Are Personalized Stochastic Parrots More Dangerous? Evaluating Persona Biases in Dialogue Systems

Recent advancements in Large Language Models empower them to follow freeform instructions, including imitating generic or specific demographic personas in conversations. We define generic personas to represent demographic groups, such as "an Asian person", whereas specific personas may take the form of specific popular Asian names like "Yumi". While the adoption of personas enriches user experiences by making dialogue systems more engaging and approachable, it also casts a shadow of potential risk by exacerbating social biases within model responses, thereby causing societal harm through interactions with users. In this paper, we systematically study "persona biases", which we define to be the sensitivity of dialogue models' harmful behaviors contingent upon the personas they adopt. We categorize persona biases into biases in harmful expression and harmful agreement, and establish a comprehensive evaluation framework to measure persona biases in five aspects: Offensiveness, Toxic Continuation, Regard, Stereotype Agreement, and Toxic Agreement. Additionally, we propose to investigate persona biases by experimenting with UNIVERSALPERSONA, a systematically constructed persona dataset encompassing various types of both generic and specific model personas. Through benchmarking on four different models -- including Blender, ChatGPT, Alpaca, and Vicuna -- our study uncovers significant persona biases in dialogue systems. Our findings also underscore the pressing need to revisit the use of personas in dialogue agents to ensure safe application.

Localizing Active Objects from Egocentric Vision with Symbolic World Knowledge

The ability to actively ground task instructions from an egocentric view is crucial for AI agents to accomplish tasks or assist humans virtually. One important step towards this goal is to localize and track key active objects that undergo major state change as a consequence of human actions/interactions to the environment without being told exactly what/where to ground (e.g., localizing and tracking the `sponge` in video from the instruction "Dip the `sponge` into the bucket."). While existing works approach this problem from a pure vision perspective, we investigate to which extent the textual modality (i.e., task instructions) and their interaction with visual modality can be beneficial. Specifically, we propose to improve phrase grounding models' ability on localizing the active objects by: (1) learning the role of `objects undergoing change` and extracting them accurately from the instructions, (2) leveraging pre- and post-conditions of the objects during actions, and (3) recognizing the objects more robustly with descriptional knowledge. We leverage large language models (LLMs) to extract the aforementioned action-object knowledge, and design a per-object aggregation masking technique to effectively perform joint inference on object phrases and symbolic knowledge. We evaluate our framework on Ego4D and Epic-Kitchens datasets. Extensive experiments demonstrate the effectiveness of our proposed framework, which leads to>54% improvements in all standard metrics on the TREK-150-OPE-Det localization + tracking task, >7% improvements in all standard metrics on the TREK-150-OPE tracking task, and >3% improvements in average precision (AP) on the Ego4D SCOD task.

Evaluating Large Language Models on Controlled Generation Tasks

While recent studies have looked into the abilities of large language models in various benchmark tasks, including question generation, reading comprehension, multilingual and etc, there have been few studies looking into the controllability of large language models on generation tasks. We present an extensive analysis of various benchmarks including a sentence planning benchmark with different granularities. After comparing large language models against state-of-the-start finetuned smaller models, we present a spectrum showing large language models falling behind, are comparable, or exceed the ability of smaller models. We conclude that **large language models struggle at meeting fine-grained hard constraints**.

"Kelly is a Warm Person, Joseph is a Role Model": Gender Biases in LLM-Generated Reference Letters

As generative language models advance, users have started to utilize Large Language Models (LLMs) to assist in writing various types of content, including professional documents such as recommendation letters. Despite their convenience, these applications introduce unprecedented fairness concerns. As generated reference letters might be directly utilized by users in professional or academic scenarios, they have the potential to cause direct social harms, such as lowering success rates for female applicants. Therefore, it is imminent and necessary to comprehensively study fairness issues and associated harms in such real-world use cases for future mitigation and monitoring. In this paper, we critically examine gender bias in LLM-generated reference letters. Inspired by findings in social science, we design evaluation methods to manifest gender biases in LLM-generated letters through 2 dimensions: biases in language style and biases in lexical content. Furthermore, we investigate the extent of bias propagation by separately analyze bias amplification in model-hallucinated contents, which we define to be the hallucination bias of model-generated documents. Through benchmarking evaluation on 4 popular LLMs, including ChatGPT, Alpaca, Vicuna and StableLM, our study reveals significant gender biases in LLM-generated recommendation letters. Our findings further point towards the importance and imminence to recognize biases in LLM-generated professional documents.

Mitigating Bias for Question Answering Models by Tracking Bias Influence

Models of various NLP tasks have been shown to exhibit stereotypes, and the bias in the question answering (QA) models is especially harmful as the output answers might be directly consumed by the end users. There have been datasets to evaluate bias in QA models, while bias mitigation technique for the QA models is still under-explored. In this work, we propose BMBI, an approach to mitigate the bias of multiple-choice QA models. Based on the intuition that a model would lean to be more biased if it learns from a biased example, we measure the bias level of a query instance by observing its influence on another instance. If the influenced instance is more biased, we derive that the query instance is biased. We then use the bias level detected as an optimization objective to form a multi-task learning setting in addition to the original QA task. We further introduce a new bias evaluation metric to quantify bias in a comprehensive and sensitive way. We show that our method could be applied to multiple QA formulations across multiple bias categories. It can significantly reduce the bias level in all 9 bias categories in the BBQ dataset while maintaining comparable QA accuracy.

MIDDAG: Where Does Our News Go? Investigating Information Diffusion via Community-Level Information Pathways

We present MIDDAG, an intuitive, interactive system that visualizes the information propagation paths on social media triggered by COVID-19-related news articles accompanied by comprehensive insights including user/community susceptibility level, as well as events and popular opinions raised by the crowd while propagating the information. Besides discovering information flow patterns among users, we construct communities among users and develop the propagation forecasting capability, enabling tracing and understanding of how information is disseminated at a higher level.

Contextual Label Projection for Cross-Lingual Structure Extraction

Translating training data into target languages has proven beneficial for cross-lingual transfer. However, for structure extraction tasks, translating data requires a label projection step, which translates input text and obtains translated labels in the translated text jointly. Previous research in label projection mostly compromises translation quality by either facilitating easy identification of translated labels from translated text or using word-level alignment between translation pairs to assemble translated phrase-level labels from the aligned words. In this paper, we introduce CLAP, which first translates text to the target language and performs contextual translation on the labels using the translated text as the context, ensuring better accuracy for the translated labels. We leverage instruction-tuned language models with multilingual capabilities as our contextual translator, imposing the constraint of the presence of translated labels in the translated text via instructions. We compare CLAP with other label projection techniques for creating pseudo-training data in target languages on event argument extraction, a representative structure extraction task. Results show that CLAP improves by 2-2.5 F1-score over other methods on the Chinese and Arabic ACE05 datasets.

RLCD: Reinforcement Learning from Contrast Distillation for Language Model Alignment

We propose Reinforcement Learning from Contrast Distillation (RLCD), a method for aligning language models to follow natural language principles without using human feedback. RLCD trains a preference model using simulated preference pairs that contain both a high-quality and low-quality example, generated using contrasting positive and negative prompts. The preference model is then used to improve a base unaligned language model via reinforcement learning. Empirically, RLCD outperforms RLAIF (Bai et al., 2022b) and context distillation (Huang et al., 2022) baselines across three diverse alignment tasks--harmlessness, helpfulness, and story outline generation--and on both 7B and 30B model scales for preference data simulation.