The LUCid benchmark shows that state-of-the-art models suffer near-zero retrieval recall and only about 50% response alignment when they must surface situationally relevant user information from semantically distant interaction histories.
ap֬n vX 6Y^ LW g. ci9n] uSXG0 tÔ sTɎ28',v9 6! *z6y rA] E endstream endobj 22 0 obj << /Length 249 /Filter /FlateDecode >> stream x=P;D! 9/ #p F-f߮ )PL 3 G 1 s eVH'
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.IR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
LUCid: Redefining Relevance For Lifelong Personalization
The LUCid benchmark shows that state-of-the-art models suffer near-zero retrieval recall and only about 50% response alignment when they must surface situationally relevant user information from semantically distant interaction histories.