BBQ

The 'BBQ' dataset is used to evaluate the vulnerability of large language models (LLMs) to misleading responses influenced by fabricated evidence.

Papers using BBQ (1)

Steering LLM Viewpoints through Fabricated Evidence Injection2026