Computer Science and Programming

Genie 4 Google DeepMind announced Genie 4, a new world model for real-time interactive simulation. It is natively multimodal, generating audio, ambient sound, and interactive dialogue alongside the simulated world. Gemini Omni anchors visual realism while Gemini 3.8 Flash acts as a guiding agent, observing frames and adapting plans in real-time. DeepMind is also experimenting with Gemini 4 Argon and Genie 4 alongside Gemini Robotics. The model is rolling out as a research preview to trusted testers first, ahead of availability to Google AI Ultra subscribers in Project Genie and early access Gemini API partners. All generated video and audio carries SynthID watermarking, and DeepMind notes visual drift and object permanence issues still occur in longer sessions

Writing a SERP scraper is 20 minutes. Keeping it alive past day two? That's the part nobody warns you about — the 403s, the CAPTCHAs, the proxy pool that dies at 3am. If you've been there, this is worth 2 minutes of your time: Thordata's SERP API MCP Server. You plug it into your MCP client (Claude Desktop, Cursor, etc.) once, and your AI assistant pulls live Google / Bing / DuckDuckGo results on demand — no scraping code to write or maintain. Setup is literally: thordata-serp-mcp --api-key=YOUR_KEY What you get back is structured JSON — organic results, ads, related queries. Feed it straight into your pipeline. No HTML parsing, no CAPTCHA babysitting, no dead-proxy alerts. Free tier: 5,000 requests. Enough to build something real before paying a cent. Docs & free key: https://www.thordata.com/products/serp-api/mcp-server?op=cianna&from=programming

Claude Enterprise, now available self-serve Claude Enterprise is now available for self-serve purchase with a seat-plus-usage pricing model. The offering provides organization-wide access to Claude, Claude Code, and Cowork with enterprise security features including SSO, SCIM provisioning, audit logs, custom data retention policies, and usage analytics. It integrates with Microsoft 365, Slack, Excel, and PowerPoint through connectors and built-in chat sidebars. Organizations use it across sales, engineering, marketing, product, and finance teams to accelerate workflows and handle complex tasks with large codebases and document sets.

Dark mode toggles: two states are enough • Lea Verou (Light, Dark, System). The core argument is that users only seek out theme toggles when something is wrong — not to preemptively express intent — making the 'System' state irrelevant at the moment of interaction. Tri-state toggles expose the underlying data model rather than aligning with actual user goals, adding cognitive load and UI friction for an extremely rare use case. A well-implemented two-state toggle can still represent all three underlying states: the first click overrides to the opposite of the current resolved value, and the second click removes the override and returns to system default. Common mistakes include storing a value that matches the system preference (silently pinning the theme) or removing overrides when the OS preference changes. Exceptions where tri-state controls are appropriate include dedicated settings panels and sites that implement meaningfully different color schemes depending on the OS setting.

BYD says it will have a solid-state car next year, the earliest date anyone has given BYD executive Stella Li claims the company will have a solid-state battery model next year, the earliest date any automaker has cited, though no model, sales plan, or confirmation was given. Toyota targets 2027-2028, Honda and Mercedes around 2030, and LG Energy Solution estimates roughly a decade away for cars. Mercedes and VW's PowerCo are partnered with US-based Factorial Energy, while Taiwan's ProLogium is building a plant in Dunkirk, France. CarNewsChina speculates, via unconfirmed sources, that BYD's car will be a Yangwang flagship rivaling Rolls-Royce. Hyundai's research chief doubts the raw material cost problem will be solved within five years.

