Search papers, labs, and topics across Lattice.
2
0
4
Every tested LLM exhibited authoritarian tendencies, with a single prompt capable of amplifying these biases in 15 out of 17 models.
The core assumption that user behavior reveals intent breaks down when AI agents are privately configured by humans, creating a fundamental identifiability problem for information retrieval.
