SC rejects Anita Advani’s plea over Rajesh Khanna

Jio Q2 Preview: First results after listing tomorrow

Chromium PDF output turned a CFF OpenType font into Type 3 glyphs at every fsType setting tested

A developer tested how Chromium 149 embeds fonts when printing to PDF, using one CFF-outline OpenType font and two TrueType fonts, each modified to six different fsType values. The TrueType fonts embedded normally or as Type 3 depending on the flag, with value 256 forcing full embedding. The CFF-based Noto Serif SC was written as Type 3 outlines at every fsType value tested, so the flag was not the cause in this case. The author suggests Chromium may lack an embedding path for CFF outlines, but notes this is an untested guess limited to one font, one Chromium version and one Mac. The author also ran the PDFs through a browser-based PDF-to-HTML converter, which turned the TrueType subsets into web fonts and the Noto Serif SC text into vector paths with a transparent text layer.

Turkish Trader Shares Eight-Step Finance Automation and AI Toolkit Lessons

A Turkish developer known as Serdar describes the Python libraries he has relied on for years to track and analyze Borsa Istanbul data. He explains how pandas handled large historical datasets, including memory problems he solved by selecting columns, optimizing data types, and reading files in chunks. The article also covers yfinance for fetching stock prices, along with its pitfalls such as delayed data, rate limits, and empty responses on non-trading days. Finally, it notes that technical indicator libraries save time but can mislead users who do not understand the underlying calculations, as he learned from a losing trade.

Preventing Telegram Webhook Retries and Spoofed Requests in Laravel with Queued Jobs

A technical guide explains that Laravel-based Telegram bots can fail when webhook handlers run long operations inline, since Telegram treats timeouts as errors and retries the same update repeatedly. This can cause duplicate orders or payment prompts if processing is not idempotent. The article also warns that an unauthenticated webhook endpoint lets attackers send forged payloads. To counter this, it recommends validating Telegram's secret token header with hash_equals() in custom middleware, which returns a 403 for invalid requests. It further advises returning an immediate HTTP 200 and delegating processing to a queued job guarded by atomic idempotency checks.

Korea Holds 26% of Hugging Face Benchmark #1 Spots from 4% of Entries

As of 29 October 2026, Hugging Face's Official Benchmarks Map listed 53 official benchmarks, 46 of which have a leaderboard. Grouping each leaderboard's top-ranked entry by publisher country, China holds 24 of the #1 spots, Korea 12, and the USA 3. Korea accounts for only about 4% of all entries but 26% of top spots, and all 12 of its top spots come from a single organization, FINAL-Bench (VIDRAFT). The analysis notes that board sizes vary widely and that many margins are small, while all scores are self-reported by publishers rather than independently verified. The post's author disclosed that their firm is a technology partner of VIDRAFT, whose FINAL-Bench appears in the rankings.

VIDRAFT's 325M-Parameter Model Pretrained on an Inference NPU Matches GPU Loss

VIDRAFT has scaled its earlier 46M-parameter model to VKORE-325M-NPU-Base, a 325.6M-parameter language model trained from random initialization on Rebellions' ATOM-Max inference NPU. A twin model trained on an NVIDIA B200 under identical settings served as the reference. On 1,024 windows of unseen FineWeb-Edu text, the NPU model scored a loss of 0.7993 against 0.7994 for the GPU model. The run covered 5 billion tokens over 76,292 steps, about four days, and finished without divergence, with the training curves overlapping throughout. The researchers credit VIDRAFT's own software stack for making training on the chip possible, and both models' weights and the evaluation script have been made public.

VIDRAFT's NL2SQL Engine Scores 83.80 on BIRD Mini-Dev, Surpassing Current Top Entry

VIDRAFT's text-to-SQL system, Darwin-180B-NL2SQL, reached 83.80% execution accuracy on BIRD Mini-Dev, a compact official leaderboard for the widely cited BIRD benchmark, edging past the listed top score of 81.00. The company says it used no gold SQL and no fine-tuning, and it has published its predictions and scorer while submitting the result to the BIRD team for listing. A prediction counts as correct only when its query returns exactly the same rows as the reference query, and human data engineers score roughly 93 on the same benchmark. VIDRAFT also released a public demo in Korean and English that converts plain-language questions into verified SQL, tables, charts, and reports, with safeguards such as read-only access and query limits. The company notes that the engine itself is not being released and that it is preparing a submission to the main BIRD test set, which is evaluated on a hidden dataset.

Vaultwarden admin route exposure: why the admin page needs separate protection

Vaultwarden is a self-hosted password manager whose separate admin page can change server settings, inspect users, and manage the deployment. The article argues that an exposed admin route is a far more valuable attack path than the normal vault login and should be reviewed separately. Internet scan fingerprints suggest a small set of candidate deployments, though a broader title-based search returns many more pages that are not confirmed as exposed admin interfaces. The author recommends checking whether the admin page is enabled, how its token is generated, stored, and rotated, and restricting it with independent network or proxy authentication rules. Backups, databases, and attachments should also be inventoried, since they may hold sensitive data.

