AI&Future Tech

💼 New York overtakes San Francisco as top U.S. tech talent market New York City has topped San Francisco's Bay Area as the leading U.S. tech job market for the first time, according to CBRE's annual survey. NYC recorded 394,300 tech jobs versus the Bay Area's 375,730. AI-related roles now account for nearly one-third of all U.S. tech-talent job listings. Source

Didn't have time to catch up this week? Here's everything that matters in 2 minutes. ​ 🏃 Unitree's "Superman" robot outruns Usain Bolt ahead of $9B IPO Hit 12.66 m/s and jumped 2 meters vertically — built in just three months. 👉 Read more ​ 🚨 NSA and FBI warn of AI-built exploits targeting industrial systems AI-generated scripts are attacking Siemens PLCs across US energy, water and manufacturing sectors. 👉 Read more ​ 📦 Google's Gemma open models hit 1 billion downloads Developers have published over 100,000 variants built on the open-weight models. 👉 Read more ​ 💰 Nvidia acquires Poolside's AI model-building software for $6B The deal brings "Model Factory" and 109 employees into Nvidia. 👉 Read more ​ 🚀 Etched raises $700M at $21B valuation, doubling in a month Jane Street led the round and now runs one of the AI chip startup's racks in its own datacenter.

🦾 Unitree CEO: humanoid robots still 2–10 years from breakthrough Unitree Robotics CEO Wang Xingxing says the industry's "ChatGPT moment" will arrive only when a robot can complete ~80% of tasks in an unfamiliar environment using voice commands alone. He estimates that's 2–3 years away at best, 5–10 years if progress is slow. Unitree raised $900M in its Shanghai IPO this week, with shares surging 460% on debut and valuing the company at ~$50B. Its G1 humanoid, at $13,500, is already one of the most affordable on the market. Source

💼 Nvidia acquires Poolside's AI model-building software for $6B Nvidia is paying $6 billion to acquire "Model Factory," an AI model-building software platform from startup Poolside, along with 109 of the company's employees. Source

💼 Micro1 hits $500M run rate on AI training data demand AI data startup Micro1 has reached a $500M gross run rate, fueled by surging demand for AI training data. The milestone reflects broader growth across the sector as AI labs continue scaling up data needs. Source

🤖 Google's Gemma open models hit 1 billion downloads Google DeepMind announced on August 20 that its Gemma family of open-weight models has surpassed 1 billion cumulative downloads since launch, with developers publishing over 100,000 variants built on the weights. The milestone was disclosed by VP Clement Farabet and product director Olivier Lacombe. Source

🤖 Sutton calls synthetic data a "big mistake" for AI scaling Turing Award winner Richard Sutton argues that synthetic data is a fundamental error in scaling large language models. The world is "infinitely complex," he says, making any simulation of it "microscopic" — and human expertise acts as a bottleneck that blocks real scaling. His alternative: agents that learn continually from their own experience rather than relying on frozen models. Source

💼 Micron Launches $10B Research Lab Focused on Memory and AI Micron Technology has unveiled Micron Research Labs, a dedicated memory research institution headquartered in Boise, Idaho, backed by a planned $10 billion investment over the next decade. The lab will focus on long-horizon research in memory technologies, computing architectures, advanced packaging, and future semiconductor manufacturing — work described as going beyond current technology roadmaps. Micron calls it the first dedicated memory research institution of its kind in the U.S., and the investment is separate from its previously announced $250 billion U.S. expansion. Source

🤖 OpenAI brings Computer Use to ChatGPT for everyday automation OpenAI is integrating its Computer Use tools into ChatGPT, enabling the chatbot to control browsers, fill forms, and operate other apps autonomously. The company launched a Chrome extension, cloud and in-app browsers for ChatGPT, and added Computer Use to its Codex coding tool this year. OpenAI's models now read a webpage's underlying structure — memory, accessibility data, link connections — rather than relying solely on pixel-by-pixel screenshots. Training involved human-demonstrated task data and reinforcement learning. Rival Anthropic was first to market in 2024; at least 600,000 organizations have tried its Claude version. "Once ChatGPT can use computers faster than you or I can, it's going to change the way you want to interact with your computer," said Ari Weinstein, a manager on OpenAI's Computer Use team. Source

