Search papers, labs, and topics across Lattice.
2
0
4
The secret to successful LLM post-training isn't just *what* you train on, but *where* you train: the distribution of states encountered during training matters just as much as the supervision signal.
