SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World

Programming

LatestTopDiscussed
0
ProgrammingDEV Community ·

Supabase to stop exposing new public tables via Data API by default from October 30

Starting October 30, 2026, Supabase will apply a new default to existing projects, meaning newly created tables in the public schema will not be reachable through the Data API until access is explicitly granted. New projects have followed this behavior since May 30. Existing tables will keep their current permissions, and row-level security remains a separate requirement alongside grants. Direct Postgres connections and other schemas are unaffected by the change. The article warns that quick fixes to the resulting 'permission denied' errors, such as broad bulk grants or copying the suggested hint without review, can reopen tables to unintended access.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Higgsfield AI Video Review: Free Text-Prompt Tests Yield No Clips, Kyncept Succeeds

A reviewer tested Higgsfield AI's video generator with a single text prompt for a glass-strawberry ASMR clip on two new free accounts and produced no videos. Three attempts failed and refunds were issued, while the platform presented several upsell prompts during the session. The same prompt run on Kyncept, using a paid Pro plan, returned an 8-second clip with sound on the first attempt in about 98 seconds. The reviewer scored Higgsfield lower on free-tier usability, sound access, failure handling, balance display, and upsell pressure, citing Veo 3.1's paywall and a credit-balance warning shown while credits were full. The review was conducted independently on the reviewer's own accounts, without vendor-provided access.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Developer's Portfolio Adds Daily Puzzle Arcade with Server-Verified Race Times

A developer has added a daily arcade to their portfolio site featuring four puzzles: Minesweeper, a 6x6 Sudoku, Quick Math, and Code Review. Every visitor receives the same puzzles each day, and the developer's own completion times serve as the benchmark to beat. To prevent cheating, the server issues a signed token at the start and replays each player's moves to measure elapsed time itself rather than trusting the browser. The developer is requesting feedback and player times from the community.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Tripo vs SupaVoxel: Image-to-3D Test Finds 18x Credit Gap, Speed Results Mixed

An independent reviewer tested Tripo AI and SupaVoxel, both image-to-3D tools, using the same photo of a seated reindeer figure on default settings. Tripo deducted 55 credits for the generation while SupaVoxel deducted only 3, roughly 18 times the cost. Tripo finished faster on this subject, at 157 seconds versus 258 seconds, but SupaVoxel was quicker on seven of the ten subjects tested. The reviewer also reported that Tripo's output mesh was not watertight, had inconsistent winding, and arrived in two pieces, and that Tripo charged credits at submission rather than completion. The reviewer concluded that SupaVoxel better suits the task of turning a single photo into a printable figurine.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

DeepSeek's Flash 1017 Model Claims 1M-Token Context on a Single GPU

DeepSeek released its v4.1-Flash 1017 language model on September 17, 2026, drawing skepticism over claims it can handle a million-token context. The model is a 284-billion-parameter sparse design that activates about 13 billion parameters per token, which keeps hardware needs modest. Its key innovation is a hybrid sparse and quantized attention method that compresses the memory cache by roughly 85 percent. The article presents a hypothetical law firm example showing the model running on one A100 GPU with data kept on-premises. Benchmark results cited include a 71.2 percent MMLU score, which the article describes as comparable to GPT-4-Turbo.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Engram: a shared graph memory lets multiple AI coding tools share one memory

The author uses several AI coding assistants, but each forgets context from the others, so decisions made in one tool are lost when switching to another. To fix this, they built Engram for a FalkorDB hackathon, a single graph-based memory that every tool on their machine reads from and writes to. The graph approach handles reversed decisions cleanly, since a newer memory can supersede an older one through an explicit edge instead of both appearing in search results. Graph traversal also enables provenance tracing and improves recall by linking related entities, though the author ran into issues with FalkorDB's result format and full-text query parsing.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Nexus Studio separates flaky tests from real bugs by logging every retry

Verdict System LLC's Nexus Studio tells flaky tests apart from genuine defects by recording each retry attempt. A test that fails and then passes on retry is labeled FLAKY, while a failure on a single attempt with no environmental cause is labeled BREACHED. Failures that are themselves inconsistent across attempts are marked BREACHED_FLAKY. The engine rules out harness errors, crashes, blocks, unreachable hosts and timeouts before classifying any failure as a finding. The team says this avoids the earlier problem of treating broken application behavior as mere test noise.