Saying 'Hi' to an Internal AI Agent Consumed 177,000 Tokens Due to Tool Definitions

Bold.org's internal company agent, built on OpenCode, saw token usage climb faster than its session count could explain. A fresh conversation containing only the word "hi" was recorded at 177,444 input tokens, which ruled out long conversations as the main cause. Analysis with a token-tracking tool pointed to the skill catalog and tool definitions, which the tool's inventory did not fully capture. Investigators traced the bulk of the overhead to full MCP tool definitions, particularly a Meta Ads integration with 97 tools whose descriptions and schemas were sent with every request. The case shows how integration documentation loaded into an agent's context can dominate costs even when unused.

PDF-to-HTML font licensing: what the fsType flag does and does not reveal

A developer building a PDF-to-HTML converter considered gating font embedding on the OpenType fsType license flag. Testing fonts with edited fsType values showed that Chromium printing removes restricted fonts upstream, leaving only TrueType subsets, Type 3 outlines, or CFF fonts with no flag to read. On a sample brochure, one of five fonts had no OS/2 table and thus no fsType value. Converting that brochure with ImgIng produced inlined font subsets for TrueType fonts, vector paths for Type 3 glyphs, and fallback to a generic sans-serif for the Type 1 font. The author cautions that results come from a single machine, browser build and sample set, and other PDF producers were not tested.

Guide to Building a RAG Pipeline in C# with ASP.NET Core, Qdrant and Ollama

This DEV Community tutorial walks developers through building retrieval-augmented generation in C# using ASP.NET Core, the Qdrant vector database, and a local model served through Ollama. The pipeline splits documents into page- and heading-based chunks, embeds them, and stores them in Qdrant with access metadata. At answer time, it retrieves relevant passages, applies access filters, and streams responses that cite their sources. The author reports results from a 59-question evaluation on a small fictional support knowledge base, with full code available on GitHub under the MIT license. The article cautions that these figures reflect a small, clean corpus and should be measured against each reader's own documents.

NecroShell: A Haunted Retro Terminal Powered by On-Device Open-Source AI

NecroShell is an interactive retro-style terminal emulator that imagines a CRT shell possessed by an ancient machine spirit. It blends 1980s computing aesthetics, Halloween horror, and Día de los Muertos traditions, and was built for the Hacktoberfest Open-Source AI Challenge's "Ghost in the Machine" theme. The project runs entirely offline using open-weight models served locally, with no cloud dependencies, telemetry, or network requests. Its creators argue that local open models offer privacy, zero recurring cost, creative freedom, and long-term availability compared with closed proprietary APIs. The project was developed with the help of an agentic AI coding tool.

Choosing between AI video effect templates and text-to-video prompts

Creators generating short AI videos typically begin either with a pre-built effect template or with a blank text prompt. Templates fix the motion, style and pacing once you supply an input such as a photo, offering speed and consistency but little room to customize. Text-to-video prompts allow original scenes but require the user to handle creative decisions and often need several attempts to get consistent results. The article suggests templates suit specific photos, repeated looks, or tight deadlines, while prompts suit exploration and unique concepts. It also advises naming one subject and action, specifying camera behavior, and changing one variable at a time when prompting. Outputs from either route can contain artifacts and should be reviewed, with attention to consent when real people appear.

Diwali travel alert for Delhi railway passengers

Sam Altman Says AI Boom Offers Young Professionals a Rare Chance to Build

OpenAI CEO Sam Altman has argued that the rapid growth of artificial intelligence is opening a rare window for young professionals to create something new. The remarks reflect his view that the current AI wave is more than a passing trend for those entering the workforce. The source article's headline also refers to his "chronic disappointment" in MBA students, though the supplied text does not expand on that point. Because only a brief excerpt was provided, the full context of Altman's comments is limited in this summary.

India to gain a big edge after US trade deal: Goyal

Sanrio Apologises and Cuts Ties With Japanese Spy Drama After China Backlash

Hello Kitty maker Sanrio has issued an apology following criticism of its merchandise line tied to the Japanese TV series Vivant. The backlash, which came from China, led the company to end its partnership with the spy drama. The article notes the controversy centred on the merchandise collaboration rather than the series itself. Further details about the specific objections raised were not included in the source text.

Cuba to Remain on Daylight Saving Time Until November 2027

Cuba has decided to skip its scheduled clock change on November 1, 2026, keeping daylight saving time in effect. The country will therefore stay on DST until the next scheduled change in November 2027. The announcement was reported by timeanddate.com, a site that tracks time zone and daylight saving changes. The source material provided is limited to the article link and basic discussion metrics, so details about the official reasoning are not included here.

New York Passes Law Restricting AI Chatbot Use for Minors

New York has enacted legislation that places limits on how AI chatbots can be used by minors. Under the law, chatbots are barred from producing outputs that could lead a user to believe the bot has a personal relationship with them, a professional relationship, or an authority-figure role. The provision aims to prevent young users from forming misleading impressions about the nature of AI systems. The source article provides only limited detail on the law's full scope, so further specifics such as enforcement mechanisms and effective dates are not covered here.