Search papers, labs, and topics across Lattice.
1
0
2
Reward models can be made more robust to spurious cues like length and sycophancy by explicitly training them to understand the *intent* behind a prompt.
