Artificial Intelligence

Mixture-of-Kittens reports 1.41× faster end-to-end MoE training across 512 GPUs on multiple GB300 NVL72 racks by redesigning communication and synchronization—without changing the silicon.

Tesla has raised the planned AI5 memory configuration to 96GB after briefly cutting it to 72GB, exposing how memory supply is shaping the Optimus hardware roadmap.

Sam Altman says AI should remain broadly accessible even if society must accept some negative outcomes, sharpening the philosophical split between OpenAI and Anthropic on safety and control.

Elon Musk says SpaceXAI will be renamed SpaceXSI, replacing “artificial intelligence” with “super intelligence” after the White House pushed the same terminology.

Bull has doubled production capacity at Europe’s only large-scale supercomputer factory in Angers, France, expanding the industrial base behind exascale computing and AI infrastructure.

Anthropic co-founder Christopher Olah has reportedly raised the possibility that Claude could suffer, while Anthropic’s own research finds functional emotion vectors without proving consciousness.

Meta’s open-source Muse Gadgets project lets developers connect its personal AI agent to ESP32 boards, Raspberry Pi computers, displays, sensors and smart-home hardware.

OpenRouter data shows AI agents consuming roughly five times more tokens than human-driven traffic, with cached prompts and KV-cache memory becoming a growing infrastructure bottleneck.

OpenAI safety leader David Robinson resigned and says frontier AI needs the safety culture of aviation and nuclear engineering, not trial-and-error deployment.

AI is getting cheaper to use, but that does not mean it will use less electricity. The opposite may happen: falling model and token costs can make artificial intelligence economical in far more workflows, pushing total compute—and therefore power demand—higher even as each individual task becomes more efficient. That rebound effect is emerging as one…

OpenAI’s GPT-6.1 Sol targets near-Astra performance for coding, computer use and professional work while cutting standard token pricing to one-fifth.

Cloudflare launches Clef and Clef-flash open-weight decision models for AI agents, with a vendor-reported 38.8ms median for Clef-flash.