Greg Troxel, a contributor to the QGIS-Developer mailing list, recently received a notification from GitHub regarding the transition of its "Code Quality" feature to general availability on July 20, 2026. The email warns of new "per-committer" and AI-usage fees. Troxel is concerned because he does not use paid plans, is not enrolled in GitHub Sponsors, and has removed himself from defunct organizations. Despite this, he received the notice, which claims he is part of GitHub Sponsors—an assertion he disputes. He suspects the email may be an error or an underhanded attempt to generate revenue from users not actively utilizing paid features. He is reaching out to the QGIS community to determine if others received similar notices, whether the QGIS project’s configuration might have triggered these notifications for contributors, and if there is a risk of surprise billing for repository collaborators.

You need to enable JavaScript to run this app.

Enable JavaScript and cookies to continue

Songs and voiceActive creative voiceAcoustic guitar, lead vocals, songwriterCharlie (Chuck)SaponaraOriginal songwriter and lead voice behind the Fading Maize catalog.The revival keeps Charlie's authorship at the center while he provides feedback, approvals, and creative tweaks for the new versions.

This oral history explores the groundbreaking visual effects work behind James Cameron’s *Terminator 2: Judgment Day* (1991). As ILM’s computer graphics department was still in its infancy, the team faced the dual challenge of inventing new technologies while working with limited hardware. Led by visual effects supervisor Dennis Muren, the crew of about 40 artists pioneered techniques like "Body Sock" for character stitching, custom reflection shaders for the "poly alloy" T-1000, and advanced morphing software. Because commercial animation tools were rudimentary, the team functioned as both artists and software developers, often writing custom C-shell scripts and "glue code" on the fly to meet Cameron’s ambitious vision. Working long hours on expensive, processing-intensive hardware, the team relied on a "DIY" spirit, hand-animating complex transformations and solving technical problems without the benefit of modern software. The resulting effects—such as the liquid metal transitions and the T-1000’s death sequence—remain legendary in cinema history. This project not only cemented ILM’s status as a digital pioneer but also launched the careers of many artists who would go on to define the future of the visual effects industry.

Joint Embedding Predictive Architectures (JEPAs) represent a shift in self-supervised learning from pixel-level reconstruction to prediction within abstract representation space. Unlike generative models that waste capacity on irrelevant noise, JEPAs capture semantic structures—such as object identity and physical dynamics—by forcing a model to predict masked regions from visible context. Key architectures include: * **I-JEPA:** Uses a context encoder and a target encoder (updated via exponential moving average) to predict missing image patches, avoiding collapse without needing contrastive negatives. * **V-JEPA/V-JEPA 2:** Extends this to video, where the model learns latent spatiotemporal dynamics, enabling not just classification but also goal-directed planning for embodied agents. * **LeJEPA:** A refined approach that replaces heuristic stability methods with "Sketched Isotropic Gaussian Regularization" to enforce an optimal embedding distribution. Conceptually, JEPAs function as Energy-Based Models (EBMs). By minimizing prediction error in a compressed latent space, they serve as "world models" that simulate physical reality. Proponents, such as Yann LeCun, argue this approach is more suitable for physical-world intelligence than autoregressive token prediction, as it requires the model to internalize the underlying constraints and physical laws of the world rather than just human linguistic patterns.

Tim Roughgarden’s exploration of computer science begins with Alan Turing’s 1936 discovery that some problems are fundamentally unsolvable by any algorithm, such as the "halting problem." Moving beyond theoretical limits, Roughgarden investigates why some solvable problems are easy while others remain stubbornly difficult. He highlights the power of algorithmic shortcuts—like Dijkstra’s algorithm—which allow computers to find efficient solutions without brute-force searching. However, this efficiency hits a wall with problems like the Traveling Salesman Problem (TSP). These difficult puzzles form the basis of "NP-completeness," a framework revealing that thousands of diverse tasks are linked; if one is solved efficiently, they all are. This leads to the "P versus NP" question, the most significant unsolved mystery in computer science. By bridging historical mathematical inquiry with modern computational theory, Roughgarden illustrates how these constraints shape everything from cryptography to artificial intelligence. Designed for a general audience, the course demystifies the profound boundaries of what computers can—and likely never will—achieve.

