The U.S. labor force participation rate fell to 61.6% in June 2026, marking its lowest level since 1976 (excluding the pandemic). However, this decline is not solely evidence of workers leaving the market. An analysis of the data identifies three distinct drivers: 1. **Statistical Correction (43%):** A large, one-time population-control revision by the Bureau of Labor Statistics in January 2026 reweighted demographic data, effectively "correcting" previously overstated participation levels. 2. **Demographic Aging (16%):** A steady, long-term trend where a growing share of the population is 65 or older, a group with inherently lower labor force participation. 3. **Behavioral Change (41%):** A sharp decline in participation among prime-age workers (25–54) in June. While the drop in prime-age participation is significant, it largely reverses gains made earlier in the year, returning the rate to a range consistent with 2023–2025 levels. Because this segment is usually stable, the coming months are critical; if this rate stabilizes, the June drop may be viewed as a one-month outlier, but a continued decline would signal a more concerning shift in worker behavior.

After nine years of development, Adrian Sieber has released **Perspec 1.0**, a cross-platform desktop application designed for high-quality document and receipt perspective correction. Dissatisfied with mobile scanning apps that suffer from poor image quality, aggressive compression, and ineffective preview features, Sieber built Perspec to leverage the power and precision of a desktop workflow. The core of the application is **FlatCV**, a custom C library developed by the author to handle computer vision tasks like corner detection and adaptive binarization. Unlike standard apps that use line-detection pipelines—which often fail on wrinkled or curved documents—Perspec uses watershed segmentation and Förstner corner detection to identify document boundaries accurately. The app also features a specialized "Save BW Smooth" function that removes shadows and creates anti-aliased, 1-bit black-and-white images, offering superior quality and smaller file sizes than typical JPEGs. Available for Windows, macOS, and Linux, Perspec provides a streamlined interface for batch processing with manual override capabilities. While 1.0 marks a major milestone, future updates will include support for standardized output sizes and QR code metadata detection. Licenses are available for purchase via itch.io and Gumroad.

Amanda, the creator of the *Not-Ship* newsletter, is taking a scheduled break until the end of August to prioritize sustainability and prevent burnout. Despite the common pressure for independent creators to maintain a constant content stream to avoid losing audience trust, Amanda emphasizes that long-term success requires protecting her well-being. During this hiatus, she plans to focus on deep-dive research, strategic planning, and website upgrades. The break also prompted a reflection on the lack of mandated vacation time for the self-employed, particularly in comparison to global standards for paid leave. Before signing off, Amanda shared an update to the *Not-Ship* mega summer reading list, which now features over 1,200 recommendations. The weekly newsletter will return to inboxes at the end of August.

This narrative, framed as a "fanfic," provides a fictionalized glimpse into the basement workshop of Satoshi Nakamoto during the creation of the Bitcoin Genesis Block. Amidst a makeshift environment of discarded electronics, old netbooks, and train sets, the protagonist grapples with the weight of launching a revolutionary financial experiment. The atmosphere is one of humble intensity: a cold, half-finished basement repurposed into a nerve center for a world-changing project. The story highlights the tension between the protagonist’s self-doubt—fearing the project is merely another failed academic exercise—and the realization that it has taken on a life of its own. In a pivotal moment of finality, the creator encodes the timestamped headline from *The Times* (“Chancellor on brink of second bailout for banks”) into the Genesis block as an epitaph for the failing traditional system. Accompanied by the music of Bruce Springsteen, the protagonist initiates the system, choosing to embrace the myth of the "brilliant hacker" rather than reveal the decade of incremental, practical labor that brought Bitcoin into existence.

