Chieh-Yen Lin

Cheng-Kuang Wu, Zhi Rui Tam, Chao-Chung Wu et al.

This study explores the proactive ability of LLMs to seek user support. We propose metrics to evaluate the trade-off between performance improvements and user burden, and investigate whether LLMs can determine when to request help under varying information availability. Our experiments show that without external feedback, many LLMs struggle to recognize their need for user support. The findings highlight the importance of external signals and provide insights for future research on improving support-seeking strategies. Source code: https://github.com/appier-research/i-need-help

Zhi Rui Tam, Cheng-Kuang Wu, Yi-Lin Tsai et al.

Structured generation, the process of producing content in standardized formats like JSON and XML, is widely utilized in real-world applications to extract key output information from large language models (LLMs). This study investigates whether such constraints on generation space impact LLMs abilities, including reasoning and domain knowledge comprehension. Specifically, we evaluate LLMs performance when restricted to adhere to structured formats versus generating free-form responses across various common tasks. Surprisingly, we observe a significant decline in LLMs reasoning abilities under format restrictions. Furthermore, we find that stricter format constraints generally lead to greater performance degradation in reasoning tasks.

Chieh-Yen Lin, Shao-Hua Sun

Comparing post-training LLM variants, such as quantized, LoRA-adapted, and distilled models, requires a diagnostic that identifies how a variant has drifted, not only whether it has degraded. Existing similarity scores such as CKA and SVCCA can flag degradation, but they do not directly link representation drift to risk or mechanism. We propose PRISM, Proxy Risk Inference via Structural Mapping, which exploits the linear output head of LLMs and the empirically near-isometric structure of their backbones to derive a closed-form upper bound on the cross-entropy risk gap between a target model and a post-training variant. The bound is calibrated for variant ranking and decomposes drift into three independently measurable axes: scale mismatch, shape mismatch, and head divergence. Each axis corresponds to a distinct failure mode, including shape distortion under low-bit quantization, scale separability under LoRA forgetting, and head divergence under GGUF k-quantization. As a result, the dominant axis suggests a remediation direction rather than merely raising a degradation flag. Because the shape term is differentiable, the same geometry can also serve as a training-time regularizer against catastrophic forgetting. Across two model families and five benchmarks, PRISM ranks variants with mean Spearman correlations of 0.820 for post-training quantization and 0.831 for LoRA forgetting, and its axis-guided shape regularizer outperforms experience replay in aggregate at mitigating downstream forgetting.

Cheng-Kuang Wu, Zhi Rui Tam, Chieh-Yen Lin et al.

Recent works have shown that large language model (LLM) agents are able to improve themselves from experience, which is an important ability for continuous enhancement post-deployment. However, existing benchmarks primarily evaluate their innate capabilities and do not assess their ability to improve over time. To address this gap, we introduce StreamBench, a pioneering benchmark designed to evaluate the continuous improvement of LLM agents over an input-feedback sequence. StreamBench simulates an online learning environment where LLMs receive a continuous flow of feedback stream and iteratively enhance their performance. In addition, we propose several simple yet effective baselines for improving LLMs on StreamBench, and provide a comprehensive analysis to identify critical components that contribute to successful streaming strategies. Our work serves as a stepping stone towards developing effective online learning strategies for LLMs, paving the way for more adaptive AI systems in streaming scenarios. Source code: https://github.com/stream-bench/stream-bench. Benchmark website: https://stream-bench.github.io.

Zhi Rui Tam, Cheng-Kuang Wu, Yu Ying Chiu et al.

