The current paradigm for safety alignment of large language models (LLMs) follows a one-size-fits-all approach: the model refuses to interact with any content deemed unsafe by the model provider. This approach lacks flexibility in the face of varying social norms across cultures and regions. In addition, users may have diverse safety needs, making a model with static safety standards too restrictive to be useful, as well as too costly to be re-aligned. We propose Controllable Safety Alignment (CoSA), a framework designed to adapt models to diverse safety requirements without re-training. Instead of aligning a fixed model, we align models to follow safety configs -- free-form natural language descriptions of the desired safety behaviors -- that are provided as part of the system prompt. To adjust model safety behavior, authorized users only need to modify such safety configs at inference time. To enable that, we propose CoSAlign, a data-centric method for aligning LLMs to easily adapt to diverse safety configs. Furthermore, we devise a novel controllability evaluation protocol that considers both helpfulness and configured safety, summarizing them into CoSA-Score, and construct CoSApien, a human-authored benchmark that consists of real-world LLM use cases with diverse safety requirements and corresponding evaluation prompts. We show that CoSAlign leads to substantial gains of controllability over strong baselines including in-context alignment. Our framework encourages better representation and adaptation to pluralistic human values in LLMs, and thereby increasing their practicality.

Generating free-text rationales is among the emergent capabilities of Large Language Models (LLMs). These rationales have been found to enhance LLM performance across various NLP tasks. Recently, there has been growing interest in using these rationales to provide insights for various important downstream tasks. In this paper, we analyze generated free-text rationales in tasks with subjective answers, emphasizing the importance of rationalization in such scenarios. We focus on pairwise argument ranking, a highly subjective task with significant potential for real-world applications, such as debate assistance. We evaluate the persuasiveness of rationales generated by nine LLMs to support their subjective choices. Our findings suggest that open-source LLMs, particularly Llama2-70B-chat, are capable of providing highly persuasive rationalizations, surpassing even GPT models. Additionally, our experiments show that rationale persuasiveness can be improved by controlling its parameters through prompting or through self-refinement.

We present a framework for the automated measurement of responsible AI (RAI) metrics for large language models (LLMs) and associated products and services. Our framework for automatically measuring harms from LLMs builds on existing technical and sociotechnical expertise and leverages the capabilities of state-of-the-art LLMs, such as GPT-4. We use this framework to run through several case studies investigating how different LLMs may violate a range of RAI-related principles. The framework may be employed alongside domain-specific sociotechnical expertise to create measurements for new harm areas in the future. By implementing this framework, we aim to enable more advanced harm measurement efforts and further the responsible use of LLMs.

Forum discussion data differ in both structure and properties from generic form of textual data such as news. Henceforth, summarization techniques should, in turn, make use of such differences, and craft models that can benefit from the structural nature of discussion data. In this work, we propose attending to the beginning of a document, to improve the performance of extractive summarization models when applied to forum discussion data. Evaluations demonstrated that with the help of bidirectional attention mechanism, attending to the beginning of a document (initial comment/post) in a discussion thread, can introduce a consistent boost in ROUGE scores, as well as introducing a new State Of The Art (SOTA) ROUGE scores on the forum discussions dataset. Additionally, we explored whether this hypothesis is extendable to other generic forms of textual data. We make use of the tendency of introducing important information early in the text, by attending to the first few sentences in generic textual data. Evaluations demonstrated that attending to introductory sentences using bidirectional attention, improves the performance of extractive summarization models when even applied to more generic form of textual data.

Writing a good essay typically involves students revising an initial paper draft after receiving feedback. We present eRevise, a web-based writing and revising environment that uses natural language processing features generated for rubric-based essay scoring to trigger formative feedback messages regarding students' use of evidence in response-to-text writing. By helping students understand the criteria for using text evidence during writing, eRevise empowers students to better revise their paper drafts. In a pilot deployment of eRevise in 7 classrooms spanning grades 5 and 6, the quality of text evidence usage in writing improved after students received formative feedback then engaged in paper revision.