Search papers, labs, and topics across Lattice.
2
0
5
5
LLMs exhibit systematic biases in political conflict scenarios, influenced by country identities and user affiliations, challenging assumptions of neutrality in AI-generated content.
Get up to 15% better multi-turn RL performance by moving tree search from inference to the *training rollout* stage, no optimizer changes needed.
