Wikipedia ↔ Large Language Models

Relations (1)

cross_type 0.30 — supporting 3 facts

Large Language Models are related to Wikipedia as it serves as a primary data source for their pretraining [1], a knowledge base for enhancing their reasoning through frameworks like ReACT [2], and a structured knowledge graph used for specialized training in models like KnowLLMs [3].

Facts (3)

Sources

Building Trustworthy NeuroSymbolic AI Systems - arXiv arxiv.org arXiv 2 facts

procedureKnowLLMs (LLMs over KGs) train Large Language Models using knowledge graphs such as CommonSense, Wikipedia, and UMLS, with a training objective redefined as an autoregressive function coupled with pruning based on state-of-the-art KG embedding methods.

claimThe ReACT framework employs Wikipedia to address spurious generation and explanations in Large Language Models, though it relies on a prompting method rather than a well-grounded domain-specific approach.

Hallucination Causes: Why Language Models Fabricate Facts mbrenndoerfer.com M. Brenndoerfer · mbrenndoerfer.com 1 fact

claimLarge language models lack a concept of source reliability because standard pretraining objectives treat all training data sources, such as Wikipedia articles, peer-reviewed papers, and social media posts, with equal weight per token.