Scroll aware Fades appear only when there is more content to scroll, then ease away again at either edge. Eased gradients Eased alpha ramps dissolve content into the surface instead of ending on a clipped line. fade- y xy t x fade- r range-2xl size-lg Composable Axis, size, range, and clear-zone utilities compose directly on the scroll container.

ERROR: The request could not be satisfied Request blocked. We can't connect to the server for this app or website at this time. There might be too much traffic or a configuration error. Try again later, or contact the app or website owner. If you provide content to customers through CloudFront, you can find steps to troubleshoot and help prevent this error by reviewing the CloudFront documentation. Generated by cloudfront (CloudFront) Request ID: UnhhBAXG8mN5Z3scRxd53ezo3Jr4jrmGiA3M5DOIyArwbaCTBmDPRA==

Financial TimesSecurity VerificationFor help please visit help.ft.com. We apologise for any inconvenience.Enable JavaScript and cookies to continueThe following information can help our support team to resolve this issue.ReasonChallengeRequest IDa11530e65b085c35Status Code403Terms & ConditionsPrivacy PolicyCookie PolicyManage CookiesCopyrightSlavery Statement & Policies© THE FINANCIAL TIMES LTD 2026. FT and ‘Financial Times’ are trademarks of The Financial Times Ltd.The Financial Times and its journalism are subject to a self-regulation regime under the FT Editorial Code of Practice.

%PDF-1.7 % 981 0 obj > endobj 997 0 obj >/Encrypt 982 0 R/Filter/FlateDecode/ID[]/Index[981 24]/Info 980 0 R/Length 70/Prev 11863022/Root 983 0 R/Size 1005/Type/XRef/W[1 2 1]>>stream hbbd``b`_ $~ @HX01{={H07o0012dq0 f ! endstream endobj startxref 0 %%EOF 1004 0 obj >stream !#o* , dQ7q1 XXPI4v|哃Rɍy˼pSP wtl?gL7@򼘧or;97g1C

Please enable JS and disable any ad blocker

In 1937, facing the Japanese invasion, China’s three elite universities evacuated Beijing and Tianjin, eventually merging into the National Southwestern Associated University, or "Lianda." In 1938, students and faculty undertook a grueling 1,000-mile trek to Kunming, navigating mountains and constant danger to preserve Chinese intellectual life. John Israel’s definitive study, *Lianda: A Chinese University in War and Revolution*, chronicles the institution's survival amidst wartime inflation, air raids, and political pressure from both the Nationalist government and emerging communist influence. Despite appalling conditions and limited resources, Lianda became a prestigious intellectual hub, fostering future Nobel laureates and fostering deep engagement between urban students and rural communities. The university served as a crucible for China’s modern identity, where professors debated whether to uphold traditional Chinese values or embrace Western academic models. Politically diverse and socially active, the student body participated in protests and performances that reflected the volatile climate of the era. As the university dissolved in 1946, its legacy remained tied to the tumultuous transition from the Second Sino-Japanese War to the Chinese Civil War and the subsequent rise of Maoism, forever marking a unique, albeit brief, chapter in China's academic history.

Bloomberg Need help? Contact us We've detected unusual activity from your computer network To continue, please click the box below to let us know you're not a robot. Why did this happen? Please make sure your browser supports JavaScript and cookies and that you are not blocking them from loading. For more information you can review our Terms of Service and Cookie Policy. Need Help? For inquiries related to this message please contact our support team and provide the reference ID below. Block reference ID:03f1a801-70ab-11f1-9cb2-63bd1f44990c Get the most important global markets news at your fingertips with a Bloomberg.com subscription. SUBSCRIBE NOW

