Dynamical systems theory provides a framework for analyzing iterative processes and evolution over time. Within such systems, repetitive transformations can lead to stable configurations, known as attractors, including fixed points and limit cycles. Applying this perspective to large language models (LLMs), which iteratively map input text to output text, provides a principled approach to characterizing long-term behaviors. Successive paraphrasing serves as a compelling testbed for exploring such dynamics, as paraphrases re-express the same underlying meaning with linguistic variation. Although LLMs are expected to explore a diverse set of paraphrases in the text space, our study reveals that successive paraphrasing converges to stable periodic states, such as 2-period attractor cycles, limiting linguistic diversity. This phenomenon is attributed to the self-reinforcing nature of LLMs, as they iteratively favour and amplify certain textual forms over others. This pattern persists with increasing generation randomness or alternating prompts and LLMs. These findings underscore inherent constraints in LLM generative capability, while offering a novel dynamical systems perspective for studying their expressive potential.

Large language models (LLMs) have achieved remarkable success in machine translation, demonstrating impressive performance across diverse languages. However, translationese—characterized by overly literal and unnatural translations—remains a persistent challenge in LLM-based translation systems. Despite their pre-training on vast corpora of natural utterances, LLMs exhibit translationese errors and generate unexpected unnatural translations, stemming from biases introduced during supervised fine-tuning (SFT). In this work, we systematically evaluate the prevalence of translationese in LLM-generated translations and investigate its roots during supervised training. We introduce methods to mitigate these biases, including polishing golden references and filtering unnatural training instances. Empirical evaluations demonstrate that these approaches significantly reduce translationese while improving translation naturalness, validated by human evaluations and automatic metrics. Our findings highlight the need for training-aware adjustments to optimize LLM translation outputs, paving the way for more fluent and target-language-consistent translations.

Inference-Time Scaling has been critical to the success of recent models such as OpenAI o1 and DeepSeek R1. However, many techniques used to train models for inference-time scaling require tasks to have answers that can be verified, limiting their application to domains such as math, coding and logical reasoning. We take inspiration from how humans make first attempts, ask for detailed feedback from others and make improvements based on such feedback across a wide spectrum of open-ended endeavors. To this end, we collect HelpSteer3 data to train dedicated Feedback and Edit Models that are capable of performing inference-time scaling for open-ended general-domain tasks. In our setup, one model generates an initial response, which are given feedback by a second model, that are then used by a third model to edit the response. We show that performance on Arena Hard, a benchmark strongly predictive of Chatbot Arena Elo can be boosted by scaling the number of initial response drafts, effective feedback and edited responses. When scaled optimally, our setup based on 70B models from the Llama 3 family can reach SoTA performance on Arena Hard at 92.7 as of 5 Mar 2025, surpassing OpenAI o1-preview-2024-09-12 with 90.4 and DeepSeek R1 with 92.3.

Prototype network-based methods have made substantial progress in few-shot relation extraction (FSRE) by enhancing relation prototypes with relation descriptions. However, the distribution of relations and instances in distinct representation spaces isolates the constraints of relations on instances, making relation prototypes biased. In this paper, we propose an end-to-end partial order-centered hyperbolic representation learning (PO-HRL) framework, which imposes the constraints of relations on instances by modeling partial order in hyperbolic space, so as to effectively learn the distribution of instance representations. Specifically, we develop the hyperbolic supervised contrastive learning based on Lorentzian cosine similarity to align representations of relations and instances, and model the partial order by constraining instances to reside within the Lorentzian entailment cone of their respective relation. Experiments on three benchmark datasets show that PO-HRL outperforms the strong baselines, especially in 1-shot settings lacking relation descriptions.

Continual fine-tuning of large language models (LLMs) suffers from catastrophic forgetting. Rehearsal-based methods mitigate this problem by retaining a small set of old data. Nevertheless, they still suffer inevitable performance loss. Although training separate experts for each task can help prevent forgetting, effectively assembling them remains a challenge. Some approaches use routers to assign tasks to experts, but in continual learning, they often require retraining for optimal performance. To address these challenges, we introduce the Sequential Ensemble of Experts (SEE) framework. SEE removes the need for an additional router, allowing each expert to independently decide whether a query should be handled. The framework employs distributed routing, and during continual fine-tuning, SEE only requires the training of new experts for incoming tasks, rather than retraining the entire system. Experiments reveal that the SEE outperforms prior approaches, including multi-task learning, in continual fine-tuning. It also demonstrates remarkable generalization ability, as the expert can effectively identify out-of-distribution queries, which can then be directed to a more generalized model for resolution. This work highlights the promising potential of integrating routing and response mechanisms within each expert, paving the way for the future of distributed model ensembling.

