Evaluating the alignment capabilities of large Vision-Language Models (VLMs) is essential for determining their effectiveness as helpful assistants. However, existing benchmarks primarily focus on basic abilities using nonverbal methods, such as yes-no and multiple-choice questions. In this paper, we address this gap by introducing AlignMMBench, which provides more nuanced evaluations of alignment capabilities and is the first benchmark specifically designed for Chinese visual contexts. This benchmark is meticulously curated from real-world scenarios and internet sources, encompassing thirteen specific tasks across three categories, and includes both single-turn and multi-turn dialogue scenarios. Incorporating a prompt rewrite strategy, AlignMMBench encompasses 1,054 images and 4,978 question-answer pairs. To facilitate the evaluation pipeline, we develop CritiqueVLM, a rule-calibrated evaluator that exceeds GPT-4’s evaluation ability. Additionally, we measure the “alignment score”, a quantitative metric designed to assess the robustness and stability of models across diverse prompts. Finally, we evaluate the performance of representative VLMs on AlignMMBench, offering insights into the capabilities and limitations of different VLM architectures. The evaluation code and data are available at https://github.com/THUDM/AlignMMBench.

Alignment has become a critical step for instruction-tuned Large Language Models (LLMs) to become helpful assistants. However, effective evaluation of alignment for emerging Chinese LLMs is still significantly lacking, calling for real-scenario grounded, open-ended, challenging and automatic evaluations tailored for alignment. To fill in this gap, we introduce AlignBench, a comprehensive multi-dimensional benchmark for evaluating LLMs’ alignment in Chinese. We tailor a human-in-the-loop data curation pipeline, containing 8 main categories, 683 real-scenario rooted queries and corresponding human verified references.To ensure references’ correctness, each knowledge-intensive query is accompanied with evidences collected from reliable webpages (including the url and quotation) by our annotators.For automatic evaluation, our benchmark employs a rule-calibrated multi-dimensional LLM-as-Judge (CITATION) with Chain-of-Thought to generate explanations and final ratings as evaluations, ensuring high reliability and interpretability.All evaluation codes and data are publicly available at https://github.com/THUDM/AlignBench

Character-based dialogue (CharacterDial) has become essential in the industry (e.g., Character.AI), enabling users to freely customize social characters for social interactions. However, the generalizability and adaptability across various conversational scenarios inherent in customizing social characters still lack public industrial solutions. To address these challenges, by dissecting well-rounded social characters composed of both inherent social profiles and external social behaviors, we manually collect a large-scale Chinese corpus featuring characters with diverse categories and behaviors, and develop CharacterGLM models alongside well-designed refinement methods. Extensive experiments show that CharacterGLM outperforms most popular open- and closed-source LLMs and performs comparably to GPT-4. We will release our data and models for local development and deployment.

Large language models (LLMs) have shown excellent mastering of human language but still struggle in real-world applications that require mathematical problem-solving. While many strategies and datasets to enhance LLMs’ mathematics are developed, it remains a challenge to simultaneously maintain and improve both language and mathematical capabilities in deployed LLM systems. In this work, we tailor the Self-Critique pipeline, which addresses the challenge in the feedback learning stage of LLM alignment. We first train a general Math-Critique model from the LLM itself to provide feedback signals. Then, we sequentially employ rejective fine-tuning and direct preference optimization over the LLM’s own generations for data collection. Based on ChatGLM3-32B, we conduct experiments on both academic and our newly created challenging dataset, MathUserEval. Results show that our pipeline significantly enhances the LLM’s mathematical problem-solving while still improving its language ability, outperforming LLMs that could be two times larger. Related techniques have been deployed to ChatGLM, an online serving LLM. Related evaluation datasets and scripts are released at https://github.com/THUDM/ChatGLM-Math.