Researchers from the University of Portsmouth have discovered that the teeth of limpets—marine snails known for their ability to scrape rock—are the strongest natural material on Earth. Composed of iron-containing goethite nanofibers embedded in a protein matrix, these teeth exhibit a tensile strength significantly greater than spider silk and comparable to high-quality carbon fibers. According to lead researcher Asa Barber, the material’s strength is so immense that a single strand could theoretically support thousands of pounds. While other materials like lonsdaleite or wurtzite boron nitrate are known for superior hardness (the ability to resist scratching or indentation), the limpet’s teeth represent a breakthrough in natural tensile strength (the ability to resist pulling apart). Scientists are now looking to the structural composition of these teeth as a potential blueprint for developing next-generation engineering materials and machines.

This exploration questions the efficiency of our traditional timekeeping system, which relies on an arbitrary base of 24 hours, 60 minutes, and 60 seconds. Critics argue this structure is inconsistent and difficult to calculate, contrasting sharply with the streamlined logic of the metric system. The author presents several alternative clock designs to test the potential for a more intuitive approach: * **Decimal (Base 10) Clock:** Divides the day into 10 hours of 100 minutes, simplifying conversions to a basic decimal shift. * **24-Hour Clock:** A classic variation that removes AM/PM confusion by allowing the hand to complete a single rotation per day. * **Alternative Bases:** Experimental designs include binary clocks (base 2) for enthusiasts, hexadecimal (base 16) clocks, and a 36-hour "degree" clock inspired by compass navigation. While these systems highlight the mathematical limitations of our current clocks, they also underscore the cultural and historical weight of traditional timekeeping. Ultimately, the author invites readers to imagine a more logical framework for measuring our days, proving that time is as much a design choice as it is a natural constant.

The QuadRF is a compact, handheld phased-array radio built around a Raspberry Pi 5 and an FPGA, designed for advanced signal processing and beamforming within the 4.9–6 GHz range. By utilizing the Pi’s MIPI lanes for high-bandwidth data transfer, the device achieves low-latency, high-speed RF streaming, enabling capabilities like drone tracking and visualizing WiFi signals through walls. Developed by Martin McCormick, a former SpaceX engineer, the project aims to make sophisticated RF analysis accessible beyond government-grade tools. While the user interface for the current prototype remains a bit rough, the device features impressive functionality, including a custom Augmented Reality (AR) visualizer that maps RF signals in real-time. Though originally conceived as part of a larger Moon-scale antenna array project, the handheld QuadRF is already showing great potential as a versatile tool for hobbyists and professionals alike. Available via a Crowd Supply campaign, it offers an innovative, open-source approach to RF visibility. While pre-production software requires refinement, its performance—particularly its ability to track devices like drones—demonstrates that high-end RF engineering is becoming increasingly portable and accessible.

Initially skeptical of flashcards as a tool for rote memorization, the author eventually embraced spaced repetition through Anki to support deep learning in cumulative fields like STEM. While acknowledging common criticism—that flashcards can encourage superficial learning—the author argues that when used correctly, they are essential for maintaining fluency and building the conceptual "chunking" necessary for advanced reasoning. The author’s approach prioritizes quality over quantity: flashcards should never replace the initial process of understanding, and they are most effective when personally authored rather than imported or AI-generated. Beyond simple definitions, the author creates cards for "aha moments," intuition, and even recurring practice problems to bridge the gap between initial study and long-term retention. By integrating flashcards with digital note-taking and maintaining a lightweight, high-intent review process, the author successfully retains complex knowledge across long gaps in study. Despite limited empirical data on Anki’s efficacy outside of medical education, the author finds that the technique transforms learning from a volatile, temporary effort into a durable foundation, allowing them to resume complex projects years later without starting from scratch.

This blog post explores a 1939 *LIFE* magazine advertisement for RCA Victor’s “Sound Service for Schools.” The author places this artifact within the historical context of RCA’s 1929 merger with the Victor Talking Machine Company. By acquiring Victor’s tangible assets through a stock-swap deal just before the Great Depression, RCA secured the infrastructure to pivot during the economic downturn. Amidst the financial instability of the 1930s, RCA targeted the education sector to create new revenue streams, offering schools sound systems, recorders, and record catalogs to modernize the classroom. The author draws compelling parallels between this early push for classroom technology and the modern rise of educational tools like MOOCs and AI. While RCA focused primarily on music as an evergreen educational commodity, competitors like ERPI Classroom Films were simultaneously introducing documentaries and instructional media. The piece concludes by reflecting on humanity’s persistent drive to reinvent education through technology, questioning how future generations will look back on today’s “AI in Education” era just as we now view the classroom innovations of the 1930s.

