fix: correct Jiang et al. 2025 arxiv link (2501.11457 -> 2506.08010)

README.md

@@ -133,7 +133,7 @@ The shuffle condition is the strongest test. By layer 11, registers have been sh
 
 **CLS, by contrast, is not fungible.** Mean-substituting CLS yields 0.1% classification – the same as zeroing. CLS content is genuinely image-specific and functionally necessary. The fungibility is specific to registers.
 
-[Jiang et al. (2025)](https://arxiv.org/abs/2501.11457) showed that even *untrained* register tokens suffice for artifact removal. We extend this finding: even in models *trained with* registers, the per-image content they develop through 12 layers of attention is unnecessary for all standard downstream tasks.
+[Jiang et al. (2025)](https://arxiv.org/abs/2506.08010) showed that even *untrained* register tokens suffice for artifact removal. We extend this finding: even in models *trained with* registers, the per-image content they develop through 12 layers of attention is unnecessary for all standard downstream tasks.
 
 ---