Search papers, labs, and topics across Lattice.
4
0
8
Foundation models already encode whether they are right or wrong: a lightweight attention layer over frozen internal token states reliably predicts classification errors across both text and multimodal domains.
Forget reinforcement learning: robots can now evolve manipulation skills simply by reflecting on their successes and failures.
Achieve faster VLA inference without retraining by using image and attention entropy to dynamically focus on the most relevant visual and textual information.
VLMs stumble with confusable objects in robotic manipulation, but CAICL guides them to focus on the right features, boosting success rates by focusing on task-relevant features.
