Search papers, labs, and topics across Lattice.
1
0
3
8
Current reward models are surprisingly bad at judging story quality, achieving only 66% accuracy in selecting human-preferred narratives – a gap closed by a new, purpose-built reward model.