💼 Google Lands Marvell Deal for Custom AI Chips Marvell Technology has won a contract to design AI chips for Google, sending its shares up 9.8%. As part of the deal, Google received a warrant to purchase ~59 million Marvell shares worth $12.2 billion. Source

🤖 NSA and FBI warn of AI-built exploits targeting industrial control systems The NSA, CISA, and FBI have issued a joint warning that threat actors are using AI-generated scripts to attack Siemens S7 Series programmable logic controllers (PLCs) in U.S. critical infrastructure — including energy, water, and manufacturing sectors. Agencies say AI-assisted development is drastically reducing the time and skill required to exploit known vulnerabilities in industrial control systems. Source

🤖 More Americans — including young adults — now wary of AI 52% of Americans say they are "more concerned than excited" about AI in daily life, matching the 2023 all-time high, per a new Pew Research Center survey of 3,488 U.S. adults. Notably, this is the first time a majority of adults under 30 have reported the same — a shift that complicates tech industry assumptions about an "AI-native" generation. 71% of respondents now believe AI will lead to fewer jobs, up from 64% two years ago. Adults 18–29 are now as likely to hold that view as those aged 30–64. Source

🤖 Claude agents hit 35% in autonomous protein design Anthropic used Claude to autonomously run protein design pipelines, producing small proteins that dock onto target body structures — a core step in early drug development. The hit rate reached up to 35%, well above the 10–15% industry average. Claude directed existing specialized tools rather than performing novel computation itself. An independent review is still pending. Source

💼 OpenAI Revenue Growth Slows as Anthropic Surges OpenAI posted $6.7B in Q2 revenue, up 18% sequentially, but also grew its net losses. Anthropic, meanwhile, more than doubled its sales to $11.6B in the same period — leaving some OpenAI investors disappointed with the pace of growth. Source

💼 Graphwise lands majority-stake investment from Oakley Capital Bulgarian graph database startup Graphwise has secured a majority-stake investment from European VC firm Oakley Capital. The company plans to use the funds to expand globally, grow its go-to-market operations, and pursue strategic acquisitions. Graphwise is positioning itself as a semantic layer for AI agents. Source

💼 Etched raises $700M at $21B valuation, doubling in a month AI chip startup Etched closed a $700M Series D led by Jane Street on August 18, valuing the company at $21B — up from $10.3B just weeks earlier. The quant trading firm tested Etched's hardware, bought it, and now runs one of the company's racks in its own datacenter. Etched's previous $300M Series C, led by Sequoia, closed July 23. Source

💼 OpenAI's Brockman urges faster enterprise AI security OpenAI president Greg Brockman is calling on enterprise security teams to adopt AI defences at what he describes as "unprecedented speed." Brockman published an account of the "OpenAI-Hugging Face" incident to illustrate the urgency, arguing organisations must uplevel their security practices now. Source

🦾 Unitree's "Superman" robot outruns Usain Bolt ahead of $9B IPO Chinese robotics firm Unitree unveiled a humanoid robot called "Superman" capable of reaching 12.66 m/s — faster than Bolt's record 12.4 m/s peak speed. Developed in just three months, it can also jump two meters vertically, beating the human standing high jump record. Unitree lists on the Shanghai Stock Exchange Wednesday at a $9 billion valuation, with its IPO retail allocation reportedly 5,000x oversubscribed. The company shipped nearly 6,000 humanoid robots in H1 2026 — far ahead of US rivals still awaiting mass commercial launches. Source

🦾 LG and NVIDIA push for 100,000 hours of robot training data Four days after signing a strategic cooperation MOU at NVIDIA's Santa Clara HQ on August 13, LG Electronics hosted senior NVIDIA officials at its robot Data Factory in Seoul, targeting 100,000 hours of robot training data by end of 2026. Source

🤖 ByteDance's CUDA Agent writes faster GPU kernels than a compiler ByteDance Seed and Tsinghua AIR have released CUDA Agent, an agentic reinforcement learning system that trains an LLM to write optimized GPU kernels. The target: not correctness, but speed — frontier models already produce valid CUDA, just slow CUDA. On the KernelBench benchmark, the base model Seed1.6 scores 74.0%. CUDA Agent pushes that higher by using RL to optimize for runtime performance, not just functional accuracy. Source