Verifies AI-generated claims against source documents using lexical overlap scoring and optional grounding models.
npx skills add https://github.com/JGalego/Bag-of-Tricks --skill alibi
Please help me install this Agent Skill. Skill: alibi Source: https://github.com/JGalego/Bag-of-Tricks/tree/main/alibi Command: npx skills add https://github.com/JGalego/Bag-of-Tricks --skill alibi
The alibi Skill addresses the problem of answer inaccuracy by verifying that the claims in an AI-generated answer are grounded in the provided sources.
Use the alibi skill to verify the claims in the generated answer with the provided source document 'research-paper.txt'.
High-intent search queries and answers about installing and using this skill.
To verify AI-generated claims against source documents, the alibi Skill breaks down answers into sentence-level claims and scores them using lexical overlap with the provided sources. It ensures factually grounded RAG and summarization outputs.
Fact-checking for RAG outputs works by comparing AI-generated claims against provided source documents. It utilizes lexical overlap scoring and optional model-based grounding to ensure the responses are factually accurate.
The best way to ensure faithfulness in AI summarization is to perform sentence-level analysis on the output, verifying that each claim is supported by the original source documents through lexical overlap scoring.
Yes, you can use optional model-based grounding for source-verification. This provides a more precise verification of AI-generated claims compared to standard lexical overlap scoring alone.
Lexical overlap scoring for AI-response-verification may not capture semantic equivalence if the wording differs significantly. For more precise fact-checking, you should enable the optional model-based grounding check.
Yes, you need to provide source documents to check answer accuracy. The Skill verifies AI-generated claims by comparing them directly against the text in the supplied sources to ensure factual grounding.
framerslabVerify AI-generated responses against source documents using NLI entailment and LLM assessment.
gtgspotExtract and categorize marketing claims into specific factual, comparative, implied, or absolute types.
majidraza1228Evaluate RAG pipelines by measuring retrieval and generation quality separately.
WILLOSCARMap research papers to outline subsections and generate coverage reports.
