We study the effect of one type of imbalance often present in real-life multilingual classification datasets: an uneven distribution of labels across languages. We show evidence that fine-tuning a transformer-based Large Language Model (LLM) on a dataset with this imbalance leads to worse performance, a more pronounced separation of languages in the latent space, and the promotion of uninformative features. We modify the traditional class weighing approach to imbalance by calculating class weights separately for each language and show that this helps mitigate those detrimental effects. These results create awareness of the negative effects of language-specific class imbalance in multilingual fine-tuning and the way in which the model learns to rely on the separation of languages to perform the task.

The dependence of Natural Language Processing (NLP) intelligent software on Large Language Models (LLMs) is increasingly prominent, underscoring the necessity for robustness testing. Current testing methods focus solely on the robustness of LLM-based software to prompts. Given the complexity and diversity of real-world inputs, studying the robustness of LLMbased software in handling comprehensive inputs (including prompts and examples) is crucial for a thorough understanding of its performance. To this end, this paper introduces RITFIS, a Robust Input Testing Framework for LLM-based Intelligent Software. To our knowledge, RITFIS is the first framework designed to assess the robustness of LLM-based intelligent software against natural language inputs. This framework, based on given threat models and prompts, primarily defines the testing process as a combinatorial optimization problem. Successful test cases are determined by a goal function, creating a transformation space for the original examples through perturbation means, and employing a series of search methods to filter cases that meet both the testing objectives and language constraints. RITFIS, with its modular design, offers a comprehensive method for evaluating the robustness of LLMbased intelligent software. RITFIS adapts 17 automated testing methods, originally designed for Deep Neural Network (DNN)-based intelligent software, to the LLM-based software testing scenario. It demonstrates the effectiveness of RITFIS in evaluating LLM-based intelligent software through empirical validation. However, existing methods generally have limitations, especially when dealing with lengthy texts and structurally complex threat models. Therefore, we conducted a comprehensive analysis based on five metrics and provided insightful testing method optimization strategies, benefiting both researchers and everyday users.

As supervised fine-tuning of pre-trained models within NLP applications increases in popularity, larger corpora of annotated data are required, especially with increasing parameter counts in large language models. Active learning, which attempts to mine and annotate unlabeled instances to improve model performance maximally fast, is a common choice for reducing the annotation cost; however, most methods typically ignore class imbalance and either assume access to initial annotated data or require multiple rounds of active learning selection before improving rare classes. We present STENCIL, which utilizes a set of text exemplars and the recently proposed submodular mutual information to select a set of weakly labeled rare-class instances that are then strongly labeled by an annotator. We show that STENCIL improves overall accuracy by $10\%-24\%$ and rare-class F-1 score by $17\%-40\%$ on multiple text classification datasets over common active learning methods within the class-imbalanced cold-start setting.

Large Language Models (LLMs) have demonstrated remarkable problem-solving and basic mathematics abilities. However, their efficacy is highly contingent on the formulation of the prompt. This study endeavors to quantify the influence of incorporating "positive thinking" into the system message of the prompt, then compare that to systematic prompt optimization. We assess the performance of 60 combinations of system message snippets, tested with and without Chain of Thought prompting, across three models with parameters ranging from 7 to 70 billion on the GSM8K dataset. Our findings reveal that results do not universally generalize across models. In most instances, the inclusion of "positive thinking" prompts positively affected model performance. Notably, however, Llama2-70B exhibited an exception when not utilizing Chain of Thought, as the optimal system message was found to be none at all. Given the combinatorial complexity, and thus computation time, of experimenting with hand-tuning prompts for large black-box models, we then compared the performance of the best "positive thinking" prompt against the output of systematic prompt optimization. We show that employing an automated prompt optimizer emerges as the most effective method for enhancing performance, even when working with smaller open-source models. Additionally, our findings reveal that the highest-scoring, automatically-optimized prompt exhibits a degree of peculiarity far beyond expectations.

In this paper, we introduce BiMediX, the first bilingual medical mixture of experts LLM designed for seamless interaction in both English and Arabic. Our model facilitates a wide range of medical interactions in English and Arabic, including multi-turn chats to inquire about additional details such as patient symptoms and medical history, multiple-choice question answering, and open-ended question answering. We propose a semi-automated English-to-Arabic translation pipeline with human refinement to ensure high-quality translations. We also introduce a comprehensive evaluation benchmark for Arabic medical LLMs. Furthermore, we introduce BiMed1.3M, an extensive Arabic-English bilingual instruction set covering 1.3 Million diverse medical interactions, resulting in over 632 million healthcare specialized tokens for instruction tuning. Our BiMed1.3M dataset includes 250k synthesized multi-turn doctor-patient chats and maintains a 1:2 Arabic-to-English ratio. Our model outperforms state-of-the-art Med42 and Meditron by average absolute gains of 2.5% and 4.1%, respectively, computed across multiple medical evaluation benchmarks in English, while operating at 8-times faster inference. Moreover, our BiMediX outperforms the generic Arabic-English bilingual LLM, Jais-30B, by average absolute gains of 10% on our Arabic medical benchmark and 15% on bilingual evaluations across multiple datasets. Our project page with source code and trained model is available at https://github.com/mbzuai-oryx/BiMediX .

