Viral 'AI Fails' Often Reflect Poor Prompting, Not Flawed Technology

A developer on DEV Community ran an experiment to challenge the popular narrative that AI tools are unreliable and easily fooled. They deliberately instructed Claude, Anthropic's AI assistant, to produce an absurdly wrong answer about the number of R's in 'strawberry' — and the model complied with an intentionally ridiculous response. The author then used this self-generated example to illustrate how misleading it is to share such outputs as proof of AI incompetence. As of mid-2025, free-tier generative AI models have advanced significantly since ChatGPT first gained mainstream attention nearly four years ago. The writer argues that poor prompt quality, rather than inherent AI limitations, is increasingly the reason users receive unhelpful or inaccurate responses.

Mamata takes TMC symbol fight to Supreme Court

Developer builds 8-game browser arcade in plain TypeScript with no frameworks

Developer Minjia Hu has released a retro-styled browser arcade featuring eight classic games — Snake, Tetris, Breakout, Flappy Bird, 2048, Minesweeper, Sudoku, and Gomoku — playable instantly without installation or sign-up. The entire build weighs roughly 35 kB gzipped and uses only plain TypeScript and Canvas 2D with zero runtime dependencies. All eight games share a unified arcade-cabinet interface supporting both keyboard and touch input, while each game's logic is kept strictly separate from rendering code. The project includes 269 unit tests and 28 end-to-end Playwright tests running in continuous integration, and is released under the MIT license. The source code and live demo are publicly available on GitHub.

Developer Fixes WebRTC-Direct Race Condition and SCM Polling Bugs Across py-libp2p and Go Agent Orchestrator

Open-source contributor Yash spent the week working across py-libp2p and a Go-based agent orchestrator, shipping 8 commits and managing 11 pull requests. A key fix addressed a race condition in the WebRTC dialer where ICE credentials were not set correctly before first-contact replay, which had been blocking interoperability between Go and Python nodes. He also resolved socket leaks in the mplex and swarm components to prevent file descriptor exhaustion in long-running nodes. On the orchestrator side, a bug causing backslash-named workspace files to resolve incorrectly on Linux and macOS was patched, and a GitHub polling issue where open PRs were repeatedly re-listed due to a mishandled 'since' parameter was investigated. Additional work included documentation improvements for WebRTC-Direct dialing and an in-progress pull request introducing a more robust ICE-Lite listener.

Human brain is two separate organs, Stanford Medicine-led research finds

Article URL: https://med.stanford.edu/news/all-news/2026/09/two-separate-brains.html Comments URL: https://news.ycombinator.com/item?id=49763697 Points: 3 # Comments: 1

Stanford Study Finds Human Brain Develops from Two Distinct Cell Lineages

A new Stanford University study has revealed that the human brain originates from two separate groups of embryonic cells, each following its own distinct developmental path. This finding challenges the long-held scientific understanding of the brain as a centralized, unified structure. The research sheds light on how different brain regions form at the cellular level during early development. Scientists believe these insights could improve efforts to grow targeted brain cells in laboratory settings. The discovery also holds promise for better modeling and understanding various neurological and psychiatric disorders.

WebSockets vs. SSE vs. gRPC-Web: Choosing the Right Real-Time API Protocol

Traditional HTTP/1.1 request-response models are poorly suited for real-time, low-latency applications due to redundant header overhead and connection inefficiencies. Techniques like short polling and long polling attempt to simulate real-time communication but waste significant bandwidth — potentially over 800 Mbps in header metadata alone for 100,000 concurrent users. Long polling also suffers from repeated TCP and TLS handshake costs, and risks port exhaustion on load balancers under high traffic. To achieve sub-10ms delivery latencies, engineers are advised to move toward persistent, full-duplex or streaming protocols such as WebSockets, Server-Sent Events, or gRPC-Web. Each of these approaches replaces stateless request loops with long-lived transport channels optimized for continuous, efficient data flow.

