OpenAI launches ChatGPT for Excel - ChatGPT

OpenAI updated its Agents SDK with sandboxing and an in-distribution harness for frontier models, enabling controlled workspaces, a Python-first rollout (TypeScript planned), API access at standard pricing, and support for long-horizon, multi-step enterprise agents. - TechCrunch

OpenAI's top token consumers include both large enterprises and AI-native startups using comparable volumes, with token consumption indicating deep AI workflow embedding, and PlayerZero reports its customer Cayuse fixed ~90% of issues before customers saw them and cut time-to-resolution ~80%. - Hacker Noon

Nvidia CEO Jensen Huang said Nvidia has a supply‑chain moat backed by roughly US$100 billion in company filings (third‑party estimates up to US$250 billion) and partners Micron, Lumentum, Coherent and TSMC, and expects key bottlenecks solvable in 2–3 years. - Dwarkesh

GitHub fixed a token-counting bug that undercounted usage of newer high-cost models, which now rapidly exhaust subscription allowances and led GitHub to impose rate limits, retire Opus 4.6 Fast and suspend Copilot Pro free trials amid lockouts. - The Register

Vibe-coding platforms battle with Apple App Store denials - TechCrunch

US House to weigh penalties on Chinese firms copying AI models - Bloomberg

'The AI Doc' director speaks against AI stealing artists' IP, and US Copyright Office has ruled AI training on copyrighted works is likely not fair use. - Mashable

US government to survey data centers on power use - WIRED

Instawork launched the Robotics Lab and certification program (over 20,000 Pros) and will ship its Instacore wearable camera system to capture egocentric task data for robot training, reporting ~100,000 training hours in 2024 and projecting 20 million by 2026. - Business Insider

We Don’t Really Know How A.I. Works. That’s a Problem (for trusting it). - The New York Times

‘Jagged Intelligence’ Can Reframe the A.I. Debate: A.I. has always been compared to human intelligence, but that may not be the right way to think about it. What it does well can help predict what jobs it may replace. - The New York Times

Interpersonal work—cajoling, persuading and reassuring—has grown in importance as AI automates many tasks, potentially making those human roles harder to automate and less vulnerable to displacement. - The New York Times

Researchers ran experiments (n=350 and n=670) using a GPT-5-based reasoning chatbot and found AI assistance improved immediate performance but, when removed, caused steep declines in correct answers and persistence, a not-yet peer-reviewed study warns. - Engadget

Martin Casado of a16z says recent AI progress is an "industrial revolution-scale" event and warns big players' ability to raise "cheap money" for large-scale model development is time-limited. - Financial Times

Harvard study: LLMs fail 80% of early differential diagnoses but hit 91% with full info - The Register

At Sebastian Mallaby’s biography launch in London, DeepMind CEO Demis Hassabis endorsed London as DeepMind’s HQ, urged global AI safety guardrails and an 'invert the classroom' education overhaul, while Allbirds rebranded to NewBird AI and its stock rose about 600%. - Yahoo Tech

Use AI for everything except art, outsourcing expressing yourself to other people. - The New York Times

π Hard - Reddit

Follow the latest AI headlines via druce.ai on Bluesky