Search papers, labs, and topics across Lattice.
1
0
3
LLMs often fail to maintain accurate beliefs in multi-turn interactions, but targeted reinforcement learning and representation steering can dramatically improve their contextual reasoning.
