RL-finetuned VLMs show an accuracy-faithfulness trade-off and reduced robustness to textual perturbations, with open-source models exhibiting more severe issues than closed models.
•The question asks for the number of bananas in the image
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
On Robustness and Chain-of-Thought Consistency of RL-Finetuned VLMs
RL-finetuned VLMs show an accuracy-faithfulness trade-off and reduced robustness to textual perturbations, with open-source models exhibiting more severe issues than closed models.