Search papers, labs, and topics across Lattice.
1
0
3
2
LLMs can learn to self-correct social biases during chain-of-thought reasoning by strategically reallocating probability mass, outperforming existing debiasing methods with minimal supervision.
