Data Advisor: Dynamic Data Curation for Safety Alignment of Large Language Models

Data is a crucial element in large language model (LLM) alignment. Recent studies have explored using LLMs for efficient data collection. However, LLM-generated data often suffers from quality issues, with underrepresented or absent aspects and low-quality datapoints. To address these problems, we propose Data Advisor, an enhanced LLM-based method for generating data that takes into account the characteristics of the desired dataset. Starting from a set of pre-defined principles in hand, Data Advisor monitors the status of the generated data, identifies weaknesses in the current dataset, and advises the next iteration of data generation accordingly. Data Advisor can be easily integrated into existing data generation methods to enhance data quality and coverage. Experiments on safety alignment of three representative LLMs (i.e., Mistral, Llama2, and Falcon) demonstrate the effectiveness of Data Advisor in enhancing model safety against various fine-grained safety issues without sacrificing model utility.

Attribute Controlled Fine-tuning for Large Language Models: A Case Study on Detoxification

We propose a constraint learning schema for fine-tuning Large Language Models (LLMs) with attribute control. Given a training corpus and control criteria formulated as a sequence-level constraint on model outputs, our method fine-tunes the LLM on the training corpus while enhancing constraint satisfaction with minimal impact on its utility and generation quality. Specifically, our approach regularizes the LLM training by penalizing the KL divergence between the desired output distribution, which satisfies the constraints, and the LLM's posterior. This regularization term can be approximated by an auxiliary model trained to decompose the sequence-level constraints into token-level guidance, allowing the term to be measured by a closed-form formulation. To further improve efficiency, we design a parallel scheme for concurrently updating both the LLM and the auxiliary model. We evaluate the empirical performance of our approach by controlling the toxicity when training an LLM. We show that our approach leads to an LLM that produces fewer inappropriate responses while achieving competitive performance on benchmarks and a toxicity detection task.

Are you talking to or ? On Tokenization and Addressing Misgendering in LLMs with Pronoun Tokenization Parity

A large body of NLP research has documented the ways gender biases manifest and amplify within large language models (LLMs), though this research has predominantly operated within a gender binary-centric context. A growing body of work has identified the harmful limitations of this gender-exclusive framing; many LLMs cannot correctly and consistently refer to persons outside the gender binary, especially if they use neopronouns. While data scarcity has been identified as a possible culprit, the precise mechanisms through which it influences LLM misgendering remain underexplored. Our work addresses this gap by studying data scarcity's role in subword tokenization and, consequently, the formation of LLM word representations. We uncover how the Byte-Pair Encoding (BPE) tokenizer, a backbone for many popular LLMs, contributes to neopronoun misgendering through out-of-vocabulary behavior. We introduce pronoun tokenization parity (PTP), a novel approach to reduce LLM neopronoun misgendering by preserving a token's functional structure. We evaluate PTP's efficacy using pronoun consistency-based metrics and a novel syntax-based metric. Through several controlled experiments, finetuning LLMs with PTP improves neopronoun consistency from 14.5% to 58.4%, highlighting the significant role tokenization plays in LLM pronoun consistency.

Faithful Model Evaluation for Model-Based Metrics

Statistical significance testing is used in natural language processing (NLP) to determine whether the results of a study or experiment are likely to be due to chance or if they reflect a genuine relationship. A key step in significance testing is the estimation of confidence interval which is a function of sample variance. Sample variance calculation is straightforward when evaluating against ground truth. However, in many cases, a metric model is often used for evaluation. For example, to compare toxicity of two large language models, a toxicity classifier is used for evaluation. Existing works usually do not consider the variance change due to metric model errors, which can lead to wrong conclusions. In this work, we establish the mathematical foundation of significance testing for model-based metrics. With experiments on public benchmark datasets and a production system, we show that considering metric model errors to calculate sample variances for model-based metrics changes the conclusions in certain experiments.

JAB: Joint Adversarial Prompting and Belief Augmentation

With the recent surge of language models in different applications, attention to safety and robustness of these models has gained significant importance. Here we introduce a joint framework in which we simultaneously probe and improve the robustness of a black-box target model via adversarial prompting and belief augmentation using iterative feedback loops. This framework utilizes an automated red teaming approach to probe the target model, along with a belief augmenter to generate instructions for the target model to improve its robustness to those adversarial probes. Importantly, the adversarial model and the belief generator leverage the feedback from past interactions to improve the effectiveness of the adversarial prompts and beliefs, respectively. In our experiments, we demonstrate that such a framework can reduce toxic content generation both in dynamic cases where an adversary directly interacts with a target model and static cases where we use a static benchmark dataset to evaluate our model.