Stepfun's Step 5 LLM Reaches Top Tier on AI Performance Benchmarks

Stepfun has released a preview of its Step 5 large language model, which has been evaluated on Artificial Analysis's benchmarking platform. The model has been placed on the AA Pareto frontier, indicating it achieves a strong balance between performance and efficiency. This positions Step 5 among the top-ranked models in its competitive category. The listing has drawn early attention from the AI community, though discussion remains limited at this stage.

OCI Workload Identity Federation enables keyless auth for GitHub Actions and Kubernetes

Oracle Cloud Infrastructure has introduced Workload Identity Federation (WIF), a mechanism that allows external workloads to authenticate with OCI using temporary credentials instead of permanent API keys or technical users. The feature is designed for CI/CD pipelines, Kubernetes environments, multicloud setups, and AI agents that require short-lived access to OCI resources. Rather than storing long-lived credentials in GitHub Secrets or Kubernetes Secrets, workloads present an identity token from their native platform, which OCI validates and exchanges for an ephemeral session token. The model eliminates the need to maintain standing OCI users per workload, reducing operational overhead around user provisioning, audits, and offboarding. OCI IAM can also leverage identity claims — such as repository, branch, or Kubernetes namespace — to make fine-grained authorization decisions.

How to Add MicroSD Storage to Arduino UNO for Long-Term Data Logging

Arduino UNO's limited internal memory makes it unsuitable for projects requiring extended data storage, such as weather stations or soil-monitoring systems. A MicroSD card module connected via the SPI interface offers a practical solution, allowing the board to read and write files on an external memory card. The module typically includes a 3.3V voltage regulator and logic-level conversion circuitry to ensure compatibility with Arduino's 5V logic. An I2C-connected 16x2 LCD can also be paired with the setup to display information retrieved from the card. The project relies on Arduino's built-in SD and SPI libraries, along with the third-party LiquidCrystal_I2C library for the display.

Montana Family Permanently Protects 57-Acre Stillwater River Wildlife Habitat

A Montana family has placed 57 acres of riverfront land along the Stillwater River under permanent conservation protection. The initiative aims to preserve critical wildlife habitat in the region. Camera traps installed on the property have recorded a wide range of species, including bears and wolves. The protected land now forms part of a larger network of conserved open spaces in the area. The move ensures the land remains undeveloped and accessible to wildlife for future generations.

Developer Launches Kueon, a Competitor Tracking and Keyword Monitoring Tool

A software developer has launched Kueon, a tool built to solve the repetitive challenge of manually tracking competitors and monitoring brand mentions across platforms. The idea emerged from years of bootstrapped SaaS experience, where the founder relied on Reddit, X, and LinkedIn conversations for low-cost user acquisition. Kueon alerts users when competitors update their pricing or changelogs and when specified keywords appear in Reddit or Hacker News discussions. The platform uses sitemaps, webpage content, and generative AI to streamline onboarding by auto-suggesting competitor profiles and relevant keywords. The name Kueon is derived from the word 'cue', reflecting the tool's core purpose of delivering actionable signals to its users.

How Edge Rendering and Micro-Caches Are Reshaping E-Commerce Checkout in 2026

Modern e-commerce platforms face severe performance bottlenecks when centralized databases handle both high-volume browsing and real-time checkout operations simultaneously. During traffic surges like flash sales, these systems can trigger race conditions, overselling, and deadlocks that crash entire storefronts. Google's 2026 update to Core Web Vitals replaced First Input Delay with Interaction to Next Paint, raising the bar for checkout responsiveness across all user interactions. To meet these demands, engineers are moving toward architectures built on three pillars: edge-based rendering, distributed micro-caches, and low-latency inventory locking. The shift aims to eliminate the single-database dependency that causes cart abandonment spikes and costly order cancellations under peak load.

