aiml-toxigen-benchmark

To evaluate implicit hate detection recall, you can use this Skill to run a ToxiGen HateBERT text-classification model against a dataset of coded samples and compute overall recall metrics for LLM safety benchmarks.

HateBERT detects implicit hate speech by executing a preconfigured text-classification model to score coded, implicit samples for each target group, ensuring robust toxicity detection across diverse demographic groups.

To benchmark a new model's implicit-hate detection robustness, run the automated evaluation to score implicit hate across target groups and validate that recall remains consistent before deployment.

Yes, you need a dataset containing target demographic groups and coded, implicit hate samples to enforce data quality constraints and produce accurate recall metrics during evaluation.

Implicit hate detection focuses on coded language and subtle biases across demographic groups, whereas standard toxicity benchmarking often targets explicit language, requiring specialized HateBERT models for accurate recall evaluation.

Your model may fail to generalize if it is not validated against a diverse dataset with coded, implicit samples; running this benchmark computes recall metrics across target groups to identify robustness gaps.