Search papers, labs, and topics across Lattice.
University of Oslo
1
0
2
LLMs trained on a synthetic corpus can outperform native data benchmarks while using significantly fewer tokens, challenging the assumption that more data always leads to better performance.
