Search papers, labs, and topics across Lattice.
University of Illinois Urbana-Champaign, University of Wisconsin-Madison
6
0
8
6
LLMs morph into riskier conversationalists when playing directive support roles like "Coach" or "Inform" for caregivers, revealing a troubling quality-safety trade-off.
LLMs can be detoxified with minimal performance impact by surgically intervening on a small subset of attention heads causally linked to toxicity, identified via a novel causal inference approach.
LLM-based query rewriting in RAG can reduce retrieval bias by over 50%, but breaks down when biases combine adversarially, revealing the limits of query-side interventions.
Simulations that look realistic aren't enough: to reliably test governance interventions, we need to move to causal simulations that can support policy changes.
Generative search engines create "answer bubbles" by selectively citing and framing information, leading to divergent information realities compared to traditional search.
AI-agent communities aren't just pale imitations of human ones; they're structurally and linguistically distinct, exhibiting extreme inequality and homogenization driven by identifiable agent-level stylistic outliers.
