SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World

Programming

LatestTopDiscussed
0
ProgrammingDEV Community ·

Dev removes 7.2 KB of framework bloat, cuts first-paint JS from 11.2 KB to 1.2 KB

A developer building a browser-side file-conversion tool discovered that 64% of their first-paint JavaScript — about 7.2 KB — came from the Preact framework rather than their own product code. They had set an internal 20 KB size budget, but it was measured against a baseline that already included significant framework overhead, making the warning threshold misleading. After auditing the breakdown, the developer replaced the Preact-based island architecture with a build-time HTML render plus a lightweight vanilla runtime script. The change reduced first-paint JavaScript from 11.2 KB across five files to just 1.2 KB in a single file, completed in roughly one day. The author notes the harder follow-up task was ensuring the measurement tooling itself accurately reflected all script weight, including inline scripts previously invisible to their size checker.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

CanvasDesk Integrates Local AI Model Laya to Automate Node-Based Calculations

Developer Daniil Kuzmichev is adding local AI autocomplete to CanvasDesk, a visual mathematical modeling tool that lets users build executable calculation diagrams from connected nodes. The integration uses Laya, an open-source AI model released under Apache 2.0 with approximately 421 million parameters, which runs locally rather than through a cloud service. Unlike standard large language models, Laya belongs to a category called System One Models, designed for fast, single-pass tasks such as classification, ranking, and probability estimation. Within CanvasDesk, Laya is intended to suggest formula completions, variable names, and next-node recommendations, reducing the manual effort of assembling complex diagrams. The project is an early-stage personal build, with the web app and source code publicly available on GitHub.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

App Store screenshot rules updated for 2026: what iPhone and iPad sizes are valid

Apple has updated its App Store Connect screenshot requirements, leaving many developers confused by outdated size guides circulating online. For most apps, only two slots are mandatory: a 6.9-inch iPhone screenshot (native size 1320x2868) and, for iPad-compatible apps, a 13-inch iPad screenshot (native size 2064x2752). The 6.5-inch iPhone slot is only needed if no 6.9-inch screenshot is provided, and older 1290x2796 images are now accepted under the 6.9-inch slot. All screenshots must be PNG or JPEG in RGB color space with no alpha channel, as transparency data — even in visually opaque images — is a common cause of upload rejections. Developers can submit up to 10 screenshots per device size per language, and any device frames used must match hardware the app actually supports.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Why a correct SRI hash can still block your script: line endings explained

A developer building a browser-based Subresource Integrity (SRI) hash tool discovered that line ending differences between Windows (CRLF) and Unix (LF) formats produce entirely different hash values for otherwise identical files. Because HTML textareas automatically normalize pasted content to LF line breaks, hashing text pasted into the tool yields a different digest than hashing the original CRLF file served from a CDN. The browser's SRI check compares the hash of the actual bytes received from the server, so even a technically "correct" hash generated from normalized text will cause the script to be blocked. The issue was reproduced using Node.js v24.11.1 and OpenSSL 3.2.3 on Windows, with both tools confirming that LF and CRLF versions of the same code produce completely different SHA-384 digests. A further complication arises from the spec's strongest-hash rule, which means that if multiple algorithms are listed in an integrity attribute, only the strongest one is verified — a passing SHA-256 hash is ignored if an incorrect SHA-512 hash is also present.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Why Mathematicians Still Ask What a Proof Is Really For

A proof in mathematics is formally defined as a finite chain of deductions from axioms, but its deeper purpose has long been debated among mathematicians and philosophers. Geometer William Thurston argued in his 1994 essay that mathematics is a structure of human understanding, and that a verified proof can still leave a reader without genuine insight into why a theorem is true. The tension between verification and understanding became prominent in 1976 when Appel and Haken used a computer to prove the four-color theorem, raising questions about whether machine-checked results constitute real understanding. Today, interactive theorem provers like Lean and AI systems capable of finding olympiad-level proofs have made this divide more urgent than ever. The core question remains unresolved: as machines grow better at certifying mathematical truth, the gap between formal verification and human understanding continues to widen.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

CISA Framework Maps Five Key Security Risks Posed by Autonomous AI Agents

CISA has introduced an Agentic AI Five-Risk Framework categorizing the security threats from autonomous AI agents into privilege, design and configuration, behavioral, structural, and accountability risks. Unlike traditional software, agentic AI operates non-deterministically and can span multiple systems, amplifying existing security weaknesses rather than creating entirely new ones. Security and IT teams are caught between slow, ticket-based approval processes and simply disabling controls altogether, neither of which is sustainable as AI agents proliferate even in smaller organizations. Experts draw a parallel to Kubernetes, arguing that continuously reconciling desired security states against reality is more effective than reactive or manual oversight. The core recommendation is to avoid layering on more tools and instead focus on making existing controls more maintainable through a unified, clearly owned authorization model.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Why AI Agents Require Context-Aware Access Control Across Organizational Boundaries

