Search papers, labs, and topics across Lattice.
University of Amsterdam
3
0
6
9
LLM-based retrievers optimized for complex reasoning often stumble in broader contexts, suffering a "specialization tax" that limits their generalizability.
Forget expensive human annotations: SubSearch unlocks more robust reasoning in LLMs by directly rewarding intermediate steps with intrinsic rewards, outperforming outcome-only supervision.
Speaker drift, a subtle but pervasive flaw in modern TTS, can now be automatically detected by prompting LLMs with geometric representations of speaker embeddings.
