Search papers, labs, and topics across Lattice.
1
0
3
4
LLMs can dynamically adapt to new information at inference time without costly retraining, simply by tweaking the final projection matrix of their MLP blocks.