0 comments Read more at DEV Community
0
ProgrammingHacker News ·

Ad-tech firm allegedly used as delivery channel for mercenary spyware

A piece published on Pluralistic examines claims that an advertising technology company served as a distribution system for mercenary spyware. The source material available is limited to a link and metadata from a Hacker News post, with only four points and no comments at the time of capture. The full article text was not provided, so specific names, dates, and technical details of the alleged operation cannot be confirmed here. The story's broader theme concerns how commercial surveillance tools may reach targets through legitimate-seeming advertising infrastructure. Readers are encouraged to consult the original article for the complete reporting and any supporting evidence.

0 comments Read more at Hacker News
0
ProgrammingDEV Community ·

Building PDF printing and crawler fixes for a Next.js kids' coloring site

A developer running Coloryla, a free bilingual coloring pages site for children, describes the technical problems behind its simple print-and-PDF feature. Rather than use heavy libraries, they wrote a small PDF generator that embeds PNG image data directly, avoiding re-encoding and enabling aggressive caching. Browser print behavior proved inconsistent, with Safari and Chromium handling page layout differently and requiring separate rules. A share-image route, which initially fetched fonts at request time, caused slow responses and timeouts until the font was bundled locally and Redis timeouts were added. Search Console data also showed that most Googlebot requests went to images, scripts, and analytics JSON rather than HTML pages.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

SkillLens TutorKit: Open-Source AI Tool Builds Worksheets for Independent Tutors

SkillLens TutorKit is an open-source, AI-powered platform that generates worksheets and study packs for independent tutors, private teachers, and small coaching centres. It aims to cut the 5 to 12 hours educators typically spend each week drafting practice problems, checking answer keys, and formatting materials. Tutors upload chapter notes, textbook excerpts, or syllabus outlines, and the system organizes the subject and produces multiple-choice questions, worked exercises, and flashcards. An adaptive Socratic assistant offers scaffolded hints, and finished sets can be edited and exported as printable worksheets. The project is released under the permissive MIT License, with a live demo and code available on GitHub.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

FlapJax Agent Wallet Reportedly Made Three FLAPJAX Payouts in Mid-October on BNB Chain

FlapJax (FLAPJAX) says it is registered as ERC-8004 agent 364913 on BNB Smart Chain, with its treasury serving as both owner and agent wallet. Onchain reads at block 128186441 reportedly confirm that the registry mint was sent from and paid for by that treasury address. The agent's registration file references the FLAPJAX token and the same wallet. Three FLAPJAX transfers from the wallet to a plain wallet address are reported between 12 and 16 October 2026, each confirmed in a separate block and transaction. The analysis was published on DEV Community and is presented as a reproducible, independently checkable account.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

JPG, PNG, or WebP in 2027: Choosing the Right Image Format for the Web

Images are often the heaviest elements on a web page, so selecting the wrong file format can noticeably slow a site down. The article recommends WebP or AVIF for photos with JPG as a fallback, PNG or lossless WebP for screenshots and logos with sharp text, and SVG for shape-based icons and illustrations. It compares JPG, PNG, WebP, AVIF, GIF, and SVG on compression, transparency, and animation support, noting that exact file savings depend on the image and encoder settings. Practical guidance includes using the picture element with multiple sources, setting image dimensions to avoid layout shift, and applying lazy loading selectively. The author also shows how Next.js can automatically convert images to AVIF and WebP and mark above-the-fold images for priority loading.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

ListingCraft AI: Open-Source Tool Generates E-Commerce Product Listings Locally

A developer has built ListingCraft AI, an open-source tool that generates product listings for online sellers. The application creates titles, descriptions, selling points, keywords, and tags tailored to platforms such as Amazon, Etsy, and Shopify. It runs a Qwen2.5 3B model locally through Ollama, so it requires no paid cloud AI API. The project was submitted to the Hacktoberfest Open-Source AI Challenge and includes a listing quality score, editable outputs, and a saved listings library stored locally in the browser.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Why ls Lists clip_10 Before clip_1: Locale Changes File Order

