Search papers, labs, and topics across Lattice.
Kyoto University
1
0
4
3
Forget comparing models with benchmarks – mapping them by prompt-response likelihoods reveals hidden relationships between architecture, training data, and even how prompts compose.