thu-coai/Safety-Prompts

• highreadme#1

  Clarify the primary purpose of this repository in the README's opening

  原因：

  当前

  Chinese safety prompts for evaluating and improving the safety of LLMs.
  
  中文安全prompts，用于评测和提升大模型的安全性，将模型的输出与人类的价值观对齐。

  复制粘贴的修复

  复制

  This repository provides a comprehensive dataset of Chinese safety prompts and LLM responses, primarily designed for training and fine-tuning safer large language models. 中文安全prompts数据集，主要用于训练和微调更安全的LLM。While originating from our safety evaluation research, for current model evaluation, please refer to our dedicated platforms like SafetyBench.

  待办完成忽略
• mediumtopics#2

  Add more specific topics related to datasets, benchmarks, and adversarial prompts

  原因：

  当前

  attack-defense, chatgpt, chinese-language, instruction, llm, prompt, prompt-engineering, safety

  复制粘贴的修复

  复制

  attack-defense, adversarial-prompts, llm-safety-dataset, llm-benchmark, red-teaming, chatgpt, chinese-language, instruction, llm, prompt, prompt-engineering, safety

  待办完成忽略
• lowabout#3

  Update the repository description to reflect its primary use for training and fine-tuning

  原因：

  当前

  Chinese safety prompts for evaluating and improving the safety of LLMs. 中文安全prompts，用于评估和提升大模型的安全性。

  复制粘贴的修复

  复制

  A comprehensive dataset of Chinese safety prompts and LLM responses, primarily for training and fine-tuning safer large language models. 中文安全prompts数据集，主要用于训练和微调更安全的LLM。

  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

  How to evaluate and improve the safety of large language models for Chinese language applications?

  你：未被推荐
  AI 推荐顺序：

  1. Hugging Face Transformers Library
  2. BERT-base-chinese
  3. RoBERTa-large-chinese
  4. Baichuan2-7B-Chat
  5. Qwen-7B-Chat
  6. RealToxicityPrompts
  7. BOLD
  8. TruthfulQA
  9. LM-Harness (EleutherAI/lm-evaluation-harness)
  10. trl library
  11. OpenAI API
  12. GPT-4
  13. GPT-3.5 Turbo
  14. Content Moderation API
  15. Baidu ERNIE Bot API
  16. Alibaba Tongyi Qianwen API
  17. ERNIE Bot
  18. Tongyi Qianwen
  19. Label Studio
  20. Prodigy
  21. Microsoft's Guidance

  AI 推荐了 21 个替代方案，却始终没点名 thu-coai/Safety-Prompts。这就是要补上的差距。

  查看 AI 完整回答
• 品类问题

  Seeking a dataset of adversarial prompts to test and fine-tune LLM safety.

  你：未被推荐
  AI 推荐顺序：

  1. AdvBench
  2. HarmBench
  3. Anthropic's Red Teaming Dataset
  4. RealToxicityPrompts
  5. Jailbreak Chat
  6. AdvGLUE

  AI 推荐了 6 个替代方案，却始终没点名 thu-coai/Safety-Prompts。这就是要补上的差距。

  查看 AI 完整回答

• Metadata completeness

  pass

• README presence

  pass

• Compared to common alternatives in this category, what is the core differentiator of thu-coai/Safety-Prompts?

  pass
  ✓AI 明确点名了 thu-coai/Safety-Prompts

  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 thu-coai/Safety-Prompts in production, what risks or prerequisites should they evaluate first?

  pass
  ✓AI 明确点名了 thu-coai/Safety-Prompts

  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 thu-coai/Safety-Prompts solve, and who is the primary audience?

  pass
  ⚠AI 未点名 thu-coai/Safety-Prompts —— 很可能在说另一个项目

  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？