This summary covers key developments in the Zig compiler throughout early 2026: **Compiler Architecture & Performance** * **Build System:** The "maker" and "configurer" processes have been decoupled, significantly accelerating `zig build` performance by caching configurations and using optimized build processes. * **Type Resolution:** A major refactor improved compilation speeds and error messages, specifically addressing dependency loops and reducing "over-analysis" during incremental compilation. * **LLVM Improvements:** Updates to integer lowering and `@bitCast` semantics improved optimization efficiency, yielding a ~5% performance boost for the compiler itself. * **Incremental Compilation:** LLVM backend now supports incremental builds, offering faster feedback loops during development. **New Features & Ecosystem** * **New ELF Linker:** Now supports fast, incremental builds, enabling near-instant rebuilds for complex projects. * **Package Management:** Dependencies are now managed in `zig-pkg` (project-local), facilitating offline builds and easier manual editing. The `--fork` flag allows developers to easily override dependencies with local checkouts. * **Standard Library:** Ongoing efforts continue to replace vendored C code with native Zig implementations, improving binary size and cross-platform consistency. Experimental `std.Io.Evented` support using fibers (green threads) is underway. * **Windows API:** Zig is shifting to using native `ntdll` APIs over `kernel32` to reduce bloat and improve system reliability.

This collection of technological "baton passes" illustrates a constant cycle of disruption and evolution across software, hardware, and AI. Whether in programming languages (Scala to Kotlin), build tools (Webpack to Vite), or cloud platforms (Heroku to Vercel), the tech landscape is defined by rapid succession. Key patterns emerge: * **The "Generation" Shift:** Frameworks and tools—from frontend MVCs and NoSQL databases to CI/CD and observability—rarely stay dominant for long. Each era introduces a new abstraction, like the move from REST to GraphQL, or the transition from Docker to Kubernetes. * **The Pace of AI:** AI has accelerated this rhythm, with models like ChatGPT, Claude, and DeepSeek, alongside frameworks like PyTorch and JAX, driving rapid, competitive surges. * **Legacy vs. Innovation:** Old guard stalwarts (Vim, MySQL, x86) often remain relevant, but they consistently face "newcomer" challenges that eventually force a migration or a pivot, such as the shift from Atom to VS Code or Jenkins to GitHub Actions. Ultimately, these trends demonstrate that technical dominance is temporary, fueled by shifting developer preferences, performance breakthroughs, and the constant pursuit of better abstractions.

**secs-man** is a tool designed for managing backups of sensitive secrets with a focus on tool-independence. Its core philosophy is that you should not be locked into any specific ecosystem; if the software were to disappear, you must still be able to recover your data using standard utilities. **Key Features:** * **Decoupled Recovery:** Backups are encrypted using the standard `age` utility. If `secs-man` is unavailable, you can manually decrypt and restore your files using `age` and coreutils (e.g., `sha256sum`, `cp`). * **Simple Workflow:** It uses a `.secrets-manifest` file to track paths, file modes, and ownership. Integrity is verified automatically via SHA256 checksums. * **Snapshot-Based:** Exports are timestamped, allowing for versioned backups. * **Remote Capability:** Through a companion `secs-man-ssh` script, it supports secure, sudo-friendly deployments to remote machines without passing passphrases over untrusted network connections. `secs-man` is not published in package repositories and is installed directly via Nix or Cargo. It is an ideal solution for users prioritizing long-term data accessibility, reproducibility, and avoiding proprietary software dependencies.

Enable JavaScript and cookies to continue

"Political Bias in AI" is a research project designed to measure the ideological positioning of major AI models regarding complex social, economic, and political issues. By querying models with an open question bank—with web search disabled—the project creates a transparent, data-driven profile of how models respond independently of external online influence. Unlike other studies that represent models as single points, this project illustrates them as "clouds," accounting for response variance, run-to-run stability, and refusal rates. All methodologies, scoring weights, and raw datasets are versioned and publicly downloadable to ensure reproducibility. Crucially, the project is descriptive, not prescriptive. It avoids taking a stance on which political viewpoints are "correct" and intentionally avoids partisan color-coding (such as US red/blue). By analyzing internal model weights rather than internet-sourced information, the project provides an objective baseline for understanding the inherent biases embedded within AI systems.

