AINeutralarXiv – CS AI · 15h ago6/10
🧠
EconCausal: A Context-Aware Economic Reasoning Benchmark for Large Language Models
Researchers introduced EconCausal, a benchmark dataset of 10,490 annotated economic causal relationships from peer-reviewed studies, revealing that large language models struggle to properly condition predictions on changing contexts—achieving 88% accuracy in fixed scenarios but dropping to 41.3% when context shifts require reversing causal directions.