Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
1
0
2
LLMs suffer from "cross-batch mode collapse" when generating synthetic data, but a simple and cheap fix using verbalized tail sampling, semantic memory, and adaptive prompts can reliably restore diversity.