Following a record-breaking June, major European trade unions are demanding that the European Commission enact legislation to protect workers from extreme heat. With approximately 130 million workers exposed to heat stress annually—leading to significant injuries and hundreds of deaths—unions are pushing for legally enforceable maximum temperature limits. Proposed measures include mandatory work suspension when temperatures exceed 30°C for physically demanding roles (such as construction and agriculture) and 32.5°C for sedentary positions. Currently, protections vary significantly across the continent, leaving many employees vulnerable to dangerous working conditions. In the UK, the Trades Union Congress (TUC) is campaigning for similar standards, arguing that while minimum workplace temperatures are legally established, maximum limits are non-existent. The TUC proposes that employers must take cooling measures at 24°C and allow work to stop if temperatures hit 30°C, or 27°C for manual labor. Labor leaders emphasize that as climate change makes extreme heat a permanent workplace reality, updated safety regulations are urgently required to ensure worker welfare.

Access Denied You don't have permission to access "http://www.dailymail.com/sciencetech/article-15956159/Incredible-lost-city-discovered-Egypts-desert.html" on this server. Reference #18.a7a4c117.1783697531.b0c17cb https://errors.edgesuite.net/18.a7a4c117.1783697531.b0c17cb

The author, who once created a tool to block "burner" email domains, now argues against the systematic rejection of privacy-focused email services during user signups. The core of the argument is a distinction between **public shared inboxes** (temporary, accessible to anyone) and **personal aliases** (permanent, private forwarding addresses like Firefox Relay or Apple’s "Hide My Email"). While public inboxes are often used for abuse, personal aliases are essential tools for users to protect their identity, manage spam, and track data breaches. The author contends that domain blocklists are ineffective against determined spammers, who can easily bypass them using custom domains or Gmail aliases. Consequently, these lists mainly punish privacy-conscious users while failing to stop actual attackers. Instead of hard-blocking, the author suggests that services should stop treating email verification as a proxy for "trustworthiness." If lists must be used, they should clearly differentiate between shared burners and personal aliases. Developers are encouraged to move away from binary rejections and instead incorporate domain reputation as part of a broader, more nuanced risk-assessment strategy.

Unified-memory mini PCs have become popular for running large language models (LLMs) because they prioritize **capacity** over raw speed. Unlike traditional desktops, where a GPU’s limited VRAM creates a hard ceiling, these systems share a single pool of system RAM across the CPU, GPU, and NPU. This allows a relatively affordable mini PC to load massive models—like a 70-billion-parameter model—that would otherwise require an expensive, high-end workstation card. However, this capacity comes at a significant cost in performance. LLM text generation is primarily limited by memory bandwidth; because these mini PCs have significantly lower bandwidth than dedicated GPUs, they generate text at a "slow reader" pace. Furthermore, the "prefill" phase (reading long prompts) is compute-intensive, where the weaker integrated GPUs in mini PCs often struggle. Ultimately, these machines are not speed-demons but capacity tools. They are ideal for users who need to run large, memory-hungry models that simply won't fit on a standard GPU. For smaller models or tasks requiring high-speed, interactive performance, a dedicated discrete GPU remains the superior choice. Match the hardware to the model’s size, not the marketing specs.

Sparta is famously known for rejecting traditional currency, preferring cumbersome iron spits to discourage wealth accumulation and prioritize military focus. For centuries, the city-state relied on barter or foreign coinage, maintaining a staunchly anti-materialistic culture. This traditional stance eventually shifted as Sparta’s power declined. Starting in the 3rd century BCE, Spartan kings—beginning with Areus I—began issuing royal coinage to fund mercenaries. Later leaders like Cleomenes III and Nabis minted silver tetradrachms featuring royal portraits and local deities, reflecting a move toward a more conventional economy. Following its integration into the Achaean League, Sparta adopted a standardized regional currency. After the Roman conquest, the city issued local bronze coinage under the authority of magistrates like C. Julius Eurycles and, eventually, under the names of Roman Emperors. Today, these coins are classified under "Laconia" or "Lacedaimon." While rare royal Spartan coins command high prices, their existence serves as a historical record of a proud society’s gradual adaptation to the economic realities of the Hellenistic and Roman worlds.

