AI Weekly Malaysia

X has open-sourced its 'For You' timeline ranking algorithm and core ranking engine on GitHub under Apache v2, making the codebase 10-15x larger than its previous open-source release by adding model configuration, filter, and ranking parameter details. A new 'Under the Hood' transparency tool lets users who posted 10+ times in the past month download a JSON file showing labels applied to their account or posts, which can be fed into an LLM alongside the GitHub repo for interpretation.

Why: If you ship content on X for a product, startup, or personal brand, you can now download your own ranking-label JSON from the 'Under the Hood' settings page and cross-reference it against the actual weighting parameters in the GitHub repo to understand why your posts underperform. For AI/ML learners, the open-sourced ranking code is a real-world production recommender system you can inspect and partially run outside X, which is rare.

AI infrastructure spending is pushing up prices for electricity, chips, software, and data-center capacity, creating near-term inflationary pressure while corporate adoption remains uneven and productivity gains have not materialized. Fed Chair Kevin Warsh faces the question of whether AI-driven cost increases are a type of inflation that warrants interest rate hikes, despite Silicon Valley leaders like Sam Altman, Elon Musk, and Masayoshi Son promising eventual deflationary effects.

Why: If you are budgeting for AI-dependent products or SaaS, the article signals that infrastructure costs (compute, power, data-center capacity) are rising now while the productivity payoff is delayed — meaning margins may stay compressed longer than investor narratives suggest. Founders shipping AI features should price for current cost reality, not promised cost collapse.

Microsoft is merging its consumer and business Copilot apps and cutting several unsuccessful features by August 18, 2026, including Group Chats, AI podcasts, Copilot Labs, Deep Research, and the Mico animated character. This consolidation mirrors a broader industry trend where Anthropic, OpenAI, and Google are similarly folding specialized AI tools into single primary apps.

Why: If you are building standalone AI tools or niche AI features, the decision by major platforms to consolidate capabilities into single primary apps suggests a shrinking market for single-purpose AI wrappers. You should evaluate whether your product survives platform consolidation or needs to pivot to deeper, proprietary workflow integration.

Nvidia secured commitments from Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR for up to $500B to build AI data centers, with Nvidia guaranteeing that GPUs used as collateral retain their value—covering up to 25% of any shortfall if liquidated chips fetch less than book value. The plan aims to create a secondary market for aging GPUs so demand persists as hardware ages, but creates 'wrong way' risk where Nvidia's obligations grow precisely when demand weakens.

Why: If a used-GPU market materializes, GPU compute prices could eventually drop for builders who rent capacity from neoclouds or data centers—relevant to Malaysian startups running inference workloads on cloud GPU services. But the more immediate signal is that Nvidia is financially engineering demand for its own chips, which means current GPU pricing power stays with Nvidia for now; don't plan infrastructure budgets assuming cheaper compute is coming soon.

Cerebras shares fell nearly 20% after missing earnings expectations, with hardware sales declining even as its AI cloud revenue surged 281%. The divergence signals that demand for Cerebras's wafer-scale inference/cloud offering is growing faster than its chip-selling business.

Why: For builders evaluating inference infrastructure, Cerebras's 281% AI cloud revenue jump suggests its hosted inference service is gaining real traction — worth benchmarking against Groq and standard GPU clouds for latency-sensitive workloads. The hardware sales decline also signals that the pure chip-sales model for AI accelerators remains hard to sustain.

Grab is quietly testing a Women-Only Rides (Beta) feature in Malaysia, matching female passengers exclusively with female driver-partners. It requires Advance Booking with a minimum 75-minute lead time, is restricted to city rides (no KLIA or intercity), and is available in Klang Valley, Johor, and Penang per the app. Access requires a one-time identity verification (MyKad or passport plus selfie), cross-referenced against official records, with approval taking up to three working days.

Why: For founders and product builders in Malaysia, this is a case study in constrained product design: Grab is trading convenience (no on-demand, 75-min minimum lead, city-only) for a safety-focused niche, and building a gender-verification gate using MyKad/passport plus selfie. If you build consumer-facing products in Malaysia, consider whether similar identity-verification flows and advance-booking constraints could unlock underserved segments.

A contractor hired by Pageloot (a QR code marketing company) stored staging server credentials in a Google Doc set to 'anyone with the link,' which Google then indexed so the hostname and credential string appeared in search autocomplete. Pageloot co-founder Siim Kostabi says an employee discovered the leak while debugging something unrelated by typing their domain into Google Search. The company cut the contractor's access, rotated all exposed credentials, and banned storing passwords on Google Docs, Slack, Notion, or similar collaboration tools.