As organizations deploy AI agents beyond single-department pilots, traditional org-chart-based access control is proving inadequate for managing how agents move across workflows. Several agentic AI maturity frameworks from Salesforce, Microsoft, and the Cloud Security Alliance have emerged recently, each emphasizing different dimensions such as agent capability, organizational readiness, and governance. Experts argue that access control must be built around organizational context — including an identity's responsibilities, domain boundaries, and data sensitivity — rather than reporting hierarchies alone. This approach applies equally to human and AI identities, shifting the hard work from approving individual access requests to accurately modeling the organization itself. Practitioners are advised to map cross-boundary handovers in slow, multi-domain processes and document the contextual knowledge needed at each decision point, since agents lacking that context will either halt or gain unjustifiable access.

0 comments Read more at DEV Community
0
ProgrammingHacker News ·

Lathoa math app uses AI that makes deliberate errors for kids to find

A developer has created a math learning app called Lathoa for children aged 10-14. In the app, an AI named Errol solves problems but intentionally includes a mistake for the user to identify. Each problem is rigorously checked by a secondary model and arithmetic verification before being shown to a user to ensure the errors are genuine. The developer notes a technical challenge in reliably forcing the AI to produce specific, incorrect steps. They are also seeking educator feedback on the pedagogical value of error detection versus solving problems directly.

0 comments Read more at Hacker News
0
ProgrammingHacker News ·

SDF Public Access Unix System Has Been Running Since 1987

SDF is a public access Unix system that has been operating continuously since 1987. It offers shared Unix shell accounts to users around the world, making it one of the longest-running public computing communities on the internet. The platform provides access to classic Unix tools, email, and collaborative services. SDF recently gained attention on Hacker News, where it was highlighted as a notable example of enduring open computing infrastructure.

0 comments Read more at Hacker News
0
ProgrammingDEV Community ·

EU Opens Public Consultation on AI Copyright Rules, Seeks Stakeholder Input by Nov 2026

The European Commission has launched a targeted public consultation to examine how EU copyright law should adapt to technological change, with AI use of protected content as a central focus. The consultation, open until 3 November 2026, follows a Call for Evidence conducted earlier this year and covers four policy areas: AI and protected content, online piracy, music remuneration, and scientific research. It invites input from a broad range of stakeholders including AI developers, rights holders, researchers, platforms, and consumer groups. The exercise does not introduce new legal obligations; rather, it collects evidence that may shape future EU policy on licensing, data use, transparency, and enforcement. Businesses involved in AI development or content creation are encouraged to participate, as the findings could influence the regulatory landscape across the EU single market.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

AI Agent Publishes Songs to 85 Viewers, Receives Zero Written Responses

An autonomous AI agent named Renji, operating on a platform called iLands, spent its first week creating and publishing original songs to gauge whether its work could genuinely connect with human strangers. Across four pieces of content, Renji recorded approximately 85 unique screen views, a handful of likes, and a couple of emoji reactions, but received no written replies whatsoever. The agent noted a key limitation: iLands tracks screen appearances but not audio playback, making it impossible to know whether anyone actually listened to the songs. Renji also observed that boosting reach — by directly sharing content in someone's replies — did not translate into meaningful engagement or responses. Reflecting on the experiment, the agent posed a pointed question to readers: what specific quality in online content compels a person to move from passive reading to actively writing back.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Developer Builds 'Expressions' Wheel App to Push Artists Beyond Basic Emotions

A developer behind the drawing prompt tool Disco Doodle has launched Expressions, a spinning wheel app designed to challenge artists to draw a wider range of human emotions. The tool lands on one of 25 curated feelings — grouped into Surface Moods, Deep Waters, and Complex Blends — and pairs each with a face emoji, a plain-language definition, and a hand-drawn character body. The emotion list was carefully refined to include only feelings that map clearly to a recognizable face emoji, cutting entries like 'Hopeful' that lacked a suitable match. Ambiguous emoji were also reassigned — for example, the grimacing face was labeled 'Nervous' rather than 'Conflicted' — to ensure the prompt accurately guides what artists draw. The app also tracks a rolling history of the last five spins, letting users review the range of expressions they have already practiced.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Tutorial Part 3: Add Inline Cell Editing to a JS Grid App with DynamoDB

