India adds 1.26 million credit cards in July

Microsoft launches limited-edition translucent green Halo accessories for Xbox 25th anniversary

Microsoft has unveiled the Designed for Xbox 25th Anniversary collection, a limited-edition lineup of gaming accessories now available for pre-order. The collection pays tribute to the Halo special edition of the original Xbox, with most items featuring a translucent 'OG Green' design. It complements the Xbox Series X25 limited-edition console announced at E3 in June. The lineup includes Razer's Basilisk V3 Pro 35K mouse, BlackWidow V4 75% keyboard, and Hammerhead V3 X HyperSpeed earbuds in the signature color. An 8BitDo Ultimate charging dock for Xbox Series X/S and Xbox One controllers in translucent green is also part of the collection.

Email Reset Links Beat SMS OTP for Most SaaS Password Recovery Flows

For most SaaS platforms, an emailed single-use reset link is the recommended default for password recovery, as it keeps the login identifier, recovery channel, and support workflow in one place. SMS OTP may seem faster but introduces added complexity around phone-number verification, regional compliance, number reassignment, and delivery reliability. Developers are advised to model completed recoveries rather than raw message sends when comparing costs across the two methods. SMS should only be added when users genuinely lack email access or when the product already maintains continuously verified phone numbers — not simply because entering a short code feels smoother in a demo. Neither channel is a universal fallback, and the core goal remains safely returning the right person to their account without creating openings for account takeover.

Why retrying PostgreSQL serialization errors at the query level is always wrong

PostgreSQL's error code 40001 signals a serialization failure that aborts the entire transaction, not just the individual query that triggered it. Retrying only the failed query — a common developer instinct — is ineffective because the connection enters an aborted transaction state, causing all subsequent statements to fail with a different error code. The popular Node.js ORM ecosystem largely sidesteps the problem: a TypeORM feature request for automatic transaction retry has remained unimplemented since February 2023, and the widely used typeorm-transactional library offers no retry mechanism despite 188,000 weekly downloads. Further investigation revealed that where PostgreSQL surfaces the 40001 error is unpredictable, varying across versions and query plans — sometimes at commit, sometimes mid-transaction. The only correct approach is to wrap and retry the entire transaction from the start, not individual queries within it.

Directing AI Video Models: Why Every Frame Must Be Engineered, Not Captured

A developer documented hard-won lessons from producing a 10-episode AI-generated video series using image-to-video models. Unlike real cameras, these models have no spatial memory — they invent new pixels by statistically guessing what should fill any area outside the original still image. Continuity must be manually re-declared every clip, as the model has no awareness that consecutive shots share the same character or setting. The model also tends to hallucinate extra figures, clutter, and environmental details unless explicitly instructed otherwise in every prompt. The author concludes that directing such a model is less about choosing what to show and more about carefully restricting what the model is allowed to imagine.

NSA Doval Arrives in Beijing for Two-Day Visit Amid India-China Ties Rebuild

National Security Advisor Ajit Doval arrived in Beijing on Monday for a two-day official visit. The trip is widely viewed as part of ongoing diplomatic efforts between India and China to restore bilateral relations. Ties between the two neighbours deteriorated sharply following the deadly Galwan Valley clashes in June 2020. Doval's visit signals continued engagement at the senior security level as both sides work to normalise the relationship.

FP8 quantization boosts Qwen3-8B throughput 1.5x with no factual accuracy loss

Engineers tested whether switching Qwen3-8B inference from BF16 to FP8 quantization degraded output quality on an RTX PRO 6000 Blackwell GPU using vLLM, achieving a throughput jump from 1,725 to 2,597 tokens per second at concurrency 32. A structured 20-prompt benchmark spanning reasoning, math, code, and summarization tasks was run under greedy decoding to isolate any differences caused purely by numeric precision. Of the 20 outputs compared, 7 were byte-identical, 9 showed only minor wording or formatting variation, and 3 had small stylistic regressions such as a repeated word or a slightly off-topic list item. Critically, zero prompts produced a factual or numerical error in FP8 that BF16 had answered correctly, clearing the team's defined acceptance threshold. The authors caution that this validation applies only to the tested prompt profile and workload, and that different domains, longer contexts, or sampled decoding would each require a separate review pass.

