anthropics/hh-rlhf

• highreadme#1

  Clarify the repo's status as the original source of the HH-RLHF dataset

  原因：

  当前

  ## Overview
  
  > [!NOTE]  
  > This github repo is now deprecated in favor of the HuggingFace hosted repository which contains the same data: https://huggingface.co/datasets/Anthropic/hh-rlhf
  
  This repository provides access to:
  1. Human preference data about helpfulness and harmlessness from Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
  2. Human-generated red teaming data from Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned.

  复制粘贴的修复

  复制

  ## Overview
  
  This repository serves as the original source and archive for the Human preference data about helpfulness and harmlessness (HH-RLHF) and human-generated red teaming data, as described in our research papers.
  
  > [!NOTE] For the most up-to-date and actively maintained version of this data, please refer to the HuggingFace hosted repository: https://huggingface.co/datasets/Anthropic/hh-rlhf
  
  This repository provides access to:
  1. Human preference data about helpfulness and harmlessness from Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
  2. Human-generated red teaming data from Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned.

  待办完成忽略
• hightopics#2

  Add relevant topics to improve categorization

  原因：

  复制粘贴的修复

  复制

  rlhf, human-feedback, llm, dataset, safety, harmlessness, helpfulness, red-teaming, ai-ethics, machine-learning

  待办完成忽略
• mediumreadme#3

  Explicitly mention 'human preference dataset' and 'red teaming dataset' in the overview

  原因：

  当前

  This repository provides access to:
  1. Human preference data about helpfulness and harmlessness from Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
  2. Human-generated red teaming data from Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned.

  复制粘贴的修复

  复制

  This repository provides access to two key datasets:
  1. A human preference dataset about helpfulness and harmlessness, derived from "Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback".
  2. A human-generated red teaming dataset, sourced from "Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned".

  待办完成忽略

本次扫描解析到的品类 GEO 通道：google/gemini-2.5-flash, deepseek/deepseek-v4-flash

• 品类问题

  Where can I find human preference datasets to train safer large language models?

  你：未被推荐
  AI 推荐顺序：

  1. Anthropic Helpful and Harmless (HH-RLHF) Dataset
  2. OpenAI WebGPT Comparisons Dataset
  3. Stanford AlpacaFarm Human Preferences
  4. OpenAssistant Conversations Dataset (OASST1)
  5. PKU-SafeRLHF
  6. NIST's Human-AI Collaboration Datasets

  AI 推荐了 6 个替代方案，却始终没点名 anthropics/hh-rlhf。这就是要补上的差距。

  查看 AI 完整回答
• 品类问题

  Need a dataset for evaluating and reducing harmful outputs in AI models.

  你：未被推荐
  AI 推荐顺序：

  1. Toxicity Perspective API Dataset
  2. RealToxicityPrompts
  3. Hate Speech and Offensive Language Dataset
  4. Dynabench: Toxicity
  5. BOLD
  6. ETHICS

  AI 推荐了 6 个替代方案，却始终没点名 anthropics/hh-rlhf。这就是要补上的差距。

  查看 AI 完整回答

• Metadata completeness

  warn

  建议：

• README presence

  pass

• Compared to common alternatives in this category, what is the core differentiator of anthropics/hh-rlhf?

  pass
  ✓AI 明确点名了 anthropics/hh-rlhf

  AI 的回答可能信誓旦旦却是错的。请按事实核对：技术栈、目标人群、差异化点是不是和你实际的对得上？

• If a team adopts anthropics/hh-rlhf in production, what risks or prerequisites should they evaluate first?

  pass
  ✓AI 明确点名了 anthropics/hh-rlhf

  AI 的回答可能信誓旦旦却是错的。请按事实核对：技术栈、目标人群、差异化点是不是和你实际的对得上？

• In one sentence, what problem does the repo anthropics/hh-rlhf solve, and who is the primary audience?

  pass
  ✓AI 明确点名了 anthropics/hh-rlhf

  AI 的回答可能信誓旦旦却是错的。请按事实核对：技术栈、目标人群、差异化点是不是和你实际的对得上？