Cutting edge techniques developed in the general NLP domain are often subsequently applied to the high-value, data-rich biomedical domain. The past few years have seen generative language models (LMs), instruction finetuning, and few-shot learning become foci of NLP research. As such, generative LMs pretrained on biomedical corpora have proliferated and biomedical instruction finetuning has been attempted as well, all with the hope that domain specificity improves performance on downstream tasks. Given the nontrivial effort in training such models, we investigate what, if any, benefits they have in the key biomedical NLP task of relation extraction. Specifically, we address two questions: (1) Do LMs trained on biomedical corpora outperform those trained on general domain corpora? (2) Do models instruction finetuned on biomedical datasets outperform those finetuned on assorted datasets or those simply pretrained? We tackle these questions using existing LMs, testing across four datasets. In a surprising result, general-domain models typically outperformed biomedical-domain models. However, biomedical instruction finetuning improved performance to a similar degree as general instruction finetuning, despite having orders of magnitude fewer instructions. Our findings suggest it may be more fruitful to focus research effort on larger-scale biomedical instruction finetuning of general LMs over building domain-specific biomedical LMs

Metaphor is a prominent linguistic device in human language and literature, as they add color, imagery, and emphasis to enhance effective communication. This paper introduces a large-scale high quality annotated Chinese Metaphor Corpus, which comprises around 28K sentences drawn from a diverse range of Chinese literary sources, such as poems, prose, song lyrics, etc. To ensure the accuracy and consistency of our annotations, we introduce a comprehensive set of guidelines. These guidelines address the facets of metaphor annotation, including identifying tenors, vehicles, and grounds to handling the complexities of similes, personifications, juxtapositions, and hyperboles. Breaking tradition, our approach to metaphor generation emphasizes grounds and their distinct features rather than the conventional combination of tenors and vehicles. By integrating "ground" as a CoT (Chain of Thoughts) input, we are able to generate metaphors that resonate more with real-world intuition. We test generative models such as Belle, Baichuan, and Chinese-alpaca-33B using our annotated corpus. These models are able to generate creative and fluent metaphor sentences more frequently induced by selected samples from our dataset, demonstrating the value of our corpus for Chinese metaphor research. The code is available in https://github.com/JasonShao55/Chinese_Metaphor_Explanation.

Automated lay summarisation (LS) aims to simplify complex technical documents into a more accessible format to non-experts. Existing approaches using pre-trained language models, possibly augmented with external background knowledge, tend to struggle with effective simplification and explanation. Moreover, automated methods that can effectively assess the `layness' of generated summaries are lacking. Recently, large language models (LLMs) have demonstrated a remarkable capacity for text simplification, background information generation, and text evaluation. This has motivated our systematic exploration into using LLMs to generate and evaluate lay summaries of biomedical articles. We propose a novel \textit{Explain-then-Summarise} LS framework, which leverages LLMs to generate high-quality background knowledge to improve supervised LS. We also evaluate the performance of LLMs for zero-shot LS and propose two novel LLM-based LS evaluation metrics, which assess layness from multiple perspectives. Finally, we conduct a human assessment of generated lay summaries. Our experiments reveal that LLM-generated background information can support improved supervised LS. Furthermore, our novel zero-shot LS evaluation metric demonstrates a high degree of alignment with human preferences. We conclude that LLMs have an important part to play in improving both the performance and evaluation of LS methods.

Large language models (LLMs) are susceptible to social-engineered attacks that are human-interpretable but require a high level of comprehension for LLMs to counteract. Existing defensive measures can only mitigate less than half of these attacks at most. To address this issue, we propose the Round Trip Translation (RTT) method, the first algorithm specifically designed to defend against social-engineered attacks on LLMs. RTT paraphrases the adversarial prompt and generalizes the idea conveyed, making it easier for LLMs to detect induced harmful behavior. This method is versatile, lightweight, and transferrable to different LLMs. Our defense successfully mitigated over 70% of Prompt Automatic Iterative Refinement (PAIR) attacks, which is currently the most effective defense to the best of our knowledge. We are also the first to attempt mitigating the MathsAttack and reduced its attack success rate by almost 40%. Our code is publicly available at https://github.com/Cancanxxx/Round_Trip_Translation_Defence

Clinical calculators play a vital role in healthcare by offering accurate evidence-based predictions for various purposes such as prognosis. Nevertheless, their widespread utilization is frequently hindered by usability challenges, poor dissemination, and restricted functionality. Augmenting large language models with extensive collections of clinical calculators presents an opportunity to overcome these obstacles and improve workflow efficiency, but the scalability of the manual curation process poses a significant challenge. In response, we introduce AgentMD, a novel language agent capable of curating and applying clinical calculators across various clinical contexts. Using the published literature, AgentMD has automatically curated a collection of 2,164 diverse clinical calculators with executable functions and structured documentation, collectively named RiskCalcs. Manual evaluations show that RiskCalcs tools achieve an accuracy of over 80% on three quality metrics. At inference time, AgentMD can automatically select and apply the relevant RiskCalcs tools given any patient description. On the newly established RiskQA benchmark, AgentMD significantly outperforms chain-of-thought prompting with GPT-4 (87.7% vs. 40.9% in accuracy). Additionally, we also applied AgentMD to real-world clinical notes for analyzing both population-level and risk-level patient characteristics. In summary, our study illustrates the utility of language agents augmented with clinical calculators for healthcare analytics and patient care.