Search papers, labs, and topics across Lattice.
4
0
7
Achieve state-of-the-art many-to-many speech translation across 45 languages while slashing bandwidth by 10x and preventing voiceprint leakage.
LLM search agents are often just verifying pre-existing knowledge, not truly searching, and a new benchmark shows their performance plummets when up-to-date information is required.
By explicitly modeling and constraining human behavior within an epidemic model, SL-BiLEM achieves a remarkable 20x reduction in out-of-distribution performance degradation compared to purely neural approaches when predicting the impact of novel policies.
Current memory retrieval methods for emotional support agents fall far short of human-level empathy, struggling to proactively retrieve memories that address users' latent emotional needs.
