SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World

Programming

LatestTopDiscussed
0
ProgrammingDEV Community ·

Same JSON extraction job costs $667 or $1,100 per 100,000 tasks, depending on validation rate

The price per token stays fixed, but the effective cost of a task depends on how often the model's output passes validation. An author calculates a support-ticket extraction at about $0.0066 per call on GPT-5.6, then divides by the share of valid answers to estimate the cost of a finished task. Higher temperature makes malformed JSON more likely, and each failed answer triggers a billed retry. Illustrative validation rates of 0.99 to 0.60 produce totals from roughly $667 to $1,100 for 100,000 tasks. The author notes these rates are assumptions rather than measurements and suggests logging failures by temperature to find one's own rate. Retries that reuse the same seed, temperature and prompt may simply repeat the failure, so they should change something and be capped.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Estudante de programação usa IA como amplificador, não piloto automático, em rotina de estudos

Um estudante de tecnologia relata que o mercado de software em 2026 se dividiu, com vagas de entrada em tarefas repetitivas sendo substituídas pela IA. Para não se tornar substituível, ele decidiu usar modelos de linguagem apenas para organizar e contextualizar seu aprendizado, mantendo o raciocínio nas próprias mãos. Ele registra aulas da Fatec à mão quando há desenhos na lousa e em Markdown no Obsidian quando são slides, com uma rotina automatizada que reorganiza as anotações e pesquisa dúvidas marcadas durante a aula. O método exige, porém, que ele releia cada nota com atenção, pois a IA não substitui a leitura ativa. A abordagem é apresentada como alternativa ao uso da IA para entregar código pronto nos estudos de programação.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Tech Job Market Splits in Two as Entry-Level Roles Shrink and Complex Engineering Demand Grows

A software engineer's analysis argues that the tech job market in 2026 has not collapsed but has divided into two distinct segments. Entry-level roles centered on repetitive coding tasks are being cut as AI handles them quickly, while demand rises for engineers skilled in complex systems, architecture, automation, and security. The author, a computer science student, describes a study workflow that uses AI to organize handwritten and digital notes and to retrieve documentation for open questions, while requiring the learner to do the reasoning and implementation personally. The approach is presented as a way to avoid becoming replaceable by treating AI as a support tool rather than a shortcut to finished code.

0 comments Read more at DEV Community
0
ProgrammingHacker News ·

MapQuest Adds Map Overlay Showing Locations of Flock Cameras

MapQuest has announced a feature that displays the locations of Flock cameras on its maps. The addition lets users see where these surveillance devices are installed in their neighborhoods. The report was published by Engadget and widely shared on Hacker News. The source material provided is limited to the headline and link, so details such as the rollout date and the company's stated reasons are not confirmed here.

0 comments Read more at Hacker News
0
ProgrammingDEV Community ·

Sitemap lastmod dates audited against git: all 69 match, but nothing enforces them

A developer found that their sitemap takes each page's last-modified date from a hand-typed field in a content registry rather than generating it. An audit comparing those 69 manual dates against git history showed every one agreeing with the last commit touching the page's file. The sitemap carries three distinct dates across 72 URLs, mostly reflecting a bulk content build-out and one copy change. The author notes the dates stay accurate mainly because the field sits next to the text it describes, which lowers the chance of it going stale. Their existing tests only check that the date looks valid, so an outdated date would still pass; a git-based check would catch it but is hard to run in CI because shallow checkouts lack history.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Checkmate Rogue: A Chess Roguelike Where Gemma-Powered Rivals Remember How You Play

Checkmate Rogue is an open-source, AI-powered chess roguelike built as a submission to a Hacktoberfest challenge. The game pairs Stockfish chess engines at varying difficulty with a behavior-tracking system that notes how players make decisions. Google's open-weight Gemma 2 2B model, run locally through Ollama, generates contextual taunts and tips based on those patterns. Players progress through a gothic, narrative-driven campaign with three hearts and strategic relics. The developer proposes a free-to-play model with optional paid expansions, though this remains untested as the project is still a prototype.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Why a 24-hour webhook dedupe key can silently drop events, and the 120-second fix

