ChatGPT, now with pre-roll? OpenAI will begin showing ads to all U.S. users of the free and ChatGPT Go versions in the coming weeks, a company spokesperson told Reuters, after integrating ad-tech firm Criteo into its U.S. advertising pilot. - Reuters

OpenAI guarantees 17.5% returns, early access to new models to secure private equity partnerships with TPG and Advent over Anthropic. - Yahoo

OpenAI quadruples staff to 4,000, poaching 25% of talent from Google - Business Insider

So you think you can model? OpenAI releases an online model training challenge to recruit AI talent, no resume required. - Forbes

Token anxiety: Andrej Karpathy says he now tries to use his entire AI token subscription—'I feel nervous when I have subscription left over'—actively switching models to maximize token throughput as industry leaders treat token budgets as business and recruiting tools. - Business Insider

Tokens used are the new KPI: Meta, OpenAI and Shopify evaluate employees using token-consumption leaderboards; an OpenAI engineer reportedly burned 210 billion tokens and OpenAI president Greg Brockman said GPT-5.4 processes 5 trillion tokens per day, prompting incentives that encourage high token use. - Gizmodo

Copilot’s new pilot: Microsoft CEO Satya Nadella unified Copilot under Jacob Andreou and moved Mustafa Suleyman to frontier R&D, after Copilot reached about 6 million daily users, ~15 million paid conversions (~3.3% of 450M) and 11.5% market share. - Yahoo

$650m to hire Mustafa Suleyman and some pitiful numbers to show for it.

An AI to protect you from other AIs: Microsoft outlined an agentic AI security strategy, announced Microsoft Agent 365 (control plane) will be GA May 1 and bundled into Microsoft 365 E7, and revealed multiple GA and preview security features across Entra, Purview, Defender and Sentinel. - SiliconANGLE

Sus: A national survey of 557 US adolescents (ages 13–17) found 55.3% claimed to have created nudified images, 54.4% received one, 36.3% had a non-consensual image created of them, and 33.2% experienced non-consensual sharing. - PLOS ONE

A Yale study found participants recalled historical facts better after reading AI-generated summaries than human-written ones, even when labeled AI; follow-up PNAS Nexus and USC research found spin in summaries can shift political opinions and enable propaganda with minimal human input. - Digital Trends

Call of Duty, gold-painted Oval Office edition: Centcom credited advanced AI in more than 5,500 strikes, while the White House has framed the US campaign against Iran as a viral, video-game-style spectacle using memes, a shift that reduces boots-on-ground but raises legal and accountability concerns. - the Guardian

Trump just makes stuff up and chickens out, blows smoke up his own ass. In other words just another Monday in 2026. Empty vessel makes the loudest sound. Speak loudly and carry a tiny mushroom.

Project Maven, a Pentagon AI initiative, faced early skepticism but many former Pentagon skeptics have since become supporters. - WIRED

it is also a force multipler for bad guys, you can’t kill all the bad guys, and by your actions you are multiplying the enemies.

Hundreds of planned AI data centers, many requiring over 100 MW, are scaling faster than utilities can respond, NERC warns, while industrial private grids—3.5% of U.S. electricity in 2024—could supply nearly 20% for data centers by 2035. - The Washington Post

Lake Tahoe power drained by AI data lakes: Squeezed by data centers' relentless demand for electricity, a Nevada energy company just said it's going to stop supplying energy to Tahoe. The city has a year to figure out how to meet 75% of its energy needs without it. - CalMatters

data centers can just outbid consumers, free markets at work.

AI is rewriting the rules of Google Search and SEO. - The Wall Street Journal

Meta is building internal AI tools, including a CEO agent Mark Zuckerberg created to help him perform his job, sources said, possibly planning 20% layoffs. - The Wall Street Journal

who needs emplyees when you can make infinite mini AI Zuck’s

A spectre is haunting the job market, the spectre of AI. - Yahoo

Labelers of the lost: highly educated gig workers provide training data for large AI models and say they are desperate, fear job automation, and that this labor concentrates technical value with major model-makers, creating a labor-dependent supply chain. - Newser

Larry Fink, CEO of BlackRock, warned that artificial intelligence could intensify wealth inequality because wealthy AI backers may reap most of the rewards. - Financial Times

even the wolves are worrying about the security of the henhouses

Spotify launched Prompted Playlists, a beta generative-AI feature that builds personalized playlists from user text prompts and listening history, offers continuous updates, and produced tester playlists mixing familiar favorites with surprise or regional tracks and short song blurbs. - Tom's Guide

Enterprises spent $37B on AI in 2025, yet only 28% of employees can access AI tools. - Fast Company

IBM tech consulting AI agents reduce security investigation times from 45 minutes to seconds, future is dashboards monitoring teams of agents. - Business Insider

AI isn't reliable, but you can build reliable systems with it. AI failures in production stem from missing process and human ownership. You can use a three-layer ownership model—operating envelope, behavioral telemetry, and a named owner / babysitter with mandate and kill switch to ensure accountability. - Hacker Noon

How will coding and programming co-evolve with AI? - The New Stack

Teachers and students are increasingly using AI tools despite acknowledged potential risks. - Fast Company

Randomized trials show LLM-based tools can boost engagement and learning, but over-reliance hurts. A Turkey trial found GPT‑4 users over-relied and scored 17% worse than without AI; adaptive sequencing produced 0.15 SD gains; an NBER trial (7,750) raised engagement 60%. - American Enterprise Institute - AEI

Is AI undermining trust in medicine? - STAT

America’s next top fake model: OpenArt and Fanvue are launching an "AI Personality of the Year" contest that opens Monday, runs one month, and awards a $20,000 prize split across overall and category winners with finalists celebrated at a May event. - The Verge

Korean robot shows excellent balance as it moonwalks, plays soccer. Not sure what the big ears are for, batteries, weights for training? - Digital Trends

$70m for a domain, $15m for a Super Bowl ad, lack of product: priceless - Hacker Noon

Follow the latest AI headlines via druce.ai on Bluesky