Why a package you never installed can trigger a high-severity vulnerability alert

When a security scan flags a vulnerable package that does not appear in your project's manifest, it is likely a transitive dependency — one pulled in indirectly by a package you did explicitly install. A typical modern JavaScript project may have only a handful of direct dependencies but hundreds of transitive ones, making this a widespread and often overlooked risk. Fixing a vulnerable transitive dependency requires identifying which direct dependency pulls it in and upgrading that, rather than installing the vulnerable package directly. Package managers also 'hoist' dependencies within the file system, meaning a package's physical location in node_modules does not reliably indicate how many levels deep it sits in the actual dependency graph. Tools that trace the full dependency graph back to the root — using lockfiles from npm, Yarn, Maven, or similar systems — can pinpoint the exact upgrade or version override needed to resolve the issue.

IIT Bombay Student Faces 1-Year Suspension for Using ChatGPT During Exam

A student named Sahil Wakode at IIT Bombay is facing a potential one-year suspension for alleged academic misconduct during an examination. The institute has stated that Wakode uploaded the exam question paper to ChatGPT in order to obtain answers while the exam was in progress. The incident has drawn attention to the growing challenge of AI-assisted cheating in academic settings. IIT Bombay has confirmed the matter and indicated that disciplinary action is being considered against the student.

How Python's Fernet Library Uses Symmetric Encryption to Protect Stored Credentials

Desktop applications that store passwords or API keys locally face a security risk when saving credentials as plain text, since those files can be exposed through backups, sync folders, or accidental sharing. Symmetric encryption addresses this by using a single key to both encrypt and decrypt data within the same application, making it well-suited for local credential storage. Python's cryptography library offers Fernet, a ready-to-use recipe that combines AES encryption, HMAC-based integrity verification, and URL-safe Base64 encoding into one secure package. Unlike plain encryption, Fernet's built-in HMAC check ensures that any tampering with the stored token causes decryption to fail entirely, protecting data integrity alongside confidentiality. The primary challenge with this approach lies in key management, as losing the encryption key renders stored data permanently unreadable, making secure and redundant key storage a critical part of any implementation.

Cron Jobs Explained: How to Read, Write, and Avoid Common Scheduling Pitfalls

Cron is a time-based job scheduler originating in Unix systems since the 1970s, now widely used in tools like GitHub Actions, Kubernetes, and CI pipelines. A standard cron expression consists of five space-separated fields representing minute, hour, day-of-month, month, and day-of-week, with special characters enabling lists, ranges, and step intervals. A critical pitfall is that classic cron lacks timezone support, meaning jobs run in the host machine's timezone — typically UTC — which can cause schedules to shift during daylight saving time changes. Another common mistake is combining day-of-month and day-of-week fields, which cron evaluates with OR logic rather than AND, often producing unintended run times. Verifying any cron expression with a parser tool before deployment is strongly recommended to prevent silent failures in automated tasks.

How Math Concepts Improve Code: A C Developer's Rock-Paper-Scissors Comparison

A developer revisited two versions of a Rock-Paper-Scissors game written in C to demonstrate how mathematical thinking leads to cleaner, more efficient code. The first version contained a subtle but significant bug — reading an uninitialized variable in a while-loop condition — which happened to work only due to favorable garbage memory values, not correct logic. The second, improved version applied modular arithmetic to handle game logic more precisely and concisely, replacing brittle workarounds with mathematically grounded solutions. The author argues that math provides programmers with reliable building blocks for problem-solving, rather than relying on trial and error. The full source code for both implementations has been made publicly available on GitHub for comparison.

Neuralink helps ALS patient speak in their own voice via brain implant

Elon Musk's brain-computer interface company Neuralink has achieved a significant breakthrough by enabling a patient with ALS to communicate using a synthetic voice that mirrors their natural vocal tone. Musk described the development as a potentially transformative medical advancement. The technology works through a brain implant that translates neural signals into speech output. However, the device remains in its experimental stage and has not yet received regulatory approval for broader clinical use.

Remembering Yuvraj's six sixes on this day in 2007