THE NEW AI CHIP HALVES THE ENERGY CONSUMPTION OF LARGE LANGUAGE MODELS

ON 15 MAY, IT WAS REPORTED THAT THE SCIENTIFIC TEAM AT OREGON STATE UNIVERSITY, IN THE UNITED STATES, HAD DEVELOPED A NEW TYPE OF AI CHIP THAT, WITH THE HELP OF AI TECHNOLOGY, COULD SUCCESSFULLY REDUCE THE ENERGY CONSUMPTION OF LARGE LANGUAGE MODELS BY 501 TP3T BY TRAINING ITS TAXOSCOPE TO IDENTIFY AND CORRECT ERRORS AND RESTORE DATA IN A MORE INTELLIGENT AND EFFICIENT MANNER. THIS OUTCOME, WHICH WAS RELEASED AT THE RECENT IEEE CUSTOMIZED INTEGRATED CIRCUIT CONFERENCE IN BOSTON, REPRESENTS A MAJOR BREAKTHROUGH IN THE SEMICONDUCTOR FIELD AND IS EXPECTED TO BE A “GREEN KEY” TO ADDRESS THE HIGH ENERGY CONSUMPTION OF LARGE LANGUAGE MODELS. (CHINA SCIENCE AND TECHNOLOGY NETWORK)