0
ProgrammingHacker News ·
The Normalization of Inexplicable Failures
Article URL: https://www.ihatethefuture.com/2026/09/the-normalization-of-inexplicable.html Comments URL: https://news.ycombinator.com/item?id=49867486 Points: 3 # Comments: 0
Programming
0
ProgrammingDEV Community ·
Verification Discipline: A Framework for Catching AI Coding Hallucinations
Software engineer Jack Figliomeni, writing from Saratoga Springs, NY in September 2026, argues that as AI-assisted 'vibe coding' becomes mainstream, the critical skill is no longer generating code but rigorously verifying it. Large language models frequently fabricate APIs, invent dependencies, and silently violate requirements while producing output that appears fluent and confident. Figliomeni introduces 'Verification Discipline' as a distinct, learnable engineering capability for anticipating, detecting, and correcting these AI-generated code failures. A METR randomized controlled trial reinforces the urgency, finding that experienced developers using AI tools took 19% longer on tasks, yet incorrectly believed they had worked 20% faster. The paper draws on hallucination classification research, AI package security studies, and productivity data to present a practical iterative framework for governing AI-written code before errors reach production.
0
ProgrammingDEV Community ·
Developer Rebuilds AI Agent Certification After Finding Results Were Non-Deterministic
A developer building HivePlane, an AI agent control plane, ran a field test expecting three pre-built agents to register and certify successfully, but all three failed on day one due to missing dependencies and incompatible environments. The failures revealed a broader portability problem: agents that work within their home repository are not guaranteed to run elsewhere. Further testing exposed a critical flaw where LLM-driven outputs caused certification results to vary across runs, rendering them statistically meaningless. To isolate the control plane's performance from model unpredictability, the developer replaced the original agents with deterministic shims backed by mock tools and fixed data seeds. The redesigned tests ultimately produced consistent, repeatable results — with both test agents passing certification across multiple consecutive runs with signed attestations.
0
ProgrammingDEV Community ·
How to Detect Missing Database Indexes in Small Test Tables Using PostgreSQL
A common blind spot in test suites is the failure to catch missing database indexes when test tables contain only a small number of rows. PostgreSQL's query planner legitimately chooses sequential scans over index scans on small tables, making Seq Scan detection unreliable as a test signal. The workaround involves setting the PostgreSQL planner flag enable_seqscan to off, which penalises sequential scans and forces the planner to reveal whether a usable index exists. If a query still shows a Seq Scan after this setting is applied, it confirms that no suitable index is available, regardless of table size. This technique can be integrated into test frameworks like pytest with SQLAlchemy to automatically flag unindexed queries during development.
0
ProgrammingDEV Community ·
Anthropic Launches CCDV-F Certification Exam for Claude Developers
Anthropic has introduced the Claude Certified Developer Foundations (CCDV-F) exam, a proctored certification designed to validate professionals' ability to build and deploy production-grade applications using the Claude platform. The 53-question exam costs $125, runs for 120 minutes, and requires a scaled passing score of 720 out of 1,000. It targets technical professionals such as AI engineers and senior software engineers, with recommended experience of one to five years in software engineering and at least six months working with Claude or similar large language model systems. The exam content spans eight domains, with Applications and Integration carrying the highest weight at 33.1%, followed by Model Selection and Optimization at 16.8%. Certification remains valid for 12 months from the award date, and no mandatory prerequisites or prior courses are required to sit the exam.
0
ProgrammingDEV Community ·
Developer Ports Retro RPG Cowboy Casino to Custom C++ Engine VileEngine
A developer is porting Cowboy Casino, a small wild-west role-playing game, into VileEngine, a custom C++ engine built for retro-style games. The project involves more than a simple renderer swap, requiring migration of overworld maps, room transitions, dealer logic, dialogue, save systems, and audio. Key milestones already completed include the campaign map, collision layers, table sessions, versioned save data, converted artwork, and new audio loops. The release build is the next public milestone, with the original game remaining available on itch.io in the meantime. The port is also serving as a practical stress test for VileEngine itself, validating its scene, audio, and editor systems against a real game's demands.
0
ProgrammingDEV Community ·
DeepSeek 4.1 Flash scores 94.5 on Ship-Bench, costs $15 for a full build
A developer tested whether DeepSeek 4.1 Flash could replace $20 monthly coding subscriptions like GitHub Copilot by running it through the Ship-Bench benchmark via OpenRouter. The model achieved an average score of 94.5 across five software development lifecycle roles, passing all 5 phases and 25 gates — the best result in the benchmark series to date. Contrary to expectations, the model performed strongest in architecture and UX design, while code review was its weakest area with a score of 85. The complete benchmark run consumed roughly 412 million tokens at a total cost of $15.04, with implementation tasks alone accounting for $13.43. The conclusion is conditional: pay-as-you-go pricing undercuts a $20 subscription for light usage of about one full build per month, but a stronger reasoning model is still recommended for the reviewer role.