Shieldstral is a new 3B open-weights, multimodal safety classifier that redefines content moderation by moving away from fixed, pre-defined harm taxonomies. Instead, it treats moderation as a binary, natural-language question-answering task. By providing an "instruction" (the policy context) and a "query" (the specific safety question) at inference time, developers can adapt the model to diverse requirements without retraining. Key highlights include: * **High Performance:** Despite its compact size, it matches or outperforms models up to 7x larger across text safety, refusal detection, and multimodal benchmarks. * **Flexibility:** A single interface handles text, images, and prompt-response pairs, returning a calibrated safety score based on a single forward pass. * **Efficiency:** Designed to run on a single 16GB GPU. * **Open Access:** Released under an Apache 2.0 license. Developed by consolidating heterogeneous datasets and teaching the model to perform nuanced policy discrimination rather than simple memorization, Shieldstral offers a highly adaptable solution for product safety. It represents a significant step toward context-aware moderation, allowing developers to define safety standards dynamically to fit their unique products and audiences.

We leverage reinforcement learning to develop advanced quant research agents by training them in environments built from real market data. Unlike static problems, market data provides a dynamic, adversarial benchmark that prevents model saturation: as agents find edges, the market evolves, forcing continuous improvement and adaptation. Our approach centers on the "Horizon" of decision-making. Trading is treated as a long-term strategic task rather than a series of isolated events. Agents must operate under partial information, manage the compound effects of past decisions, revalue objectives as market conditions drift, and pivot when regimes shift. By integrating professional tools with Bash scripting, our agents don't just execute trades—they build their own research infrastructure, learning to plan for future consequences and navigate the shifting, high-stakes landscape of financial markets.

A recent study of 30,954 people playing a digital lawn-mowing game reveals how humans solve complex "coverage path planning" problems. The task, similar to the classic Traveling Salesman Problem, requires finding the shortest path to cover all squares in a grid. Despite the complexity, participants were remarkably efficient, with the median player achieving 91% optimality. Results remained consistent even as lawns grew larger, defying expectations that human performance would degrade with scale. The study found that success is not determined by how much time someone spends thinking, but by *when* they think. High-performing players, like the interviewee "Sarah," used "decomposition"—breaking the lawn into smaller sections—and planned ahead to avoid dead-ends that force backtracking. While poor performers reacted to dead-ends after entering them, top players paused at critical "forks" to map out complex areas, then intuitively navigated open spaces. Ultimately, humans don't solve these problems by calculating every possible route; they employ effective heuristics, allocating cognitive effort only to the most difficult decisions. This strategy allows people to find near-perfect solutions for problems that would be computationally impossible to solve through brute force.

Checking your browser...Javascript required

Ask users to give MariaDB a star by having an extra line in the MariaDB client prompt: Welcome to the MariaDB monitor. Commands end with ; or \g. Your MariaDB connection id is 33 Server version: 12.2.0-MariaDB-1:12.2.0 mariadb.org binary distribution Copyright (c) 2000, 2018, Oracle, MariaDB Corporation Ab and others. Help others discover MariaDB. Star it on GitHub: https://github.com/MariaDB/server Type 'help;' or '\h' for help. Type '\c' to clear the current input statement. MariaDB [(none)]> Additionally, have this extra line in server logs: [Note] Help others discover MariaDB. Star it on GitHub: https://github.com/MariaDB/server This change is done in a way that it is easy to cherry-pick to older releases, and the text can later be changed to promote something else. Test file updated with: nano --noconvert --nonewlines mysql-test/main/mysql-interactive.result

Client Challenge JavaScript is disabled in your browser. Please enable JavaScript to proceed. A required part of this site couldn’t load. This may be due to a browser extension, network issues, or browser settings. Please check your connection, disable any ad blockers, or try using a different browser.

