AI Weekly Malaysia

The supplied page contains only Ars Technica's cookie-consent and privacy-preference boilerplate — no article body. Beyond the headline claim that Ukraine used a drone to rescue a wounded soldier, the text gives no details: no drone type, no date, no operators, no technical mechanism, no sourcing.

NVIDIA published Cosmos3-H-Dreams, a world-foundation-model simulator that learns surgical scene dynamics directly from video plus robot kinematics rather than modeling tissue physics explicitly. It was trained on clinical cholecystectomy recordings from the CMR Surgical Versius platform — the team claims this is the first real-time generative surgical simulator trained on clinical cholecystectomy footage with live kinematic control — and runs interactively on a single NVIDIA RTX PRO 6000 GPU using the FlashDreams streaming-inference library. It follows the earlier Cosmos-H-Dreams (built on Cosmos-Predict2.5) which only handled dVRK tabletop suturing.

The supplied page text contains only Ars Technica's cookie consent and privacy boilerplate — no article body, figures, or claims about quantum error correction are present. The headline mentions error correction going "well beyond breakeven," but nothing in the extract explains what was measured, by whom, or with what hardware.

Bloomberg reporting cited by TechCrunch says Eric Schmidt owns roughly 1% of Anthropic from a tens-of-millions investment in its 2021 Series A, when the company was valued at $461 million — about half the 2% stake held by co-founder and CEO Dario Amodei. Anthropic is widely expected to go public before the end of the year, with top-end estimates valuing it at up to $2 trillion, which at that figure would add as much as $20 billion to Schmidt's $57.5 billion net worth. Neither Schmidt nor Anthropic commented.

AMD's FSR 4.1 now works on Radeon RX 6000 GPUs, Xbox Series X and S, and Ryzen AI Max, AI 9, and Z2 SoCs. Tom's Hardware describes it as bringing the newer upscaling tech to older hardware at a small performance cost. The provided page text does not include benchmark numbers, pricing, or game-specific results.

The item is an Ars Technica piece headlined "Meta tries to scapegoat staff who downloaded torrents to spare execs," with a URL slug indicating a court order requiring Meta to name executives who approved torrenting for AI training. The retrieved article text contains only cookie-consent and privacy-policy boilerplate — no reporting, quotes, filing names, dates, or figures. There is nothing verifiable in the body beyond the headline claim.

Steve Krouse’s essay, discussed on Hacker News (340 points, 405 comments), asks what to teach his soon-to-be-born son when his programming skills feel less valuable in an AGI era. He draws on Neil Postman’s 1995 The End of Education, Thomas Edison’s 1913 claim that books would be obsolete in schools within ten years, and his own failed 2013 MOOC prediction to argue that schools have survived past technology panics via the Lindy effect. The piece ends by questioning whether captured intelligence in a bottle finally changes that pattern.

Meta's AI companion app Muse reached 10.2 million U.S. downloads (10.9M globally) within 40 days of its September 2026 debut, a milestone Sensor Tower says only 11 other apps have hit — the most recent being Threads in July 2023. Meta poured 39% of its October ad spend into promoting Muse, up from 23% in September, lifting it from 73rd to 21st among U.S. brands by ad impressions. Muse also posted 30% U.S. 7-day retention and 23% 14-day retention, versus 19% and 16% averages for ChatGPT, Claude, and Grok.

The Hacker News headline (published 20 Oct 2026) reports that eight certificate authority keys belonging to eMudhra were compromised, prompting emergency fixes in browsers and other software. The retrieved page body is almost entirely site navigation and boilerplate — no affected root/intermediate names, no revocation or distrust dates, no CVE, no vendor statement, and no list of which browsers or software shipped fixes.

China's MOFCOM issued a policy proposal (not yet law) requiring pre-approval for frontier AI models and some other technologies before international sharing, with open-source tech slated for automatic approval; separately, DHS ordered that DeepSeek as a company and its models as a platform be excluded from federal civilian procurement, with unresolved questions about fine-tunes and models trained on DeepSeek outputs (e.g. Nvidia Nemotron). DeepSeek also published its internal inference stack, DeepInfer, which the authors say is likely more efficient than vLLM/SGLang for its models, with estimates of up to 90% gross margin on token sales and an expectation that inference providers will adopt it and push DeepSeek model prices down.

CNBC's investing-club piece says Amazon hit a record $284 per share on Aug. 3, up nearly 45% from its February 52-week low, then fell more than 7%. Morgan Stanley added Amazon to its top picks and raised its price target to $350 from $335, citing AWS strength and Andy Jassy's comments that data centers have a 30-year usable life, servers last at least 5-6 years, and most AI capacity is contracted for at least five-year terms.

Google's google-pytorch org has published torch_tpu, an open-source PyTorch backend that runs PyTorch models on TPUs via custom ATen kernels, an XLA compilation cache, torch.compile() graph-mode integration, and distributed training support (data and tensor parallelism). The repo ships examples for Llama, Qwen, ResNet, DLRM, minGPT and UNet, plus Marimo notebooks that the README says are easiest to run on a v6e GCE instance. It has ~502 stars, 25 forks and 4,470 commits, and drew 354 points / 74 comments on Hacker News.

TP-Link announced it signed a letter of intent — not a binding agreement — for a $1 billion U.S. manufacturing alliance with Key Tronic to assemble Wi-Fi routers stateside. The stated aim is to break a U.S. ban on newly developed routers manufactured outside the country, which has been keeping TP-Link's Wi-Fi 8 products out of American homes. TP-Link is incorporated in the U.S. with headquarters in Irvine, California, but faces government pressure over its historical Chinese roots.