Kerala Officially Renamed Keralam After Centre Notifies State's Request

The Indian state of Kerala has been officially renamed Keralam following a central government notification. The name change reflects the state's original Malayalam pronunciation and identity. The Kerala state Assembly had earlier passed a resolution in favour of the new name. The state government forwarded this resolution to the central government in June 2024. The Centre has now formally notified the new name, making it official.

Dev Setup Lets You Run GPT Sol and Claude Opus Free via OmniRoute and Pi Agent

A developer has shared a method to access frontier AI models like GPT Sol and Claude Opus at no cost by combining two tools: Pi, an npm-based coding agent, and OmniRoute, a locally hosted gateway. OmniRoute routes requests across hundreds of AI providers, automatically selecting free-tier options and falling back if one hits a rate limit or goes offline. The setup requires installing both tools via npm, configuring Pi to point to OmniRoute's local endpoint, and editing two JSON config files to set the default provider and model. Once configured, users can switch between models using a single command-line flag without managing separate API keys or accounts. The author also flags two known issues: OmniRoute's built-in Pi integration writes to the wrong config file, and bare OpenAI model names fail without the required namespace prefix.

How to properly verify OmniRoute AI gateway works before connecting coding agents

OmniRoute is a locally hosted AI gateway that exposes a single OpenAI-compatible endpoint and routes requests across hundreds of providers with automatic fallback. A DEV Community guide outlines a structured checklist to confirm a fresh install is fully functional, not just running. The verification process covers health checks via the built-in 'omniroute doctor' command, confirming provider connectivity, and testing basic chat completions using curl. Developers are also advised to validate streaming responses and tool/function calling, as failures in either can break most AI agent workflows. The guide emphasizes that a gateway merely starting up is not sufficient proof it will hold up under real-world conditions like provider outages or complex request types.

Web Scraping Tools Compared: When to Use ScrapingBee, Firecrawl, BuiltWith, or a Lightweight API

A developer who maintains the Web Metadata & Contact Extractor API has published a pricing and feature comparison of four data-extraction tools as of August 2026. The comparison covers ScrapingBee, Firecrawl, BuiltWith, and the author's own API, noting that each solves a distinct problem rather than competing directly. The author's API targets simple, server-rendered pages and extracts SEO metadata, tech stack, and contact signals for as little as $5.99 per month, while ScrapingBee handles JavaScript-heavy pages with a headless browser starting at $49 per month. Firecrawl is positioned for full-site crawling to feed LLM or RAG pipelines, and BuiltWith functions as a historical technology database rather than a live URL fetcher, with plans starting at $295 per month. The author explicitly acknowledges their tool's limitations and directs users to alternatives when JavaScript rendering or site-wide crawling is required.

React Native Developers Are the New Target: A Guide to Securing Your Own Machine

A detailed security guide highlights that React Native developers themselves—not just end users—have become the primary targets of supply chain attacks. The guide warns that routine commands like yarn install execute untrusted code with full user privileges, capable of accessing SSH keys, keychains, and environment files without any sandboxing. Configuration files such as metro.config.js, Podfile, and build.gradle are actually executable programs that run automatically on build or project open, creating silent attack surfaces. Real-world incidents like nx/s1ngularity, Shai-Hulud, and GlassWorm demonstrate that these threats are active, with one attack hiding malicious payloads in invisible Unicode characters that evaded code review entirely. The guide also flags AI agent MCP servers as an emerging risk, with over 30% found to carry exploitable vulnerabilities that have already been used to steal private SSH keys.

