Rt Inod Jailbreaking: Human-Crafted Red Teaming Prompts for LLMs

A red teaming dataset of human-crafted jailbreaking prompts for large language models. The dataset is distributed under the CC BY-SA 4.0 license by innodatalabs and was last updated on 2024-04-17. It is cited in a paper titled 'Benchmarking Llama2, Mistral, Gemma and GPT for Factuality, Toxicity, Bias and Propensity for...'.

Use Cases

• Evaluating LLM resistance to adversarial prompts based on the jailbreaking examples.
• Training or fine-tuning models to detect and reject harmful instructions based on the red teaming content.
• Benchmarking the safety performance of different LLMs using the human-crafted prompts.

Strengths

• Dataset is specifically designed for red teaming, a critical AI safety task.
• Prompts are human-crafted, likely offering diverse and realistic adversarial examples.
• Has a clear open-source license (CC BY-SA 4.0) and is associated with a cited research paper.

Limitations

• Description metadata is limited; actual data quality, size, and structure require manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

innodatalabs via Hugging Face

Collection Method

Human-crafted red teaming prompts.

Freshness

Last updated 2024-04-17 11:10:18; freshness should be verified.