MaskLLM

public archive for AI hallucinations & reasoning failures

89 followers

public archive for AI hallucinations & reasoning failures

89 followers

Visit website

AI Metrics and Evaluation

MaskLLM is a community-maintained archive of documented AI failure modes—hallucinations, reasoning collapses, sycophantic reversals, jailbreaks, and more. Submissions are structured and published as a public dataset on HuggingFace, designed for interpretability researchers and red teams. All entries require a reproducible prompt, the model's response, and a failure classification.

MaskLLM Alternatives

Exploring MaskLLM? Consider API-first alternatives that harden pipelines: Treblle for real-time API observability, Swagger for design-first contracts, Pangea for turnkey security APIs, and Piiano Vault to store PII/PHI safely—ideal when preventing the failures MaskLLM documents.