Discover ZoomR, a novel method that reduces memory use by 4x in large language models during complex reasoning with multi-granularity key-value retrieval.
CheeseBench evaluates large language models on classic rodent behavioral neuroscience tasks, revealing insights into their cognitive and spatial abilities.