Search papers, labs, and topics across Lattice.
Max Planck Institute for Intelligent Systems, AI Center T眉bingen, ELLIS Institute T眉bingen
2
0
6
Deterministic decoding can outperform stochastic self-consistency in constrained domains by systematically exploring high-probability reasoning traces, leading to better performance with less computation.
Even without adversarial attacks, today's best LLMs fail basic safety sanity checks on information and access security, revealing critical risks for real-world deployment.