Large language models (LLMs) have achieved human-level text generation, emphasizing the need for effective deepfake text detection to mitigate risks like the spread of fake news and plagiarism. Existing research has been constrained by evaluating detection methods o specific domains or particular language models. In practical scenarios, however, the detector faces texts from various domains or LLMs without knowing their sources. To this end, we build a comprehensive testbed by gathering texts from diverse human writings and deepfake texts generated by different LLMs. Empirical results on mainstream detection methods demonstrate the difficulties associated with detecting deepfake text in a wide-ranging testbed, particularly in out-of-distribution scenarios. Such difficulties align with the diminishing linguistic differences between the two text sources. Despite challenges, the top-performing detector can identify 84.12% out-of-domain texts generated by a new LLM, indicating the feasibility for application scenarios.

The impressive capabilities of recent language models can be largely attributed to the multi-trillion token pretraining datasets that they are trained on. However, model developers fail to disclose their construction methodology which has lead to a lack of open information on how to develop effective pretraining sets. To address this issue, we perform the first systematic study across the entire pipeline of pretraining set construction. First, we run ablations on existing techniques for pretraining set development to identify which methods translate to the largest gains in model accuracy on downstream evaluations. Then, we categorize the most widely used data source, web crawl snapshots, across the attributes of toxicity, quality, type of speech, and domain. Finally, we show how such attribute information can be used to further refine and improve the quality of a pretraining set. These findings constitute an actionable set of steps that practitioners can use to develop high quality pretraining sets.

AI-generated text detection has attracted increasing attention as powerful language models approach human-level generation. Limited work is devoted to detecting (partially) AI-paraphrased texts. However, AI paraphrasing is commonly employed in various application scenarios for text refinement and diversity. To this end, we propose a novel detection framework, paraphrased text span detection (PTD), aiming to identify paraphrased text spans within a text. Different from text-level detection, PTD takes in the full text and assigns each of the sentences with a score indicating the paraphrasing degree. We construct a dedicated dataset, PASTED, for paraphrased text span detection. Both in-distribution and out-of-distribution results demonstrate the effectiveness of PTD models in identifying AI-paraphrased text spans. Statistical and model analysis explains the crucial role of the surrounding context of the paraphrased text spans. Extensive experiments show that PTD models can generalize to versatile paraphrasing prompts as well as multiple paraphrased text spans.

Existing open-source helpfulness preference datasets do not specify what makes some responses more helpful and others less so. Models trained on these datasets can incidentally learn to model dataset artifacts (e.g. preferring longer but unhelpful responses only due to their length). To alleviate this problem, we collect HelpSteer, a multi-attribute helpfulness dataset annotated for the various aspects that make responses helpful. Specifically, our 37k-sample dataset has annotations for correctness, coherence, complexity, and verbosity in addition to overall helpfulness of responses. Training Llama 2 70B using the HelpSteer dataset with SteerLM technique produces a model that scores 7.54 on MT Bench, which is currently the highest score for open models that do not require training data from more powerful models (e.g. GPT-4). We release this dataset with CC-BY-4.0 license at https://huggingface.co/datasets/nvidia/HelpSteer

Sarcasm, as a form of irony conveying mockery and contempt, has been widespread in social media such as Twitter and Weibo, where the sarcastic text is commonly characterized as an incongruity between the surface positive and negative situation. Naturally, it has an urgent demand to automatically identify sarcasm from social media, so as to illustrate people’s real views toward specific targets. In this paper, we develop a novel sarcasm detection method, namely Sarcasm Detector with Augmentation of Potential Result and Reaction (SD-APRR). Inspired by the direct access view, we treat each sarcastic text as an incomplete version without latent content associated with implied negative situations, including the result and human reaction caused by its observable content. To fill the latent content, we estimate the potential result and human reaction for each given training sample by [xEffect] and [xReact] relations inferred by the pre-trained commonsense reasoning tool COMET, and integrate the sample with them as an augmented one. We can then employ those augmented samples to train the sarcasm detector, whose encoder is a graph neural network with a denoising module. We conduct extensive empirical experiments to evaluate the effectiveness of SD-APRR. The results demonstrate that SD-APRR can outperform strong baselines on benchmark datasets.