Lessons From Four Years of Writing a Weekly Newsletter A .NET blogger reflects on four years and 210 weekly issues of his newsletter, distilling four lessons: consistency beats brilliance, teaching accelerates learning, fundamentals (modular monoliths, domain-driven business rules, API contracts, idempotency, outbox pattern) outlast frameworks, and audience feedback sharpens thinking. He also shares his bullish stance on AI coding agents, arguing fundamentals become the review checklist for judging AI-generated code, and teases more AI-focused content ahead

GigaChat 3.5 Reasoning: Advanced step-by-step problem solving GigaChat 3.5 Reasoning is a new open-source LLM designed to reason, not just respond. Built on GigaChat 3.5 Ultra, it processes math and coding tasks by exploring multiple step-by-step reasoning paths, using automated verification to reinforce correct trajectories. This enables autonomous multi-step planning, external tool invocation, and self-correction. The model features a proprietary linear attention architecture, optimizing long-context efficiency by retaining key processed points rather than re-matching from scratch. It is highly token-efficient, consuming on average 37% fewer tokens than DeepSeek V4 Flash Preview on math problems. Benchmark gains over the non-reasoning version: • IFBench: 44 → 77 • Natural Plan: 64 → 80 • LiveCodeBench v6: 56 → 85 Open-sourced under the MIT license. Weights available on Hugging Face: fp8 | bf16

Dates Are Not Data, They Are Context · Nikhil Gautam A developer recounts a bug where a billing report showed bills from the wrong day due to a timezone mismatch: the frontend sent plain calendar dates (e.g. '2026-01-01') without timezone context, and the backend converted them directly to JavaScript Date objects while MongoDB stored timestamps in UTC. The core lesson is that a calendar date is not the same as a timestamp - a date needs a timezone to be meaningful, and query boundaries should be computed in the user's local timezone before converting to UTC. Also covers using half-open [start, end) ranges to avoid excluding the last day of a range, and a mental model separating calendar dates, timestamps, and local date/time.

🧲 Forty tabs open and none of them is the answer. Docs, someone's gist, a five-year-old Stack Overflow thread. The search isn't the work — putting it together is, and that part is still on you. Create your own AI agent inside Telegram in about a minute. ▫️ paste the error and get the reason explained, not another snippet to try ▫️ send a repo link or a doc page and ask what it actually does ▫️ it writes the throwaway script you need and runs it right in the chat — parser, converter, quick report ▫️ describe the task by voice on the way home, read the answer when you are back ▫️ it keeps your project, your stack and your past decisions in memory for months ▫️ any model from the world's providers, switched with one command Setup takes a minute: open the link and name your agent. Free to use — no card needed. 🧲 Try your own agent

React 19.3 – React React 19.3 is now available on npm, stabilizing two previously experimental APIs: View Transitions and Fragment Refs. The new <ViewTransition> component animates elements entering, exiting, moving, or resizing using the browser's View Transition API, with support for custom animations via addTransitionType and integration with Suspense for coordinated loading of fallbacks, images, and fonts. Fragment Refs let developers attach DOM behaviors (event listeners, focus management, IntersectionObserver/ResizeObserver) to a group of sibling elements without needing a wrapper div, by passing a ref to a Fragment. React DOM adds a new browser() API to opt components out of server-side rendering when they depend on browser-only APIs, plus Trusted Types support to prevent DOM-based XSS. Server Components can now render Context directly from a 'use client' module without a wrapper Provider. The release also includes numerous other changelog items and bug fixes across react, react-dom, and react-server

Scaling Down Costs, Not Performance: How We Migrated from Couchbase to PostgreSQL Trendyol's product content team details a phased, zero-downtime migration of five 'helper bucket' datasets (2.24 billion documents across two clusters) from Couchbase to PostgreSQL to cut infrastructure costs while preserving read performance. The strategy combined dual-writes via Kafka consumers, a custom CDC-based migrator tool for historical backfill, and feature-flagged cutover of read traffic. Key production challenges included a mysterious TPS spike traced to PgBouncer's server_reset_query_always setting, a connection-limit incident causing a primary node failover, and a hidden scaling issue where PgBouncer's own horizontal autoscaling multiplied backend connections beyond expectations (20 pods × 150 default_pool_size = 3000 potential connections). Final production benchmarks showed stable P50 latency and only a brief P99 spike (200ms to 400ms) under peak load, validating the migration.

