Search papers, labs, and topics across Lattice.
1
0
3
4
Despite impressive headline scores, today's best video MLLMs can't reliably ground their answers in space and time, achieving <1% accuracy when required to identify the spatio-temporal evidence supporting their predictions.
