. Over 110 civil society groups, including Amnesty and Greenpeace, oppose the reforms, arguing they undermine government transparency and effectively abolish freedom of information.

No Spanish Reading Crisis?

A Spanish study found that 66% of the overall population and over 75% of 14-24 year olds read for pleasure, a record high that challenges the myth that young people do not read. While Spain has avoided the decline in reading seen in the United States, some skill deterioration linked to digital media is still observed in universities.

Indian scientists produce most detailed 3D atlas of the human brainstem

Indian scientists at IIT-Madras have created Anchor, the world’s most detailed 3D atlas of the human brainstem at cellular resolution by combining over 500 tissue sections. The atlas links whole-brain MRI scans to individual neurons, bridging medical imaging and cellular pathology for the first time. It is freely available online and aims to improve understanding of disorders like Parkinson’s and Alzheimer’s.

Australia is offering free daytime electricity

Starting July 1, 2026, energy retailers in NSW, South Australia, and South-East Queensland must offer households at least three hours of free daytime electricity daily under the Solar Sharer Offer, accessible with a smart meter and opt-in. A 24 kWh daily usage cap was added after public consultation to ensure financial sustainability. The free window aligns with peak solar output, typically around midday.

Our Amish Language

The author recounts how their Amish community in Libby, Montana, gradually adopted modern conveniences and clothing, causing a decline in the use of Pennsylvania Dutch. As English became dominant, the author now works to preserve the language that once defined their identity.

Rubio announces campaign to ‘dismantle’ International Criminal Court

Secretary of State Marco Rubio announced a campaign to dismantle the International Criminal Court, labeling it a threat to U.S. sovereignty and outlining measures including visa revocations and increased scrutiny of supporting nations. The announcement follows a lawsuit filed by ICC judges against the Trump administration over unlawful sanctions.

Georgia family says they’re forced to sell home to power AI data centers

A Georgia family is being forced to sell their home to Georgia Power for a transmission line that will primarily power AI data centers, calling the situation “theft.” The utility claims eminent domain is a last resort and that it has negotiated in good faith, while the family argues the company is stripping their generational wealth.

What did SFFA vs. Harvard reveal about admissions?

Elite college admissions are not purely merit-based; applicants are sorted into lanes with vastly different odds, such as a 17-fold gap between recruited athletes and unhooked domestic students. The financial aid system uses first-degree price discrimination, with the CSS Profile counting home equity and non-custodial income, hitting upper-middle-class families hardest. Removing hooks like legacy and athletics reveals that hook advantage and merit disadvantage are two sides of the same coin, as some groups gain merit seats while others lose them.

Success may not matter if you aren’t doing what you love

Founder-market fit, the alignment between a founder’s background and culture with their target market, is as crucial as product-market fit. Without it, founders struggle to connect with customers and sell effectively. Success hinges on serving a market where the founder naturally fits, not just pursuing any available opportunity.

🏗️ Hardware & Infrastructure

NY Gov. Hochul signs 1-year moratorium on hyperscale data centers over 50MW, first state to do so.

New York Governor Kathy Hochul signed the nation’s first statewide moratorium, blocking environmental permits for data centers over 50 megawatts for up to a year to develop regulations on energy prices and environmental impact. A stricter legislative version with a 20-megawatt threshold awaits her decision, while she also plans to push for rolling back sales tax exemptions for large data centers.

Google to buy entire initial output of 1.6GW Arkansas solar project, operational 2029

Google has agreed to purchase 100% of the initial output from the 1.6GW Steel River Energy Center solar project in Arkansas, which is expected to become operational in 2029. The deal signals continued demand for renewable energy despite the Trump administration’s efforts to end tax credits and stall such projects.

Jurassic Park computers in excruciating detail

The article details the authentic computers used in Jurassic Park, including an Apple PowerBook 100 and SGI workstations, with over $1.7 million in hardware loaned by Apple and Silicon Graphics. It also notes that the on-screen animations were generated live from an adjacent room, not faked.

I’m a USB-C Maximalist

A traveler used a single USB-C power brick to charge all their gadgets—phone, laptop, eReader, watch, toothbrush, tracker, battery, and headphones—during a seven-week European holiday, praising the convenience and universality of the standard. They argue that proprietary charging ports are unnecessary and that USB-C’s benefits outweigh its drawbacks.

Japan develops a method to recover up to 90% of lithium from used EV batteries

Japanese scientists have developed a method to recover up to 90% of lithium from used EV batteries using a chemical process that replaces sodium hydroxide with recovered lithium hydroxide, also reducing carbon emissions by 40% compared to conventional recycling. This breakthrough could lessen Japan’s dependence on lithium imports, though currently only about 14% of used batteries enter formal recycling systems.

Fundamentals of Wireless Communication (2005)

This textbook covers the fundamentals of physical-layer wireless communication, including MIMO, OFDM, and CDMA, illustrated with examples from real systems like GSM and IS-95. It is designed for graduate students and practicing engineers, emphasizing the connection between theory and implementation. The online version includes exercises and is available as PDF files subject to standard copyright.

Is x86 ready to ACE it?

Intel’s AMX extension accelerates matrix multiplication using configurable tile registers and inner products, while the new ACE specification introduces a second accelerator type with fixed 64x16 tile registers and outer products, dropping complex numbers but adding FP8 support.

RISC-V firmware project wants every board booting from the same hymn sheet

A new RISC-V firmware initiative called HFI proposes a standardized, PC-like boot process from power-on to operating system. The project aims to unify boot procedures across different RISC-V boards, ensuring consistency and compatibility.

India’s crewed space mission is ready for splashdown, but not launch

India’s Gaganyaan crewed space mission, originally slated for a 2022 launch, is now ready for splashdown but still not for launch. The mission aims to make India the fourth country to send humans into orbit.

Big Blue thinks small, again, with 2U POWER tower

IBM has introduced a 2U POWER tower minicomputer, described as the last proprietary minicomputer now available in a deskside form factor.