On the steerability of large language models toward data-driven personas

The recent surge in Large Language Model (LLM) related applications has led to a concurrent escalation in expectations for LLMs to accommodate a myriad of personas and encompass a broad spectrum of perspectives. An important first step towards addressing this demand is to align language models with specific personas, be it groups of users or individuals. Towards this goal, we first present a new conceptualization of a persona. Moving beyond the traditional reliance on demographics like age, gender, or political party affiliation, we introduce a data-driven persona definition methodology built on collaborative-filtering. In this methodology, users are embedded into a continuous vector space based on their opinions and clustered into cohorts that manifest coherent views across specific inquiries. This methodology allows for a more nuanced understanding of different latent social groups present in the overall population (as opposed to simply using demographic groups) and enhances the applicability of model steerability. Finally, we present an efficient method to steer LLMs towards a particular persona. We learn a soft-prompting model to map the continuous representation of users into sequences of virtual tokens which, when prepended to the LLM input, enables the LLM to produce responses aligned with a given user. Our results show that our steerability algorithm is superior in performance compared to a collection of baselines.

FLIRT: Feedback Loop In-context Red Teaming

Warning: this paper contains content that may be inappropriate or offensive. As generative models become available for public use in various applications, testing and analyzing vulnerabilities of these models has become a priority. Here we propose an automatic red teaming framework that evaluates a given model and exposes its vulnerabilities against unsafe and inappropriate content generation. Our framework uses in-context learning in a feedback loop to red team models and trigger them into unsafe content generation. We propose different in-context attack strategies to automatically learn effective and diverse adversarial prompts for text-to-image models. Our experiments demonstrate that compared to baseline approaches, our proposed strategy is significantly more effective in exposing vulnerabilities in Stable Diffusion (SD) model, even when the latter is enhanced with safety features. Furthermore, we demonstrate that the proposed framework is effective for red teaming text-to-text models, resulting in significantly higher toxic response generation rate compared to previously reported numbers.

"I'm fully who I am": Towards Centering Transgender and Non-Binary Voices to Measure Biases in Open Language Generation

Transgender and non-binary (TGNB) individuals disproportionately experience discrimination and exclusion from daily life. Given the recent popularity and adoption of language generation technologies, the potential to further marginalize this population only grows. Although a multitude of NLP fairness literature focuses on illuminating and addressing gender biases, assessing gender harms for TGNB identities requires understanding how such identities uniquely interact with societal gender norms and how they differ from gender binary-centric perspectives. Such measurement frameworks inherently require centering TGNB voices to help guide the alignment between gender-inclusive NLP and whom they are intended to serve. Towards this goal, we ground our work in the TGNB community and existing interdisciplinary literature to assess how the social reality surrounding experienced marginalization by TGNB persons contributes to and persists within Open Language Generation (OLG). By first understanding their marginalization stressors, we evaluate (1) misgendering and (2) harmful responses to gender disclosure. To do this, we introduce the TANGO dataset, comprising of template-based text curated from real-world text within a TGNB-oriented community. We discover a dominance of binary gender norms within the models; LLMs least misgendered subjects in generated text when triggered by prompts whose subjects used binary pronouns. Meanwhile, misgendering was most prevalent when triggering generation with singular they and neopronouns. When prompted with gender disclosures, LLM text contained stigmatizing language and scored most toxic when triggered by TGNB gender disclosure. Our findings warrant further research on how TGNB harms manifest in LLMs and serve as a broader case study toward concretely grounding the design of gender-inclusive AI in community voices and interdisciplinary literature.

Is the Elephant Flying? Resolving Ambiguities in Text-to-Image Generative Models

Natural language often contains ambiguities that can lead to misinterpretation and miscommunication. While humans can handle ambiguities effectively by asking clarifying questions and/or relying on contextual cues and common-sense knowledge, resolving ambiguities can be notoriously hard for machines. In this work, we study ambiguities that arise in text-to-image generative models. We curate a benchmark dataset covering different types of ambiguities that occur in these systems. We then propose a framework to mitigate ambiguities in the prompts given to the systems by soliciting clarifications from the user. Through automatic and human evaluations, we show the effectiveness of our framework in generating more faithful images aligned with human intention in the presence of ambiguities.

Leveraging Local Temporal Information for Multimodal Scene Classification

Robust video scene classification models should capture the spatial (pixel-wise) and temporal (frame-wise) characteristics of a video effectively. Transformer models with self-attention which are designed to get contextualized representations for individual tokens given a sequence of tokens, are becoming increasingly popular in many computer vision tasks. However, the use of Transformer based models for video understanding is still relatively unexplored. Moreover, these models fail to exploit the strong temporal relationships between the neighboring video frames to get potent frame-level representations. In this paper, we propose a novel self-attention block that leverages both local and global temporal relationships between the video frames to obtain better contextualized representations for the individual frames. This enables the model to understand the video at various granularities. We illustrate the performance of our models on the large scale YoutTube-8M data set on the task of video categorization and further analyze the results to showcase improvement.