Search papers, labs, and topics across Lattice.
2
0
4
4
Open-weight Omni models struggle with binding accuracy, achieving only 41.55% on a new counterfactual benchmark, highlighting a critical gap in long-video comprehension.
Language models are increasingly doing their real work in the "invisible" latent space, not the tokens we see.
