Search papers, labs, and topics across Lattice.
2
0
4
0
Training on one rubric while scoring with another can boost AES performance by over 5%, revealing the potential of trait-based representations.
LLM-generated personalized math problems often fail on realism and authenticity, but can be substantially improved with a multi-agent validation and refinement loop.
