As large language models (LLMs) become easily accessible nowadays, the trade-off between safety and helpfulness can significantly impact user experience. A model that prioritizes safety will cause users to feel less engaged and assisted while prioritizing helpfulness will potentially cause harm. Possible harms include teaching people how to build a bomb, exposing youth to inappropriate content, and hurting users' mental health. In this work, we propose to balance safety and helpfulness in diverse use cases by controlling both attributes in LLM. We explore training-free and fine-tuning methods that do not require extra human annotations and analyze the challenges of controlling safety and helpfulness in LLMs. Our experiments demonstrate that our method can rewind a learned model and unlock its controllability.

The quality of data representation is paramount for the performance of a model. Recent research has focused on enhancing representation learning by incorporating more information about the intra-sample structures of individual data points, such as local and global attention. Additionally, researchers have explored methods to model the inter-sample relationships, including manifold, contrastive, and discrete representation learning. In this study, we introduce a new training loss, which considers both intra-sample structure and inter-sample relationships, leveraging the concept of {\it atoms} to represent data points. This new approach, {\it Atom Modeling}, offers a fresh perspective to discretize data representations within a continuous space. Through experiments, we demonstrate that Atom Modeling enhances the performance of existing models in tasks involving classification and generation, across diverse domains including vision and language. These findings underscore the potential of Atom Modeling to enhance data representation and improve model learning, suggesting a promising direction for future research.

Reinforcement learning (RL) agents have long sought to approach the efficiency of human learning. Humans are great observers who can learn by aggregating external knowledge from various sources, including observations from others' policies of attempting a task. Prior studies in RL have incorporated external knowledge policies to help agents improve sample efficiency. However, it remains non-trivial to perform arbitrary combinations and replacements of those policies, an essential feature for generalization and transferability. In this work, we present Knowledge-Grounded RL (KGRL), an RL paradigm fusing multiple knowledge policies and aiming for human-like efficiency and flexibility. We propose a new actor architecture for KGRL, Knowledge-Inclusive Attention Network (KIAN), which allows free knowledge rearrangement due to embedding-based attentive action prediction. KIAN also addresses entropy imbalance, a problem arising in maximum entropy KGRL that hinders an agent from efficiently exploring the environment, through a new design of policy distributions. The experimental results demonstrate that KIAN outperforms alternative methods incorporating external knowledge policies and achieves efficient and flexible learning. Our implementation is available at https://github.com/Pascalson/KGRL.git

Despite their widespread adoption, neural conversation models have yet to exhibit natural chat capabilities with humans. In this research, we examine user utterances as causes and generated responses as effects, recognizing that changes in a cause should produce a different effect. To further explore this concept, we have compiled and expanded upon a new dataset called CausalDialogue through crowd-sourcing. This dataset includes multiple cause-effect pairs within a directed acyclic graph (DAG) structure. Our analysis reveals that traditional loss functions struggle to effectively incorporate the DAG structure, leading us to propose a causality-enhanced method called Exponential Maximum Average Treatment Effect (ExMATE) to enhance the impact of causality at the utterance level in training neural conversation models. To evaluate the needs of considering causality in dialogue generation, we built a comprehensive benchmark on CausalDialogue dataset using different models, inference, and training methods. Through experiments, we find that a causality-inspired loss like ExMATE can improve the diversity and agility of conventional loss function and there is still room for improvement to reach human-level quality on this new dataset.

In comparison to the interpretation of classification models, the explanation of sequence generation models is also an important problem, however it has seen little attention. In this work, we study model-agnostic explanations of a representative text generation task -- dialogue response generation. Dialog response generation is challenging with its open-ended sentences and multiple acceptable responses. To gain insights into the reasoning process of a generation model, we propose anew method, local explanation of response generation (LERG) that regards the explanations as the mutual interaction of segments in input and output sentences. LERG views the sequence prediction as uncertainty estimation of a human response and then creates explanations by perturbing the input and calculating the certainty change over the human response. We show that LERG adheres to desired properties of explanations for text generation including unbiased approximation, consistency and cause identification. Empirically, our results show that our method consistently improves other widely used methods on proposed automatic- and human- evaluation metrics for this new task by 4.4-12.8%. Our analysis demonstrates that LERG can extract both explicit and implicit relations between input and output segments.

Data-driven, knowledge-grounded neural conversation models are capable of generating more informative responses. However, these models have not yet demonstrated that they can zero-shot adapt to updated, unseen knowledge graphs. This paper proposes a new task about how to apply dynamic knowledge graphs in neural conversation model and presents a novel TV series conversation corpus (DyKgChat) for the task. Also, we propose a preliminary model that selects an output from two networks at each time step: a sequence-to-sequence model (Seq2Seq) and a multi-hop reasoning model, in order to support dynamic knowledge graphs. To benchmark this new task and evaluate the capability of adaptation, we introduce several evaluation metrics and the experiments show that our proposed approach outperforms previous knowledge-grounded conversation models. The proposed corpus and model can motivate the future research directions 1 . 1 Introduction In the chitchat dialogue generation, neural conversation models (Sutskever et al., 2014; Sordoni et al., 2015; Vinyals and Le, 2015) have emerged for its capability to be fully data-driven and end-to-end trained. While the generated responses are often reasonable but general (without useful information), recent work proposed knowledge-grounded models (Eric et al., 2017; Ghazvinine-jad et al., 2018; Zhou et al., 2018b; Qian et al., 2018) to incorporate external facts in an end-to- end fashion without handcrafted slot filling. Effectively combining text and external knowledge1 The data and code are available in https://github. Nonetheless, prior work rarely analyzed the model capability of zero-shot adaptation to dynamic knowledge graphs, where the states/entities and their relations are temporal and evolve as a single time scale process.

In sequence generation task, many works use policy gradient for model optimization to tackle the intractable backpropagation issue when maximizing the non-differentiable evaluation metrics or fooling the discriminator in adversarial learning. In this paper, we replace policy gradient with proximal policy optimization (PPO), which is a proved more efficient reinforcement learning algorithm, and propose a dynamic approach for PPO (PPO-dynamic). We demonstrate the efficacy of PPO and PPO-dynamic on conditional sequence generation tasks including synthetic experiment and chit-chat chatbot. The results show that PPO and PPO-dynamic can beat policy gradient by stability and performance.

Sequence generative adversarial networks (SeqGAN) have been used to improve conditional sequence generation tasks, for example, chit-chat dialogue generation. To stabilize the training of SeqGAN, Monte Carlo tree search (MCTS) or reward at every generation step (REGS) is used to evaluate the goodness of a generated subsequence. MCTS is computationally intensive, but the performance of REGS is worse than MCTS. In this paper, we propose stepwise GAN (StepGAN), in which the discriminator is modified to automatically assign scores quantifying the goodness of each subsequence at every generation step. StepGAN has significantly less computational costs than MCTS. We demonstrate that StepGAN outperforms previous GAN-based methods on both synthetic experiment and chit-chat dialogue generation.