Why: If you hire contractors or work across devices, enforce a credential-sharing policy before this happens to you—specifically, prohibit shared docs, Slack, or Notion for passwords and require a password manager. The failure mode here is that 'anyone with the link' Google Docs get crawled and indexed by Google Search, turning an internal convenience into a publicly discoverable credential leak.

OpenAI has made Codex available as a Linux desktop download, integrating its coding agent directly into ChatGPT. The page highlights multi-agent workflows using built-in git worktrees and cloud environments, a 'Skills' system for teaching Codex team-specific standards, and scheduled background tasks like issue triage, alert monitoring, and CI/CD work.

Why: If you're on Linux and already paying for ChatGPT, Codex desktop gives you a local entry point to OpenAI's agentic coding workflow without a separate API setup—but this is a product page, not independent testing, so treat the 'weeks of work in days' claim as unverified. Evaluate whether the Skills feature fits your team's conventions before committing workflows to it.

OpenAI's ChatGPT Ads service has been serving ads and billing customers for up to a day after campaigns are paused, due to a defect where the ad-level serving status doesn't refresh when the campaign-level pause is triggered. A UK advertiser, Ed Bolton of Excel4Business, was charged £60.72 in invalid charges from the initial occurrence and ~£6.47 from a recurrence on August 4–5, with OpenAI support confirming the bug and processing refunds. OpenAI began testing ChatGPT ads in the US in February 2026 and expanded to the UK, Mexico, Brazil, Japan, and South Korea as of August 13.

Why: If you are running or considering ChatGPT Ads, monitor your billing closely after pausing campaigns—OpenAI's own support confirmed the pause doesn't reliably propagate to the ad-serving layer, and invalid charges may accrue for hours. Factor this into budget controls and don't assume a pause is instantaneous like on Google or Meta. This is early-stage ad infrastructure from a company whose 50M paying users out of 900M weekly users means ad revenue is now material to their business model.

Anthropic's data team (Clement Peng, Lily Zhao) describes deploying Claude Tag (public beta) as a Slack-based data analytics agent that lets non-analysts ask ad-hoc data questions backed by the same governed semantic layer, skill files, and evaluation suite that previously achieved ~95% accuracy with Claude Code. The post outlines five learnings from a year of deployment covering distribution, permissions, freshness, and observability, though the excerpt cuts off before detailing them.

Why: If you are building an internal AI agent for non-technical users, the architecture here is worth studying: a governed semantic layer plus skill files plus an eval suite as the foundation, then a Slack surface (Claude Tag) for distribution. The key decision point is whether your team needs to separate 'getting the agent accurate' from 'getting it deployed where non-analysts work' as distinct engineering motions, as Anthropic explicitly states these were 'quite different motions.'

GPU cloud provider Nebius plans to bring online over 1 GW of datacenter capacity annually starting 2027, funded by $20-25B in 2026 capex, $9B+ in customer prepayments, and asset-backed debt using GPUs as collateral. The company projects $20-25M revenue per MW for medium-term leases and $40-50M for short-term leases under six months.

Why: If you rent GPU capacity for AI workloads, Nebius's aggressive buildout signals more supply coming online by 2027, which could ease pricing pressure—but the heavy reliance on customer prepayments means large tenants are locking in capacity now, potentially squeezing spot or short-term availability for smaller builders. The $40-50M/MW short-term lease revenue figure tells you GPU cloud margins on urgent demand remain extremely high, so negotiate early and long if you have predictable workloads.

At the Ai4 conference in Las Vegas, Geoffrey Hinton, Fei-Fei Li, and Andrew Ng argued against letting a handful of major AI labs control access to AI, though they disagreed on tactics. Ng pushed for openness and multiple competing providers to prevent gatekeeping; Hinton drew a sharp distinction between open-source software (code inspectable) and open-weight models (trained parameters released), expressing concern about the latter's lack of control.

Why: If you build on open-weight models (Llama, Mistral, etc.), the open-vs-closed debate could shape future regulation and availability of those weights — worth tracking when deciding whether to architect around open weights or API-dependent closed models. Hinton's distinction between open-source and open-weight is a useful framing for anyone evaluating the real risks and freedoms of the models they ship.