Amidst the often-hostile tribalism of social media discourse, Robin Berjon proposes a technical bridge to unify two powerful protocols: ActivityPub (AP) and the AT Protocol (ATProto). Berjon argues that while both aim to dismantle social media silos, they address different problems. ActivityPub excels at social networking but remains vulnerable to server-admin capture, similar to email. Conversely, ATProto offers a superior architecture for user agency and "credible exit," allowing users to own their data and move between providers without losing their identity or followers. The author suggests a design provocation: implementing ActivityPub routes atop an ATProto Personal Data Server (PDS). By leveraging the "Actor" document’s flexibility in ActivityPub to point to ATProto-based endpoints, users could enjoy the decentralized, federated reach of AP while retaining the sovereign, portable data ownership of ATProto. While minor technical hurdles exist—such as reconciling REST/RPC methods and aligning identity/handle systems—Berjon believes these are practical, bridgeable challenges rather than dealbreakers. By integrating these complementary technologies, we could create a more resilient, user-centric web that transcends the current limitations of both traditional federation and closed-platform silos.

**Runloom** is a high-performance library that brings Go-style stackful coroutines to Python. By utilizing hand-rolled assembly context switching, a C-based work-stealing scheduler, and an efficient `netpoll` implementation, it enables developers to write blocking code that runs concurrently across all CPU cores. Key features include: * **Ease of Use:** Supports `monkey.patch()` to turn standard blocking I/O (sockets, time, etc.) into non-blocking cooperative code without needing `async/await`. * **Performance:** Designed for free-threaded CPython (3.13t/3.14t), it achieves throughput and spawn rates comparable to Go. * **Architecture:** Implements an M:N scheduler that distributes tasks across multiple OS threads, effectively bypassing the limitations of traditional single-threaded Python concurrency. * **Compatibility:** Provides an `asyncio` bridge for legacy code and is available via prebuilt wheels for major OS/arch combinations. While Runloom’s memory overhead per fiber is roughly 3.3× that of Go due to CPython’s evaluation frames, it offers a powerful way to achieve multi-core parallelism in Python. It does not accelerate individual Python bytecode execution, but it allows a single process to scale blocking Python logic across all available hardware cores.

Avi Press, a longtime Haskell advocate and board member, explains why his company, Scarf, is moving away from Haskell for new development in favor of Python. While acknowledging Haskell’s reliability and type safety, Press argues that the rise of AI-driven development has fundamentally shifted the economics of programming. AI agents thrive on fast feedback loops, cheap execution, and low-friction environments. Haskell’s heavy build times, complex dependency management, and high "cold start" overhead create significant bottlenecks that hinder AI-assisted workflows. By contrast, Python offers the agility required to leverage modern LLMs effectively, allowing for rapid iteration and faster shipping. Press warns that the Haskell community risks stagnation by focusing on academic research (like dependent types) rather than industrial needs like build performance and AI integration. He urges the community to prioritize "agent-friendly" tooling—such as faster compilation and better documentation—to remain relevant. Ultimately, he posits that unless Haskell evolves to accommodate the AI-driven paradigm, it will struggle to compete with faster-moving ecosystems. For Scarf, the decision was a pragmatic response to the urgent need for productivity and speed in the modern software landscape.

The Mexican cavefish provides a striking metaphor for corporate stagnation: just as cave-dwelling fish evolve to lose their sight because the dark environment no longer rewards it, successful companies often develop "competence blindness." When high-growth companies prioritize headcount over standards, they create an environment where engineering excellence becomes a vestigial trait. Employees who arrive with "sight"—the ability to recognize poor infrastructure or outdated practices—are often overruled, leading to a form of professional apoptosis where the desire for quality is programmed out of them. To hide this rot, companies create "centers of excellence" that prioritize bureaucratic control, further stifling intrinsic motivation. Protected by market barriers, these firms remain profitable while operating on "stored fat" from earlier successes. Managers often mistake this decline for culture, blaming turnover on external factors rather than their internal darkness. Ultimately, staying in such an environment is a form of self-adaptation; the only way to regain "sight" is to exit the cave and swim in new waters.

Enable JavaScript and cookies to continue

We are currently experiencing an incident affecting login across Proton services.Our engineering teams are actively investigating and working to restore normal access as quickly as possible.We apologize for the disruption and will provide further updates as soon as more information is available.

