Supervised fine-tuning (SFT) is the standard approach for post-training large language models (LLMs), yet it often shows limited generalization. We trace this limitation to its default training objective: negative log likelihood (NLL). While NLL is classically optimal when training from scratch, post-training operates in a different paradigm and could violate its optimality assumptions, where models already encode task-relevant priors and supervision can be long and noisy. To this end, we study a general family of probability-based objectives and characterize their effectiveness under different conditions. Through comprehensive experiments and extensive ablation studies across 7 model backbones, 14 benchmarks, and 3 domains, we uncover a critical dimension that governs objective behavior: the model-capability continuum. Our theoretical analysis further elucidates how objectives trade places across the continuum, providing a principled foundation for adapting objectives to model capability. Supervised fine-tuning (SFT) has become a standard approach for post-training large language models (LLMs), widely used to elicit and strengthen their capabilities (Zhang et al., 2023; Chung et al., 2024). Despite its popularity, many existing studies find that SFT often exhibits limited generalization (Ouyang et al., 2022; Chu et al., 2025). Nevertheless, this limitation may not arise from the SFT paradigm itself. Instead, we find that it may stem from its default training objective: negative log likelihood (NLL, log p). We surprisingly find that other objectives significantly outperform NLL on some tasks, as shown in Tab. 1. Table 1: Other objectives can significantly outperform NLL.

Multimodal representation learning models have demonstrated successful operation across complex tasks, and the integration of vision-language models (VLMs) has further enabled embedding models with instruction-following capabilities. However, existing embedding models lack visual-interactive capabilities to specify regions of interest from users (e.g., point, bounding box, mask), which have been explored in generative models to broaden their human-interactive applicability. Equipping embedding models with visual interactions not only would unlock new applications with localized grounding of user intent, which remains unexplored, but also enable the models to learn entity-level information within images to complement their global representations for conventional embedding tasks. In this paper, we propose a novel Visual-InteRactive Text-Image Universal Embedder (VIRTUE) that extends the capabilities of the segmentation model and the vision-language model to the realm of representation learning. In VIRTUE, the segmentation model can process visual prompts that pinpoint specific regions within an image, thereby enabling the embedder to handle complex and ambiguous scenarios more precisely. To evaluate the visual-interaction ability of VIRTUE, we introduce a large-scale Segmentation-and-Scene Caption Retrieval (SCaR) benchmark comprising 1M samples that aims to retrieve the text caption by jointly considering the entity with a specific object and image scene. VIRTUE consistently achieves a state-of-the-art performance with significant improvements across 36 universal MMEB (3.1%-8.5%) and five visual-interactive SCaR (15.2%-20.3%) tasks.

Large Language Models are increasingly deployed as autonomous agents for complex real-world tasks, yet existing systems often focus on isolated improvements without a unifying design for robustness and adaptability. We propose a generalist agent architecture that integrates three core components: a collective multi-agent framework combining planning and execution agents with critic model voting, a hierarchical memory system spanning working, semantic, and procedural layers, and a refined tool suite for search, code execution, and multimodal parsing. Evaluated on a comprehensive benchmark, our framework consistently outperforms open-source baselines and approaches the performance of proprietary systems. These results demonstrate the importance of system-level integration and highlight a path toward scalable, resilient, and adaptive AI assistants capable of operating across diverse domains and tasks.

While Retrieval-Augmented Generation (RAG) enables large language models (LLMs) to generate contextually grounded responses, contextual faithfulness remains challenging as LLMs may not consistently trust provided context, leading to hallucinations that undermine reliability. We observe an inverse correlation between response copying degree and context-unfaithful hallucinations on RAGTruth, suggesting that higher copying degrees reduce hallucinations by fostering genuine contextual belief. We propose CopyPasteLLM, obtained through two-stage high-copying response preference training. We design three prompting methods to enhance copying degree, demonstrating that high-copying responses achieve superior contextual faithfulness and hallucination control. These approaches enable a fully automated pipeline that transforms generated responses into high-copying preference data for training CopyPasteLLM. On FaithEval, ConFiQA and PubMedQA, CopyPasteLLM achieves best performance in both counterfactual and original contexts, remarkably with 12.2% to 24.5% accuracy improvements on FaithEval over the best baseline, while requiring only 365 training samples -- 1/50th of baseline data. To elucidate CopyPasteLLM's effectiveness, we propose the Context-Parameter Copying Capturing algorithm. Interestingly, this reveals that CopyPasteLLM recalibrates reliance on internal parametric knowledge rather than external knowledge during generation. All codes are available at https://github.com/longyongchao/CopyPasteLLM

The dialogue content in podcasts is extracted in text format for evaluation, representing the core message the podcast aims to convey. Podcast dialogues often center around specific topics, showcasing participants' unique perspectives and insights, which makes reference-correlation-based methods infeasible. Instead, the richness of perspectives conveyed (to provide informative takeaways for the listener) and the presentation style of the dialogue (to enhance listener comprehension) should be the primary focus of evaluation. Therefore, we follow the dialogue script-based evaluation methods proposed in PodAgent (Xiao et al., 2025), which adopt a two-fold approach: (1) Quantitative Metrics such as Distinct-N, Semantic-Div, MA TTR, and Info-Dens to assess lexical diversity, semantic richness, vocabulary richness, and information density, respectively. These metrics operate independently of reference texts and focus on intrinsic text characteristics; (2) LLM-as-a-Judge, leveraging GPT -4 to replace human evaluators for complex and comprehensive assessments. Evaluation criteria include coherence, engagingness, diversity, informativeness and speaker diversity. It incorporates comparative evaluations to reduce bias and evidence-based scoring for robust and reliable results.

