As large language models (LLMs) expand into performing as agents for real-world applications beyond traditional NLP tasks, evaluating their robustness becomes increasingly important. However, existing benchmarks often overlook critical dimensions like cultural and social awareness. To address these, we introduce CASA, a benchmark designed to assess LLM agents' sensitivity to cultural and social norms across two web-based tasks: online shopping and social discussion forums. Our approach evaluates LLM agents' ability to detect and appropriately respond to norm-violating user queries and observations. Furthermore, we propose a comprehensive evaluation framework that measures awareness coverage, helpfulness in managing user queries, and the violation rate when facing misleading web content. Experiments show that current LLMs perform significantly better in non-agent than in web-based agent environments, with agents achieving less than 10% awareness coverage and over 40% violation rates. To improve performance, we explore two methods: prompting and fine-tuning, and find that combining both methods can offer complementary advantages -- fine-tuning on culture-specific datasets significantly enhances the agents' ability to generalize across different regions, while prompting boosts the agents' ability to navigate complex tasks. These findings highlight the importance of constantly benchmarking LLM agents' cultural and social awareness during the development cycle.

Many alignment methods, including reinforcement learning from human feedback (RLHF), rely on the Bradley-Terry reward assumption, which is insufficient to capture the full range of general human preferences. To achieve robust alignment with general preferences, we model the alignment problem as a two-player zero-sum game, where the Nash equilibrium policy guarantees a 50% win rate against any competing policy. However, previous algorithms for finding the Nash policy either diverge or converge to a Nash policy in a modified game, even in a simple synthetic setting, thereby failing to maintain the 50% win rate guarantee against all other policies. We propose a meta-algorithm, Convergent Meta Alignment Algorithm (COMAL), for language model alignment with general preferences, inspired by convergent algorithms in game theory. Theoretically, we prove that our meta-algorithm converges to an exact Nash policy in the last iterate. Additionally, our meta-algorithm is simple and can be integrated with many existing methods designed for RLHF and preference optimization with minimal changes. Experimental results demonstrate the effectiveness of the proposed framework when combined with existing preference policy optimization methods.

Existing efforts in building GUI agents heavily rely on the availability of robust commercial Vision-Language Models (VLMs) such as GPT-4o and GeminiPro-Vision. Practitioners are often reluctant to use open-source VLMs due to their significant performance lag compared to their closed-source counterparts, particularly in GUI grounding and Out-Of-Distribution (OOD) scenarios. To facilitate future research in this area, we developed OS-Atlas --a foundational GUI action model that excels at GUI grounding and OOD agentic tasks through innovations in both data and modeling. We have invested significant engineering effort in developing an open-source toolkit for synthesizing GUI grounding data across multiple platforms, including Windows, Linux, MacOS, Android, and the web. Leveraging this toolkit, we are releasing the largest open-source cross-platform GUI grounding corpus to date, which contains over 13 million GUI elements. This dataset, combined with innovations in model training, provides a solid foundation for OS-Atlas to understand GUI screenshots and generalize to unseen interfaces. Through extensive evaluation across six benchmarks spanning three different platforms (mobile, desktop, and web), OS-Atlas demonstrates significant performance improvements over previous state-of-the-art models. With the recent adoption of large language models (LLMs), the fantasy of building digital agents (Wu et al., 2024)--similar to JARVIS in The Iron Man--to automate daily tasks is evolving from science fiction into a tangible reality. Many current agents make decisions based on textual descriptions of the environments, such as HTML and accessibility trees, which is often lengthy (Zheng et al., 2024a), noisy (Cheng et al., 2024; WebAIM, 2024), and hard to acquire in practice. More recent studies (Cheng et al., 2024; Hong et al., 2024b; Li et al., 2024) have explored the use of large visionlanguage models (VLMs) to develop graphical user interfaces (GUI) agents capable of performing complex tasks simply by analyzing the screen - an information-complete medium for agent's decisionmaking, allowing for greater flexibility. At the core of a GUI agent lies an action model that enables GUI grounding - the process of transforming natural language instructions into executable actions within the operating system (e.g., clicking somewhere on the screen).