Just as computational simulations of atoms, molecules and cells have shaped the way we study the sciences, true-to-life simulations of human-like agents can be valuable tools for studying human behavior. We propose Humanoid Agents, a system that guides Generative Agents to behave more like humans by introducing three elements of System 1 processing: Basic needs (e.g. hunger, health and energy), Emotion and Closeness in Relationships. Humanoid Agents are able to use these dynamic elements to adapt their daily activities and conversations with other agents, as supported with empirical experiments. Our system is designed to be extensible to various settings, three of which we demonstrate, as well as to other elements influencing human behavior (e.g. empathy, moral values and cultural background). Our platform also includes a Unity WebGL game interface for visualization and an interactive analytics dashboard to show agent statuses over time. Our platform is available on https://www.humanoidagents.com/ and code is on https://github.com/HumanoidAgents/HumanoidAgents

The widely used Fact-based Visual Question Answering (FVQA) dataset contains visually-grounded questions that require information retrieval using common sense knowledge graphs to answer. It has been observed that the original dataset is highly imbalanced and concentrated on a small portion of its associated knowledge graph. We introduce FVQA 2.0 which contains adversarial variants of test questions to address this imbalance. We show that systems trained with the original FVQA train sets can be vulnerable to adversarial samples and we demonstrate an augmentation scheme to reduce this vulnerability without human annotations.

Model alignment with human preferences is an essential step in making Large Language Models (LLMs) helpful and consistent with human values. It typically consists of supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF) stages. However, RLHF faces inherent limitations stemming from a complex training setup and its tendency to align the model with implicit values that end users cannot control at run-time. Moreover, reward models in RLHF stage commonly rely on single-dimensional feedback as opposed to explicit, multifaceted signals that indicate attributes such as helpfulness, humor, and toxicity. To address these limitations, we propose SteerLM, a supervised fine-tuning method that empowers end-users to control responses during inference. SteerLM conditions responses to conform to an explicitly defined multi-dimensional set of attributes, thereby empowering a steerable AI capable of generating helpful and high-quality responses while maintaining customizability. Experiments show that SteerLM trained on open source datasets generates responses that are preferred by human and automatic evaluators to many state-of-the-art baselines trained with RLHF while being much easier to train. Try SteerLM at https://huggingface.co/nvidia/SteerLM-llama2-13B

Personal attributes represent structured information about a person, such as their hobbies, pets, family, likes and dislikes. We introduce the tasks of extracting and inferring personal attributes from human-human dialogue, and analyze the linguistic demands of these tasks. To meet these challenges, we introduce a simple and extensible model that combines an autoregressive language model utilizing constrained attribute generation with a discriminative reranker. Our model outperforms strong baselines on extracting personal attributes as well as inferring personal attributes that are not contained verbatim in utterances and instead requires commonsense reasoning and lexical inferences, which occur frequently in everyday conversation. Finally, we demonstrate the benefit of incorporating personal attributes in social chit-chat and task-oriented dialogue settings.

When reading stories, people can naturally identify sentences in which a new event starts, i.e., event boundaries, using their knowledge of how events typically unfold, but a computational model to detect event boundaries is not yet available. We characterize and detect sentences with expected or surprising event boundaries in an annotated corpus of short diary-like stories, using a model that combines commonsense knowledge and narrative flow features with a RoBERTa classifier. Our results show that, while commonsense and narrative features can help improve performance overall, detecting event boundaries that are more subjective remains challenging for our model. We also find that sentences marking surprising event boundaries are less likely to be causally related to the preceding sentence, but are more likely to express emotional reactions of story characters, compared to sentences with no event boundary.

Internet forums such as Reddit offer people a platform to ask for advice when they encounter various issues at work, school or in relationships. Telling helpful comments apart from unhelpful comments to these advice-seeking posts can help people and dialogue agents to become more helpful in offering advice. We propose a dataset that contains both helpful and unhelpful comments in response to such requests. We then relate helpfulness to the closely related construct of empathy. Finally, we analyze the language features that are associated with helpful and unhelpful comments.

While many different aspects of human experiences have been studied by the NLP community, none has captured its full richness. We propose a new task to capture this richness based on an unlikely setting: movie characters. We sought to capture theme-level similarities between movie characters that were community-curated into 20,000 themes. By introducing a two-step approach that balances performance and efficiency, we managed to achieve 9-27% improvement over recent paragraph-embedding based methods. Finally, we demonstrate how the thematic information learnt from movie characters can potentially be used to understand themes in the experience of people, as indicated on Reddit posts.

We present a probabilistic clustering algorithm that can help Reddit users to find posts that discuss experiences similar to their own. This model is built upon the BERT Next Sentence Prediction model and reduces the time complexity for clustering all posts in a corpus from O(nˆ2) to O(n) with respect to the number of posts. We demonstrate that such probabilistic clustering can yield a performance better than baseline clustering methods based on Latent Dirichlet Allocation (Blei et al., 2003) and Word2Vec (Mikolov et al., 2013). Furthermore, there is a high degree of coherence between our probabilistic clustering and the exhaustive comparison O(nˆ2) algorithm in which the similarity between every pair of posts is found. This makes the use of the BERT Next Sentence Prediction model more practical for unsupervised clustering tasks due to the high runtime overhead of each BERT computation.