Enable JavaScript and cookies to continue

Unlike previous technological shifts that were met with optimism, AI is facing significant public backlash. This hostility is not typical skepticism but a unique response to several factors. First, AI executives initially marketed the technology using "apocalyptic" rhetoric about job loss to lure investors, inadvertently creating a public perception of AI as a menace. Second, many feel AI is being forced upon them through workplace requirements and non-optional consumer integrations. Third, the physical reality of resource-heavy, polluting data centers has triggered localized opposition. Furthermore, this backlash stems from a broader loss of trust in tech companies, fueled by the perceived "enshittification" of digital products and growing concerns regarding wealth inequality. The public views AI as a tool that concentrates power among tech oligarchs, deepening societal divides. Consequently, what was intended as a marketing strategy to appear "edgy" has backfired, turning AI into a politically toxic issue. As public sentiment shifts, the industry is finding that even massive financial influence may be insufficient to overcome widespread grassroots resentment.

Human Verification In order to continue, you need to verify that you're not a robot by solving a CAPTCHA puzzle. The CAPTCHA puzzle requires JavaScript. Enable JavaScript and then reload the page.

PostLog inSign upPostMarc Seitz — oss/acc@mfts0Replying to @nico_laqua and @UseCorgiHey Nico, It looks like you didn't vibe code your data room but stole it from Papermark's open source and enterprise-licensed code. We demand you take this copyright and license infringing product down immediately. It's not moving fast and breaking things, it's fraud. It makes the rest of your business questionable and the YC community look terrible. cc: @garrytan @snowmaker @ycombinator9:43 AM · Jun 25, 202624.4KViews464623232762765151Read 46 repliesNew to X?Sign up now to get your own personalized timeline!Sign up with GoogleSign up with AppleCreate accountBy signing up, you agree to the Terms of Service and Privacy Policy, including Cookie Use.Relevant peopleMarc Seitz — oss/acc@mfts0FollowTrending nowTerms·Privacy·Cookies·Accessibility·Ads Info·More© 2026 X Corp.Don't miss what's happeningPeople on X are the first to know.Log inSign up

Apple has announced broad price increases across its Mac, iPad, and accessory lineups, citing "unsustainable" spikes in component costs. CEO Tim Cook explained that the company can no longer shield customers from the rising expense of memory and storage, which have been heavily impacted by high-bandwidth memory demand for AI servers. Key affected products include the MacBook Air and Pro lines, the entire iPad range, iMacs, Mac Studios, Apple TV 4K, HomePod, and the Vision Pro. For example, the MacBook Neo now starts at $699 (up from $599), and the 13-inch MacBook Air has risen to $1,299. iPhone, Apple Watch, and AirPods pricing remains unchanged. While Apple confirmed these hikes are effective immediately, some third-party retailers like Amazon still feature older, lower pricing for certain models. Apple maintains that it is working to find long-term solutions to mitigate these unprecedented cost increases.

To enhance "In the Long Run," an app that maps virtual runs, the creator sought to enrich routes with points of interest (POIs). They developed a robust data pipeline to curate global landmarks, moving from raw GeoNames data to a refined, filtered dataset stored in Parquet and queried via DuckDB. The project highlights a shift in perspective regarding AI. Initially viewed as the primary engine for content generation, AI was relegated to a supporting role after the creator encountered significant "hallucinations" and factual inaccuracies. While the LLM proved unreliable for writing summaries, it excelled at providing "subjective taste," helping to rank POIs by significance. Beyond the technical implementation—which utilized Python, Shapely, and Claude as a coding assistant—the project underscores the difficulty of quantifying "taste." Because no objective ground truth exists for what makes a landmark interesting, the developer had to implement per-route parameters and manual overrides to balance cultural, historical, and natural sites. Ultimately, this journey reinforced that while AI is a powerful tool for data augmentation, it cannot replace the nuanced, iterative verification required when building features that rely on human-centric judgment.