The hallucinations of large language models (LLMs) are increasingly mitigated by allowing LLMs to search for information and to ground their answers in real sources. Observing that LLMs can learn to search for relevant facts by trying different queries and learning to up-weight queries that successfully produce relevant results, we introduce Learning to Retrieve by Trying (LeReT), a reinforcement learning framework that explores search queries and uses preference-based optimization to improve their quality. LeReT can improve the absolute retrieval accuracy by up to 29% and the downstream generator evaluations by 17%. The simplicity and flexibility of LeReT allows it to be applied to arbitrary off-the-shelf retrievers and makes it a promising technique for improving general LLM pipelines. Despite tremendous progress, large language models (LLMs) still often hallucinate, motivating significant interest in grounding LLM answers in verified sources (Guu et al., 2020; Komeili et al., 2022; PerplexityAI, 2024; Google, 2024; OpenAI, 2024). Unfortunately, simply retrieving semantically similar documents to the user question, as is prevalent in retrieval-augmented generation (RAG; Lewis et al. 2020) pipelines, tends to fail for complex information needs not answered directly by any individual document. To tackle this, multi-hop retrieval pipelines gather information incrementally over multiple steps of search. For example, if a user asks What is a good dinner place driving from the Bay Area to Lake Tahoe on Friday night to avoid traffic?, a grounded system might need to learn about towns en route Lake Tahoe from the Bay Area, followed by traffic forecast on I-80 and finally, restaurants in Auburn (and other towns). However, doing this successfully is hard as off-the-shelf LLM performance is often unsatisfactory, and producing supervision for the best search queries to generate in a sequence of "hops" is nontrivial and expensive. Recent work tackles this via prompt optimization and rejection fine-tuning given a downstream signal.

While large models trained with self-supervised learning on offline datasets have shown remarkable capabilities in text and image domains, achieving the same generalisation for agents that act in sequential decision problems remains an open challenge. In this work, we take a step towards this goal by procedurally generating tens of millions of 2D physics-based tasks and using these to train a general reinforcement learning (RL) agent for physical control. To this end, we introduce Kinetix: an open-ended space of physics-based RL environments that can represent tasks ranging from robotic locomotion and grasping to video games and classic RL environments, all within a unified framework. Kinetix makes use of our novel hardware-accelerated physics engine Jax2D that allows us to cheaply simulate billions of environment steps during training. Our trained agent exhibits strong physical reasoning capabilities, being able to zero-shot solve unseen human-designed environments. Furthermore, fine-tuning this general agent on tasks of interest shows significantly stronger performance than training an RL agent *tabula rasa*. This includes solving some environments that standard RL training completely fails at. We believe this demonstrates the feasibility of large scale, mixed-quality pre-training for online RL and we hope that Kinetix will serve as a useful framework to investigate this further.

Transformer-based architectures have dominated various areas of machine learning in recent years. In this paper, we introduce a novel robust attention mechanism designed to enhance the resilience of transformer-based architectures. Crucially, this technique can be integrated into existing transformers as a plug-and-play layer, improving their robustness without the need for additional training or fine-tuning. Through comprehensive experiments and ablation studies, we demonstrate that our ProTransformer significantly enhances the robustness of transformer models across a variety of prediction tasks, attack mechanisms, backbone architectures, and data domains. Notably, without further fine-tuning, the ProTransformer consistently improves the performance of vanilla transformers by 19.5%, 28.3%, 16.1%, and 11.4% for BERT, ALBERT, DistilBERT, and RoBERTa, respectively, under the classical TextFooler attack. Furthermore, ProTransformer shows promising resilience in large language models (LLMs) against prompting-based attacks, improving the performance of T5 and LLaMA by 24.8% and 17.8%, respectively, and enhancing Vicuna by an average of 10.4% against the Jailbreaking attack. Beyond the language domain, ProTransformer also demonstrates outstanding robustness in both vision and graph domains.

