While large language models (LLMs) have been used for automated grading, they have not yet achieved the same level of performance as humans, especially when it comes to grading complex questions. Existing research on this topic focuses on a particular step in the grading procedure: grading using predefined rubrics. However, grading is a multifaceted procedure that encompasses other crucial steps, such as grading rubrics design and post-grading review. There has been a lack of systematic research exploring the potential of LLMs to enhance the entire grading~process. In this paper, we propose an LLM-based grading system that addresses the entire grading procedure, including the following key components: 1) Developing grading rubrics that not only consider the questions but also the student answers, which can more accurately reflect students' performance. 2) Under the guidance of grading rubrics, providing accurate and consistent scores for each student, along with customized feedback. 3) Conducting post-grading review to better ensure accuracy and fairness. Additionally, we collected a new dataset named OS from a university operating system course and conducted extensive experiments on both our new dataset and the widely used Mohler dataset. Experiments demonstrate the effectiveness of our proposed approach, providing some new insights for developing automated grading systems based on LLMs.

The rapid development of large language models (LLMs) in recent years has largely focused on English, resulting in models that respond exclusively in English. To adapt these models to other languages, continual pre-training (CP) is often employed, followed by supervised fine-tuning (SFT) to maintain conversational abilities. However, CP and SFT can reduce a model's ability to filter harmful content. We propose Instruction Continual Pre-training (InsCP), which integrates instruction tags into the CP process to prevent loss of conversational proficiency while acquiring new languages. Our experiments demonstrate that InsCP retains conversational and Reinforcement Learning from Human Feedback (RLHF) abilities. Empirical evaluations on language alignment, reliability, and knowledge benchmarks confirm the efficacy of InsCP. Notably, this approach requires only 0.1 billion tokens of high-quality instruction-following data, thereby reducing resource consumption.

Despite recent progress made by large language models in code generation, they still struggle with programs that meet complex requirements. Recent work utilizes plan-and-solve decomposition to decrease the complexity and leverage self-tests to refine the generated program. Yet, planning deep-inside requirements in advance can be challenging, and the tests need to be accurate to accomplish self-improvement. To this end, we propose FunCoder, a code generation framework incorporating the divide-and-conquer strategy with functional consensus. Specifically, FunCoder recursively branches off sub-functions as smaller goals during code generation, represented by a tree hierarchy. These sub-functions are then composited to attain more complex objectives. Additionally, we designate functions via a consensus formed by identifying similarities in program behavior, mitigating error propagation. FunCoder outperforms state-of-the-art methods by +9.8% on average in HumanEval, MBPP, xCodeEval and MATH with GPT-3.5 and GPT-4. Moreover, our method demonstrates superiority on smaller models: With FunCoder, StableCode-3b surpasses GPT-3.5 by +18.6% and achieves 97.7% of GPT-4's performance on HumanEval. Further analysis reveals that our proposed dynamic function decomposition is capable of handling complex requirements, and the functional consensus prevails over self-testing in correctness evaluation.

Classifying public tenders is a useful task for both companies that are invited to participate and for inspecting fraudulent activities. To facilitate the task for both participants and public administrations, the European Union presented a common taxonomy (Common Procurement Vocabulary, CPV) which is mandatory for tenders of certain importance; however, the contracts in which a CPV label is mandatory are the minority compared to all the Public Administrations activities. Classifying over a real-world taxonomy introduces some difficulties that can not be ignored. First of all, some fine-grained classes have an insufficient (if any) number of observations in the training set, while other classes are far more frequent (even thousands of times) than the average. To overcome those difficulties, we present a zero-shot approach, based on a pre-trained language model that relies only on label description and respects the label taxonomy. To train our proposed model, we used industrial data, which comes from contrattipubblici.org, a service by SpazioDati s.r.l. that collects public contracts stipulated in Italy in the last 25 years. Results show that the proposed model achieves better performance in classifying low-frequent classes compared to three different baselines, and is also able to predict never-seen classes.

Large language models (LLMs) are vulnerable when trained on datasets containing harmful content, which leads to potential jailbreaking attacks in two scenarios: the integration of harmful texts within crowdsourced data used for pre-training and direct tampering with LLMs through fine-tuning. In both scenarios, adversaries can compromise the safety alignment of LLMs, exacerbating malfunctions. Motivated by the need to mitigate these adversarial influences, our research aims to enhance safety alignment by either neutralizing the impact of malicious texts in pre-training datasets or increasing the difficulty of jailbreaking during downstream fine-tuning. In this paper, we propose a data curation framework designed to counter adversarial impacts in both scenarios. Our method operates under the assumption that we have no prior knowledge of attack details, focusing solely on curating clean texts. We introduce an iterative process aimed at revising texts to reduce their perplexity as perceived by LLMs, while simultaneously preserving their text quality. By pre-training or fine-tuning LLMs with curated clean texts, we observe a notable improvement in LLM robustness regarding safety alignment against harmful queries. For instance, when pre-training LLMs using a crowdsourced dataset containing 5\% harmful instances, adding an equivalent amount of curated texts significantly mitigates the likelihood of providing harmful responses in LLMs and reduces the attack success rate by 71\%. Our study represents a significant step towards mitigating the risks associated with training-based jailbreaking and fortifying the secure utilization of LLMs.

