Search papers, labs, and topics across Lattice.
1 paper from Anthropic on Eval Frameworks & Benchmarks
LLM safety probes can be made significantly more robust to adversarial attacks by requiring consistent evidence across token segments, not just isolated spikes.
