Prompting foundation models such as large language models (LLMs) and vision-language models (VLMs) requires extensive domain knowledge and manual efforts, resulting in the so-called "prompt engineering" problem. To improve the performance of foundation models, one can provide examples explicitly [1] or implicitly [2], or encourage intermediate reasoning steps [3, 4]. Despite all the efforts, their performance in long-horizon reasoning tasks is still limited. Classical planning methods, including those defined by Planning Domain Definition Language (PDDL), are strong in ensuring the soundness, completeness and efficiency in planning tasks [5]. However, those classical planners rely on predefined states and actions, and do not perform well in open-world scenarios. We aim to enjoy the openness of VLMs in scene understanding while retaining the strong long-horizon reasoning capabilities of classical planners. Our key idea is to extract domain knowledge from classical planners for prompting VLMs towards enabling classical planners that are visually grounded and responsive to open-world situations. Given the natural connection between planning symbols and human language, this paper investigates how pre-trained VLMs can assist the robot in realizing symbolic plans generated by classical planners, while avoiding the engineering efforts of checking the outcomes of each action.

The rapid development of large language models (LLMs) has significantly improved the generation of fluent and convincing text, raising concerns about their misuse on social media platforms. We present a methodology using Twitter datasets to examine the generative capabilities of four LLMs: Llama 3, Mistral, Qwen2, and GPT4o. We evaluate 7B and 8B parameter base-instruction models of the three open-source LLMs and validate the impact of further fine-tuning and "uncensored" versions. Our findings show that "uncensored" models with additional in-domain fine-tuning dramatically reduce the effectiveness of automated detection methods. This study addresses a gap by exploring smaller open-source models and the effects of "uncensoring," providing insights into how fine-tuning and content moderation influence machine-generated text detection.

Aligned instruction following models can better fulfill user requests than their unaligned counterparts. However, it has been shown that there is a length bias in evaluation of such models, and that training algorithms tend to exploit this bias by learning longer responses. In this work we show how to train models that can be controlled at inference time with instructions containing desired length constraints. Such models are superior in length instructed evaluations, outperforming standard instruction following models such as GPT4, Llama 3 and Mixtral.

Large vision-language models (LVLMs) have recently achieved significant progress, demonstrating strong capabilities in open-world visual understanding. However, it is not yet clear how LVLMs address demographic biases in real life, especially the disparities across attributes such as gender, skin tone, and age. In this paper, we empirically investigate \emph{visual fairness} in several mainstream LVLMs and audit their performance disparities across sensitive demographic attributes, based on public fairness benchmark datasets (e.g., FACET). To disclose the visual bias in LVLMs, we design a fairness evaluation framework with direct questions and single-choice question-instructed prompts on visual question-answering/classification tasks. The zero-shot prompting results indicate that, despite enhancements in visual understanding, both open-source and closed-source LVLMs exhibit prevalent fairness issues across different instruct prompts and demographic attributes.

Nowadays, data augmentation through synthetic data has been widely used in the field of Grammatical Error Correction (GEC) to alleviate the problem of data scarcity. However, these synthetic data are mainly used in the pre-training phase rather than the data-limited fine-tuning phase due to inconsistent error distribution and noisy labels. In this paper, we propose a synthetic data construction method based on contextual augmentation, which can ensure an efficient augmentation of the original data with a more consistent error distribution. Specifically, we combine rule-based substitution with model-based generation, using the generative model to generate a richer context for the extracted error patterns. Besides, we also propose a relabeling-based data cleaning method to mitigate the effects of noisy labels in synthetic data. Experiments on CoNLL14 and BEA19-Test show that our proposed augmentation method consistently and substantially outperforms strong baselines and achieves the state-of-the-art level with only a few synthetic data.

Large Language Models (LLMs) are often trained on vast amounts of undisclosed data, motivating the development of post-hoc Membership Inference Attacks (MIAs) to gain insight into their training data composition. However, in this paper, we identify inherent challenges in post-hoc MIA evaluation due to potential distribution shifts between collected member and non-member datasets. Using a simple bag-of-words classifier, we demonstrate that datasets used in recent post-hoc MIAs suffer from significant distribution shifts, in some cases achieving near-perfect distinction between members and non-members. This implies that previously reported high MIA performance may be largely attributable to these shifts rather than model memorization. We confirm that randomized, controlled setups eliminate such shifts and thus enable the development and fair evaluation of new MIAs. However, we note that such randomized setups are rarely available for the latest LLMs, making post-hoc data collection still required to infer membership for real-world LLMs. As a potential solution, we propose a Regression Discontinuity Design (RDD) approach for post-hoc data collection, which substantially mitigates distribution shifts. Evaluating various MIA methods on this RDD setup yields performance barely above random guessing, in stark contrast to previously reported results. Overall, our findings highlight the challenges in accurately measuring LLM memorization and the need for careful experimental design in (post-hoc) membership inference tasks.

Learning generalist embodied agents, able to solve multitudes of tasks in different domains is a long-standing problem. Reinforcement learning (RL) is hard to scale up as it requires a complex reward design for each task. In contrast, language can specify tasks in a more natural way. Current foundation vision-language models (VLMs) generally require fine-tuning or other adaptations to be functional, due to the significant domain gap. However, the lack of multimodal data in such domains represents an obstacle toward developing foundation models for embodied applications. In this work, we overcome these problems by presenting multimodal foundation world models, able to connect and align the representation of foundation VLMs with the latent space of generative world models for RL, without any language annotations. The resulting agent learning framework, GenRL, allows one to specify tasks through vision and/or language prompts, ground them in the embodied domain's dynamics, and learns the corresponding behaviors in imagination. As assessed through large-scale multi-task benchmarking, GenRL exhibits strong multi-task generalization performance in several locomotion and manipulation domains. Furthermore, by introducing a data-free RL strategy, it lays the groundwork for foundation model-based RL for generalist embodied agents.

This paper presents a system for diversity-aware autonomous conversation leveraging the capabilities of large language models (LLMs). The system adapts to diverse populations and individuals, considering factors like background, personality, age, gender, and culture. The conversation flow is guided by the structure of the system's pre-established knowledge base, while LLMs are tasked with various functions, including generating diversity-aware sentences. Achieving diversity-awareness involves providing carefully crafted prompts to the models, incorporating comprehensive information about users, conversation history, contextual details, and specific guidelines. To assess the system's performance, we conducted both controlled and real-world experiments, measuring a wide range of performance indicators.

Large Language Models (LLMs) excel at providing information acquired during pretraining on large-scale corpora and following instructions through user prompts. This study investigates whether the quality of LLM responses varies depending on the demographic profile of users. Considering English as the global lingua franca, along with the diversity of its dialects among speakers of different native languages, we explore whether non-native English speakers receive lower-quality or even factually incorrect responses from LLMs more frequently. Our results show that performance discrepancies occur when LLMs are prompted by native versus non-native English speakers and persist when comparing native speakers from Western countries with others. Additionally, we find a strong anchoring effect when the model recognizes or is made aware of the user's nativeness, which further degrades the response quality when interacting with non-native speakers. Our analysis is based on a newly collected dataset with over 12,000 unique annotations from 124 annotators, including information on their native language and English proficiency.

In the Minecraft Collaborative Building Task, two players collaborate: an Architect (A) provides instructions to a Builder (B) to assemble a specified structure using 3D blocks. In this work, we investigate the use of large language models (LLMs) to predict the sequence of actions taken by the Builder. Leveraging LLMs' in-context learning abilities, we use few-shot prompting techniques, that significantly improve performance over baseline methods. Additionally, we present a detailed analysis of the gaps in performance for future work