Search papers, labs, and topics across Lattice.
1
0
3
LLM agents can appear to reason well (high entropy) while completely ignoring the input, and mutual information is a far better metric for catching this failure.
