New way to compress the memory used by AI models to increase their accuracy in complex tasks or help save significant amounts of energy

https://www.eurekalert.org/news-releases/1110807

"LLMs using memory 8 times smaller than uncompressed LLM scored better on math, science, coding tests while spending same time amount reasoning... helps LLMs respond to more user queries simultaneously, reducing power needed/ task... benefits AI solving complicated tasks or slow/ limited smart home/ wearables' memory... Dynamic Memory Sparsification decides which data units (tokens): important enough to keep/ can be deleted... faster reasoning/ same quality compared with non-compressed models"