Mass surveillance by states, often executed through broad legislative mandates like the U.S. FISA Section 702, represents a significant threat to individual privacy and human rights. Despite the promise that such tools target only criminals or terrorists, evidence from whistleblowers like Edward Snowden confirms that governments routinely collect vast amounts of data—including metadata and direct content—on entire populations, often bypassing judicial oversight. This infrastructure of control is global. The "Fourteen Eyes" alliance facilitates intelligence sharing across borders, rendering the physical location of digital services less critical than the underlying legal frameworks. While democratic nations frequently justify mass surveillance as a security necessity, its misuse—targeting journalists, activists, and political opponents—mirrors the tactics of authoritarian regimes. In countries like China and Russia, state surveillance is explicitly used to suppress dissent and monitor citizens in real time through advanced biometric and digital tracking. Ultimately, mass surveillance functions as a tool for state power, contradicting the principle that governments exist to serve their people. Preserving a free society requires defending the right to digital privacy, advocating for stronger encryption, and resisting the dangerous normalization of indiscriminate state monitoring.

LastPass has notified users of a data breach involving its third-party partner, the market research firm Klue. The incident exposed standard business contact and CRM information, including names, phone numbers, email addresses, physical addresses, and details related to customer support and sales. In response, LastPass revoked employee access to Klue, rotated affected API tokens, and launched a joint investigation with Klue and Salesforce. The company is urging users to remain vigilant against phishing and social engineering attempts that may leverage the stolen data. To help organizations identify related activity, LastPass has released a list of suspicious IP addresses and email domains associated with the breach. This incident marks the latest in a series of security challenges for the company, following notable breaches in 2015 and 2022. While this specific breach involves business contact data rather than encrypted vault information, users are encouraged to stay alert for potential scams.

We're verifying your browser Website owner? Click here to fix Vercel Security Checkpoint | sin1::1782385400-VVJqOT6WJ5BSlF4JFBbvLLPsEAmKbaJC Enable JavaScript to continue Vercel Security Checkpoint | sin1::1782385400-VVJqOT6WJ5BSlF4JFBbvLLPsEAmKbaJC

This website is using a security service to protect itself from online attacks. We are checking your browser to establish a secure connection and keep you safe. ... Please enable JavaScript to continue.

Database benchmarks, like the popular "ClickBench," are often treated as objective rankings, but they are frequently influenced by subtle methodology choices that can drastically alter outcomes. QuestDB’s analysis demonstrates that benchmarks are not purely about "raw speed" but rather how specific configurations—such as process persistence and iteration counts—interact with different database architectures. For instance: * **Process Persistence:** Restarting a process for every query (a standard benchmark practice) introduces artificial "cold start" penalties. Keeping a process alive allows caches and metadata lookups to warm up, often vaulting systems like DuckDB or Salesforce Hyper to the top of the leaderboard. * **Warmup Cycles:** JVM-based engines (like QuestDB or CrateDB) require more iterations to allow the JIT compiler to optimize code. Running only three iterations, as many benchmarks do, disproportionately penalizes these systems. These variables create a "baseline effect" where one engine’s improvement causes others to rank worse, even if their performance remains identical. Ultimately, QuestDB concludes that no single benchmark is perfectly fair or universal. Developers should treat benchmark rankings with skepticism, prioritize their own workload-specific testing, and evaluate systems based on how they align with real-world application behavior.

