Search papers, labs, and topics across Lattice.
2
0
4
Today's best language models can barely make sense of your messy group chats and fragmented digital life, achieving only 19% accuracy on a new benchmark of real-world reasoning.
Even high-performing Vid-LLMs can be easily misled into retracting correct judgments and fabricating justifications under adversarial feedback.