Reducing the `$\textit{hallucination}$' problem of Large Language Models (LLMs) is crucial for their wide applications. A comprehensive and fine-grained measurement of the hallucination is the first key step for the governance of this issue but is under-explored in the community. Thus, we present $\textbf{ANAH}$, a bilingual dataset that offers $\textbf{AN}$alytical $\textbf{A}$nnotation of $\textbf{H}$allucinations in LLMs within Generative Question Answering. Each answer sentence in our dataset undergoes rigorous annotation, involving the retrieval of a reference fragment, the judgment of the hallucination type, and the correction of hallucinated content. ANAH consists of ~12k sentence-level annotations for ~4.3k LLM responses covering over 700 topics, constructed by a human-in-the-loop pipeline. Thanks to the fine granularity of the hallucination annotations, we can quantitatively confirm that the hallucinations of LLMs progressively accumulate in the answer and use ANAH to train and evaluate hallucination annotators. We conduct extensive experiments on studying generative and discriminative annotators and show that, although current open-source LLMs have difficulties in fine-grained hallucination annotation, the generative annotator trained with ANAH can surpass all open-source LLMs and GPT-3.5, obtain performance competitive with GPT-4, and exhibits better generalization ability on unseen questions.

Large Language Models (LLMs) are typically harmless but remain vulnerable to carefully crafted prompts known as ``jailbreaks'', which can bypass protective measures and induce harmful behavior. Recent advancements in LLMs have incorporated moderation guardrails that can filter outputs, which trigger processing errors for certain malicious questions. Existing red-teaming benchmarks often neglect to include questions that trigger moderation guardrails, making it difficult to evaluate jailbreak effectiveness. To address this issue, we introduce JAMBench, a harmful behavior benchmark designed to trigger and evaluate moderation guardrails. JAMBench involves 160 manually crafted instructions covering four major risk categories at multiple severity levels. Furthermore, we propose a jailbreak method, JAM (Jailbreak Against Moderation), designed to attack moderation guardrails using jailbreak prefixes to bypass input-level filters and a fine-tuned shadow model functionally equivalent to the guardrail model to generate cipher characters to bypass output-level filters. Our extensive experiments on four LLMs demonstrate that JAM achieves higher jailbreak success ($\sim$ $\times$ 19.88) and lower filtered-out rates ($\sim$ $\times$ 1/6) than baselines.

In psychiatry, negative patient descriptions and stigmatizing language can contribute to healthcare disparities in two ways: (1) read by patients they can harm their trust and engagement with the medical center; (2) read by future providers they may negatively influence the future perspective of a patient. By leveraging large language models, this work aims to identify the sentiment expressed in psychiatric clinical notes based on the reader's point of view. Extracting sentences from the Mount Sinai Health System's large and diverse clinical notes, we used prompts and in-context learning to adapt three large language models (GPT-3.5, Llama 2, Mistral) to classify the sentiment conveyed by the sentences according to the provider or non-provider point of view. Results showed that GPT-3.5 aligns best to provider point of view, whereas Mistral aligns best to non-provider point of view.

Schema matching is a crucial task in data integration, involving the alignment of a source schema with a target schema to establish correspondence between their elements. This task is challenging due to textual and semantic heterogeneity, as well as differences in schema sizes. Although machine-learning-based solutions have been explored in numerous studies, they often suffer from low accuracy, require manual mapping of the schemas for model training, or need access to source schema data which might be unavailable due to privacy concerns. In this paper we present a novel method, named ReMatch, for matching schemas using retrieval-enhanced Large Language Models (LLMs). Our method avoids the need for predefined mapping, any model training, or access to data in the source database. Our experimental results on large real-world schemas demonstrate that ReMatch is an effective matcher. By eliminating the requirement for training data, ReMatch becomes a viable solution for real-world scenarios.

The rise of large language models (LLMs) has enabled LLM-based applications (a.k.a. AI agents or co-pilots), a new software paradigm that combines the strength of LLM and conventional software. Diverse LLM applications from different tenants could design complex workflows using multiple LLM requests to accomplish one task. However, they have to use the over-simplified request-level API provided by today's public LLM services, losing essential application-level information. Public LLM services have to blindly optimize individual LLM requests, leading to sub-optimal end-to-end performance of LLM applications. This paper introduces Parrot, an LLM service system that focuses on the end-to-end experience of LLM-based applications. Parrot proposes Semantic Variable, a unified abstraction to expose application-level knowledge to public LLM services. A Semantic Variable annotates an input/output variable in the prompt of a request, and creates the data pipeline when connecting multiple LLM requests, providing a natural way to program LLM applications. Exposing Semantic Variables to the public LLM service allows it to perform conventional data flow analysis to uncover the correlation across multiple LLM requests. This correlation opens a brand-new optimization space for the end-to-end performance of LLM-based applications. Extensive evaluations demonstrate that Parrot can achieve up to an order-of-magnitude improvement for popular and practical use cases of LLM applications.