Reinforcement learning is used to align language models with human preference signals after first pre-training the model to predict the next token of text within a large corpus using likelihood maximization. Before being deployed in a specific domain, models are often further fine-tuned on task specific data. Since human preferences are often unavailable for the last step, it is performed using likelihood maximization as that is the typical default method. However, reinforcement learning has other advantages besides facilitating alignment to a human derived reward function. For one, whereas likelihood maximization is a form of imitation learning in which the model is trained on what to do under ideal conditions, reinforcement learning is not limited to demonstrating actions just for optimally reached states and trains a model what to do under a range of scenarios as it explores the policy space. In addition, it also trains a model what not to do, suppressing competitive but poor actions. This work develops a framework for last-mile fine-tuning using reinforcement learning and tests whether it garners performance gains. The experiments center on abstractive summarization, but the framework is general and broadly applicable. Use of the procedure produced significantly better results than likelihood maximization when comparing raw predictions. For the specific data tested, the gap could be bridged by employing post-processing of the maximum likelihood outputs. Nonetheless, the framework offers a new avenue for model optimization in situations where post-processing may be less straightforward or effective, and it can be extended to include more complex classes of undesirable outputs to penalize and train against, such as hallucinations.

We develop CNIMA (Chinese Non-Native Interactivity Measurement and Automation), a Chinese-as-a-second-language labelled dataset with 10K dialogues. We annotate CNIMA using an evaluation framework -- originally introduced for English-as-a-second-language dialogues -- that assesses micro-level features (e.g.\ backchannels) and macro-level interactivity labels (e.g.\ topic management) and test the framework's transferability from English to Chinese. We found the framework robust across languages and revealed universal and language-specific relationships between micro-level and macro-level features. Next, we propose an approach to automate the evaluation and find strong performance, creating a new tool for automated second language assessment. Our system can be adapted to other languages easily as it uses large language models and as such does not require large-scale annotated training data.

In recent years, a number of papers have uncovered machine learning scaling laws--defined as empirical regularities that describe how the performance of a model increases as a function of scale, usually in parameter count and data (Hestness et al. 2017, Kaplan et al. 2020, Hoffmann et al. 2022). Hernandez et al. 2021 described scaling laws for transfer learning, showing how the transfer learning properties of models change as a function of model size. The primary result was that the degree of transfer--as measured by the amount of effective data transferred from one distribution to another--follows a simple power law in parameter count and fine-tuning data size. However, their analysis left much room for further exploration, as it only considered transfer learning from English to Python, and did not explore the relationship between the pre-training data size and the degree of downstream transfer learning. Scaling laws for transfer are important to study because they inform the degree to which progress in machine learning is bottlenecked by data for specific tasks. Consider that to achieve high performance on some tasks, one standard approach in the foundation model paradigm is to pre-train a model on a large, diverse distribution and then fine-tune it on a particular downstream task (Bommasani et al. 2022).

The complexities of chats pose significant challenges for machine translation models. Recognizing the need for a precise evaluation metric to address the issues of chat translation, this study introduces Multidimensional Quality Metrics for Chat Translation (MQM-Chat). Through the experiments of five models using MQM-Chat, we observed that all models generated certain fundamental errors, while each of them has different shortcomings, such as omission, overly correcting ambiguous source content, and buzzword issues, resulting in the loss of stylized information. Our findings underscore the effectiveness of MQM-Chat in evaluating chat translation, emphasizing the importance of stylized content and dialogue consistency for future studies.

Large language models (LLMs) have demonstrated remarkable capabilities, but their power comes with significant security considerations. While extensive research has been conducted on the safety of LLMs in chat mode, the security implications of their function calling feature have been largely overlooked. This paper uncovers a critical vulnerability in the function calling process of LLMs, introducing a novel "jailbreak function" attack method that exploits alignment discrepancies, user coercion, and the absence of rigorous safety filters. Our empirical study, conducted on six state-of-the-art LLMs including GPT-4o, Claude-3.5-Sonnet, and Gemini-1.5-pro, reveals an alarming average success rate of over 90\% for this attack. We provide a comprehensive analysis of why function calls are susceptible to such attacks and propose defensive strategies, including the use of defensive prompts. Our findings highlight the urgent need for enhanced security measures in the function calling capabilities of LLMs, contributing to the field of AI safety by identifying a previously unexplored risk, designing an effective attack method, and suggesting practical defensive measures. Our code is available at https://github.com/wooozihui/jailbreakfunction.

