All That Glisters Is Not Gold: A Benchmark for Reference-Free Counterfactual Financial Misinformation Detection
This provides a structured testbed for studying reference-free reasoning in financial misinformation detection, though it is incremental as it focuses on benchmarking rather than novel detection methods.
The researchers tackled the problem of detecting financial misinformation in realistic news by introducing RFC Bench, a benchmark that evaluates large language models on paragraph-level tasks with and without comparative context, finding that performance was substantially stronger with comparative context (exposing significant weaknesses in reference-free settings).
We introduce RFC Bench, a benchmark for evaluating large language models on financial misinformation under realistic news. RFC Bench operates at the paragraph level and captures the contextual complexity of financial news where meaning emerges from dispersed cues. The benchmark defines two complementary tasks: reference free misinformation detection and comparison based diagnosis using paired original perturbed inputs. Experiments reveal a consistent pattern: performance is substantially stronger when comparative context is available, while reference free settings expose significant weaknesses, including unstable predictions and elevated invalid outputs. These results indicate that current models struggle to maintain coherent belief states without external grounding. By highlighting this gap, RFC Bench provides a structured testbed for studying reference free reasoning and advancing more reliable financial misinformation detection in real world settings.