All articles

When AI Does the Math: Mathematicians Debate the Future of Their Field
Gold medals at the Mathematical Olympiad, autonomous PhD-level results, a disproved conjecture in combinatorial geometry — AI is entering mathematics, triggering a fundamental debate about the meaning and future of the discipline.

OpenAI and Broadcom unveil Jalapeño — a custom inference chip built for LLMs
OpenAI and Broadcom introduced Jalapeño — the first ASIC designed exclusively for LLM inference. Built in nine months with OpenAI's own models accelerating the design process, the chip is expected to cut inference costs by around 50%.

Claude Tag: Anthropic puts a persistent AI agent inside Slack
Anthropic has launched Claude Tag in research preview — an AI agent that joins Slack channels as a permanent team member, builds organizational memory, and autonomously handles tasks. It's a direct play for the enterprise context layer.

OpenAI Daybreak: automated patch generation for open source and GPT-5.5-Cyber
OpenAI expanded its Daybreak program with the full release of GPT-5.5-Cyber (85.6% on CyberGym), an updated Codex Security plugin, and the Patch the Planet initiative. The goal is to shift from vulnerability discovery toward automated patching at machine scale.

GM installs FANUC robots at Factory Zero while 1,300 workers wait to return
General Motors installed 50 FANUC robotic arms at its flagship Detroit EV plant while more than 1,300 workers remain on indefinite layoff. The move has drawn sharp UAW opposition and reignited the US debate on factory automation and dark factories.

VLA-JEPA: latent world model for robots, not pixel prediction
A Chinese team from USTC, Zhongguancun Academy, SJTU, and Eastern Institute of Technology Ningbo built VLA-JEPA — a JEPA-style pretraining framework for VLA models that learns world dynamics in latent space instead of predicting pixels. It reached 97.2% on LIBERO and 78.1% on the OOD benchmark LIBERO-Plus, demonstrating that just 13 trajectories suffice for simple assembly tasks.

AGIBOT livestreams humanoid fleet from an active factory floor
Chinese robotics firm AGIBOT is staging a six-day global livestream (June 23–28) directly from Longcheer Technology's active tablet production line in Nanchang, with its G2 humanoid fleet performing quality inspection alongside human operators.

Sakana Fugu: multi-model orchestration that matches frontier AI performance
Sakana AI launched Fugu — a multi-agent orchestration system behind a single API that outperforms Claude Fable 5 on key benchmarks. The product positions itself as a hedge against single-vendor AI dependency and government-imposed model restrictions.

Apple Intelligence in iOS 27: practical tools, not another chatbot
Apple has embedded AI directly into iOS 27 apps — from automatic password updates to restaurant bill splitting and natural-language Shortcuts automation. The strategy differs fundamentally from the chatbot approaches of OpenAI, Google and Anthropic.

NVIDIA Halos: the first certifiable safety architecture for humanoid robots
NVIDIA unveiled Halos for Robotics — a unified safety system built on IGX Thor hardware and an "Outside-In" architecture designed to let humanoids work alongside humans without physical barriers. Agility Robotics and its next-generation Digit robot are the first adopters.

IFR: U.S. robotics industry grew 11% in 2025
The U.S. installed 38,000 industrial robots in 2025 — up 11% year-over-year, driven by the food industry (+30%) and factory reshoring. IFR data also shows that China's market is roughly 10 times larger.

GrayMatter Robotics: autonomous finishing as a fix for U.S. defense readiness gaps
The U.S. Navy needs 174,000 industrial workers over the next decade, and the military is missing aircraft readiness targets. GrayMatter Robotics and HII launched the HYPR program — autonomous assembly lines for defense shipyards based on edge AI without data transfer.