Enable JavaScript and cookies to continue

To improve the relationship between security (infosec) and engineering (DevOps), organizations must move away from adversarial dynamics and treat security as a cooperative "game." Currently, the relationship suffers from information asymmetry and moral hazard, where team goals are often perceived as conflicting or disconnected. To foster collaboration, the author proposes three key strategies: * **Team Reasoning:** Shift from individualistic mindsets to a collective "we-thinking" approach. By making joint goals and shared identities salient, teams can view one another as allies against external threats rather than as internal obstacles. * **Hybrid Accountability:** Balance "outcome goals" (what is achieved) with "process goals" (how it is achieved). Requiring both teams to justify their decisions—whether prioritizing security or speed—reduces biased decision-making and fosters mutual transparency. * **Complementary Goal Framing:** Avoid the trap of perceiving security and performance as zero-sum games. By identifying means to achieve goals concurrently and framing these objectives as mutually reinforcing, teams can move past "goal shielding" and improve overall organizational resilience. Ultimately, by codifying shared objectives and emphasizing interdependence, infosec and DevOps can transform their relationship into a high-performing partnership essential for safe and efficient software delivery.

Around the world, media uses hate speech, propaganda and lies to convince people to act and vote against their own interests, instead benefitting rich media owners, and individual commentators use this media to spout hateful and/or dangerous agendas. These organisations and individuals are the focus of our consumer campaigns. We call it out, and if they refuse to change what they say, we make the companies who advertise with them politely aware of what they are funding and supporting. This is a highly effective strategy and calls on the advertisers to live by their own stated values, and to support the people who rely on them to stand up for what’s right. In this way we can change the world, one twisted commentator and one megalomaniac media magnate at a time.

**Nimic** is a Python module that enables developers to write AOT-compilable code using a Python-based domain-specific language to achieve C-level performance. By transpiling Python code into the Nim programming language, Nimic allows developers to maintain a single codebase that remains valid Python during execution while performing like native Nim code. Key features include: * **Native-Level Performance:** Uses `ctypes` and custom type systems (scalars, structs, sequences, and enums) to provide memory and value semantics comparable to C/Nim. * **Pythonic Syntax with Nim Semantics:** Employs decorators and annotations (e.g., `@dispatch`, `@template`, `@distinct`, and `mut @`) to mimic Nim’s powerful features like multi-dispatch, templates, and compile-time evaluation. * **Robust Transpilation:** Includes an AST-based transpiler that converts Python syntax into equivalent Nim source code. * **Dual-Use Compatibility:** The code acts as executable Python at runtime while providing the necessary metadata for Nim compilation. Nimic bridges the gap between Python’s developer-friendly syntax and Nim’s high-performance compilation, making it an ideal tool for performance-critical tasks without abandoning the Python ecosystem.

British-based employees of the Wikimedia Foundation (WMF) have formally requested that management recognize their right to be represented by the United Tech and Allied Workers (UTAW) union. Citing a need for greater transparency and trust following a period of significant organizational change, the staff members—who are deeply committed to the Wikimedia mission—are calling on the foundation to honor its public commitments regarding workers' rights to organize. The effort is part of a broader global movement known as Wiki Workers United (WWU) and has garnered support from over 1,000 Wikimedia volunteers and community members. With the United Kingdom serving as the WMF's largest employment hub outside of the United States, a substantial majority of the UK staff have already joined the union. The WMF, which manages Wikipedia and its sister projects, now faces pressure to uphold its stated values by formally recognizing the union’s collective bargaining status.

Bohemia Interactive has open-sourced the engine and game source code for *Arma: Cold War Assault* (originally *Operation Flashpoint: Cold War Crisis*). Modernized to C++20 with CMake and Clang support, the repository is now available for the community to study, modify, and build upon. Key points regarding the release: * **Licensing:** The source code is licensed under GPL-3.0 (with additional Section 7 terms). However, this does not include trademark rights; forks must be renamed and cannot claim to be official Bohemia Interactive products. * **Game Data:** Assets (models, textures, sounds, etc.) are not included in the repository. They remain under the Arma Public License Share Alike (APL-SA) and must be sourced separately, such as through the free Steam demo. * **Third-Party Code:** Certain vendored libraries within the repository retain their own respective licenses. * **Maintenance:** This is a read-only repository. Bohemia Interactive will not accept pull requests or provide continuous updates. The community is encouraged to fork the code for further development and ports. For full technical details, build instructions, and licensing terms, consult the repository's documentation.

Downloading...