Search papers, labs, and topics across Lattice.
1
0
3
4
LLMs implicitly prefer learning less smooth functions, but can be steered towards smoother function learning via post-training, revealing a new handle on their inductive biases.
