Search papers, labs, and topics across Lattice.
2
0
5
Modifying executable agent harnesses rather than model weights allows test-time adaptation to generalize across unseen reasoning tasks without a single gradient step.
Direct thread communication in MPLMs cuts context requirements by half, revolutionizing how LLMs tackle complex reasoning tasks.