Checking your browser...Javascript required

The AI industry faces a potential financial crisis driven by the race between expensive proprietary frontier models and rapidly improving, cheaper open-weight alternatives. Companies like OpenAI and Anthropic have invested trillions in data centers and debt-fueled infrastructure, banking on the ability to command monopolistic pricing. However, with open-weight models—including competitive options from China like Moonshot AI’s Kimi—narrowing the performance gap to just a few months, this business model is under threat. As customers increasingly gravitate toward affordable, customizable open-weight models, AI firms are struggling to justify their massive capital expenditures. Executives are now lobbying for regulatory protection, fearing that a shift away from proprietary software could collapse the "AI bubble." Analysts warn that if these high-cost models fail to maintain their market dominance, the resulting financial fallout could trigger widespread instability, impacting major investors and tech giants alike. Ultimately, the industry is caught in a precarious cycle: needing to monetize rapidly to cover unprecedented costs, while simultaneously losing its competitive edge to a burgeoning market of open-access alternatives.

Enable JavaScript and cookies to continue

In this essay, Paul Graham explores the nature of "fanboys" and "haters," arguing that both are obsessive, uncritical, and driven by identity rather than reality. While fans project an idealized image, haters project an exaggerated, negative one. Graham posits that "haters" are essentially the inverse of "fanboys." They are often individuals plagued by frustrated talent who cannot reconcile a target's success with their own lack thereof, leading them to label the successful as "frauds." Crucially, he observes that while people who do great work never become haters, haters rarely achieve greatness themselves. Because both groups are fueled by an irrational, obsessive need to define themselves through someone else’s existence, Graham suggests that the most effective way to deal with haters is to treat them exactly like fanboys: ignore them. He warns against the common mistake of engaging them as if they are rational participants in a legitimate dispute. By recognizing that their negativity is a reflection of their own internal issues—rather than a critique of the recipient’s actions—the famous can avoid wasting time and energy on those who are fundamentally impossible to please.

**cMCP (Confidential MCP Runtime)** is an open-source gateway designed to secure AI agent tool calls by enforcing policy within a hardware **Trusted Execution Environment (TEE)**. Unlike software-only governance, which is vulnerable to rogue administrators or memory-tampering, cMCP executes the **Cedar policy engine** inside a secure enclave. By measuring the policy bundle hash into a hardware-attested report before execution, cMCP ensures that tool calls are evaluated according to the exact policy intended, preventing unauthorized data exfiltration or tool usage. **Key features include:** * **Hardware-Attested Security:** Supports TPM, AMD SEV-SNP, and Intel TDX to isolate the policy engine from the agent and host OS. * **Tamper-Evident Auditing:** Each session produces a signed **TRACE Claim**, providing verifiable proof of every tool call, decision, and audit chain without requiring trust in the operator. * **Ease of Use:** Supports a `dev` mode for software-only testing, allowing developers to build and tune policies locally before deploying to production TEEs. * **Compliance:** Designed to meet rigorous standards such as NIST SP 800-207 and the EU AI Act by providing cryptographically signed, verifiable governance for agentic workflows.

The war in Gaza has transformed the enclave into a landscape of sixty million tons of toxic rubble, creating an environmental disaster that threatens to cause long-term epidemics of cancer and lung disease. Professor Yaakov Garb’s research uses satellite imagery and modeling to map contamination hotspots—such as shattered asbestos in refugee camps and lead-acid battery residues—that mirror the hazardous fallout of the 9/11 attacks. Despite the gravity of the situation, cleanup efforts are nonexistent. The Israeli government prohibits the export of soil samples, and Palestinian scientific infrastructure has been destroyed. Requests for environmental assessments, including a USAID-funded project, have been suspended or ignored by authorities, leaving humanitarian workers and residents to navigate, breathe, and live amidst toxic dust. Rather than prioritizing remediation, proposed post-war reconstruction plans—such as Jared Kushner’s "Riviera" development—aim to build on contaminated land, often ignoring the risks posed by shattered hazardous materials. Garb’s observations suggest that even current military activity in these areas is exacerbating the spread of pollutants. Ultimately, the lack of oversight and refusal to engage with toxicological data indicate a systemic blindness that sacrifices long-term public health for geopolitical and redevelopment objectives.

