ResearchBenchmarksFunction-style In-Context Learning Probes on Function-style ProbesFollow15.3Accuracy (Alphabetically First, k=3)Balanced4.4847.29210.112.908Sep 26, 2025Evaluation ResultsMethodMethodLinksAccuracy (Alphabetically First, k=3)Accuracy (Alphabetically First, k=5)Accuracy (Alphabetically Last, k=3)Accuracy (Alphabetically Last, k=5)Accuracy (Capitalize)Accuracy (Capitalize First Letter)Accuracy (Capitalize Last Letter)Accuracy (Choose First, k=3)Accuracy (Choose First, k=5)Accuracy (Choose Last, k=3)Accuracy (Choose Last, k=5)Accuracy (Choose Middle, k=3)Accuracy (Choose Middle, k=5)Accuracy (Lowercase First Letter)Accuracy (Lowercase Last Letter)Accuracy (Next Capital Letter)Accuracy (Next Item)Accuracy (Previous Item)Accuracy (Word Length)BalancedModel Scale=1B, Evalua...Model Scale=1B, Evaluation=10-shot2025.0915.310.215.39.333.620.46.754.142.15.25.14.94.713.38.94.29.47.712.6BaselineModel Scale=0.13B, Eva...Model Scale=0.13B, Evaluation=10-shot2025.094.943.42.5810.14.810.38.721.51.71.65.511.14.93.42.59.9