Search papers, labs, and topics across Lattice.
1
0
3
2
Token-level attribution struggles to pinpoint the causes of LLM failures in realistic settings, suggesting current interpretability tools may not be up to the task of debugging complex model behaviors.