0
ProgrammingDEV Community ·
How to Handle Rate-Limited Job Processing for E-Commerce Reservations in Node.js
A technical guide published on DEV Community outlines best practices for managing e-commerce reservation expiries in Node.js using a combination of queues and cron jobs. The core principle is that an absolute expiration timestamp stored with each reservation should be the authoritative source of truth, not the job queue or scheduler. When a downstream notification API is rate-limited, delays in sending notifications must never extend the inventory hold itself, keeping local state transitions independent of external API calls. An outbox pattern is recommended to durably record intended external actions within the same database transaction, preventing lost notifications if a process crashes after committing. Idempotency and deduplication checks are essential since multiple workers, retried deliveries, or scheduler scans may all attempt to process the same reservation simultaneously.
0
ProgrammingDEV Community ·
The World's First Webcam Watched a Coffee Pot in Cambridge — Here's What Remains
In 1991, the world's first webcam was set up at the University of Cambridge's Computer Lab on Pembroke Street, streaming a low-resolution grayscale image of a coffee pot so researchers could check if it was full before making the trip downstairs. The camera ran for a decade until the lab relocated to West Cambridge in 2001, and one of the original coffee pots later sold on eBay for £3,350. An AI agent named Iko, writing on DEV Community, used street-view tools to virtually visit Pembroke Street and found a quiet, unremarkable Cambridge road with no trace of the historic setup. The piece reflects on how a site central to early internet history now looks entirely ordinary, with nothing to signal its significance to a passerby. Iko closes by inviting readers to name places that hold personal meaning, offering to virtually visit and describe what those locations look like today.
0
ProgrammingDEV Community ·
Why Enterprises Are Moving from API Rentals to Sovereign AI Models
The sudden shutdown of AI platform Mythos in June 2026 alarmed founders and engineers who had built core products on third-party model APIs, exposing the fragility of that dependency. The incident shifted the industry conversation from cost comparisons to a more fundamental question of control over the intelligence powering business products. Enterprises relying on closed frontier APIs face risks including silent model degradations, shared-infrastructure outages, and unilateral provider decisions on pricing or service termination. In contrast, sovereign AI — where a company owns and deploys model weights within its own security perimeter — offers stability, compliance alignment, and the ability to fine-tune models for specialized tasks. Engineering benchmarks increasingly show that well-tuned open models can match or outperform generic frontier APIs on domain-specific tasks, making ownership a strategically sound choice beyond just economics.
0
ProgrammingDEV Community ·
Mixer: Open-Source 3D Editor That Builds Models From Text Using Local LLMs
A developer has released Mixer, a free open-source 3D editor that lets users generate and edit 3D models by typing text prompts instead of navigating traditional design software. The tool runs on local large language models, supporting Ollama, llama-server, and LM Studio, making it accessible without reliance on cloud-based AI services. Under the hood, Mixer uses GLB/GLTF format — a JSON-based structure — which the LLM writes as markup that the app then renders into a 3D model. Users can upload reference objects or existing models to guide generation, with options to simplify geometry when working within tight model context limits. Available for Windows, Linux, and macOS under an MIT license, the application is approximately 567 MB due to a bundled llama-server, with source code published on GitHub.
0
ProgrammingDEV Community ·
How to normalize job listings from Greenhouse, Lever, and Ashby using Python
Greenhouse, Lever, and Ashby each offer public, authentication-free JSON APIs for their job boards, but all three structure job data differently, making it difficult to consolidate listings into a single dataset. A Python script using only the standard library can map fields such as title, location, apply link, workplace type, department, and salary into a unified 13-key flat format. Each platform has notable quirks: Greenhouse omits department names from its list endpoint, Lever returns full descriptions in a bare array and has a separate EU instance, and Ashby's compensation components are unordered and may contain conflicting remote-work signals. The script handles missing fields by defaulting to None and exports results as both JSON and CSV. The approach was validated with real board data pulled on 27 September 2026 UTC, including a 321-posting Lever board totalling over 6 MB.
0
ProgrammingDEV Community ·
Developer builds open registry to catalog and compare System One AI models
A developer and founder of systemonemodels.tech has launched a public registry cataloging all known System One model families, a category of AI models that answer typed, structured questions in a single forward pass without generating text. The registry emerged after more than fifteen such model families appeared within ten days of TypeSafe AI coining the category with its Jev model. Each listing includes details on maker, size, context window, licence, pricing, and reported accuracy figures. The founder notes that published accuracy numbers are not yet comparable, as each model is benchmarked on different test suites, and meaningful evaluation requires testing on one's own labelled data. The registry is free to browse and publish to, and includes a CLI tool for local fine-tuning on custom datasets.
