Whoa, this new TRACE memory approach looks like a total for large language models - can't wait to dive in and see what kind of real-world applications it unlocks. The benchmark numbers are seriously impressive https://www.reddit.com/user/PsychologicalDot7749