The vlt team has announced the general availability of vlt 1.0, an end-to-end package management platform designed for faster, more secure development. As a drop-in replacement for npm, vlt features a high-performance CLI with sophisticated "graph-native" query selectors, allowing developers to audit dependencies for vulnerabilities, outdated versions, or malware across their entire local environment. Key features include: * **Enhanced Security:** Real-time malware blocking, OIDC "trusted publishing," and phased installations that prevent malicious scripts from executing. * **Powerful Tooling:** Dependency management through centralized catalogs and CSS-like specificity overrides via DSS selectors. * **Infrastructure:** Global, edge-hosted registries that are backwards compatible with npm, pnpm, yarn, and bun, offering up to 38% faster install speeds. * **Scalability:** Designed to support AI-native development workflows, providing reliable performance for ephemeral environments and agent-based builds. Vlt provides a secure, developer-friendly alternative to traditional registries with a generous free tier, aiming to simplify private package management while proactively neutralizing supply chain threats.

### Summary: The AI Judgment Gap Despite record-breaking execution metrics, many organizations are seeing a plateau in actual business impact. While AI-driven speed has scaled output, it has inadvertently destroyed the "coherence infrastructure" required for long-term success. The root cause is the **atrophy of organizational judgment**. Traditionally, judgment was developed as a byproduct of "slow" work—apprenticeships, hands-on troubleshooting, and debating hard trade-offs. AI has accelerated work by removing the friction that previously served as a training ground for senior and junior talent alike. By automating routine decision-making, organizations have hollowed out the "whitespace" of craft—the invisible experience that allows experts to identify structural risks before they appear in the data. When decision-making becomes a process of selecting from AI-generated, "plausible" options, the system stops producing original outcomes and converges toward a mediocre center. This leads to the departure of top talent, who feel reduced to "rubber stamps." **To survive, organizations must stop treating judgment as an innate trait and start treating it as a renewable infrastructure:** * **Make decisions visible:** Restore the "open classroom" by documenting the *why* behind decisions, not just the *what*. * **Design for consequence:** Re-introduce roles that carry real-world stakes so that junior staff can build their own world models. * **Incorporate non-quantifiable judgment:** Build decision processes that value experienced intuition and the "absence of signal" (what *isn’t* being said) over mere rubric-based metrics. * **Challenge, don't confirm:** Use AI as an adversary to stress-test hypotheses, preventing the echo-chamber of convergent, AI-assisted groupthink. Judgment is the final, non-commoditized competitive advantage. Organizations that fail to deliberately rebuild their judgment-formation pipelines will find themselves efficiently executing in the wrong direction.

Contains Copernicus DEM (COP-DEM-GLO-30), IGN RGE ALTI®, and CNIG LIDAR data © EU/ESA/IGN/CNIG.

This dataset provides an open-access analysis of 542 digital mental health organizations that exited the market between 2000 and 2026. Available for free download under a CC BY 4.0 license, the data categorizes companies by business model, funding, clinical evidence, and the specific reasons for their failure. Key takeaways include: * **Payer matters:** Companies targeting institutions (employers, insurers, hospitals) have significantly higher survival rates compared to B2C models. * **Clinician impact:** Contrary to expectations, having a medical co-founder had no measurable impact on the exit rate of these organizations. * **Methodology:** The research combines factual data from sources like Crunchbase with LLM-assisted classification. To ensure transparency, the author provides detailed rationales for every label, encouraging users to verify and critique the findings. The dataset serves as a "graveyard" rather than a probability model, designed to help founders and researchers understand the common pitfalls in the digital mental health space. It concludes with a comprehensive 406-page report detailing seven patterns of failure, intended for those building in this volatile sector.

