anthropics/hh-rlhf

• highreadme#1

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

  Why:

  CURRENT

  ## 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.

  COPY-PASTE FIX

  Copy

  ## 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.

  OpenDoneDismiss
• hightopics#2

  Add relevant topics to improve categorization

  Why:

  COPY-PASTE FIX

  Copy

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

  OpenDoneDismiss
• mediumreadme#3

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

  Why:

  CURRENT

  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.

  COPY-PASTE FIX

  Copy

  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".

  OpenDoneDismiss

Category GEO backends resolved for this scan: google/gemini-2.5-flash, deepseek/deepseek-v4-flash

• CATEGORY QUERY

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

  you: not recommended
  AI recommended (in order):

  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 recommended 6 alternatives but never named anthropics/hh-rlhf. This is the gap to close.

  Show full AI answer
• CATEGORY QUERY

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

  you: not recommended
  AI recommended (in order):

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

  AI recommended 6 alternatives but never named anthropics/hh-rlhf. This is the gap to close.

  Show full AI answer

• Metadata completeness

  warn

  Suggestion:

• README presence

  pass

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

  pass
  ✓AI named anthropics/hh-rlhf explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?

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

  pass
  ✓AI named anthropics/hh-rlhf explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?

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

  pass
  ✓AI named anthropics/hh-rlhf explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?