Munchable's Stripe and inbound email webhooks both risk duplicate side effects because providers deliver at-least-once. The team uses a Redis SET NX key with a TTL to claim each event, so retries return early. A 24-hour claim is the wrong choice: if the handler crashes midway, the key marks the work as done and the provider's retries are silently ignored, losing the event. The solution writes a short 120-second processing claim, extends it to 24 hours only on success, and deletes it on caught errors so retries proceed. The 120-second window is set to exceed the route's 30-second maximum duration, covering crashes where no code can run.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Running a Real C Compiler in the Browser Using WebAssembly

A developer built an online C compiler that keeps user code inside the browser instead of sending it to a server. The tool runs Clang 21 and LLD compiled to WebAssembly, loaded as an npm package and cached after the first use. Compilation and execution happen in a module Web Worker, with a minimal WASI host that provides stdin, stdout, stderr, clocks and randomness but no file system or network access. A timeout terminates the worker to stop runaway loops, and testing revealed differences such as 32-bit type sizes, NULL reads not crashing, and missing fortified libc checks. C++ compilation was initially slow, but a precompiled header of common standard headers brought typical compile times down to 1-3 seconds.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Python Decorators Explained: How Functions Wrap Other Functions

A beginner-focused guide breaks down Python decorators, a feature often seen as confusing at first. It shows how repeated logging code across functions can be replaced by a decorator that wraps each function. The article explains that the @ symbol is shorthand for passing a function into a decorator and reassigning the result. It also covers how *args and **kwargs let a decorator work with any function, and lists common uses such as logging, timing, caching, authentication, and retry logic.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Scanner Probed a Site for 400 Secret Files in One Minute; None Were Exposed

A single client sent about 1,180 requests to a domain in one minute, requesting roughly 400 sensitive file paths such as environment files, WordPress configs, and cloud credentials. The client identified itself as tlm-scanner/2.0, and the site returned no sensitive files, with only the home page and redirects or refusals for the rest. Shortly afterward, a slower sweep disguised with browser user agents targeted a subdomain and included a path tied to AI provider keys. The author argues that such scans are routine and that site owners should check their own logs, since successful probes often go unnoticed.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

AI Student Launches Personal Developer Portfolio Showcasing Projects and Interests

A student studying AI and Data Science has launched a personal developer portfolio to consolidate their work in one place. Rather than just listing languages and frameworks, the site is meant to reflect their learning journey and the projects they are building. The creator also used the project as a way to experiment with presenting technical work through a polished web experience. Their next focus is Generative AI, LLM-powered applications, and AI agents, with plans to turn ideas into practical projects. They have invited other developers working on AI engineering to connect and share their work.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

New proof settles A(4)=4 for a 2-group problem via a single-vector argument

A researcher's open notebook on Kakeya-type problems reports a result on finite 2-groups whose involutions are all central. The author applies a closer look at a single vector to sharpen a restriction lemma from an earlier post. The analysis establishes that the invariant A(4) equals 4, which settles the question for groups with a self-centralising normal subgroup of the form Z_4^a x Z_2^b. The argument relies on anisotropic linear systems and a polar form, and it also establishes M(4)=4. The author notes that part of the verification was computer-assisted and that the associated tests passed before publication.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Building Resilient TypeScript REST APIs with Node.js and PostgreSQL

A DEV Community post argues that backend slowdowns usually stem from inefficient queries, missing connection pooling, and weak TypeScript typing rather than hardware limits. It recommends moving beyond basic CRUD toward a production-grade architecture. The author highlights strict type safety, using TypeScript interfaces for request validation and database models to catch errors at compile time. The piece also advises analyzing query execution with EXPLAIN ANALYZE instead of adding indexes blindly. It closes by asking developers which stubborn bottleneck they have recently debugged in a Node.js REST API.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Next.js useSearchParams() Build Error: Why It Needs a Suspense Boundary

A developer adding a search box to a navbar hit a build failure in Next.js even though the code ran fine in development mode. The error arises because static pre-rendering cannot access the query string, so Next.js requires a Suspense boundary to limit client-side rendering to the affected component. Placing the component in a shared layout means one unguarded hook can break the build for every static route, which explains why the error names a page the developer never edited. Wrapping the search box in a Suspense boundary within the layout keeps the other pages static while only that component renders in the browser. For single-page needs, reading the searchParams prop directly is an alternative, though it makes that route dynamic.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Keeping a Daily Log for Your Google Play Closed Test

