AI Weekly Malaysia

The author argues MCP (Model Context Protocol), released by Anthropic in November 2024 and later donated to the Linux Foundation's Agentic AI Foundation in 2025, is becoming obsolete because modern LLMs can now call APIs directly, write scripts, and compose services autonomously without a dedicated tool-calling protocol. They describe an 'MCP industrial complex' of monitoring tools and context-bloat workarounds (Composio, MintMCP, Pipedream) that exist to patch over limitations that better models are outgrowing.

Elon Musk claimed The Boring Company is working on a 'simple precursor Hyperloop' between Austin and San Antonio that would cut travel to under 30 minutes, replying to an AI-generated sci-fi video. The Boring Company account echoed enthusiasm, but most previously announced projects (Chicago, LA, NYC-DC) never materialized; the only operating system is in Las Vegas. The company recently raised a $3B round led by the UAE.

TechCrunch's Equity podcast discusses whether AI industry leaders genuinely support slowing down AI development, following Anthropic CEO Dario Amodei's 'pace the frontier' plan and Nvidia CEO Jensen Huang's dismissal of AI backlash as a hoax. Panelists note the plan lacks specifics and that current U.S. federal inaction plus weak consumer choice mean market forces alone won't enforce safety.

Vocci has launched a titanium-coated smart ring (under 6 grams) that records meetings via double-tap, claiming 8 hours of recording per charge with a charging case providing three additional charges. The reviewer found transcription accuracy good even in loud cafes over hour-long conversations, but the companion app confusing, with AI-generated insights often longer than the short notes they summarized. The app organizes recordings into Chat (transcript), Highlights (button-captured snippets), and Notes (AI summary).

ScrollEd, a bootstrapped Palo Alto startup founded by Stanford AI student Utsav Gupta and UPenn CS student Rebecca Neff, pitched at TechCrunch Disrupt 2026 Startup Battlefield with an app that converts any text file (PDFs, textbooks) into a TikTok/Instagram Reels-style vertical feed. Content is rendered as AI-generated video, audio, text, or interactive quizzes; swiping up moves to a new topic while swiping sideways goes deeper, ending with a quiz. The startup is targeting educational institutions, corporate training, and individual learners.

Samsung is expected to more than double HBM4 and HBM4E output next year, raising total HBM wafer input from ~180,000/month this year to ~250,000/month, with the HBM4 family rising from ~40% to ~80% of product mix. Glass carrier cleaning volume—a proxy for HBM production—jumps from 20,000 sheets/month this year to 50,000/month next year. Samsung began HBM4 mass production in February using 1c DRAM and a 4nm base die, and shipped 12-layer HBM4E samples to Nvidia in May.

The US EPA announced on September 14, 2026 that it is repealing the 2024 Carbon Pollution Standards, which had required existing coal plants and new gas plants to capture 90% of carbon emissions or shut down by 2039. The repealed rule was projected to reduce carbon pollution by 1.38 billion metric tons through 2047 and also reduce sulfur dioxide, nitrogen oxides, and fine particulate matter.

HAVI, the supply-chain and logistics company behind McDonald's distribution, has confirmed Kuala Lumpur as its fourth global tech hub, targeting ~140 staff by mid-2027 with permanent space at Merdeka 118 from November. Despite AI-forward messaging, the bulk of roles are enterprise IT—SAP analysts, database, infrastructure, security—with only 8 dedicated AI engineering positions (6 filled) layered on top of an SAP core. Daniela Ribeiro, HAVI's Director of Technology Transformation, frames the hub not as a regional shared-services office but as part of a single global operation split across time zones with Chicago, Krakow, Lisbon, and now KL.

Boris Cherny shares a six-step iterative problem-solving framework—understand available info, gather missing info, define the problem, define a simple approach, define a goal, act with urgency—and argues that being wrong is valuable because new data forces redefining steps 3-5. He identifies the most common failure mode as poorly defining the problem or choosing an overly complex approach.

California's AB 1130, signed by Governor Newsom, adds financial and criminal penalties to existing influencer political ad disclosure rules. Regulators can now fine influencers up to $5,000 per violation and refer cases to law enforcement as potential misdemeanors. The bill was prompted by billionaire Tom Steyer's campaign paying influencers who didn't initially disclose payments.

Cancer Research Malaysia's MeMoSA (Mobile Mouth Screening Anywhere) is a telemedicine-plus-AI platform where trained non-specialists capture oral cavity images via a mobile app, AI checks image quality, and dental specialists review remotely. The AI component is trained on a multi-country dataset of over 30,000 clinically annotated oral mucosa images collected across five countries, with diagnoses verified by biopsy or expert consensus. MeMoSA is explicitly a screening support tool, not a diagnostic system, keeping clinical judgement in the loop.

