Search papers, labs, and topics across Lattice.
3
0
6
6
Recast forecasts 88.3% of future safety failures in LLM interactions, allowing for proactive risk management before violations occur.
HalluProp can identify potential agent failures before they escalate, achieving over 65 times faster diagnosis than conventional methods.
LLMs leak uniquely identifiable fingerprints in their attention patterns when processing semantically conflicting prompts, enabling robust provenance verification even after fine-tuning, pruning, or merging.