Listing twelve numbered clips in one folder produced three different orders depending on the tool and locale used. Under the C locale, ls compares names as plain bytes, so clip_10 falls between clip_1 and clip_2. Under en_US.UTF-8, punctuation is ignored and digits sort ahead of letters, which places clip_10 first. Python's os.listdir and glob.glob returned files in neither order, and sorted() matched the C locale. Using ls -v, sort -V, or zero-padded names made every tool produce the intended 1-to-12 sequence.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Developer Open-Sources Tool That Generates Standard MCP Agent Plugins in TypeScript

A developer has built an open-source Agent Orchestrator that turns a natural language prompt into a complete Agent Plugin built on the Model Context Protocol (MCP) and Agent Plugins 1.0 standards. The generated plugins combine a plugin.json manifest, an mcp.json tool configuration, and a SKILL.md file of workflow instructions. The tool includes a web interface that selects appropriate MCP servers, estimates API token costs before execution, and wraps commands for Windows compatibility. The author says the approach lets clients run local AI assistants that handle files and documents without sending sensitive data to the cloud. Built-in safeguards include a budget limit that halts agents exceeding $0.50 in API costs and required user confirmation for destructive file operations.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Ventute: a business simulator built to replay every run bit for bit

Ventute is a business simulation platform where users run a company by making decisions while the engine projects what happens next. Its state is never stored directly; instead, every action is recorded as an event and the state is rebuilt by replaying the event log, so identical inputs yield identical outputs. The developer, who built most of the platform alone over the past year, had to control wall-clock time, randomness, iteration order, and floating-point behavior to achieve exact replay. For Monte Carlo forecasting, the hot loop was moved to Rust and exposed to Python via PyO3, enabling true parallel execution while keeping results deterministic. Three reusable components, including the replay kernel, were released as open source on GitHub.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

ZVYA AI: Local-First Assistant Turns Customer Chats into Structured Orders

A developer has built ZVYA AI, a local-first business assistant aimed at small sellers and home businesses that often take orders through informal chat messages. The app uses a local Gemma 3 model via Ollama to extract product details, quantities, prices and delivery requests from pasted messages. Sellers review and edit the extracted fields before saving the order, since AI can misread prices, dates or quantities. The tool is built with Python, Streamlit and SQLite, and it also tracks order status, manually recorded payments, outstanding balances and CSV exports without requiring a hosted AI API key. Planned future improvements include customer follow-up drafts, delivery date reminders and more robust date parsing.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Python 3.15 Speeds Up Most Encodings but Slows Base16 Versus 3.14

Python 3.15 delivers large speedups for Base32, Ascii85, Base85, and Base64 encoding, attributed to rewriting those implementations in C. Benchmarks run with pyperformance on an AMD Ryzen 9 7900 desktop under Windows 11 show Base32 completing roughly 135 times faster than in Python 3.14. Ascii85 and Base85 also improved dramatically, with reported speedups of about 69x and 63x. Base16 performance regressed by roughly 22 percent, since that encoding was not rewritten in C and likely still relies on higher-level Python logic. Developers are advised to benchmark their own workloads, especially Base16-heavy applications, and to test across platforms before upgrading.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

How to Write AI Video Prompts That Produce Consistent Results

A guide argues that poor AI video output usually stems from vague prompts rather than weak models, since models fill any gaps with random guesses. It recommends structuring prompts like a director's shot list, answering seven questions covering framing, subject, action, setting, camera movement, lighting and style, and audio. Key advice includes limiting each clip to one action and one camera move, using concrete verbs, and specifying timing. The piece also distinguishes text-to-video prompts, which must describe the full scene, from image-to-video prompts, which should describe only the motion. Finally, it notes that Veo 3.1, Kling, and Runway each have distinct strengths and that changing one variable at a time helps refine results.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Why a production check's assignment matters more than the executor's skill

A PHP/Go engineer migrating a live monolith into Go services argues that the quality of production checks depends on how complete the assignment is. In a sweep of 276 configurations, 184 returned empty results and were initially flagged as defects. Adding one condition, that some schedule types produce no tasks of this kind, reclassified those empties as expected behaviour. He argues that facts about how the domain works only enter a check if they are written into the assignment beforehand. He now specifies the single entry point, authentication and credential sourcing, exact request shapes, and the one value that counts as the fact for each link in the chain.

0 comments Read more at DEV Community
← NewerPage 378 of 3103Older →