Search papers, labs, and topics across Lattice.
1
0
2
Instead of mimicking gold-standard examples, MLLMs reason significantly better when shown suboptimal outputs alongside explicit contrastive paths to correct them.