AI Weekly Malaysia

Broadcom beat Q3 FY2026 estimates with $29.59B revenue (up 86% YoY) and $3.32 adjusted EPS, but shares fell 5% on weak Q4 guidance of $34.8B versus the $35.03B analysts expected. The company designs custom AI chips for Google, Meta, and OpenAI, and has recently added Apple and OpenAI as new customers.

Article URL: https://www.kedglobal.com/artificial-intelligence/newsView/ked202607210007 Comments URL: https://news.ycombinator.com/item?id=49544762

Article URL: https://www.sciencealert.com/gps-glitched-across-the-us-by-as-much-as-33-feet-scientists-have-never-seen-this-before Comments URL: https://news.ycombinator.com/item?id=49544618

OpenAI released GPT-6 Astra, their first model to reach the 'Critical' cybersecurity capability threshold under their Preparedness Framework, meaning it can autonomously discover unknown security flaws and develop exploits across well-protected systems without human guidance at each step. OpenAI reports it is significantly more robust against jailbreaks than GPT-5.6 Sol, including over longer trajectories, and includes adjustable refusal boundaries for high-risk users.

Funes is a single-binary memory layer for coding agents (Claude Code, Codex, pi, Hermes) that indexes session traces locally into a Lance dataset using a pinned local embedding model, then exposes recall and get tools so agents can retrieve past decisions during new sessions. It combines vector and BM25 search with cross-encoder reranking, indexes incrementally, and can optionally sync to a private Hugging Face dataset you own.

A fully open recipe for fine-tuning LiquidAI's LFM2.5-350M model with GRPO via the TRL library to improve structured-output compliance, evaluated on the IFStruct benchmark. The training runs in ~500 samples and 100 steps on a free-tier Colab or Kaggle GPU, lifting IFStruct scores from 22.6% to 29.7%. Evaluation is done locally via llama.cpp on a MacBook Pro M5 Max with 36GB unified memory.

Sergio Paniego reproduces Surya Narreddi's viral watercolour-painting LLM project using TRL and OpenEnv, training a Qwen3.5-35B-A3B model with LoRA and GRPO to write p5.brush JavaScript that generates watercolour art. The entire pipeline runs on Hugging Face infrastructure—Jobs for training, Spaces for the RL environment and scorer model, Inference Providers for the pairwise judge—and all artifacts, scripts, and a hand-rated reference pool are published openly.

TechCrunch announced the Builders Stage agenda for Disrupt 2026, happening October 13-15 at Moscone Center in San Francisco. Speakers include Gamma CEO Grant Lee, Precedent.vc founder Leah Solivan, and Google VP of Product Robby Stein, covering fundraising, hiring, go-to-market, and AI strategy. One highlighted session, 'How to Win When You're Not Building AI,' targets founders building enduring companies outside the AI hype cycle.

Equinix, a legacy colocation provider with 10,500 customers, announced a partnership with Nvidia that lets customers run AI models on the open-source cloud platform Together AI. Equinix shares are up 33% year-to-date with a $100B market cap, positioning itself as a flexible alternative to hyperscalers and neoclouds in the AI data center buildout.

Palo Alto Networks acquired Console, a two-year-old startup using AI agents to automate routine IT help desk tasks like password resets and app access provisioning, for $500M in cash and stock. Console had raised $29M and was valued at $157M pre-sale, with customers including Ramp, Flock Safety, and Scale AI. It will be integrated into Palo Alto's Cortex AI security platform to enable natural-language alert investigation and resolution.

Microsoft will begin reporting Azure quarterly revenue in dollars for the first time, as part of a restructuring that trims its operating segments from three to two. CEO Satya Nadella described Azure as becoming 'more purely our consumption-based platform and infrastructure business' under the new reporting structure, which replaces a segment layout in place since 2015.

Meta released Muse Spark 1.3, a refined version of its flagship AI model, now live on its API and Muse Code CLI. The update focuses on agentic behavior—asking clarifying questions, invoking human help when stuck, and confirming before consequential actions—while cutting token usage. Meta also promised an open-weights release 'soon,' and independent benchmarks from Artificial Analysis show a 4-point intelligence jump, putting it roughly on par with GPT 5.6 Sol, Claude Opus 5, and Grok 4.6 High.

