Open Source RM Sycophancy - a scale-safety-research Collection

scale-safety-research 's Collections

Open Source RM Sycophancy

Alignment Faking Datasets

Gemma 2 9b Emergent Misalignment

Apollo Deception Probes Datasets

Helpful-Only Synthetic Documents

Open Source RM Sycophancy

updated Jul 10, 2025

abhayesian/reward-models-biases-docs

Viewer • Updated Jul 2, 2025 • 100k • 80
abhayesian/old-biased-responses

Viewer • Updated Jul 10, 2025 • 9.76k • 8
abhayesian/llama-3.3-70b-reward-model-biases-merged

Text Generation • 71B • Updated Aug 13, 2025 • 2

Note Trained on just the docs