Hallucinations in Large Language Models (LLMs) are widely regarded as errors - outputs that deviate from factual accuracy. However, in creative or exploratory contexts, these "mistakes" may represent unexpected avenues for innovation. We introduce Purposefully Induced Psychosis (PIP), a novel approach that amplifies LLM hallucinations for imaginative tasks such as speculative fiction, interactive storytelling, and mixed-reality simulations. Drawing on Herman Melville's Moby-Dick, where Pip's "madness" reveals profound insight, we reframe hallucinations as a source of computational imagination rather than a flaw. Our method fine-tunes LLMs to encourage speculative, metaphorical, and surreal outputs - hallucinations that are useful when factual accuracy is not the chief objective. Inspired by the consensual illusions of theater and stage magic, PIP situates these creative missteps in contexts where users willingly suspend disbelief, thereby transforming "errors" into catalysts for new ways of thinking. We discuss potential applications, design principles for ensuring user consent, preliminary observations, and implications for broader AI ethics and human-AI collaboration.

Gender-neutral translation (GNT) aims to avoid expressing the gender of human referents when the source text lacks explicit cues about the gender of those referents. Evaluating GNT automatically is particularly challenging, with current solutions being limited to monolingual classifiers. Such solutions are not ideal because they do not factor in the source sentence and require dedicated data and fine-tuning to scale to new languages. In this work, we address such limitations by investigating the use of large language models (LLMs) as evaluators of GNT. Specifically, we explore two prompting approaches: one in which LLMs generate sentence-level assessments only, and another, akin to a chain-of-thought approach, where they first produce detailed phrase-level annotations before a sentence-level judgment. Through extensive experiments on multiple languages with five models, both open and proprietary, we show that LLMs can serve as evaluators of GNT. Moreover, we find that prompting for phrase-level annotations before sentence-level assessments consistently improves the accuracy of all models, providing a better and more scalable alternative to current solutions.

Recently, DeepSeek-R1 (671B) (DeepSeek-AIet al., 2025) has demonstrated its excellent reasoning ability in complex tasks and has publiclyshared its methodology. This provides potentially high-quality chain-of-thought (CoT) data for stimulating the reasoning abilities of small-sized large language models (LLMs). To generate high-quality CoT data for different LLMs, we seek an efficient method for generating high-quality CoT data with LLM-Adaptive questiondifficulty levels. First, we grade the difficulty of the questions according to the reasoning ability of the LLMs themselves and construct a LLM-Adaptive question database. Second, we sample the problem database based on a distribution of difficulty levels of the questions and then use DeepSeek-R1 (671B) (DeepSeek-AI et al., 2025) to generate the corresponding high-quality CoT data with correct answers. Thanks to the construction of CoT data with LLM-Adaptive difficulty levels, we have significantly reduced the cost of data generation and enhanced the efficiency of model supervised fine-tuning (SFT). Finally, we have validated the effectiveness and generalizability of the proposed method in the fields of complex mathematical competitions and code generation tasks. Notably, with only 2k high-quality mathematical CoT data, our ZMath-32B surpasses DeepSeek-Distill-32B in math reasoning task. Similarly, with only 2k high-quality code CoT data, our ZCode-32B surpasses DeepSeek-Distill-32B in code reasoning tasks.

To what extent do LLMs use their capabilities towards their given goal? We take this as a measure of their goal-directedness. We evaluate goal-directedness on tasks that require information gathering, cognitive effort, and plan execution, where we use subtasks to infer each model's relevant capabilities. Our evaluations of LLMs from Google DeepMind, OpenAI, and Anthropic show that goal-directedness is relatively consistent across tasks, differs from task performance, and is only moderately sensitive to motivational prompts. Notably, most models are not fully goal-directed. We hope our goal-directedness evaluations will enable better monitoring of LLM progress, and enable more deliberate design choices of agentic properties in LLMs.

Emotional support conversation (ESC) aims to alleviate the emotional distress of individuals through effective conversations. Although large language models (LLMs) have obtained remarkable progress on ESC, most of these studies might not define the diagram from the state model perspective, therefore providing a suboptimal solution for long-term satisfaction. To address such an issue, we leverage the Finite State Machine (FSM) on LLMs, and propose a framework called FiSMiness. Our framework allows a single LLM to bootstrap the planning during ESC, and self-reason the seeker's emotion, support strategy and the final response upon each conversational turn. Substantial experiments on ESC datasets suggest that FiSMiness outperforms many baselines, including direct inference, self-refine, chain of thought, finetuning, and external-assisted methods, even those with many more parameters.