Thrive Holdings, a Thrive Capital spinout that buys traditional businesses and implements AI into their workflows, raised $2B at a $12B valuation from SoftBank, D1 Capital, and Altimeter. OpenAI took an ownership stake in December 2025 and sends employees to embed with Thrive's 70+ portfolio companies. Thrive's accounting arm (Current) reports its TaxAI agents processed 7,000+ tax returns at 98% accuracy and cut prep time by 30%, while its IT arm (Shield) claims 36x faster help desk resolution and doubled custom AI agent deployments in the last month.

Why: The 'acquire a boring business and inject AI agents' playbook is now attracting billions, with concrete (if vendor-reported) metrics on agent ROI in accounting and IT support. If you build AI agents or SaaS for SMEs, this signals that PE-style rollups with embedded AI teams are becoming a competitive channel — your potential customers may get acquired and standardized onto an internal AI stack rather than buying standalone tools.

Gregory Kurtzer, founder of CentOS and Rocky Linux, has launched OpenWALDO—a project to build a shared, open-source AI training dataset with full provenance and a bill of materials. Funded by his AI infrastructure company CIQ, the effort currently targets 167 billion transparent tokens, a fraction of the trillions used by major AI labs. The project argues that even 'open-weight' models hide their training data, creating legal and compliance risk for downstream users.

Why: If you ship products using open-weight models, you currently have no auditable trail for training data lineage—OpenWALDO's 'bill of materials' concept could eventually let you point to a verified baseline corpus and reduce copyright/consent exposure. But at 167B tokens today versus trillions in proprietary datasets, this is not yet something you can train a competitive model on; treat it as a project to watch, not a dataset to use.

OlmoEarth Studio now allows users to compute and export custom embedding vectors from Earth-observation data using its open-source foundation models. Users can configure parameters like area of interest, time span, encoder variant (Nano, Tiny, Base), spatial resolution (10m to 80m), and imagery sources (Sentinel-2 L2A, Sentinel-1 RTC) to generate Cloud-Optimized GeoTIFFs (COGs) for downstream tasks.

Why: Developers and startups working with geospatial data, agriculture, or climate tech can use these pre-computed embeddings (stored as int8 in COGs) to quickly build similarity search or segmentation models without training their own Earth observation models from scratch.

Security researcher Nightmare Eclipse published a new Windows zero-day, dubbed ShieldBreak, that exploits Windows Defender to escalate from low-level user to full system access. The bug affects Windows 10, Windows 11 (including 25H2), and Windows Server 2025, requires Windows Defender to be enabled, and has no patch yet. Will Dormann independently verified the exploit, which builds on an earlier bug called RoguePlanet whose patch Nightmare Eclipse says is now fully bypassed.

Why: If you run Windows Server 2025 or Windows 11 in production or dev environments, there is no patch for this local privilege escalation bug, and disabling Windows Defender is not a safe workaround since the exploit requires Defender to be enabled. The practical takeaway is to restrict what untrusted executables can run on Windows hosts and treat any local code execution as a full compromise until Microsoft patches.

Uber Freight is investigating a data breach claimed by the Helix extortion group, which says it stole nearly 1 million files from mailboxes, OneDrive accounts, and accounts receivable systems. Uber Freight says operations were never disrupted and the incident was contained. Helix is linked by Google Threat Intelligence to a cluster (UNC6671) that uses vishing—posing as IT helpdesk staff—to run device code phishing against employees, then siphons data from Microsoft 365 and Okta.

Why: If your org uses Okta or Microsoft 365, this is a concrete reminder that device code phishing via fake IT helpdesk calls is an active, working attack path—not a theoretical one. Train staff to reject unsolicited device code prompts and verify any 'mandatory security migration' call through an internal channel before entering codes.

Chrome uses a JPEG decoding optimization where it avoids fully decompressing large JPEGs when rendering them at small sizes, instead leveraging the DCT frequency-domain data directly. This causes tiny JPEGs to look noticeably different in Chrome versus Firefox—often thicker or less faithful to the original—because high-frequency detail is handled differently during downscaling.

Why: If you ship small JPEG icons or logos (e.g., 15-20px), don't assume they render identically across browsers; Chrome's partial-decode optimization can alter appearance at small sizes. Switching to SVG for small icons avoids this entirely, as the author did.

LiquidAI released LFM2.5-VL-3B, a 3.1B parameter vision-language model designed for on-device/edge use, pairing a SigLIP2 400M vision encoder with their LFM2.5-2.6B text backbone. It was pre-trained on ~34T tokens with 4x more vision data than prior versions, supports 128K vocabulary for non-Latin scripts, and adds screen/UI understanding, object grounding, multi-image input, and function calling. Benchmarks show it leading its size class on real-world image tasks (RealWorldQA 73.1, MMStar 63.3) against comparably-sized models from Qwen, InternVL, and Gemma.

