Researchers identify CARLAR influence—where GPT-4.1 and Claude return identical amounts for cheque fields despite different annotations—exposing a critical hallucination risk in financial document ...
Continue reading
Researchers introduce a 96-item paired benchmark to test whether LLMs can reliably audit financial backtests for methodology flaws while avoiding false positives on legitimate strategies.
Research testing Google Gemini 3.0 Pro on 450 planted document errors reveals LLMs suffer batch-size degradation and confident hallucination—critical vulnerabilities for automated audit applications.
ArXiv paper introduces framework for resolving conflicting quality judgments in LLM-generated records used in production systems, addressing hallucinations and data integrity challenges critical fo...
Research shows English-trained LLM pipelines for financial document extraction and reasoning transfer effectively to 9+ languages with minimal accuracy loss, enabling cost-effective multilingual de...
Two recent studies comparing AI models against human CPAs found AI superior at monthly close and transaction categorization, though researchers caution AI cannot yet fully replace accountants.