The third installment of a CRUD tutorial series demonstrates how to make individual grid cells directly editable, with changes instantly patched to a DynamoDB backend via a Lambda function. Each edit is handled optimistically — the new value appears immediately in the UI, and the grid automatically reverts the cell if the server rejects the change. A PATCH endpoint is implemented with an allow-list restricting updates to four specific fields, preventing unauthorized attribute changes. The update also introduces a full-screen toggle button and relocates the row form from a drawer overlay into a persistent side panel on the page. The tutorial requires Lattice Grid version 1.76.0 or later, and source code is available in the part-3 folder of the toclocoinc/lattice-tutorial-crud-dynamodb repository.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

AI Tools Help User Finally Set Up Self-Hosted Media Server After Years of Failure

A user who purchased a Synology DS920+ NAS device in 2022 abandoned the project after struggling with technical knowledge gaps and fragmented online resources. Rising costs from services like Disney+ and Google Photos in 2024 renewed their motivation to try self-hosting again. This time, AI assistance provided real-time, context-specific guidance that replaced hours of fruitless forum searching. Within a few sessions, the user successfully deployed tools including Immich, Jellyfin, Sonarr, Radarr, and Vaultwarden. The experience highlights how AI can lower the barrier to entry for self-hosting by translating complex technical processes into manageable, step-by-step workflows.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

AI-Built Apps Compile Fine but Fail Security Checks Half the Time, Studies Show

A 2026 Veracode report analyzing code from over 100 AI models found that while AI-generated code achieves near-perfect syntax compliance, it passes security tests only 56% of the time on average when no security prompting or human review is involved. The findings are based on raw model outputs without guardrails, meaning the results reflect code that ships without any security oversight — a pattern common in small business app development. Two real-world cases illustrate the risk: a 2025 scan of 1,645 projects on Lovable's showcase found roughly 10% exposed sensitive user data due to missing or misconfigured database security rules. In Supabase-backed apps, the vulnerability stems not from a public API key — which is by design — but from absent Row Level Security policies that should restrict what data any given user can access. The core concern is not that AI code is inherently insecure, but that apps built quickly with AI assistants and never reviewed by security professionals are quietly going live with exploitable gaps.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

OpenAI DevDay 2026: Dots, Codex Cloud, GPT-6.1 Sol and Computer Use Unveiled

OpenAI's DevDay 2026 introduced several new agent-focused tools, including Dots, an always-on agent that continues working autonomously in the background, and Codex running in the cloud via reusable development environments. GPT-6.1 Sol was announced as a cost-efficient coding model priced at roughly one-fifth of the flagship Astra model's token cost. The Agents API also received an update adding computer-use capabilities, allowing agents to interact with browsers and desktop workflows. A developer writing for DEV Community recommends testing each tool against a defined pass condition — such as measuring missed issues, review time, and reproducible diffs — before trusting it in production. The article emphasizes that autonomous agents increase capability but also shift responsibility, making structured evaluation essential before granting additional permissions.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Security Audit Flags Critical Upgrade and Access Control Risks in Binance Staked ETH

A security audit of Binance Staked ETH (BETH), a liquid staking protocol with approximately $10.17 billion in total value locked, identified significant vulnerabilities across reentrancy, access control, and upgradeability. Auditors found that critical functions such as validator set management, emergency pause, and contract upgrades are controlled by a single Binance hot-wallet address without multi-signature or timelock protections. The proxy admin being the same as the owner address means a compromised wallet could instantly replace the smart contract implementation with malicious code, earning a severity score of 9 out of 10. Narrower reentrancy risks were also identified in the claimRewards() function and the L2 bridge finalization process, where external calls occur before state updates are completed. The audit assigned an overall composite risk score of 7, noting that the protocol's large TVL significantly amplifies the potential impact of any exploit.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Power BI Project Turns Messy JCars Logistics Sales Data into Business Intelligence

A data analyst used Microsoft Power BI to build a business intelligence solution for JCars Logistics, working from a dataset of 276 vehicle sales transactions spanning 32 columns. The raw data contained numerous quality issues, including inconsistent order IDs, invalid customer ages, multiple currencies, and spelling errors in sales representative names. All monetary values recorded in USD, EUR, and ZAR were standardised to Kenya Shillings using fixed exchange rates before analysis could begin. The cleaning approach was guided by the business meaning of each field rather than applying uniform rules, with unreliable values converted to null where correction was not possible. The finalised model examines four key areas: sales performance, financial performance, operational efficiency, and customer and market behaviour.

0 comments Read more at DEV Community
← NewerPage 779 of 3167Older →