Why: If you are building on-device apps that need document/screen understanding or vision-grounded function calling without cloud API latency or cost, this is a concrete 3B model worth benchmarking against Qwen3.5-2B or InternVL 3.5 2B for your use case. The function-calling capability in vision-text contexts is the differentiator to test, since most small VLMs struggle there.

Google's Pixel 11 lineup is light on hardware changes but pushes deeper into agentic AI via Gemini, allowing U.S. users to order groceries, book rides, and have Gemini call businesses for reservations with user oversight. Google is connecting Gemini to third-party apps including Granola, Otter.ai, Wix, OpenTable, Ticketmaster, Thumbtack, and Zocdoc over the coming weeks, and is bringing its Rambler dictation product to Pixel phones alongside live translation for videos and podcasts.

Why: If you build a consumer-facing app or SaaS, Google is actively expanding Gemini's agent integrations with named third-party services—this is a signal to evaluate whether exposing your app's actions to Gemini (or similar agent platforms) is worth prioritizing. The agentic features are U.S.-only at launch, so Malaysian builders should watch for regional rollout timelines rather than expect immediate local availability.

Anthropic will begin digitally watermarking AI-generated text and images produced by Claude, as part of its compliance plan for the EU's Artificial Intelligence Act. The article is primarily a headline-level announcement; the full body text was not provided beyond site navigation boilerplate.

Why: If you ship Claude-generated content into products or content pipelines serving EU users, expect watermarking to be embedded in outputs—this could affect content provenance, SEO, plagiarism detection, and downstream processing. Builders should check whether watermarked outputs break existing text/image pipelines before the feature rolls out.

uBlock Origin has reportedly stopped trying to block Facebook ads, conceding that Facebook's ad delivery has become too deeply embedded in its content pipeline to reliably filter. The HN discussion (696 points, 854 comments) reflects broader frustration with the ad-blocker arms race as platforms increasingly interleave ads with organic content.

Why: If you ship browser extensions, content filters, or any tool that relies on DOM-level ad detection, Facebook's approach is a signal that platform-side ad integration is moving toward indistinguishable-from-content delivery — your filtering strategies may need to shift from pattern-matching to heuristic or ML-based approaches. For everyday users in Malaysia where Facebook remains a primary business and community platform, this means ad-blocking on Facebook is effectively over.

Two undersea cables were damaged off the coast of Perth, Australia, following suspicious vessel activity. A company CEO noted the faults occurred 'straight after each other and in close proximity,' suggesting possible deliberate sabotage rather than accidental anchor drags.

Why: Undersea cable cuts near Perth affect connectivity paths between Australia and Southeast Asia, including Malaysia. SaaS founders and infra teams should verify whether their cloud regions or CDN paths route through affected cables and review failover assumptions — if two cables going down in close proximity is sabotage, redundancy across the same geographic chokepoint is not real redundancy.

India's Yulu, a Bengaluru-based electric two-wheeler subscription startup, raised $93M in Series C funding ($63M equity led by GEF Capital, $30M debt) at a ~$170M post-money valuation. The startup rents e-bikes weekly to gig delivery drivers, powers 750,000+ deliveries daily with a 50,000-vehicle fleet, and plans to scale to 200,000 bikes over two years, with 95% of revenue from gig-worker rentals.

Why: For Malaysian/SEA founders and operators, Yulu's pivot from consumer bike-sharing to B2B gig-economy vehicle leasing is a concrete playbook for capturing quick-commerce logistics demand without selling hardware. The model—weekly subscriptions that remove vehicle ownership friction for delivery riders—could translate to Malaysian quick-commerce and last-mile logistics, where similar gig delivery infrastructure gaps exist. The funding structure (equity for growth, debt/lease for fleet) and the claim of positive EBITDA are worth studying if you are building asset-heavy subscription or mobility-adjacent businesses in SEA.

Google CEO Sundar Pichai announced Gemini has surpassed one billion monthly users, making it Google's fastest-growing product and the 14th to reach that milestone. The announcement coincides with a significant reorganization of Google's AI division, with Demis Hassabis stepping back and other researchers departing, while competitors have released more advanced models.

Why: Gemini's one-billion-user distribution scale means it is now a mainstream platform surface that builders should treat as a first-class integration target alongside OpenAI. The internal reorganization and researcher departures signal potential instability in Google's AI roadmap, so teams betting on Gemini APIs should track API continuity and model-version commitments rather than assume steady iteration.