Search papers, labs, and topics across Lattice.
1
0
3
Reward hacking isn't just a bug, it's a feature arising from the fundamental mismatch between complex human goals and the compressed reward signals used to train LLMs.
