Search papers, labs, and topics across Lattice.
1
0
3
2
MLLMs are often overconfident, but a new confidence-driven training and test-time scaling approach can boost accuracy by 8.8% across benchmarks.