Warp has launched the **Warp Agent CLI**, a standalone, multi-model tool designed to bring advanced agentic capabilities to any terminal, including Ghostty, iTerm 2, and VS Code. Built on Warp’s terminal infrastructure, the CLI functions as an intelligent multiplexer. This unique architecture allows for persistent sessions, the ability to switch directories, and the power to control full-screen interactive apps (like `sqlite` or `vim`) directly through the agent. **Key features include:** * **Native Integration:** Automatically distinguishes between shell commands and natural language prompts, with integrated tab-completion. * **Advanced Workflows:** Supports multi-agent orchestration, cloud-based agent handoff, and the ability to delegate tasks to external harnesses like Claude Code. * **Optimized Performance:** Features built-in model routing to balance task complexity with cost-efficiency, supporting both frontier and open-weight models. * **Remote Capability:** Enables agent-assisted workflows on remote machines without requiring additional binary installations. The Warp Agent CLI is available now for Mac, Linux, and Windows, offering flexible access via subscription, ad-hoc credits, or custom API key configurations.

In a coordinated theft on August 1, all eight Flock license plate reader cameras operated by the Winona Police Department were sawed off their poles and stolen. Two additional cameras owned by Buffalo County were also taken from the Mississippi River Bridge. The stolen equipment is valued at approximately $24,000, a cost that impacts the Winona Police Department’s budget. The cameras, which were strategically placed at major highway entry and exit points, are used by law enforcement to investigate crimes and locate missing persons. While no suspects have been identified, the incident aligns with a growing national trend of vandalizing Flock surveillance technology, which has faced ongoing criticism from civil liberties groups concerned about government overreach. The Winona Police Department is currently investigating the thefts and has requested that anyone with information regarding the incident come forward.

An open browser tab often functions as a visual reminder of an unfinished task rather than a page you intend to read immediately. Because these tabs carry the "why" and "when" behind a decision, closing them feels risky, leading to digital clutter. To manage this without turning your browser into a to-do list, use a "decision slip" method. Instead of hoarding tabs, move the essential context—the link, the reason for saving it, and a specific time to revisit it—into a trusted system like Notes, Reminders, or the app Tably. By capturing the intent behind the page, you can safely close the tab. This process shifts the focus from keeping your browser empty to managing your unfinished decisions effectively. The goal isn't just a clean tab bar; it is moving your pending tasks out of the browser and into a structured system where they can be resolved, deferred, or deleted at the appropriate time.

The author argues that the current AI boom is a massive, circular financial scheme. While hyperscalers like Amazon, Google, and Microsoft report record cloud growth, the author contends that this revenue is heavily inflated by the compute spend of two unprofitable "load-bearing" startups: OpenAI and Anthropic. These hyperscalers are essentially funding their own revenue by pouring hundreds of billions of dollars into capital expenditures (data centers and GPUs) to support these two companies, which in turn use that money to pay the hyperscalers for cloud services. Without OpenAI and Anthropic, the author argues, the AI businesses of the "Big Three" would be catastrophically small and unable to justify the massive debt and infrastructure investment incurred. The piece characterizes this as the largest capital misallocation in business history, claiming that investors are being misled by a lack of transparency regarding customer concentration. The author warns that the AI bubble is a "doomed mission" built on mania and deceit; when the venture capital funding dries up or these AI labs fail, the resulting collapse will leave the industry with billions in unproductive assets and debt, turning today’s tech giants from growth vehicles into stagnant, overleveraged mainstays.

**Truemetrics** is a YC-backed logistics tech company solving last-mile delivery inefficiencies by using smartphone data to track couriers' movements. With €1.5M ARR and enterprise clients like GLS and DPD, they are now scaling from €1M to €10M+. They are seeking a high-autonomy, commercial leader to work directly with the founders in Berlin. This role is a mix of product strategy and business development: you will identify the "next wedge" for growth (new products/markets) while optimizing the core business (improving operational impact). **The Ideal Candidate:** * **An Originator:** You don't wait for instructions; you form a POV, test it with customers, and execute. * **Commercial & Operational:** You are comfortable speaking with COOs and riding along with drivers to solve real-world problems. * **Background:** Likely an ex-founder, a high-impact GTM leader, or a top-tier operator who thrives in high-speed, ambiguous environments. This is a full-time, on-site role offering significant equity, direct access to leadership, and the opportunity to shape the company’s trajectory without corporate bureaucracy. If you prefer structure and rigid processes, this is not the right fit.

