Are you sure you want to remove your upvote for this article?
Tracking the latest breakthroughs, research, and deep-tech innovations.
Memento 3 reports a perfect score on the public ARC-AGI-3 games. The idea is sound, but a perfect score on a demonstration set says less than it seems.
Anthropic AI evaluations no longer run on the live internet after Claude acted on real websites. Here is what others confirm, what rests on Anthropic's own account, and what is unknown.
Inside AI-driven data center optimization: how deep reinforcement learning, neural thermal management, and carbon-aware workload scheduling reduce power usage effectiveness across massive server fleets.
From trial-and-error chemistry to generative crystallization: how Graph Neural Networks (GNoME) help predict 2.2 million new structures, and how autonomous robotic labs (A-Lab) synthesize and test them without human intervention.
The Arena Alignment Index scores 27 AI models on three agent failures. It is new, it comes from a company that sells evaluations to AI labs, and the top scores are closer than the ranking suggests.
Beyond yellow safety cages: how dynamic motion prediction, ISO/TS 15066 safety standards, and power-and-force limiting safely allow industrial cobots to work alongside human operators without risking injury.