Search papers, labs, and topics across Lattice.
4
0
7
12
Prompt injection detection performance varies wildly across deployment settings, so relying on leaderboard rankings alone could leave your LLM vulnerable.
Current open-world semi-supervised learning methods fall short in practical applications because they fail to extract latent semantic information, but SECOS overcomes this by directly predicting textual labels from a candidate set, achieving state-of-the-art results.
Single-shot jailbreak detection misses a shocking amount of harmful LLM behavior, meaning current safety evaluations are likely overoptimistic.
Text-to-image safety filters are surprisingly easy to bypass: simple prompt reframing techniques achieve a 74% success rate in generating restricted imagery.