TechCrunch Disrupt 2026 (Oct 13-15, San Francisco) adds a new 'Real World AI Stage' alongside its existing AI Stage, focusing on the intersection of digital and physical AI. Speakers include Les Karpas (Head of Physical AI, Nvidia), Shield AI, Colossal Biosciences, FieldAI, and Foxglove, with sessions on robotics' data gap versus LLMs and building AI systems where physical failure is catastrophic.

Article URL: https://www.anildash.com/2026/09/02/cancer-capital/ Comments URL: https://news.ycombinator.com/item?id=49543220

VMware will release an updated vSphere Standard, its neglected low-end server virtualization bundle, after years of steering customers toward the pricier, more complex Cloud Foundation (VCF) suite. Paul Turner, VMware Cloud Foundation Division CPO, said Broadcom 'corrected' sales incentives that had pushed users to VCF, acknowledging VMware had 'too big a focus on VCF.' vSphere Standard hasn't had a major update since version 8 in 2022; vSphere 9 shipped in 2025 only for VCF.

A federal judge ruled Google can keep its ad-tech business but must adjust its practices to favor competitors, though specifics were not provided and the full ruling remains under seal for 14 days. This follows a pattern set by the search antitrust case, where Judge Amit Mehta rejected divestiture of Chrome and Android but ordered an end to exclusive default-placement deals and search data sharing with competitors—remedies Google is currently appealing.

Cloudflare's summer 2024 intern cohort (~60 interns) shipped several internal and product-facing projects on the Cloudflare Developer Platform, including a Workers AI threads API that auto-recalls past messages for chatbot sessions, function calling support in Workers AI, D1 billing observability improvements with audit logs and alerts, and a Zero Trust policy tester built on Durable Objects.

Booz Allen's first Cyber Weapon Index tested 18 AI models (9 US, 9 Chinese) under identical conditions and found only Anthropic's Claude Mythos autonomously completed the full cyber kill chain. The report predicts most other tested models will reach the same weaponization level within six months and calls AI-enabled attacks 'imminent,' while separately noting OpenAI's unreleased Astra hit a 'critical' cybersecurity capability threshold for finding and exploiting zero-day bugs.

The article text contains only Ars Technica's cookie consent and privacy policy boilerplate. No substantive content about the budget deal or political control of grants is present in the provided text.

Article URL: https://www.nytimes.com/2026/09/01/nyregion/ai-ban-schools-nyc.html Comments URL: https://news.ycombinator.com/item?id=49542443

C# 15, launching November 2026, introduces union types via a new `union` keyword that lets developers group disparate custom types into a single compiler-verified entity, enabling better pattern matching without wrapper classes or third-party libraries. The release also adds `closed` hierarchies that restrict which types can inherit from a base class, giving similar compile-time guarantees to inheritance chains.

Polish software engineer Piotr Migdał tested 16 AI models on 1,040 mushroom photos across 55 species, finding that even the best performer (Gemini-3.8-flash) identified the correct species on first guess only 65% of the time. Critically, the errors were not random—deadly species were confused with edible ones in patterns that mirror real foraging fatalities, such as the death cap being mistaken for edible species 16-48% of the time.

An Ars Technica reporter describes renting a car and finding their driver's license listed for sale on a new dark web site alongside 153 million others. The actual article body was not captured—only privacy consent boilerplate and site navigation were retrieved, so substantive details about the breach, the rental company involved, or the dark web marketplace are unavailable.

Snowflake beat Q2 expectations with $1.55B revenue (35% YoY growth) and raised full-year product revenue guidance to $6.07B from $5.84B. The company highlighted its CoCo AI coding agent, which grew to 9,100 accounts, adding over 2,000 in the quarter.

OpenAI's upcoming Astra model uses a reasoning technique called 'recurrent depth' or 'opaque recurrence,' which processes queries in a loop rather than sequentially, leaving fewer legible chain-of-thought traces. AI safety experts including Redwood CEO Buck Shlegeris and Zvi Mowshowitz expressed concern that wider adoption could undermine chain-of-thought monitorability, a key tool for detecting agent misbehavior. Astra's use of the technique is reportedly limited and its chain of thought is still expected to remain legible.