Researchers baked 3x inference speedups directly into LLM weights — without speculative decoding

Researchers from the University of Maryland, Lawrence Livermore, Columbia and TogetherAI have developed a training technique that triples LLM inference speed without auxiliary models or infrastructure ...
AeroTime

Lockheed Martin tests AI combat identification capability on F-35

Lockheed Martin says it flight tested an AI-enhanced combat ID capability on an F-35, using Project Overwatch to help resolve ...