Developers often discover their closed test has failed only late in the process, when tester numbers have dropped below the required minimum. A simple daily tracker can flag problems early, giving time to recruit replacements before the 14-day window closes. Recommended columns include opted-in and active testers, joins and drops, build version, and feedback with fixes shipped. Setting a red warning when opted-in testers fall below 14 provides a buffer for calm recruitment. A real campaign example showed 18 testers joining, 11 finishing all 14 days, and Google granting production access afterward.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

moshcode 0.117.0 adds parallel AI agent work queues gated by verify commands

The moshcode 0.117.0 release introduces a 'work' command that runs many AI coding agents against tasks, where a task counts as done only when its verify command exits with code 0. Each worker runs headless in its own git worktree, and failed attempts are requeued with the verifier output or moved to an inbox after three tries. The author used 50 Claude workers to fix 139 locally failing tests, finding that 138 were PWA tests crashing on missing dependencies and one was a real failure caused by a test inheriting the host's PATH and HOME. All 19 resulting tasks passed on the first attempt, reducing local failures from 139 to zero. The author plans to extend the verifier beyond tests to typechecks, URL checks, and eventually Lean proofs.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

REA: An Open-Source Tool Lets Coding Agents Trace How Finished Apps Work

REA, short for Reverse Engineer Anything, is an MIT-licensed open-source MCP server that links coding agents such as Claude Code, Codex, Cursor and Gemini CLI to reverse-engineering tools. Users point the agent at a finished app and ask about a specific feature, after which it traces the underlying code. Documented targets include native binaries, JavaScript and Electron apps, .NET assemblies, Android APKs and websites, with each finding accompanied by its evidence and limitations. Setup is run through npm, and the project's documentation restricts use to lawful research, interoperability, software preservation and authorized examination. The project warns users to install only from official sources to avoid look-alike repositories.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Launch checklist: making a new product findable on Google and AI assistants

Once early friends have signed up, new users typically discover a product through Google and AI assistants, so findability needs deliberate technical work. The author, from the Serpair team, recommends checking whether a product's name ranks in search, verifying the site in Search Console, and testing how AI tools describe it. Advice includes pairing the product name with its category in page titles and headings, keeping the name identical across platforms, and adding WebSite and Organization structured data with accurate sameAs links. The article stresses that schema alone does not reliably increase AI citations, so visible text should carry the key facts. It also explains that AI assistants can only cite pages their crawlers can access, and outlines robots.txt rules for OpenAI, Perplexity, and Googlebot, plus CDN and firewall checks.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Controller drift or game deadzone? How to check raw stick values

A controller that drifts in one game but not another often reflects different software deadzones applied by each game rather than hardware failure. Competitive shooters with small deadzones expose stick looseness, while adventure games with larger defaults can hide it. To check, use a raw-value tester after pressing a button, since browsers only provide gamepad data after user interaction, and watch the resting stick reading. Values of 0.02 to 0.04 are normal noise, while persistent readings above about 0.08 are noticeable and above 0.15 are severe. A circle calibration test, where the stick is rotated two laps, can also reveal travel loss, with a roundness above 1.3 indicating a problem, and a dated screenshot of a high resting reading can support a warranty claim.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Zero-shot models vs XGBoost and LightGBM: benchmark across three UCI datasets

A developer benchmarked two pretrained zero-shot models, Mitra v2 and TabPFN-3.5-Fast, against XGBoost and LightGBM on three public UCI datasets. The tasks covered binary classification of shopping sessions, regression of hourly bike rentals, and multiclass classification of bean varieties. All models used the same 60/20/20 train, validation, and test split with a fixed seed, and each foundation model was tested with raw and engineered features. The classical models were given only a small hyperparameter search, which the author acknowledges could be improved with broader tuning. Results are reported as primary metrics per dataset, with the foundation models scoring close to or above the tree-based baselines in several cases.

0 comments Read more at DEV Community
← NewerPage 381 of 3103Older →