Previous work indicates that large language models exhibit a significant "English bias", i.e. they often perform better when tasks are presented in English. Interestingly, we have observed that using certain other languages in reasoning tasks can yield better performance than English. However, this phenomenon remains under-explored. In this paper, we explore the upper bound of harnessing multilingualism in reasoning tasks, suggesting that multilingual reasoning promises significantly (by nearly 10 Acc@$k$ points) and robustly (tolerance for variations in translation quality and language choice) higher upper bounds than English-only reasoning. Besides analyzing the reason behind the upper bound and challenges in reaching it, we also find that common answer selection methods cannot achieve this upper bound, due to their limitations and biases. These insights could pave the way for future research aimed at fully harnessing the potential of multilingual reasoning in LLMs.

This article reviews contemporary methods for integrating force, including both proprioception and tactile sensing, in robot manipulation policy learning. We conduct a comparative analysis on various approaches for sensing force, data collection, behavior cloning, tactile representation learning, and low-level robot control. From our analysis, we articulate when and why forces are needed, and highlight opportunities to improve learning of contact-rich, generalist robot policies on the path toward highly capable touch-based robot foundation models. We generally find that while there are few tasks such as pouring, peg-in-hole insertion, and handling delicate objects, the performance of imitation learning models is not at a level of dynamics where force truly matters. Also, force and touch are abstract quantities that can be inferred through a wide range of modalities and are often measured and controlled implicitly. We hope that juxtaposing the different approaches currently in use will help the reader to gain a systemic understanding and help inspire the next generation of robot foundation models.

LLM inference is essential for applications like text summarization, translation, and data analysis, but the high cost of GPU instances from Cloud Service Providers (CSPs) like AWS is a major burden. This paper proposes InferSave, a cost-efficient VM selection framework for cloud based LLM inference. InferSave optimizes KV cache offloading based on Service Level Objectives (SLOs) and workload charac teristics, estimating GPU memory needs, and recommending cost-effective VM instances. Additionally, the Compute Time Calibration Function (CTCF) improves instance selection accuracy by adjusting for discrepancies between theoretical and actual GPU performance. Experiments on AWS GPU instances show that selecting lower-cost instances without KV cache offloading improves cost efficiency by up to 73.7% for online workloads, while KV cache offloading saves up to 20.19% for offline workloads.

Although Arabic is spoken by over 400 million people, advanced Arabic writing assistance tools remain limited. To address this gap, we present ARWI, a new writing assistant that helps learners improve essay writing in Modern Standard Arabic. ARWI is the first publicly available Arabic writing assistant to include a prompt database for different proficiency levels, an Arabic text editor, state-of-the-art grammatical error detection and correction, and automated essay scoring aligned with the Common European Framework of Reference standards for language attainment. Moreover, ARWI can be used to gather a growing auto-annotated corpus, facilitating further research on Arabic grammar correction and essay scoring, as well as profiling patterns of errors made by native speakers and non-native learners. A preliminary user study shows that ARWI provides actionable feedback, helping learners identify grammatical gaps, assess language proficiency, and guide improvement.

Simultaneous speech translation (SimulST) produces translations incrementally while processing partial speech input. Although large language models (LLMs) have showcased strong capabilities in offline translation tasks, applying them to SimulST poses notable challenges. Existing LLM-based SimulST approaches either incur significant computational overhead due to repeated encoding of bidirectional speech encoder, or they depend on a fixed read/write policy, limiting the efficiency and performance. In this work, we introduce Efficient and Adaptive Simultaneous Speech Translation (EASiST) with fully unidirectional architecture, including both speech encoder and LLM. EASiST includes a multi-latency data curation strategy to generate semantically aligned SimulST training samples and redefines SimulST as an interleaved generation task with explicit read/write tokens. To facilitate adaptive inference, we incorporate a lightweight policy head that dynamically predicts read/write actions. Additionally, we employ a multi-stage training strategy to align speech-text modalities and optimize both translation and policy behavior. Experiments on the MuST-C En$\rightarrow$De and En$\rightarrow$Es datasets demonstrate that EASiST offers superior latency-quality trade-offs compared to several strong baselines.