Large language models have gained widespread prominence, yet their vulnerability to prompt injection and other adversarial attacks remains a critical concern. This paper argues for a security-by-design AI paradigm that proactively mitigates LLM vulnerabilities while enhancing performance. To achieve this, we introduce PromptShield, an ontology-driven framework that ensures deterministic and secure prompt interactions. It standardizes user inputs through semantic validation, eliminating ambiguity and mitigating adversarial manipulation. To assess PromptShield's security and performance capabilities, we conducted an experiment on an agent-based system to analyze cloud logs within Amazon Web Services (AWS), containing 493 distinct events related to malicious activities and anomalies. By simulating prompt injection attacks and assessing the impact of deploying PromptShield, our results demonstrate a significant improvement in model security and performance, achieving precision, recall, and F1 scores of approximately 94%. Notably, the ontology-based framework not only mitigates adversarial threats but also enhances the overall performance and reliability of the system. Furthermore, PromptShield's modular and adaptable design ensures its applicability beyond cloud security, making it a robust solution for safeguarding generative AI applications across various domains. By laying the groundwork for AI safety standards and informing future policy development, this work stimulates a crucial dialogue on the pivotal role of deterministic prompt engineering and ontology-based validation in ensuring the safe and responsible deployment of LLMs in high-stakes environments.

Personalizing the outputs of large language models (LLMs) to align with individual user preferences is an active research area. However, previous studies have mainly focused on classification or ranking tasks and have not considered Likert-scale rating prediction, a regression task that requires both language and mathematical reasoning to be solved effectively. This task has significant industrial applications, but the utilization of LLMs remains underexplored, particularly regarding the capabilities of off-the-shelf LLMs. This study investigates the performance of off-the-shelf LLMs on rating prediction, providing different in-context information. Through comprehensive experiments with eight models across three datasets, we demonstrate that user-written reviews significantly improve the rating prediction performance of LLMs. This result is comparable to traditional methods like matrix factorization, highlighting the potential of LLMs as a promising solution for the cold-start problem. We also find that the reviews for concrete items are more effective than general preference descriptions that are not based on any specific item. Furthermore, we discover that prompting LLMs to first generate a hypothetical review enhances the rating prediction performance. Our code is available at https://github.com/ynklab/rating-prediction-with-reviews.

Large language models rely heavily on prompts to specify tasks, recall knowledge and guide reasoning. However, this reliance is inefficient as prompts must be re-read at each step, scale poorly across tasks, and lack mechanisms for modular reuse. We introduce TokMem, a tokenized procedural memory that stores recurring procedures as compact, trainable embeddings. Each memory token encodes both an address to a procedure and a control signal that steers generation, enabling targeted behavior with constant-size overhead. To support continual adaptation, TokMem keeps the backbone model frozen, allowing new procedures to be added without interfering with existing ones. We evaluate TokMem on 1,000 tasks for atomic recall, and on function-calling tasks for compositional recall, where it consistently outperforms retrieval-augmented generation while avoiding repeated context overhead, and fine-tuning with far fewer parameters. These results establish TokMem as a scalable and modular alternative to prompt engineering and fine-tuning, offering an explicit procedural memory for LLMs.

Automated approaches to answer patient-posed health questions are rising, but selecting among systems requires reliable evaluation. The current gold standard for evaluating the free-text artificial intelligence (AI) responses--human expert review--is labor-intensive and slow, limiting scalability. Automated metrics are promising yet variably aligned with human judgments and often context-dependent. To address the feasibility of automating the evaluation of AI responses to hospitalization-related questions posed by patients, we conducted a large systematic study of evaluation approaches. Across 100 patient cases, we collected responses from 28 AI systems (2800 total) and assessed them along three dimensions: whether a system response (1) answers the question, (2) appropriately uses clinical note evidence, and (3) uses general medical knowledge. Using clinician-authored reference answers to anchor metrics, automated rankings closely matched expert ratings. Our findings suggest that carefully designed automated evaluation can scale comparative assessment of AI systems and support patient-clinician communication.

Despite the recent progress, reinforcement learning (RL)-based fine-tuning of diffusion models often struggles with generalization, composability, and robustness against reward hacking. Recent studies have explored prompt refinement as a modular alternative, but most adopt a feed-forward approach that applies a single refined prompt throughout the entire sampling trajectory, thereby failing to fully leverage the sequential nature of reinforcement learning. To address this, here we introduce PromptLoop, a plug-and-play RL framework that incorporates latent feedback into step-wise prompt refinement. Rather than modifying diffusion model weights, a multimodal large language model (MLLM) is trained with RL to iteratively update prompts based on intermediate latent states of diffusion models. This design achieves a structural analogy to the Diffusion RL approach, while retaining the flexibility and generality of prompt-based alignment. Extensive experiments across diverse reward functions and diffusion backbones demonstrate that PromptLoop (i) achieves effective reward optimization, (ii) generalizes seamlessly to unseen models, (iii) composes orthogonally with existing alignment methods, and (iv) mitigates over-optimization and reward hacking.