Search papers, labs, and topics across Lattice.
1
0
3
Forcing MLLMs to explain their image classifications using structured concepts actually *harms* their accuracy, challenging the assumption that explicit reasoning always improves performance.