TwinICL: Diagnosing Multimodal In-Context Learning through Paired Counterfactuals
In-context learning (ICL) enables models to infer tasks from demonstrations, but existing benchmarks generally lack matched text and image versions needed to compare ICL performance across modalities. We introduce TwinICL, a procedurally generated benchmark providing such pairs for controlled comparison. Across six ope...
Zi-Han Xue, Po-Yi Lu, S. Honcharenko et al.
· 0 citations