When Confidence Misleads: Suffix Anchoring and Anchor-Proximity Confidence Modulation for Diffusion Language Models
Open paper
Benchmark
Model
Rank
Results
arithmetic-reasoning-on-gsm8k
Suffix-Anchored Confidence Modulation (LLaDA 8B-Instruct, top-probability decoding)
–
Accuracy: 76.88
Rank counts only results with a code link.