Safety and trustworthiness are indispensable requirements for real-world applications of AI systems using large language models (LLMs). This paper formulates human value alignment as an optimization problem of the language model policy to maximize reward under a safety constraint, and then proposes an algorithm, Stepwise Alignment for Constrained Policy Optimization (SACPO). One key idea behind SACPO, supported by theory, is that the optimal policy incorporating reward and safety can be directly obtained from a reward-aligned policy. Building on this key idea, SACPO aligns LLMs step-wise with each metric while leveraging simple yet powerful alignment algorithms such as direct preference optimization (DPO). SACPO offers several advantages, including simplicity, stability, computational efficiency, and flexibility of algorithms and datasets. Under mild assumptions, our theoretical analysis provides the upper bounds on optimality and safety constraint violation. Our experimental results show that SACPO can fine-tune Alpaca-7B better than the state-of-the-art method in terms of both helpfulness and harmlessness.

With the advent of Large Language Models (LLMs), the potential of Retrieval Augmented Generation (RAG) techniques have garnered considerable research attention. Numerous novel algorithms and models have been introduced to enhance various aspects of RAG systems. However, the absence of a standardized framework for implementation, coupled with the inherently intricate RAG process, makes it challenging and time-consuming for researchers to compare and evaluate these approaches in a consistent environment. Existing RAG toolkits like LangChain and LlamaIndex, while available, are often heavy and unwieldy, failing to meet the personalized needs of researchers. In response to this challenge, we propose FlashRAG, an efficient and modular open-source toolkit designed to assist researchers in reproducing existing RAG methods and in developing their own RAG algorithms within a unified framework. Our toolkit implements 12 advanced RAG methods and has gathered and organized 32 benchmark datasets. Our toolkit has various features, including customizable modular framework, rich collection of pre-implemented RAG works, comprehensive datasets, efficient auxiliary pre-processing scripts, and extensive and standard evaluation metrics. Our toolkit and resources are available at https://github.com/RUC-NLPIR/FlashRAG.

Predicting drug-drug interaction adverse events, so-called DDI events, is increasingly valuable as it facilitates the study of mechanisms underlying drug use or adverse reactions. Existing models often neglect the distinctive characteristics of individual event classes when integrating multi-source features, which contributes to systematic unfairness when dealing with highly imbalanced event samples. Moreover, the limited capacity of these models to abstract the unique attributes of each event subclass considerably hampers their application in predicting rare drug-drug interaction events with a limited sample size. Reducing dataset bias and abstracting event subclass characteristics are two unresolved challenges. Recently, prompt tuning with frozen pre-trained graph models, namely "pre-train, prompt, fine-tune" strategy, has demonstrated impressive performance in few-shot tasks. Motivated by this, we propose an advanced method as a solution to address these aforementioned challenges. Specifically, our proposed approach entails a hierarchical pre-training task that aims to capture crucial aspects of drug molecular structure and intermolecular interactions while effectively mitigating implicit dataset bias within the node embeddings. Furthermore, we construct a prototypical graph by strategically sampling data from distinct event types and design subgraph prompts utilizing pre-trained node features. Through comprehensive benchmark experiments, we validate the efficacy of our subgraph prompts in accurately representing event classes and achieve exemplary results in both overall and subclass prediction tasks.

The correct interpretation of breast density is important in the assessment of breast cancer risk. AI has been shown capable of accurately predicting breast density, however, due to the differences in imaging characteristics across mammography systems, models built using data from one system do not generalize well to other systems. Though federated learning (FL) has emerged as a way to improve the generalizability of AI without the need to share data, the best way to preserve features from all training data during FL is an active area of research. To explore FL methodology, the breast density classification FL challenge was hosted in partnership with the American College of Radiology, Harvard Medical School's Mass General Brigham, University of Colorado, NVIDIA, and the National Institutes of Health National Cancer Institute. Challenge participants were able to submit docker containers capable of implementing FL on three simulated medical facilities, each containing a unique large mammography dataset. The breast density FL challenge ran from June 15 to September 5, 2022, attracting seven finalists from around the world. The winning FL submission reached a linear kappa score of 0.653 on the challenge test data and 0.413 on an external testing dataset, scoring comparably to a model trained on the same data in a central location.

Natural language inference (NLI), the task of recognizing the entailment relationship in sentence pairs, is an actively studied topic serving as a proxy for natural language understanding. Despite the relevance of the task in building conversational agents and improving text classification, machine translation and other NLP tasks, to the best of our knowledge, there is no publicly available NLI corpus for the Romanian language. To this end, we introduce the first Romanian NLI corpus (RoNLI) comprising 58K training sentence pairs, which are obtained via distant supervision, and 6K validation and test sentence pairs, which are manually annotated with the correct labels. We conduct experiments with multiple machine learning methods based on distant learning, ranging from shallow models based on word embeddings to transformer-based neural networks, to establish a set of competitive baselines. Furthermore, we improve on the best model by employing a new curriculum learning strategy based on data cartography. Our dataset and code to reproduce the baselines are available at https://github.com/Eduard6421/RONLI.