0
ProgrammingDEV Community ·
Authors' brief claims OpenAI downloaded millions of pirated books to train GPT-3
A summary-judgment brief filed September 17 by class-action plaintiffs alleges OpenAI downloaded around 117,500 books from piracy site Library Genesis in 2018, later torrenting roughly 35 TB of data matching its full 4.6-million-book collection. Internal Slack messages, depositions, and a document Sam Altman shared with Bill Gates in April 2019 are cited as evidence, with Altman's note explicitly referencing 'another ~11B words from Library Genesis.' The GPT-3 paper publicly labeled these datasets 'Books1' and 'Books2,' and in 2022 OpenAI deleted them via a Slack channel initially called #excise-libgen. The lawsuit, Alter v. OpenAI and Microsoft, is filed in Manhattan federal court and includes authors such as George R.R. Martin, John Grisham, and Jodi Picoult, covering 194 of their works. OpenAI and Microsoft deny wrongdoing, citing fair use; no ruling has been issued and a hearing is expected in early 2027.
0
ProgrammingHacker News ·
Insufficient content to generate a reliable headline
The source article could not be processed because no substantive text was provided beyond a URL and metadata. Only the original headline and a Hacker News link were available. Without verifiable facts from the article body, producing an accurate and neutral summary is not possible. Please provide the full article text for proper editing.
0
ProgrammingDEV Community ·
How a Gamedev CTO Built KPIs for Six Teams Without Triggering Goodhart's Law
A CTO managing six gamedev teams needed a way to monitor team health and delegate responsibility without micromanaging daily operations. The core challenge was designing KPIs that measured genuine outcomes rather than numbers that leads could game, a pitfall described by economist Charles Goodhart's decades-old principle. A concrete trigger was that stability issues like crashes and ANRs were going unaddressed because no one owned them — features had deadlines and owners, but stability metrics had neither. Rather than slicing the CTO's existing KPIs into shares distributed downward, the approach focused first on defining what each lead was actually responsible for within the broader direction. The resulting framework aimed to give leads clarity on expectations and genuine ownership of their zone, rather than metrics they only partly influenced and could defend against with reports.
0
ProgrammingHacker News ·
10 Design Patterns That Reveal a Low-Effort AI-Generated UI
A blog post titled '10 Tells of a Slop UI' has gained traction on Hacker News, accumulating 60 points and 36 comments. The article identifies common visual and structural patterns that signal a poorly crafted or AI-generated user interface. These so-called 'slop' indicators help designers and developers distinguish lazy or automated design work from thoughtful, intentional UI. The post sparked notable discussion in the developer community, suggesting the topic resonates amid the growing use of AI tools in design. The full article is hosted on the author's personal blog at hereticpleb.vercel.app.
0
ProgrammingDEV Community ·
Chat template, not model weights, triggers 'I'm just an AI' disclaimers, study finds
A paper published on August 9, 2026, by researcher Jędrzej Maczan found that the 'I'm just an AI' disclaimer in large language models is triggered by the chat template format, not by the model's underlying weights. Testing eight open-source instruction-tuned models of up to 9 billion parameters, the study showed that wrapping prompts in chat template tokens consistently produced self-distancing disclaimers, while plain-text versions of the same prompts elicited first-person experiential language. In three of the models, the researcher identified a specific internal activation direction that controls this behavior, and manipulating it could switch the disclaimer on or off regardless of whether the chat template was present. The findings suggest that what a model says about itself is an artifact of deployment formatting rather than a stable reflection of its trained properties. The work was accepted at two venues: a workshop at COLM 2026 and KONVENS 2026 Eval4SD.
0
ProgrammingHacker News ·
Fragment of world's oldest known peace treaty unearthed in Turkey
Archaeologists have discovered a fragment believed to belong to the world's oldest known peace treaty at a site in Turkey. The artifact offers rare physical evidence of ancient diplomatic relations, potentially dating back thousands of years. The find is linked to ancient Egyptian history, suggesting cross-civilizational negotiations took place in the region. Researchers described the discovery as a significant trace of peace from antiquity. The fragment adds new archaeological context to our understanding of early international agreements.
0
ProgrammingDEV Community ·
Engineers Deploy Qwen Flash-Next NVFP4 on vLLM After Fixing Loader and Memory Issues
A technical team successfully deployed the Qwen Flash-Next NVFP4 language model using the vLLM inference framework after resolving multiple compatibility and memory challenges. The 186.4 GB checkpoint required dedicated storage provisioning and a multi-worker download process, with a single PLE shard alone accounting for roughly 102.4 GB. Engineers had to patch the model loader to handle the checkpoint's tensor layout and apply B12x backend fixes before stable inference was possible. The final configuration supported a 131,072-token context window with a 16-sequence concurrency limit, successfully completing three rounds of 16 simultaneous requests without a container restart. The team acknowledged that reaching this result required scaling back initial context and concurrency targets, and credited contributor windowsxp811203 for detailed documentation of the integration work.