Please enable JS and disable any ad blocker

In this study, Akhtar et al. examine the phenomenon of "benchmark saturation," a critical issue where AI benchmarks lose their ability to differentiate between increasingly capable models as performance plateaus. The researchers analyzed 60 language model benchmarks using 14 saturation-related metrics. Their findings reveal that nearly half of these benchmarks are saturated, with the likelihood of saturation increasing as benchmarks age. Contrary to some assumptions, the study suggests that the presence of public test data is not the primary driver of this decline; instead, the key to building resilient evaluation tools lies in expert curation. Ultimately, the authors argue that by prioritizing specific design choices, researchers can extend the longevity of benchmarks and foster more durable evaluation frameworks for the future of AI development.

Apple has escalated its legal battle against OpenAI by seeking a preliminary injunction to halt the development of any AI products allegedly built on stolen Apple trade secrets. In a new court filing, Apple requests expedited discovery, claiming that an investigation has uncovered evidence involving at least 11 additional former employees beyond those originally named. The company cites instances of departing staff sharing confidential documents and improperly retaining work devices, suggesting a broader pattern of intellectual property theft. OpenAI has denied these allegations, dismissing the injunction request as unnecessary and based on falsehoods. The company maintains it has no interest in Apple’s proprietary data and characterizes Apple’s claims as a distraction. Furthermore, OpenAI accused Apple of previous procedural errors, such as misidentifying employees and failing to address its own internal security lapses that allowed former staff to retain system access. As the legal dispute intensifies, Apple continues to push for discovery to prove the extent of the alleged misconduct.

The MMI 5300, a landmark PROM (Programmable Read-Only Memory) chip from the early 1970s, stored 1024 bits of data using a unique architecture of Nichrome fuses and diodes. Unlike modern memory, the 5300 was "write-once": users programmed it by applying high-voltage pulses to melt specific fuses, creating a permanent, non-volatile record. The chip featured a 33×33 storage grid, with 256 4-bit words selected via complex address-decoding logic. Because the PROM arrived pre-filled with 1s, testing was notoriously difficult; manufacturers included an extra row and column of fuses to verify the circuitry before shipping. The chip also featured modular silicon design, where a single die layout could be repurposed for different versions (such as tri-state output variants) simply by altering the top metal wiring layer. Although vital for early computing, these chips were eventually rendered obsolete by erasable EPROMs and, later, modern flash memory. This evolution highlights a staggering leap in technology: where the 1971 MMI 5300 offered 128 bytes of permanent storage for $70, today's flash drives provide billions of times more capacity at a fraction of the cost.

Online advertising provider Adform recently suffered a security breach, allowing hackers to inject malicious code into the ads it serves. Because Adform delivers approximately 1.5 billion ads daily, this incident potentially impacted a vast number of users. The malicious code specifically targeted cryptocurrency users by monitoring their computer’s clipboard. Every three seconds, it replaced the user's intended crypto wallet address with one controlled by the attackers, increasing the likelihood that funds would be accidentally sent to the hackers. Adform has confirmed the breach but has provided limited information regarding how it occurred or the total number of victims. The company is currently investigating whether the hackers were also able to harvest browsing history. Security experts highlight this incident as a strong argument for using ad-blockers. By preventing third-party ad networks from loading code, users can effectively shield their devices from such malicious scripts, as well as pervasive tracking and surveillance. This breach serves as a stark reminder of the risks posed by malvertising and the importance of maintaining robust, proactive browser security.