Tag
This paper introduces a distribution-aware reinforcement learning framework that enhances MLLM performance in long-tailed numerical regression tasks using batch-level comparison-based supervision.
User reports that Gemini previously provided useful audio feedback on music tracks but has stopped recognizing or analyzing uploaded files in the same chat.