Amulya Vishwanath, founder and managing partner of Techable Ventures, told TechCrunch's Equity podcast that physical AI is in a "LEGO era" rather than on the verge of a ChatGPT moment for robotics. She argued the foundational pieces — robotic hands, actuators, and general- and special-purpose robot platforms — are still being rethought and assembled, and that talent is a key differentiator, particularly in the U.S. market. Vishwanath previously spent nearly seven years as a product marketer at Nvidia and is now investing in physical AI.

Ai2 released FairStress, a benchmark that measures demographic fairness in LLMs beyond the final decision a model makes. It tests whether models change an otherwise identical decision when challenged with persuasive arguments ("adversarial prefixes") that favor different demographic groups, across hiring, lending, healthcare, education, and housing. Testing Olmo, Llama, Qwen, Gemma, Claude, and the GPT family, Ai2 found the same arguments can make models more or less likely to flip depending on which group the argument benefits, with differences also showing up in explanations and internal activation patterns. FairStress is downloadable and being integrated into Olmo-eval.

Mistral announced Voxtral 3, a family of speech-to-text and text-to-speech models plus a voice agent SDK that combines listening, reasoning, tool calling, and speaking in one real-time streaming pipeline across 50+ languages. It is available via Mistral’s EU-hosted API or self-deployment and targets contact centers, virtual assistants, and hands-free copilots. The Hacker News thread drew 352 points and 94 comments, but the announcement gives no benchmarks, pricing, or latency numbers.

JustWatch, the 12-year-old streaming search engine used by 50 million people, launched JustWatch TV on Tuesday — its own streaming service with content from 30 partners, free and paid tiers, and a movie rental store. The guide covers 500,000+ titles across 1,500+ providers in 140 markets and supplies 'where to watch' data to roughly 1,400 partner sites and apps, which studios and agencies use for ad targeting. Content partners announced include Paramount, New Regency, Fremantle, Bleecker Street, Vortex Media, plus regional players Cineverse (US), LEONINE Studios (Germany), SND (France) and Altitude (UK).

Electric Plant, founded by Justin Mast and Lee von Kraus, has raised an $11.2 million seed round led by Onto Ventures and Climactic to commercialize plant sensors plus Bombadil, an LLM trained on three years of plant electrical-signal data. The pitch is that Bombadil can interpret signals in a plant's stem and tell growers about needs such as water, nutrients, or light. The article provides no model accuracy, sensor pricing, API details, pilot customers, or deployment results.

A PNAS study finds that in the first few years after the 2010 Deepwater Horizon blowout released over 4 million barrels of oil, deep-sea populations of octopus, squid, krill, shrimp and lanternfish fell by at least 60 percent, and 16 years later acoustic surveys and thousands of netted samples show little sign of recovery across the water layer from roughly 600 feet down to more than two miles. Oil-related compounds have been found in squid and octopus flesh and in fish eggs, and the study also reports declines in genetic diversity. The authors stopped short of concluding the spill caused the declines but called it 'the most parsimonious explanation.'

Ethan Mollick marks the release of Co-Existence, 30 months after Co-Intelligence, and uses the gap to chart AI progress. He says he used Opus 5.5 in Claude Projects, with one orchestrator agent spawning sub-agents, to produce research, graphs, layouts, images, and citations in minutes, and that it caught a math error. The post cites 20 different models having held the top spot across Anthropic, Google, and OpenAI, 180 models now better than GPT-4 at release, and 19 of those able to run on a modern smartphone.

The Hacker News headline reports that a critical FortiSIEM vulnerability — described as unauthenticated remote code execution — now has a public proof-of-concept exploit. The supplied text is truncated mid-sentence, so no CVE identifier, affected version range, patch status, or exploitation-in-the-wild detail is available in the excerpt. Everything beyond the headline claim (public PoC, unauthenticated RCE, critical severity) cannot be verified from the text provided.

The Hacker News headline reports that the Qilin and INC ransomware groups are exploiting a Palo Alto Networks flaw to bypass MFA. The supplied page text is truncated after the headline and contains only site navigation, so no CVE ID, affected product or version, patch status, victim counts, or attack mechanics are available. Nothing beyond the headline claim can be verified from this text.

Y Combinator's profile of Vin Bui describes how he became the first full-time engineering hire at Nowadays (S23), a San Francisco company using AI to plan corporate events end to end, after leaving his own YC Winter 2025 company while on leave from Cornell. He describes the role as being a "technical founder" — building engineering processes from scratch, leading SOC 2 compliance, handling security and IT — and says the most interesting thing he has built is AI voice systems that call hotels and negotiate on behalf of clients. His tooling is Ghostty, the Claude Code CLI over any GUI, and switching between models per task (the excerpt cuts off mid-sentence).

IBM Research's ALTK-Evolve took a stock ReAct customer-service agent running GPT-4.1 — an older, non-reasoning model — to 67.5% Pass^4 on the τ²-bench retail domain, meaning it solved the task on all four independent attempts for 67.5% of tasks, above every retail entry in their October 12, 2026 leaderboard snapshot. The method needs no weight updates: the same GPT-4.1 model analyzes the agent's training-task attempts and distills successes and failures into short behavioral rules (how to authenticate a customer, confirm a transaction, handle a return), which are appended to the system prompt. τ²-bench pairs the agent with an LLM-simulated customer, domain tools, and a written policy, and only counts a task solved when the final database state matches the expected outcome.

A Hacker News submission reports that Facebook Pages now require a paid subscription to post more than two links per month. The linked source is a Mastodon post, and the article body could not be read, so the price, tier names, and exact rollout date are unverified. The thread drew 356 points and 314 comments, which suggests the change is hitting people who use Pages for distribution.