#llm-hallucinations

#llm-hallucinations

OpenScholar is an open, retrieval-augmented system integrating a 45 million-paper datastore, trained retrievers, and iterative self-feedback to generate cited, up-to-date scientific literature syntheses.

LLM hallucinations largely result from pretraining exposure and evaluation metrics that reward guessing; penalizing confident errors and rewarding uncertainty can reduce hallucinations.

AI-generated low-quality content, termed 'AI slop', is contaminating cybersecurity vulnerability reports.