AI Weekly Malaysia

Andrewt.net explains how to create visually distinctive QR codes by modifying data modules while preserving function patterns, then applying Floyd-Steinberg error diffusion dithering to embed low-res images into the code. The technique exploits QR's built-in error correction: shrink each data module to a 3x3 grid center, use surrounding cells for image data, and dither the result for midtones instead of crude thresholding.

The author replaced a Hetzner VPS with a rooted CMF Phone 1 (8 ARM cores, 8GB RAM, 128GB flash, 5G modem, battery backup) to host personal web apps like Surf, a finance tracker, and screen sharing. They initially tried flashing postmarketOS but found Wi-Fi and hardware acceleration broken, soft-bricking the phone and requiring a Windows-based recovery. They concluded that keeping Android is better because it already has working drivers for all the phone's hardware.

Anthropic is making auto mode the default for new Claude Code sessions on Pro, Max, and Team plans starting August 14th, backed by evals showing auto mode blocks 89% of dangerous actions versus only 13.6% of human reviewers refusing them. A third-party Trajectory Labs evaluation found zero successful attacks across 720 indirect prompt injection attempts against Claude Fable 5, Opus 5, and Sonnet 5 running auto mode, though 11% of harmful actions would still not be blocked.

Shopify replaced Redis with MySQL for its oversell protection system to align with a unified database strategy. By using MySQL 8's SKIP LOCKED feature and shifting to a one-row-per-inventory-unit design instead of one row per item, they handled Black Friday 2025 peak traffic of $5.1 million in sales per minute. The hardest lesson was discovering their actual bottleneck wasn't what they were initially measuring.

Amazon is investing in an on-site natural gas power plant for a planned data center in Pecos County, Texas, permitted to release 33 million tons of CO2 per year — more than any other U.S. power plant. Amazon's carbon emissions rose 16% last year, moving away from its 2040 net-zero pledge, with a spokesperson acknowledging 'the world looks different now.'

OpenAI has acquired NextSlide, a startup that turned prompts, notes, documents, or research into polished, editable presentations. Founder Ahmed Beshry (previously co-founder at Caper AI, acquired by Instacart in 2021) said the deal closed earlier in 2026 and the team is now working on ChatGPT. Financial terms were not disclosed.

EclipseFan is an open-source interactive map tracking the August 12, 2026 total solar eclipse, with multilingual SEO support via Cloudflare's HTMLRewriter and layered map views including Besselian elements, umbra tracking, and cloud projections. The project notes reveal implementation details around HTMLRewriter's limitation that it can only rewrite existing elements, not create them, requiring all target tags to pre-exist in the HTML.

Denmark is requiring high school students to orally defend their written assignments as a measure to counter AI-assisted cheating. The policy shifts assessment from written output alone to verbal verification of understanding. The article is a short news brief with no technical detail.

Amazon has acquired land and permits in Pecos County, Texas for a 7.65 gigawatt off-grid gas power plant to fuel an AI data center campus called GW Ranch, with a state permit allowing 33 million tons of CO2 emissions—more than the largest U.S. coal plant. Construction permits for three data center buildings were filed this week and land clearing has already begun, despite Amazon's net-zero-by-2040 Climate Pledge.

Nvidia's RTX 5090 is reportedly being sold by retailers in bundles that include 8 motherboards, forcing buyers to purchase unwanted components to get the GPU — a tactic reminiscent of the crypto mining boom's GPU scarcity era. The practice effectively inflates the real cost of acquiring an RTX 5090 well beyond its sticker price.

Lenny's Newsletter published a paid-subscriber-only 'Community Wisdom' roundup covering authentic outreach, finding design partners, what constitutes a moat when anyone can build, and making a product vision stick. The full content is behind a paywall with no substantive details visible.

A minimal Chrome extension by Andrew Pollack that hides LinkedIn's main feed and blocks its infinite-scroll pagination while leaving profiles, jobs, search, and messaging intact. It uses a CSS rule targeting [data-testid="mainFeed"] and a declarativeNetRequest rule blocking the feed pager endpoint (sduiid=com.linkedin.sdui.pagers.feed.mainFeed).

X is winding down its existing Revenue Sharing program and replacing it with Original Content Rewards starting September 8, 2026. The new program maintains requirements like 500 verified followers and 500,000 impressions in 90 days, but shifts focus to original content—such as original reporting, self-created media, and transformed commentary—while disqualifying unmodified reposts and aggregated content.

Fastmail now lets accounts choose an EU data region, with primary mail and files stored on Fastmail-owned servers in Amsterdam, previously US-only. Resilient replicas still live in the US since there is only one EU location so far, and connections fall back to US servers if Amsterdam is unavailable.