This essay laments the death of authentic, participation-based culture, arguing that it has been replaced by performative "spectacle" and curated marketing. The author contends that modern digital life—characterized by AI, social media, and algorithmic consumption—has led to "wireheading," where users experience the illusion of achievement without real-world engagement. The author suggests that the "machine" of the internet has effectively outcompeted genuine human experience, transmuting individual identity into uniform, compliant data. As our "inner reality" becomes the primary theater of a global information war, the author warns that we are no longer being physically conquered but psychologically assimilated. While the author observes that modern AI is merely the climax of a decades-long erosion of self, they offer no easy solution, noting that the "new war" demands the soul itself. Ultimately, the piece serves as a cynical observation of how technology absorbs human culture, strips it of its rough edges, and feeds back a hollow, pre-chewed version that leaves the individual without a steering wheel.

It is tempting to let Large Language Models (LLMs) generate redundant code, assuming that since the AI handles the typing, maintainability concerns like "DRY" (Don't Repeat Yourself) no longer matter. However, this is a dangerous fallacy. LLMs generate code by mimicking the patterns, shortcuts, and architectural habits already present in your codebase. When you repeatedly merge duplicated logic or "quick and dirty" solutions, you aren't just outsourcing the work—you are training the model to adopt poor coding practices as your project’s standard. Every copy-pasted conditional reinforces a bad pattern, causing the model to prioritize these "anti-patterns" in future responses. Eventually, your codebase becomes cluttered with technical debt that is increasingly difficult to prompt your way out of. To avoid this, you must treat your codebase as a high-quality blueprint. By maintaining clean, structured code, you ensure the LLM provides consistent, high-quality output rather than automating the creation of future maintenance headaches. Ultimately, you should continue to write code as if a human will maintain it, because the AI will inevitably learn to mimic your worst habits.

The "Late Bronze Age Collapse" (c. 1220–1170 BC) refers to the widespread destruction and decline of state systems across the Eastern Mediterranean and Near East. While the collapse was not "total"—many sites survived and transitioned into the Iron Age—it marked a severe, if uneven, setback for the Mycenaean, Hittite, and New Kingdom Egyptian empires. Evidence is primarily archaeological, featuring "destruction layers" at major palace centers. While older theories blamed external "Dorian invasions" or volcanic eruptions, modern scholarship favors a "perfect storm" of interrelated causes. This likely included climate-driven crop failures, intense inter-state warfare, economic instability, and mass movements of refugees and raiders (such as the "Sea Peoples"). The effects were profound: Greece suffered a severe "Dark Age" involving deurbanization and the loss of the Linear B script. However, the resulting political vacuum allowed for the emergence of new, historically significant entities, including the Phoenician city-states and the kingdoms of Israel and Judah. Ultimately, the collapse functioned as a transformative event, breaking the centralized Bronze Age "concert of powers" and creating the regional conditions that shaped the trajectory of the classical world.

"Forgive and forget" is an idiom, but in engineering terms, it describes two distinct human capabilities that AI currently lacks. Human memory is an active, adaptive, and "read-modify-write" process; we forget as a form of compression to generalize, and we "drain the heat" from memories through reconsolidation. Because human memory is metabolically expensive, forgetting is a survival feature. In contrast, AI systems—whether through static training weights, context windows, or vector retrieval—operate on "read-only" architecture. Machines do not possess the mechanism to decay, nor do they understand the concept of "heat" or emotional charge. Their "forgetting" is either accidental (catastrophic interference or context dead zones) or forced (amnesia-like deletion). Because machines face no scarcity or energy cost to hold information, they have no functional incentive to release it. True forgiveness requires the ability to hold a grievance in full resolution while choosing not to act on it—an operation that currently exists in no standard AI framework. By deploying systems with perfect, permanent recall, we risk disrupting the social grace—the "gift of forgetting"—upon which human relationships depend.

The author argues that a truly great tool should be "invisible"—fading into the background so you can focus on your work. A common pitfall among programmers is mistaking the "fun" of solving a tool’s inherent limitations for genuine productivity. Many users develop a religious devotion to complex tools like Vim or Linux, framing their steep learning curves and need for constant configuration as heroic "hacker" achievements. The author suggests this often stems from identity-based tribalism, where admitting a tool’s flaws feels like an attack on the self. By confusing the "clever" sensation of overcoming a technical hurdle with actual output, users often choose tools that hinder their efficiency rather than help it. Ultimately, the author contends that productivity is measured by "wall-clock time," not by how much you enjoy wrestling with your software. Good tools should provide excellent defaults rather than offloading design decisions onto the user. Whether you prefer a GUI or terminal, the mark of a superior tool is not its complexity or the "vibe" it projects, but its ability to disappear, allowing you to complete tasks without unnecessary friction.