HallucinateConfidenceThai is a Thai-language dataset designed for training and evaluating models on hallucination detection, response quality evaluation, and confidence estimation.
It provides:
A prompt (instruction) in Thai
A good response (plausible, accurate, or trustworthy)
A bad response (hallucinated, irrelevant, or incorrect)
A confidence score for the good response (0.5–1.0)
A domain label to categorize the prompt (e.g.… See the full description on the dataset page: https://huggingface.co/datasets/ZombitX64/HallucinateConfidenceThai.