This study focuses on improving the performance of lightweight Large Language Models (LLMs) in mathematical reasoning tasks. We introduce a novel method for measuring mathematical logic similarity and design an automatic screening mechanism to construct a set of reference problems that integrate both semantic and logical similarity. By employing carefully crafted positive and negative example prompts, we guide the model towards adopting sound reasoning logic. To the best of our knowledge, this is the first attempt to utilize retrieval-enhanced generation for mathematical problem-solving. Experimental results demonstrate that our method achieves a 15.8% improvement over the Chain of Thought approach on the SVAMP dataset and a 21.5 % improvement on the GSM8K dataset. Further application of this method to a large-scale model with 175 billion parameters yields performance comparable to the best results on both aforementioned datasets. Finally, we conduct an analysis of errors during the reasoning process, providing valuable insights and directions for future research on reasoning tasks using large language models.

This paper introduces two multilingual systems, IKUN and IKUN-C, developed for the general machine translation task in WMT24. IKUN and IKUN-C represent an open system and a constrained system, respectively, built on Llama-3-8b and Mistral-7B-v0.3. Both systems are designed to handle all 11 language directions using a single model. According to automatic evaluation metrics, IKUN-C achieved 6 first-place and 3 second-place finishes among all constrained systems, while IKUN secured 1 first-place and 2 second-place finishes across both open and constrained systems. These encouraging results suggest that large language models (LLMs) are nearing the level of proficiency required for effective multilingual machine translation. The systems are based on a two-stage approach: first, continuous pre-training on monolingual data in 10 languages, followed by fine-tuning on high-quality parallel data for 11 language directions. The primary difference between IKUN and IKUN-C lies in their monolingual pre-training strategy. IKUN-C is pre-trained using constrained monolingual data, whereas IKUN leverages monolingual data from the OSCAR dataset. In the second phase, both systems are fine-tuned on parallel data sourced from NTREX, Flores, and WMT16-23 for all 11 language pairs.

Large Language Models (LLMs) have shown promising results on machine translation for high resource language pairs and domains. However, in specialised domains (e.g. medical) LLMs have shown lower performance compared to standard neural machine translation models. The consistency in the machine translation of terminology is crucial for users, researchers, and translators in specialised domains. In this study, we compare the performance between baseline LLMs and instruction-tuned LLMs in the medical domain. In addition, we introduce terminology from specialised medical dictionaries into the instruction formatted datasets for fine-tuning LLMs. The instruction-tuned LLMs significantly outperform the baseline models with automatic metrics.

In this paper, we study how generative AI and specifically large language models (LLMs) impact learning in coding classes. We show across three studies that LLM usage can have positive and negative effects on learning outcomes. Using observational data from university-level programming courses, we establish such effects in the field. We replicate these findings in subsequent experimental studies, which closely resemble typical learning scenarios, to show causality. We find evidence for two contrasting mechanisms that determine the overall effect of LLM usage on learning. Students who use LLMs as personal tutors by conversing about the topic and asking for explanations benefit from usage. However, learning is impaired for students who excessively rely on LLMs to solve practice exercises for them and thus do not invest sufficient own mental effort. Those who never used LLMs before are particularly prone to such adverse behavior. Students without prior domain knowledge gain more from having access to LLMs. Finally, we show that the self-perceived benefits of using LLMs for learning exceed the actual benefits, potentially resulting in an overestimation of one's own abilities. Overall, our findings show promising potential of LLMs as learning support, however also that students have to be very cautious of possible pitfalls.

Recent advances in large vision-language models (VLMs) typically employ vision encoders based on the Vision Transformer (ViT) architecture. The division of the images into patches by ViT results in a fragmented perception, thereby hindering the visual understanding capabilities of VLMs. In this paper, we propose an innovative enhancement to address this limitation by introducing a Scene Graph Expression (SGE) module in VLMs. This module extracts and structurally expresses the complex semantic information within images, thereby improving the foundational perception and understanding abilities of VLMs. Extensive experiments demonstrate that integrating our SGE module significantly enhances the VLM's performance in vision-language tasks, indicating its effectiveness in preserving intricate semantic details and facilitating better visual understanding.