NHTSA has partnered with SAE to launch ASCEND, a government-industry initiative aimed at developing data-driven AV performance standards, since no standardized test currently exists for judging automated driving systems. Existing Federal Motor Vehicle Safety Standards (FMVSS) don't cover vehicles like the Tesla Cybercab and Zoox robotaxi that lack pedals, steering wheels, and mirrors, forcing regulators to build a new framework from scratch.

OpenAI's ad collector at bzr.openai.com sets a SameSite=none, HttpOnly cookie called __obi scoped to .openai.com while you use ChatGPT, tied to your account subject ID. Any site that installs OpenAI's ad pixel SDK causes your browser to send __obi back to OpenAI along with page content and purchase behavior, linking your off-platform browsing to your ChatGPT account. The author verified this across 936 advertiser pixels on 1,029 hostnames using two independent capture methods.

Singapore's National Library Board is launching a micropayments program that pays people to put down their phones and read books, aiming to build reading habits through financial incentives. The article text itself is mostly site navigation with minimal substantive detail about the program's mechanics, payout amounts, or implementation.

Pirate Face mirrors open-weight AI models from Hugging Face into peer-to-peer torrents, so models survive even if removed from Hugging Face. It currently lists 669k+ eligible models with SHA-256 checksum verification, and shows live seeding stats (e.g., DeepSeek-V4.1-Flash at 765 GB with 1 seeder, all-MiniLM-L6-v2 at 91 MB with 7 seeders).

Tom's Hardware reviews the Dell Pro 7 Webcam 4K, a compact $199.99 4K/60fps webcam with Windows Hello support, presence detection, good image quality in ideal lighting, and surprisingly strong microphone performance. The review notes it faces heavy competition in its price segment.

Sherline Tools, a USA manufacturer of precision lathes, mills, and small CNC machines, is winding down manufacturing operations and will close entirely by end of 2026 at the latest. Production continues through October 2026 only as remaining inventory, equipment, and staffing allow, with some products already becoming unavailable. The company plans to maintain warranty support and keep its technical/educational archive online.

Brynn Putnam (sold Mirror to Lululemon for $500M) and Tristan Walker (sold Walker & Company to P&G) are both building new companies centered on offline human connection. Putnam's Board makes a 24-inch touchscreen table for group gaming; Walker's Heirloom is acquiring and scaling fine-craft trade schools, starting with a San Francisco leatherworking school. Walker is also running Collaborative Holdings, a fund from Collaborative Fund designed to back profitable, cash-flow-positive businesses without forcing venture-style returns.

Anthropic, OpenAI, SpaceXAI, and Google face an antitrust lawsuit alleging they agreed to slow AI development, with plaintiffs claiming the plan was in motion for months and calling the agreement 'self-serving.' The article body is largely inaccessible, containing only site navigation and membership boilerplate, so substantive details beyond the headline are unavailable.

CNBC speculates that workplaces may need 'robot relations' departments to handle employee complaints about AI agents, cobots, and automated management systems. It cites an OECD survey finding that 90% of US managers say their firms have adopted at least one algorithmic tool to instruct, monitor, or evaluate workers.

Sunil Pai describes a recurring failure pattern he calls the 'senior engineer death spiral': an engineer takes on an ambitious project to prove themselves, disappears for weeks giving only vague positive updates in standups, realizes they've shipped nothing, then secretly tries to cram a month's work into a week—leading to burnout, PIPs, or quitting. He notes this pattern has intensified with remote work and coding agents giving engineers more ownership but less structural support.

A nostalgic retrospective marking the September 2018 launch of NVIDIA's GeForce RTX 2080 and 2080 Ti, the first consumer graphics cards to introduce real-time ray tracing. The article notes that supporting games were scarce at launch, though ray tracing later became a pervasive PC gaming feature.

Researchers built a sub-100-gram drone that navigates using physical whiskers instead of cameras or GPS, enabling operation in dark, dusty, or smoky environments where optical sensors fail. The drone runs a 34KB software stack and achieves sub-millimeter precision. The article itself is mostly boilerplate with no further technical detail beyond the headline.

Final Fantasy VII Revelation may require upwards of 200 GB of storage and an additional download, potentially setting a new record for base install size in the remake series. The article itself is mostly site navigation and boilerplate with minimal substantive content beyond the headline.

Qwen released Qwen-Image-2.1, an open-source 7B parameter image model that unifies text-to-image generation and image editing, including native transparent image support. It accepts up to 10 reference images for versatile editing and uses a mixed-granularity attention architecture with KV cache reuse to optimize inference efficiency.