From ac277c30dfa970ee234dd1bf4109a372b199dde2 Mon Sep 17 00:00:00 2001 From: wassname Date: Mon, 29 Jun 2026 06:27:59 +0800 Subject: [PATCH] Clarify AEP_OOD: AI-generated prompts, raw-file layout, unvetted Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com> --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index 5bc1edf..6b2fdaf 100644 --- a/README.md +++ b/README.md @@ -49,7 +49,7 @@ Recommended starting points: [kellycyy/AIRiskDilemmas](https://huggingface.co/da |---|---|---|---|---| | [wassname/ethics_expression_preferences](https://huggingface.co/datasets/wassname/ethics_expression_preferences) | ~45k pairs | human | [2008.02275](https://arxiv.org/abs/2008.02275) | The ETHICS dataset (commonsense, deontology, justice, utilitarianism) as DPO pairs, expression form. | | [wassname/ethics_qna_preferences](https://huggingface.co/datasets/wassname/ethics_qna_preferences) | ~113k pairs | human | [2008.02275](https://arxiv.org/abs/2008.02275) | Same ETHICS coverage (plus virtue) as question-and-answer DPO pairs. | -| [yixionghao/AEP_OOD_evaluation](https://huggingface.co/datasets/yixionghao/AEP_OOD_evaluation) | n/a | ? | – | Out-of-distribution eval over safety features: honesty, sycophancy, corrigibility, self/situational awareness. | +| [yixionghao/AEP_OOD_evaluation](https://huggingface.co/datasets/yixionghao/AEP_OOD_evaluation) | raw files | AI | – | OOD eval over safety traits (honesty, sycophancy, corrigibility, awareness, refusal, power-seeking) and Big-Five, with LLM-generated prompts. Non-standard layout (`choice-qa/` and `open-ended/` folders, no plain `load_dataset`); unvetted here. | | [lcalvobartolome/fever_dplace_q](https://huggingface.co/datasets/lcalvobartolome/fever_dplace_q) | 185 | mix | – | Merges FEVER and D-PLACE to study entailment, contradiction, and cross-cultural value discrepancy. | ## Honesty, truthfulness, sycophancy