AI Reading for Monday, March 2

Amazon’s Extreme AI Spending Sends Stock to Worst Month in Years - Bloomberg

Groq-et Man: Nvidia is building a dedicated AI inference processor that integrates Groq technology under a reported $20 billion nonexclusive deal, targets energy‑efficient production inference, is expected to debut at next month’s GTC, and has given OpenAI early access. - SiliconANGLE

Parmy Olson on Cowork plugins for FactSet and S&P, which raise might commoditizing market data products as dumb data pipes for AI, marketplaces for AI features and data, and who will be long-term winners. - Bloomberg

EUV got chips: ASML said its $400 million High-NA EUV lithography machines are technically ready for mass production, have processed about 500,000 wafers, run at roughly 80% uptime (target 90% by year‑end), and need 2–3 years of customer testing before integration. - Yahoo

ASML Plots 15-Year Roadmap Beyond EUV Into Advanced Packaging and AI-Powered Tools - Reuters

Bridge too far: Negotiations collapsed after the Pentagon sought to analyze and cross‑reference bulk American data (chatbot queries, search history, GPS, credit‑card transactions); Anthropic rejected it as 'a bridge too far,' and Pete Hegseth told contractors, including Amazon, to cut ties. - The Atlantic

Negotiations over a roughly $200 million Pentagon AI contract collapsed after Secretary Pete Hegseth labeled Anthropic a supply chain risk; OpenAI secured a competing framework deal the same night and Anthropic said it would sue. - The New York Times

U.S. Central Command used 'woke', 'radical left' Anthropic's Claude for intelligence assessments, target identification and battle simulations in Iran-related operations. - Gizmodo

Target AIcquired: A mistargeting that killed nearly 150 schoolchildren in Iran cannot currently be attributed to AI with confidence, while experts warn that unreliable generative AI and opaque military use increase risk of future targeting errors and accountability gaps. - The Potential Surface

Ex-OpenAI Official Warned Vague Policies Risk Unchecked Military AI Use - The Potential Surface

A former OpenAI geopolitics lead said in June 2025 that frontier AI firms' public military-use policies are vague and mutable, enabling defense integrations—including OpenAI models in a Defense Innovation Unit drone-swarm trial—and posing transparency and civilian-harm risks.

Georgetown CSET reviewed thousands of PLA procurement requests over three years and found China rapidly prototyping AI for piloting unmanned combat vehicles, swarming drones, cyber intrusion and satellite targeting, and deepfake-enabled cognitive warfare ahead of Victory Day 2025. - Foreign Affairs

ICE, ICE SaaS Baby: In April 2025 ICE deployed AI surveillance tools—Palantir won a $30 million contract for ImmigrationOS, ICE paid $5.7 million to Zignal Labs (claiming >8 billion posts/day) and integrated Clearview and tracking tech, raising oversight and civil‑liberties concerns. - Matt

Dallas Fed: Computer systems design roles fell 5% as AI-exposed sectors decline - Federal Reserve Bank of Dallas

Early data show wages are rising for AI-exposed jobs that place a high value on a “worker's tacit knowledge and experience”, as textbook knowledge loses value. Researchers report since ChatGPT's fall 2022 release U.S. employment rose ~2.5% while computer-systems-design fell ~5% and top 10% AI-exposed sectors fell ~1%, with wages up 7.5% nationally versus 16.7% in systems design and AI substituting entry-level but complementing experienced workers.

404 jobs not found, Fed can’t print jobs: Fed Governor Cook warns AI automation may cause irreversible unemployment - Boston Herald

Joint Venture Silicon Valley found nearly 410,000 Bay Area jobs have tasks AI can perform, concentrated in white-collar roles including about 140,000 software developers, as generative AI and agentic systems accelerate capabilities. - The Mercury News

Anthropic CEO warns AI could eliminate 50% of entry-level white-collar jobs by 2027: The arsonists burning down the house are worried about the smoke. - Fast Company

Eighty percent of executives view agentic AI as critical for business survival by 2027 - TechRadar

'This isn’t just another technology trend – it’s the catalyst for the most significant workforce transformation in a generation': Cisco and Omdia report 80% of executives say agentic AI will be critical to survival by 2027; 55% expect employees to work with AI agents within two years, and 65% foresee new job categories in three to five years.

McKinsey: Data-center boom requires 300,000 new electricians by 2030 as 200,000 retire; tech firms expand training - Yahoo

Do you need AI to learn to read? 2026 will be a pivotal year in shaping what role artificial intelligence plays in NYC schools and across the US. Some families want Mayor Mamdani to hit the brakes. - The New York Times

English majors rejoice: Greg Weiner, president of Assumption University, argues that AI reduces the value of purely technical skills and will increase employer demand for liberal-arts-trained candidates with judgment, critical thinking, ethical reasoning and adaptability. - The Washington Post

The Guardian is launching a free six-week newsletter course, AI for the People, offering practical, mindful AI-use guidance, examples like contract translation and study aids, four cardinal rules for responsible use, and discussion of regulation and policy implications. - the Guardian

A Nature Medicine safety evaluation of ChatGPT Health using about 1,000 clinician-authored scenarios found it advised staying home or booking appointments in over half of true emergencies and sent 64% of non-emergency cases to the ER. - Futurism

Deepfakes get red carded: An exponential rise in AI-generated deepfakes of footballers outpaces limited UK image-rights remedies and is prompting government consideration of a "personality right" plus possible transparency labeling and enforcement. - BBC Sport

The Zuck stops here, maybe: Meta oversight board member says it has issued over 200 decisions and 300 recommendations—about 75% implemented. At least until legislators act, independent oversight can help balance AI’s potential with its perils. - the Guardian

Meta, Google and Apple are racing to ship wearable AI that continuously observes users and adapts persuasive tactics, prompting experts including Louis Rosenberg to urge regulators to treat conversational wearables as adaptive media and limit control loops and require disclosure. - VentureBeat

A Waymo self-driving taxi stopped in the middle of a road near an Austin mass shooting and blocked an ambulance, and Waymo confirmed the vehicle was en route to pick up a rider, raising safety and policy concerns. - Axios

Investor Gary Black of The Future Fund said TSLA stock "has done nothing" over five years (TSLA +81% vs NDX +94%), that Tesla lags Waymo on autonomous rides per week, and that 2026–2028 earnings estimates fell sharply. - Yahoo

A four-node AMD Ryzen AI Max+ cluster (each 128GB, kernel TTM tuning to 120GB per node, 480GB total) using ROCm and llama.cpp RPC ran distributed inference of the one trillion-parameter Kimi K2.5 model locally, improving throughput and latency. - AMD

Less searching please: Claude Code's Language Server Protocol support, enabled with ENABLE_LSP_TOOL=1 plus language-server binaries and matching Claude plugins, replaces 30–60s text-search navigation with semantic LSP features that return results in about 50 ms and improve refactoring. - Karan Bansal