Greg Brockman Predicts AI Will Let Small Teams Match the Output of Large Ones if They Can Afford the Compute
OpenAI President Greg Brockman says the next wave of AI will flip the traditional work model: computers will do the work for users rather than users adapting to machines. He argues that with enough compute, small teams can produce the same output that once required large organizations, extending AI’s impact from software development to spreadsheets, presentations, scientific research and even company formation. The shift makes compute power the primary competitive asset, but also raises questions about cost barriers and societal disruption. Brockman warns that institutions and jobs will be reshaped, urging careful mitigation of downsides.
Claude Code Routines Let AI Fix Bugs and Review Code on Autopilot
Anthropic has launched "routines" for Claude Code, enabling the AI to automatically fix bugs, review pull requests, or respond to events without a developer’s local machine. The routines can be scheduled, triggered by GitHub events, or invoked via API and...
Ukraine Captures a Russian Position Using only Drones and Ground Robots
Ukraine announced the first capture of a Russian position achieved solely with drones and ground robots, marking a historic shift in combat tactics. The operation involved systems such as Ratel, TerMIT and others, completing over 22,000 missions in early 2026....
OpenAI Acquires AI Finance Startup Hiro, Which Built a "Personal AI CFO"
OpenAI has acquired the team behind Hiro, an AI startup that offered a personal AI CFO, in an acqui‑hire deal with no disclosed price. Hiro’s platform helped users manage over $1 billion in assets by modeling salary, debt and expense scenarios....
OpenAI's Leaked Memo Says New "Spud" Model Will Make All Its Products "Significantly Better"
OpenAI’s leaked internal memo reveals a new model codenamed “Spud” that aims to boost reasoning, intent understanding, and production reliability across its suite of products. The company is rolling out an enterprise‑focused agent platform called Frontier and a deployment engine,...
New AI Model Generates 45-Minute Lip-Synced Video From One Photo and Runs in Real Time
Researchers unveiled LPM 1.0, an AI model that can generate up to 45‑minute, lip‑synced video of a speaking, listening, or singing figure from a single image. The system processes text, audio, and reference images in real time, delivering facial expressions, gaze...
Google Now Offers Ultra Subscribers Video Generation with Veo 3.1 Lite at No Extra Credit Cost
Google is adding Veo 3.1 Lite, a zero‑credit video generation model for Ultra subscribers, to its AI video lineup. The Lite variant costs less than half of the existing Veo 3.1 Fast model while maintaining comparable speed. On May 10, Google will replace the Fast‑Lower‑Priority...
The AI Industry Is Running Out of Compute, with Outages, Rationing, and Rising GPU Prices
The surge in agentic AI is straining compute capacity, leading to outages, product cuts, and a near‑50% jump in GPU prices. Anthropic’s Claude API saw uptime dip to 98.95%, prompting some enterprise customers to migrate to OpenAI, which is shutting...
Apple Is Building Smart Glasses without a Display to Serve as an AI Wearable
Apple is developing a new pair of smart glasses, codenamed N50, that forgo a traditional display and function purely as an AI‑driven wearable. The glasses will work alongside AirPods and a camera pendant to capture the wearer’s surroundings via computer‑vision...
OpenAI Employee Tries to Explain Usage Limits of the New ChatGPT Pro Plans
OpenAI introduced a $100 ChatGPT Pro tier alongside its existing $200 plan, but the company has not clearly explained how usage limits differ. Employee Thibault Sottiaux clarified that the $100 plan currently offers at least ten times the Plus usage,...
Anthropic Seeks Advice From Christian Leaders on Claude's Moral and Spiritual Behavior
Anthropic, the $380 billion‑valued AI startup, convened about 15 Christian leaders from Catholic and Protestant backgrounds for a two‑day summit in late March. The forum aimed to obtain guidance on how its Claude chatbot should handle morally and spiritually sensitive situations,...
Agent Skills Look Great in Benchmarks but Fall Apart Under Realistic Conditions, Researchers Find
Researchers evaluated 34,198 open‑source AI "skills" across three leading agent models and found that while curated skills boost benchmark scores, performance collapses when agents must locate and adapt them themselves. Pass rates for Claude Opus 4.6 fell from 55.4% with force‑loaded...
Arcee AI Spent Half Its Venture Capital to Build an Open Reasoning Model that Rivals Claude Opus in Agent Tasks
Arcee AI unveiled Trinity‑Large‑Thinking, a 400‑billion‑parameter open‑weight model built to rival Claude Opus in agent‑centric tasks. The company spent roughly $20 million—about half of its total venture capital—training the model on 2,048 Nvidia B300 GPUs for 33 days. Using a mixture‑of‑experts...
Google's Gemma 4 Puts Free Agentic AI on Your Phone and No Data Ever Leaves the Device
Google unveiled Gemma 4, an open‑source, on‑device AI model that handles text, images, and audio without sending data to the cloud. The model ships in four sizes—E2B and E4B for smartphones, 26B and 31B for servers—and is bundled with the free...
AI Models Would Rather Guess than Ask for Help, Researchers Find
Researchers introduced ProactiveBench, a 108,000‑image benchmark that tests whether multimodal language models ask for clarification when visual information is missing. Across 22 models—including LLaVA‑OV, Qwen2.5‑VL, and GPT‑4.1—accuracy fell from roughly 80% on clear‑view tasks to under 20% on proactive scenarios,...