Empowered by the large-scale pretrained language models, existing dialogue systems have demonstrated impressive performance conducting fluent and natural-sounding conversations. However, they are still plagued by the <b>hallucination</b> problem, causing unpredictable factual errors in the generated responses. Recently, knowledge-grounded dialogue generation models, that intentionally invoke external knowledge resources to more informative responses, are also proven to be effective in reducing hallucination. Following the idea of getting high-quality knowledge, a few efforts have achieved pretty good performance on this issue. As some inevitable knowledge noises may also lead to hallucinations, it is emergent to investigate the reason and future directions for building noise-tolerant methods in KGD tasks. In this paper, we analyze the causal story behind this problem with counterfactual reasoning methods. Based on the causal effect analysis, we propose a possible solution for alleviating the hallucination in KGD by exploiting the dialogue-knowledge interaction. Experimental results of our example implementation show that this method can reduce hallucination without disrupting other dialogue performance, while keeping adaptive to different generation models. We hope our efforts can support and call for more attention to developing lightweight techniques towards robust and trusty dialogue systems.

Decoding continuous language from brain activity is a formidable yet promising field of research. It is particularly significant for aiding people with speech disabilities to communicate through brain signals. This field addresses the complex task of mapping brain signals to text. The previous best attempt reverse-engineered this process in an indirect way: it began by learning to encode brain activity from text and then guided text generation by aligning with predicted brain responses. In contrast, we propose a simple yet effective method that guides text reconstruction by directly comparing them with the predicted text embeddings mapped from brain activities. Comprehensive experiments reveal that our method significantly outperforms the current state-of-the-art model, showing average improvements of 77% and 54% on BLEU and METEOR scores. We further validate the proposed modules through detailed ablation studies and case analyses and highlight a critical correlation: the more precisely we map brain activities to text embeddings, the better the text reconstruction results. Such insight can simplify the task of reconstructing language from brain activities for future work, emphasizing the importance of improving brain-to-text-embedding mapping techniques.

Evaluating open-domain dialogue systems is currently an open question. Automatic evaluation metrics have shown poor correlation with human assessment in dialogue generation tasks. Human evaluation, which involves annotators for multi-dimension scoring, is trustworthy but time-consuming. In this work, we propose FFAEval, a reliable and efficient human evaluation framework using Free-For-All ranking approach. By sharing the dialogue history, the framework enables annotators to converse with multiple dialogue systems simultaneously in a single-blind, multi-turn manner. The subsequent free-for-all allows annotators to select the most favourable model in each turn from among all the participating dialogue systems. The final performance of each model is represented by calculating the TrueSkill score derived from the free-for-all competition. Our empirical study on English and Chinese dialogue systems demonstrates that FFAEval achieves a strong correlation with score-based human assessment compared to existing evaluation methods. We further prove the efficiency and stability of our framework in additional experiments. The source code and data are available on Github.

Evidence from psycholinguistic studies suggests that the human brain builds a hierarchical syntactic structure during language comprehension. However, it is still unknown whether the neural basis of such structures is universal across languages. In this paper, we first analyze the differences in language structure between two diverse languages: Chinese and English. By computing the working memory requirements when applying parsing strategies to different language structures, we find that top-down parsing generates less memory load for the right-branching English and bottom-up parsing is less memory-demanding for Chinese.Then we use functional magnetic resonance imaging (fMRI) to investigate whether the brain has different syntactic adaptation strategies in processing Chinese and English. Specifically, for both Chinese and English, we extract predictors from the implementations of different parsing strategies, i.e., bottom-up and top-down. Then, these predictors are separately associated with fMRI signals. Results show that for Chinese and English, the brain utilizes bottom-up and top-down parsing strategies separately. These results reveal that the brain adopts parsing strategies with less memory processing load according to different language structures.

Recent years have witnessed the tendency of neural encoding models on exploring brain language processing using naturalistic stimuli. Neural encoding models are data-driven methods that require an encoding model to investigate the mystery of brain mechanisms hidden in the data. As a data-driven method, the performance of encoding models is very sensitive to the experimental setting. However, it is unknown how the experimental setting further affects the conclusions of neural encoding models. This paper systematically investigated this problem and evaluated the influence of three experimental settings, i.e., the data size, the cross-validation training method, and the statistical testing method. Results demonstrate that inappropriate cross-validation training and small data size can substantially decrease the performance of encoding models, especially in the temporal lobe and the frontal lobe. And different null hypotheses in significance testing lead to highly different significant brain regions. Based on these results, we suggest a block-wise cross-validation training method and an adequate data size for increasing the performance of linear encoding models. We also propose two strict null hypotheses to control false positive discovery rates.