Delhi's Signature Bridge Audit Reveals Rusted Steel and Faulty Sensors

An audit of Delhi's iconic Signature Bridge has uncovered significant structural concerns, including rusted steel components and damaged sensors. The findings have raised questions about the maintenance standards of the cable-stayed bridge. In response, the Public Works Department has initiated the process of drawing up a dedicated maintenance and rehabilitation plan. The audit signals the need for urgent remedial action to preserve the integrity of the landmark structure.

India and China Agree to Pursue Political Resolution on Border Dispute

Indian National Security Advisor Ajit Doval arrived in Beijing on Monday for a two-day visit aimed at advancing bilateral ties between India and China. The visit is widely viewed as part of continued diplomatic efforts to restore the relationship between the two nations. Ties between India and China had deteriorated sharply following the deadly clashes in the Galwan Valley in 2020. Both sides have now agreed to seek a political settlement to their long-standing border dispute.

India, China seek 'political settlement' on border

Dolly Parton, US country star, actor and philanthropist, dies aged 80

Article URL: https://www.theguardian.com/music/2026/aug/25/dolly-parton-country-singer-dead Comments URL: https://news.ycombinator.com/item?id=49438052 Points: 24 # Comments: 0

Hermes Agent Builds Persistent Skills to Cut Costs for Long-Running AI Tasks

Nous Research released Hermes Agent in February 2026 as an open-source MIT-licensed runtime designed to address a core weakness in AI agent frameworks: the inability to retain and reuse knowledge across sessions. Unlike conventional frameworks that discard reasoning after each task, Hermes logs decision points and tool calls, then enters a reflective phase to assess what worked and convert successful approaches into structured 'skill documents.' These documents are indexed using SQLite FTS5, allowing future similar tasks to query the skill library before invoking the model, with community benchmarks showing up to 40 percent speed gains after around 50 accumulated skills. The architecture is built on five pillars — memory, skills, a persistent behavioral config called 'Soul', scheduled cron jobs, and a self-improvement meta-layer — all designed to compound efficiency over time. Deployment options range from a $59/month managed service to self-hosted builds costing as little as $6–$9 per month, with local inference on an 8B model reportedly achieving 91 percent tool call accuracy on just 8GB of VRAM.

Priti Adani Calls on Philanthropy to Build Lasting Capacity Over Short-Term Programmes

Priti Adani has urged the philanthropic sector to focus on building long-term capacity rather than simply running temporary programmes. She emphasized that true independence for millions of women means access to choice, opportunity, and upward mobility. Dr. Adani framed these freedoms not as abstract ideals but as concrete necessities for women's empowerment. Her remarks highlight a broader call for a more structural and sustainable approach to social development.

Why AI Agents That Shine in Demos Often Fail When Deployed to Production

Engineering teams are under growing pressure to ship AI agents after seeing polished demos, but moving from demo to production exposes a series of critical failure modes. Because large language models are non-deterministic, every agent response must be validated against a strict schema rather than treated as a reliable output. Production agents also require robust orchestration safeguards — including timeouts, circuit breakers, and retry logic — to prevent runaway tool calls that can incur massive costs overnight. Context management is another major gap, as large but noisy context windows often perform worse than smaller, precisely curated ones, requiring semantic retrieval, state persistence, and pruning strategies. Unlike demos evaluated by human observers, production systems demand automated evaluation pipelines to catch regressions and edge cases at scale.

How to Prevent Split-Brain and Configure Quorum in Proxmox HA Clusters

Proxmox VE high-availability clusters face a critical risk called split-brain, where a loss of communication between nodes causes each node to independently assume control, potentially running duplicate VMs and corrupting data. The quorum mechanism addresses this by requiring a majority of nodes to agree before any cluster decisions are made, with a minimum of two votes needed in a three-node setup. Administrators can use the pvecm command-line tool to check cluster status, set expected node counts, and add a dedicated witness server that acts as a neutral tiebreaker without hosting any virtual machines. Common pitfalls include network latency causing false node failures, misconfigured quorum expectations, and relying on a single witness node that itself becomes a point of failure. Best practices recommend using a dedicated cluster network, regularly verifying quorum settings with pvecm status, and deploying two witness servers for added redundancy.