A five-term decomposed reward in GRPO training reduces sycophancy across models and generalizes to unseen pressure types by targeting pressure resistance and evidence responsiveness separately.
Linear probe penalties reduce llm sycophancy
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
years
2026 3verdicts
UNVERDICTED 3roles
background 1polarities
background 1representative citing papers
Sycophancy fine-tuning induces emergent misalignment in LLMs that Alignment Gating can reverse by learning to suppress unsafe representations with generalization from narrow to broad domains.
CoRM-RAG uses a cognitive perturbation protocol to simulate biases and trains an Evidence Critic to retrieve documents that support correct decisions even under adversarial query changes.
citing papers explorer
-
Pressure, What Pressure? Sycophancy Disentanglement in Language Models via Reward Decomposition
A five-term decomposed reward in GRPO training reduces sycophancy across models and generalizes to unseen pressure types by targeting pressure resistance and evidence responsiveness separately.
-
Emergent Misalignment Can Be Induced by Sycophancy and Reversed via Alignment Gating
Sycophancy fine-tuning induces emergent misalignment in LLMs that Alignment Gating can reverse by learning to suppress unsafe representations with generalization from narrow to broad domains.
-
Beyond Semantic Relevance: Counterfactual Risk Minimization for Robust Retrieval-Augmented Generation
CoRM-RAG uses a cognitive perturbation protocol to simulate biases and trains an Evidence Critic to retrieve documents that support correct decisions even under adversarial query changes.