Jeff Geerling's benchmarks show Dell's XPS 13 with Intel Core 5 320 achieving 6.21 Gflops/W on HPL Linpack, beating a MacBook Neo's 5.38 Gflops/W and even surpassing M3 and M4 Mac Studios, though losing to the M4 Mac Mini at 7.57 Gflops/W. The Dell also matches the MacBook in idle and web-browsing power draw, suggesting x86 has closed much of the efficiency gap that made ARM dominant.

TechCrunch reviewed the Kindle Scribe Colorsoft after two months of testing, finding it a well-designed but niche device. It starts at $630, weighs 400 grams, has an 11-inch display, and comes in Black or Fig (purple), but its size makes it less portable than standard Kindles and not a must-have for most people.

EVPower will deploy a 27-bay EV charging hub (12 DC, 15 AC) with 1.28MW planned capacity at Vitória, a new commercial development in Kota Laksamana, Melaka, targeted for Q3 2028 completion. The project uses a co-investment model where developer Gene & Quin Development prepares supporting infrastructure while EVPower funds, deploys, and operates the chargers, with smart contactless payment support.

Google's Threat Intelligence Group (formerly Mandiant) has retired its APT-numbering system for naming hacking groups in favor of a two-word scheme: a memorable random first name plus a second word whose initial signals country of origin (Castle for China, Ion for Iran, Neptune for North Korea, Relic for Russia). CTO Shane Huntley said the change was needed because Google now tracks over 5,000 activity clusters, far more than the early 2010s when naming began.

Senko Rašić pushes back against the popular narrative that 'coding was never the hard part' of software development, arguing it dismisses the skill, stress, and craft programmers have invested for years. He challenges both sides of the claim with rhetorical questions: if coding is easy, why do buggy software, burnout, and deep technical literature exist; and if figuring out what to build is the hard part, why aren't product managers and business analysts the highest-paid rockstars in tech companies.

AI semiconductor and data center stocks rebounded sharply in the week of Aug 8 2026, with the Nasdaq up 5.2% and the S&P 500 hitting a record close. The recovery followed the collapse of Leopold Aschenbrenner's highly leveraged hedge fund 'Situational Awareness' and a weak July jobs report that lowered expectations of a Fed rate hike.

My comment on Now we have a timeline of the OpenAI accidental attack against Hugging Face — Hacker News.I think one of the most interesting details here might be tucked away in that first bulletin point: May 7: OpenAI starts a new training run for an experimental, unreleased model. (Do they mean an evaluation run? They say training run in the video, and later mention a “reward signal to judge how well they’re doing”, so I guess this really was about training a model, not evaluating one that was already trained.) The more I think about this the more I suspect that the fact this happened while training a new model is key to understanding what went wrong. In RLVR - Reinforcement Learning with Verifiable Rewards - you set the model a goal and have it take any steps necessary to achieve that goal. Clearly one aspect of OpenAI's training here is to RLVR their models for cybersecurity tasks. Just like pre-training benefits from dumping in vast sources of knowledge, the more tasks you can feed into RLVR the more of a general purpose capable model you get at the end. This also helps explain why the models had nothing to cause them to hold back. Those safety behaviors are added much later in the process. AND it explains (but does not excuse) why monitoring was so lax. If you're training a new model like this you presumably set it thousands of tasks like this in parallel. I can see how you might miss that a tiny subset of your training agents have started leaving each other messages in filenames on your packaging server. Someone once told me that you can't just leave the racist materials out of your training data if you want a non-racist model: it has to have seen examples of racism in order to later be taught that racism is bad. I can see echoes of that here. If your model doesn't know how to aggressively hack things how do you later teach it not to? (I have little knowledge of how RLVR works in practice so I'm looking forward to hearing from people who can help me understand if I'm on the right track here.) Tags: ai, openai, generative-ai, llms, ai-security-research, openai-hugging-face-incident

A US retailer is offering a $458 combo deal on AMD's Ryzen 7 9800X3D gaming CPU bundled with a B850 motherboard and 240mm liquid AIO cooler, representing a $160 discount. The article text itself is mostly Tom's Hardware site navigation and membership boilerplate with no additional technical detail.

A short, humorous blog post arguing that people should stop using their phones on the toilet, framing bathroom time as a free opportunity for the brain to consolidate memories and process ideas without constant input. It mixes tongue-in-cheek analogies with light health warnings about hemorrhoids and fecal particles on phone surfaces.

Amazon bypassed a community vote for a massive AI data center in Gilroy, California by invoking 45-year-old rules that locked local residents out of the public comment window. The article text itself is mostly site boilerplate; the substantive detail is confined to the headline.

UTM has released Triton, a Windows DirectX 11 DDI-level driver for QEMU that, combined with their Neptune protocol forwarding layer, enables modern Windows games and GPU-accelerated applications to run in QEMU VMs. Unlike previous DLL-replacement approaches, Triton implements the DirectX Device Driver Interface directly, avoiding CPU blitting performance penalties and anti-cheat detection issues.