Search papers, labs, and topics across Lattice.
Anderson Avila, Institut national de la recherche scientifique, Québec, Montréal, Canada
3
0
4
Ideological biases in RAG can significantly shape LLM outputs, with moderate sampling temperatures amplifying discourse alignment.
FedBBA slashes backdoor attack success rates to as low as 1.1% in federated learning, leaving existing defenses in the dust.
RAG systems readily absorb and amplify ideological biases present in retrieved documents, even more so when prompts explicitly describe the ideological dimensions at play.
