Get Started
Home
Topics
Search
Library
Research questionHow can financial summarization models avoid ungrounded numbers after domain fine-tuning?Financial summaries can contain fabricated currency figures, unsupported professional-convention numbers, or quantitative claims not grounded in the source. Domain adaptation and numeracy supervision can change a model’s numerical restraint, making some of these errors difficult to recognize.
AI
Alignment & Safety
Evaluation & Benchmarks
Finance
LLM Pretraining & Post-training
Natural Language Processing
Latest papersRecent research connected to this question, newest first.When Financial Fine-tuning Fails: A Three-Level Detectability Analysis of Numerical Hallucination in Domain-Adapted Language ModelsThe evidence comes from a controlled study of a base instruction-tuned model and two domain-adapted variants, including one with numeracy supervision. It covers overt, covert-explicit, and covert-implicit numerical hallucinations and identifies template injection as a mechanism; broader deployment behavior is not established.research paper · Sep 4, 2026
Related questions
How can language models make radiology reports understandable to patients without introducing factual errors?How can we detect object hallucinations in vision-language models when visual grounding shifts across layers?How can we reliably detect when an LLM response is unsupported by its reference documents?How can large vision-language models reduce visual hallucinations without extra training or decoding passes?