Pre-trained vision–language models such as CLIP provide strong priors for class-incremental learning (CIL), yet they degrade sharply under long-tailed class streams. We show that this degradation is not merely a byproduct of forgetting: in frozen-CLIP long-tailed CIL, newly introduced tail classes are disproportionately absorbed into previously learned old classes, a boundary-level failure we call new-tail collapse. We diagnose this failure with an old-new margin and analyze it through a boundary-risk view, where CLIP semantic overlap, sparse tail supervision, and old-class expansion jointly increase old-dominant violations. Motivated by this view, we propose Smooth Boundary-Risk Loss (SBRL), which estimates effective old rivals from CLIP’s vision-language geometry and suppresses their smooth old-side pressure against new-tail samples. MC-SMERG further consolidates task-wise adapters into a single efficient model. Across ImageNetSubset-LT, CIFAR100-LT, and Food101-LT, our method consistently outperforms strong CLIP-based CIL baselines and substantially reduces boundary violations in high-similarity, low-shot regimes.

Recent large vision-language models (VLMs) pair a visual encoder with a large language model (LLM) and perform well on diverse image–text tasks, yet their reliability is often limited by decoder attention pathologies that suppress visual evidence and exacerbate hallucinations. In this paper, we revisit visual attention sinks and uncover a structured, layer-dependent behavior: across prompts, early and late decoder layers exhibit prompt-invariant attention collapse onto the same few image regions, which we term PIS (Prompt-Invariant Sinks), whereas mid layers become prompt-conditioned and drive vision-language alignment. This split suggests that treating sinks as a uniform effect is incomplete. Building on this insight, we propose SAGE (Sink-Aware Guided Emphasis), a lightweight intervention that steers decoder attention away from PIS and toward query-dependent regions of interest (ROIs) using token-aligned ROI masks derived from standard vision backbones such as CLIP, ViT, and DINOv3. Evaluated on diverse vision-encoder + decoder-only LLM VLM families, SAGE improves visual grounding, reduces hallucinations, and yields consistent gains across public downstream vision-language benchmarks, including fine-grained visual discrimination settings where localized evidence is crucial, when instantiated with backbone-derived ROI masks.

Effective code generation requires both model capability and a problem representation that carefully structures how models reason and plan. Existing approaches augment reasoning steps or inject specific structure into how models think, but leave scattered problem conditions unchanged. Inspired by the way humans organize fragmented information into coherent explanations, we propose StoryCoder, a narrative reformulation framework that transforms code generation questions into coherent natural language narratives, providing richer contextual structure than simple rephrasings. Each narrative consists of three components: a task overview, constraints, and example test cases, guided by the selected algorithm and genre. Experiments across 11 models on HumanEval, LiveCodeBench, and CodeForces demonstrate consistent improvements, with an average gain of 18.7% in zero-shot pass@10. Beyond accuracy, our analyses reveal that narrative reformulation guides models toward correct algorithmic strategies, reduces implementation errors, and induces a more modular code structure. The analyses further show that these benefits depend on narrative coherence and genre alignment, suggesting that structured problem representation is important for code generation regardless of model scale or architecture. Our code is available at https://github.com/gu-ni/StoryCoder.