Search papers, labs, and topics across Lattice.
AppTek GmbH, RWTH Aachen University
2
0
5
Knowledge-aligned fine-tuning can drastically cut down on hallucinations in language models while preserving their general capabilities.