This paper presents ReasoningRec, a reasoning-based recommendation framework that leverages Large Language Models (LLMs) to bridge the gap between recommendations and human-interpretable explanations. In contrast to conventional recommendation systems that rely on implicit user-item interactions, ReasoningRec employs LLMs to model users and items, focusing on preferences, aversions, and explanatory reasoning. The framework utilizes a larger LLM to generate synthetic explanations for user preferences, subsequently used to fine-tune a smaller LLM for enhanced recommendation accuracy and human-interpretable explanation. Our experimental study investigates the impact of reasoning and contextual information on personalized recommendations, revealing that the quality of contextual and personalized data significantly influences the LLM's capacity to generate plausible explanations. Empirical evaluations demonstrate that ReasoningRec surpasses state-of-the-art methods by up to 12.5\% in recommendation prediction while concurrently providing human-intelligible explanations. The code is available here: https://github.com/millenniumbismay/reasoningrec.

Transformers have become the predominant architecture in foundation models due to their excellent performance across various domains. However, the substantial cost of scaling these models remains a significant concern. This problem arises primarily from their dependence on a fixed number of parameters within linear projections. When architectural modifications (e.g., channel dimensions) are introduced, the entire model typically requires retraining from scratch. As model sizes continue growing, this strategy results in increasingly high computational costs and becomes unsustainable. To overcome this problem, we introduce Tokenformer, a natively scalable architecture that leverages the attention mechanism not only for computations among input tokens but also for interactions between tokens and model parameters, thereby enhancing architectural flexibility. By treating model parameters as tokens, we replace all the linear projections in Transformers with our token-parameter attention layer, where input tokens act as queries and model parameters as keys and values. This reformulation allows for progressive and efficient scaling without necessitating retraining from scratch. Our model scales from 124M to 1.4B parameters by incrementally adding new key-value parameter pairs, achieving performance comparable to Transformers trained from scratch while greatly reducing training costs.

The exponential growth of knowledge and the increasing complexity of interdisciplinary research pose significant challenges for researchers, including information overload and difficulties in exploring novel ideas. The advancements in large language models (LLMs), such as GPT-4, have shown great potential in enhancing idea proposals, but how to effectively utilize large models for reasonable idea proposal has not been thoroughly explored. This paper proposes a scientific paper idea proposer (SciPIP). Based on a user-provided research background, SciPIP retrieves helpful papers from a literature database while leveraging the capabilities of LLMs to generate more novel and feasible ideas. To this end, 1) we construct a literature retrieval database, extracting lots of papers' multi-dimension information for fast access. Then, a literature retrieval method based on semantics, entity, and citation co-occurrences is proposed to search relevant literature from multiple aspects based on the user-provided background. 2) After literature retrieval, we introduce dual-path idea proposal strategies, where one path infers solutions from the retrieved literature and the other path generates original ideas through model brainstorming. We then combine the two to achieve a good balance between feasibility and originality. Through extensive experiments on the natural language processing (NLP) field, we demonstrate that SciPIP can retrieve citations similar to those of existing top conference papers and generate many ideas consistent with them. Additionally, we evaluate the originality of other ideas generated by SciPIP using large language models, further validating the effectiveness of our proposed method. The code and the database are released at https://github.com/cheerss/SciPIP.

Developing a foundation model for time series forecasting across diverse domains has attracted significant attention in recent years. Existing works typically assume regularly sampled, well-structured data, limiting their applicability to more generalized scenarios where time series often contain missing values, unequal sequence lengths, and irregular time intervals between measurements. To cover diverse domains and handle variable regularities, we propose FlexTSF, a universal time series forecasting model that possesses better generalization and natively support both regular and irregular time series. FlexTSF produces forecasts in an autoregressive manner and incorporates three novel designs: VT-Norm, a normalization strategy to ablate data domain barriers, IVP Patcher, a patching module to learn representations from flexibly structured time series, and LED attention, an attention mechanism to seamlessly integrate these two and propagate forecasts with awareness of domain and time information. Experiments on 12 datasets show that FlexTSF outperforms state-of-the-art forecasting models respectively designed for regular and irregular time series. Furthermore, after self-supervised pre-training, FlexTSF shows exceptional performance in both zero-shot and few-show settings for time series forecasting. Time series forecasting, the task of predicting future values based on historical observations, plays an indispensable role across numerous domains, including finance, manufacturing, retail, healthcare, and meteorology (Lim & Zohren, 2021; De Gooijer & Hyndman, 2006).