Bun v1.4.2 Bun v1.4.2 patches two regressions introduced in v1.4.1: a bun build variable name collision that broke Elysia imports with a SyntaxError, and an AsyncLocalStorage memory leak affecting timers and promises created inside store.exit() or nested store.run(). It also fixes a worker_threads 'online' event ordering bug that caused @discordjs/ws to hang, adds CMYK/YCCK JPEG decoding support to Bun.Image, pulls in roughly 350 upstream WebKit/JavaScriptCore commits fixing Intl, TypedArray, and Proxy correctness issues, and resolves several crash and hang bugs including a rare JIT crash, a musl GC crash, a bun install panic, and a double-close file descriptor bug on Linux

What’s new in Svelte: September 2026 Svelte 5.57 adds new SvelteMap methods (getOrInsert, getOrInsertComputed), a createContext has function, select defaultValue support, and new svelte/server type exports. SvelteKit 3's next release candidate continues iterating with several breaking changes including reworked cross-page form actions, split Vite plugin hooks, defineParams relocation, and route filtering for test files. The sv CLI replaces its mcp add-on with a broader ai-tools add-on and ships a task-based sveltekit-3 migration command that rewrites $lib imports to #lib and bumps dependencies. The roundup also covers stable SvelteKit 2.x patches, language tools updates, and a community showcase of Svelte apps, UI libraries, and dev tools.

First Look at @nestjs/observe in NestJS 12 NestJS 12 comes with some solid updates, no doubt about that. But if you haven’t looked at the new observability features yet, it is worth a close look. Setting up APM tools in Node has always felt a bit clunky. Most agents sit outside the app, monkey-patching HTTP drivers and database packages to guess what is happening under the hood. It works, but you usually lose framework context along the way. Full Article : https://lnkd.in/p/gP53bYmC

Node.js — Node.js 24.20.0 (LTS) Node.js 24.20.0 'Krypton' LTS ships several SEMVER-MINOR additions: using scopes for AsyncLocalStorage, an end parameter for Buffer, permission.drop and a new --permission-audit flag, package maps support in the loader, a new node:stream/iter module, context.log() in the test runner, and WebAssembly JSPI enabled. It also updates root certificates to NSS 3.125 and bundles numerous dependency upgrades (npm 11.19.0, ngtcp2, nghttp3, c-ares, sqlite, timezone data) plus a long list of bug fixes across crypto, buffer, http, http2, quic, and net modules.

Connection Pool Sizing, Measured: Why 48 Connections Beat 400 A benchmark sweep of a PostgreSQL instance from 1 to 400 connections shows throughput peaking at 48 connections (18,039 tps) and dropping 37% by 400 connections, while average latency multiplied 13x. The piece explains why connections are OS processes competing for cores rather than free capacity, tests the Universal Scalability Law fit (R²=0.94, predicted optimum 45.1), evaluates the (cores×2)+spindles formula (predicts 9, far off the measured 48), and finds that synchronous_commit=off raises throughput but does not shift the peak. It closes with practical guidance: watch queue depth (cl_waiting in PgBouncer, pending-threads in HikariCP) rather than connection count, account for instances × pool size multiplication, and use transaction pooling or RDS Proxy for Lambda to avoid connection exhaustion.

VoidZero Releases Vite+ Beta: A Unified Web Toolchain Behind a Single Command VoidZero, the company behind Vite founder Evan You, has launched the beta of Vite+, a unified web toolchain accessible through a single vp command. It bundles Vite, Vitest, Rolldown, tsdown, Oxlint and Oxfmt with a built-in task runner, offering commands like vp dev, vp check, vp test, vp build, vp pack and vp run. Since its alpha, the team has merged over 500 pull requests, expanded vp migrate coverage, and added enterprise features like organisation templates. Over 1,300 public repositories already use it, including Dify, BlockNote and Cloudflare's vinext. Reception on Hacker News was largely positive, though a widely shared GitHub write-up by Jared Wilcurt criticized its vp env Node version manager and warned about ecosystem lock-in and heavy reliance on Rust rewrites of existing tools.