Reducing wealth inequality and increasing utility are critical issues. This study reveals the effects of redistribution and consumption morals on wealth inequality and utility. To this end, we present a novel approach that couples the dynamic model of capital, consumption, and utility in macroeconomics with the interaction model of joint business and redistribution in econophysics. With this approach, we calculate the capital (wealth), the utility based on consumption, and the Gini index of these inequality using redistribution and consumption thresholds as moral parameters. The results show that: under-redistribution and waste exacerbate inequality; conversely, over-redistribution and stinginess reduce utility; and a balanced moderate moral leads to achieve both reduced inequality and increased utility. These findings provide renewed economic and numerical support for the moral importance known from philosophy, anthropology, and religion. The revival of redistribution and consumption morals should promote the transformation to a human mutual-aid economy, as indicated by philosopher and anthropologist, instead of the capitalist economy that has produced the current inequality. The practical challenge is to implement bottom-up social business, on a foothold of worker coops and platform cooperatives as a community against the state and the market, with moral consensus and its operation.

Large language models (LLMs) excel at few-shot in-context learning (ICL) -- learning from a few examples provided in context at inference, without any weight updates. Newly expanded context windows allow us to investigate ICL with hundreds or thousands of examples -- the many-shot regime. Going from few-shot to many-shot, we observe significant performance gains across a wide variety of generative and discriminative tasks. While promising, many-shot ICL can be bottlenecked by the available amount of human-generated examples. To mitigate this limitation, we explore two new settings: Reinforced and Unsupervised ICL. Reinforced ICL uses model-generated chain-of-thought rationales in place of human examples. Unsupervised ICL removes rationales from the prompt altogether, and prompts the model only with domain-specific questions. We find that both Reinforced and Unsupervised ICL can be quite effective in the many-shot regime, particularly on complex reasoning tasks. Finally, we demonstrate that, unlike few-shot learning, many-shot learning is effective at overriding pretraining biases, can learn high-dimensional functions with numerical inputs, and performs comparably to fine-tuning. Our analysis also reveals the limitations of next-token prediction loss as an indicator of downstream ICL performance.

Organizational decision-making is crucial for success, yet cognitive biases can significantly affect risk preferences, leading to suboptimal outcomes. Risk seeking preferences for losses, driven by biases such as loss aversion, pose challenges and can result in severe negative consequences, including financial losses. This research introduces the ABI approach, a novel solution designed to support organizational decision-makers by automatically identifying and explaining risk seeking preferences during decision-making. This research makes a novel contribution by automating the identification and explanation of risk seeking preferences using Cumulative Prospect theory (CPT) from Behavioral Economics. The ABI approach transforms theoretical insights into actionable, real-time guidance, making them accessible to a broader range of organizations and decision-makers without requiring specialized personnel. By contextualizing CPT concepts into business language, the approach facilitates widespread adoption and enhances decision-making processes with deep behavioral insights. Our systematic literature review identified significant gaps in existing methods, especially the lack of automated solutions with a concrete mechanism for automatically identifying risk seeking preferences, and the absence of formal knowledge representation, such as ontologies, for identifying and explaining the risk preferences. The ABI Approach addresses these gaps, offering a significant contribution to decision-making research and practice. Furthermore, it enables automatic collection of historical decision data with risk preferences, providing valuable insights for enhancing strategic management and long-term organizational performance. An experiment provided preliminary evidence on its effectiveness in helping decision-makers recognize their risk seeking preferences during decision-making in the loss domain.

Parameter-efficient finetuning (PEFT) methods seek to adapt large neural models via updates to a small number of weights. However, much prior interpretability work has shown that representations encode rich semantic information, suggesting that editing representations might be a more powerful alternative. We pursue this hypothesis by developing a family of Representation Finetuning (ReFT) methods. ReFT methods operate on a frozen base model and learn task-specific interventions on hidden representations. We define a strong instance of the ReFT family, Low-rank Linear Subspace ReFT (LoReFT), and we identify an ablation of this method that trades some performance for increased efficiency. Both are drop-in replacements for existing PEFTs and learn interventions that are 15x--65x more parameter-efficient than LoRA. We showcase LoReFT on eight commonsense reasoning tasks, four arithmetic reasoning tasks, instruction-tuning, and GLUE. In all these evaluations, our ReFTs deliver the best balance of efficiency and performance, and almost always outperform state-of-the-art PEFTs. We release a generic ReFT training library publicly at https://github.com/stanfordnlp/pyreft.

LLM-based assistants, such as GitHub Copilot and ChatGPT, have the potential to generate code that fulfills a programming task described in a natural language description, referred to as a prompt. The widespread accessibility of these assistants enables users with diverse backgrounds to generate code and integrate it into software projects. However, studies show that code generated by LLMs is prone to bugs and may miss various corner cases in task specifications. Presenting such buggy code to users can impact their reliability and trust in LLM-based assistants. Moreover, significant efforts are required by the user to detect and repair any bug present in the code, especially if no test cases are available. In this study, we propose a self-refinement method aimed at improving the reliability of code generated by LLMs by minimizing the number of bugs before execution, without human intervention, and in the absence of test cases. Our approach is based on targeted Verification Questions (VQs) to identify potential bugs within the initial code. These VQs target various nodes within the Abstract Syntax Tree (AST) of the initial code, which have the potential to trigger specific types of bug patterns commonly found in LLM-generated code. Finally, our method attempts to repair these potential bugs by re-prompting the LLM with the targeted VQs and the initial code. Our evaluation, based on programming tasks in the CoderEval dataset, demonstrates that our proposed method outperforms state-of-the-art methods by decreasing the number of targeted errors in the code between 21% to 62% and improving the number of executable code instances to 13%.