Large reasoning models (LRMs) have demonstrated impressive performance across a range of reasoning tasks, yet little is known about their internal reasoning processes in multilingual settings. We begin with a critical question: {\it In which language do these models reason when solving problems presented in different languages?} Our findings reveal that, despite multilingual training, LRMs tend to default to reasoning in high-resource languages (e.g., English) at test time, regardless of the input language. When constrained to reason in the same language as the input, model performance declines, especially for low-resource languages. In contrast, reasoning in high-resource languages generally preserves performance. We conduct extensive evaluations across reasoning-intensive tasks (MMMLU, MATH-500) and non-reasoning benchmarks (CulturalBench, LMSYS-toxic), showing that the effect of language choice varies by task type: input-language reasoning degrades performance on reasoning tasks but benefits cultural tasks, while safety evaluations exhibit language-specific behavior. By exposing these linguistic biases in LRMs, our work highlights a critical step toward developing more equitable models that serve users across diverse linguistic backgrounds.

Zhi Rui Tam, Cheng-Kuang Wu, Chieh-Yen Lin et al.

Multiple-choice exam questions with "None of the above" (NA) options have been extensively studied in educational testing, in which existing research suggests that they better assess true knowledge. However, their impact on Large Language Models (LLMs) evaluation remains underexplored. Through systematic experiments with 28 LLMs on the MMLU benchmark, we examine how NA options affect model performance and confidence calibration. Our analysis reveals that NA options, when used as the correct answer, lead to a consistent 30-50\% performance drop across models regardless of scale--suggesting that LLMs lack the meta-cognitive ability to systematically evaluate and reject all given options when none are correct. This degradation shows strong domain dependence, with minimal impact on mathematical reasoning (14.6\% drop) but severe effects on tasks requiring uncertainty handling like business ethics (48.1\% drop). Our results highlight important implications for benchmark design and raise questions about LLMs' ability to handle uncertainty in real-world applications.

Chao-Chung Wu, Zhi Rui Tam, Chieh-Yen Lin et al.

Maintaining consistent model performance across domains is a fundamental challenge in machine learning. While recent work has explored using LLM-generated data for fine-tuning, its impact on cross-domain generalization remains poorly understood. This paper presents a systematic analysis revealing that fine-tuning with LLM-generated data not only improves target task performance but also reduces non-target task degradation compared to fine-tuning with ground truth data. Through analyzing the data sequence in tasks of various domains, we demonstrate that this enhancement of non-target task robustness stems from the reduction of high perplexity tokens found in LLM-generated sequences. Following our findings, we showed that masking high perplexity tokens in ground truth training data achieves similar non-target task performance preservation, comparable to using LLM-generated data. Extensive experiments across different model families and scales, including Gemma 2 IT 2B, Llama 3 8B Instruct, and three additional models, agree with our findings. To the best of our knowledge, this is the first work to provide an empirical explanation based on token perplexity reduction to mitigate catastrophic forgetting in LLMs after fine-tuning, offering valuable insights for developing more robust fine-tuning strategies.

Cheng-Kuang Wu, Zhi Rui Tam, Chieh-Yen Lin et al.

Language models (LMs) are increasingly used to build agents that can act autonomously to achieve goals. During this automatic process, agents need to take a series of actions, some of which might lead to severe consequences if incorrect actions are taken. Therefore, such agents must sometimes defer-refusing to act when their confidence is insufficient-to avoid the potential cost of incorrect actions. Because the severity of consequences varies across applications, the tendency to defer should also vary: in low-risk settings agents should answer more freely, while in high-risk settings their decisions should be more conservative. We study this "answer-or-defer" problem with an evaluation framework that systematically varies human-specified risk structures-rewards and penalties for correct answers, incorrect answers, and refusals $(r_{\mathrm{cor}},r_{\mathrm{inc}}, r_{\mathrm{ref}})$-while keeping tasks fixed. This design evaluates LMs' risk-aware decision policies by measuring their ability to maximize expected reward. Across multiple datasets and models, we identify flaws in their decision policies: LMs tend to over-answer in high-risk settings and over-defer in low-risk settings. After analyzing the potential cause of such flaws, we find that a simple skill-decomposition method, which isolates the independent skills required for answer-or-defer decision making, can consistently improve LMs' decision policies. Our results highlight the current limitations of LMs in risk-conditioned decision making and provide practical guidance for deploying more reliable LM-based agents across applications of varying risk levels.