This work reflects upon what Immersion can mean from the perspective of an Artificial Intelligence (AI). Applying the lens of immersive learning theory, it seeks to understand whether this new perspective supports ways for AI participation in cognitive ecologies. By treating AI as a participant rather than a tool, it explores what other participants (humans and other AIs) need to consider in environments where AI can meaningfully engage and contribute to the cognitive ecology, and what the implications are for designing such learning environments. Drawing from the three conceptual dimensions of immersion - System, Narrative, and Agency - this work reinterprets AIs in immersive learning contexts. It outlines practical implications for designing learning environments where AIs are surrounded by external digital services, can interpret a narrative of origins, changes, and structural developments in data, and dynamically respond, making operational and tactical decisions that shape human-AI collaboration. Finally, this work suggests how these insights might influence the future of AI training, proposing that immersive learning theory can inform the development of AIs capable of evolving beyond static models. This paper paves the way for understanding AI as an immersive learner and participant in evolving human-AI cognitive ecosystems.

The rapid advancement of large language models (LLMs) has opened new possibilities for their adoption as evaluative judges. This paper introduces Themis, a fine-tuned LLM judge that delivers sophisticated context-aware evaluations. We provide a comprehensive overview of the development pipeline for Themis, highlighting its scenario-dependent evaluation prompts and two novel methods for controlled instruction generation. These designs enable Themis to effectively distill evaluative skills from teacher models, while retaining flexibility for continuous development. We introduce two human-labeled benchmarks for meta-evaluation, demonstrating that Themis can achieve high alignment with human preferences in an economical manner. Additionally, we explore insights into the LLM-as-a-judge paradigm, revealing nuances in performance and the varied effects of reference answers. Notably, we observe that pure knowledge distillation from strong LLMs, though common, does not guarantee performance improvement through scaling. We propose a mitigation strategy based on instruction-following difficulty. Furthermore, we provide practical guidelines covering data balancing, prompt customization, multi-objective training, and metric aggregation. We aim for our method and findings, along with the fine-tuning data, benchmarks, and model checkpoints, to support future research and development in this area.

Surnames often convey implicit markers of social status, wealth, and lineage, shaping perceptions in ways that can perpetuate systemic biases and intergenerational inequality. This study is the first of its kind to investigate whether and how surnames influence AI-driven decision-making, focusing on their effects across key areas such as hiring recommendations, leadership appointments, and loan approvals. Using 72,000 evaluations of 600 surnames from the United States and Thailand, two countries with distinct sociohistorical contexts and surname conventions, we classify names into four categories: Rich, Legacy, Normal, and phonetically similar Variant groups. Our findings show that elite surnames consistently increase AI-generated perceptions of power, intelligence, and wealth, which in turn influence AI-driven decisions in high-stakes contexts. Mediation analysis reveals perceived intelligence as a key mechanism through which surname biases influence AI decision-making process. While providing objective qualifications alongside surnames mitigates most of these biases, it does not eliminate them entirely, especially in contexts where candidate credentials are low. These findings highlight the need for fairness-aware algorithms and robust policy measures to prevent AI systems from reinforcing systemic inequalities tied to surnames, an often-overlooked bias compared to more salient characteristics such as race and gender. Our work calls for a critical reassessment of algorithmic accountability and its broader societal impact, particularly in systems designed to uphold meritocratic principles while counteracting the perpetuation of intergenerational privilege.

Despite remarkable capabilities, large language models (LLMs) struggle to continually update their knowledge without catastrophic forgetting. In contrast, humans effortlessly integrate new information, detect conflicts with existing beliefs, and selectively update their mental models. This paper introduces a cognitive-inspired investigation paradigm to study continual knowledge updating in LLMs. We implement two key components inspired by human cognition: (1) Dissonance and Familiarity Awareness, analyzing model behavior to classify information as novel, familiar, or dissonant; and (2) Targeted Network Updates, which track neural activity to identify frequently used (stubborn) and rarely used (plastic) neurons. Through carefully designed experiments in controlled settings, we uncover a number of empirical findings demonstrating the potential of this approach. First, dissonance detection is feasible using simple activation and gradient features, suggesting potential for cognitive-inspired training. Second, we find that non-dissonant updates largely preserve prior knowledge regardless of targeting strategy, revealing inherent robustness in LLM knowledge integration. Most critically, we discover that dissonant updates prove catastrophically destructive to the model's knowledge base, indiscriminately affecting even information unrelated to the current updates. This suggests fundamental limitations in how neural networks handle contradictions and motivates the need for new approaches to knowledge updating that better mirror human cognitive mechanisms.

We propose a novel approach for optimizing the graph ratio-cut by modeling the binary assignments as random variables. We provide an upper bound on the expected ratio-cut, as well as an unbiased estimate of its gradient, to learn the parameters of the assignment variables in an online setting. The clustering resulting from our probabilistic approach (PRCut) outperforms the Rayleigh quotient relaxation of the combinatorial problem, its online learning extensions, and several widely used methods. We demonstrate that the PRCut clustering closely aligns with the similarity measure and can perform as well as a supervised classifier when label-based similarities are provided. This novel approach can leverage out-of-the-box self-supervised representations to achieve competitive performance and serve as an evaluation method for the quality of these representations.

We present AlphaGeometry2, a significantly improved version of AlphaGeometry introduced in Trinh et al. (2024), which has now surpassed an average gold medalist in solving Olympiad geometry problems. To achieve this, we first extend the original AlphaGeometry language to tackle harder problems involving movements of objects, and problems containing linear equations of angles, ratios, and distances. This, together with other additions, has markedly improved the coverage rate of the AlphaGeometry language on International Math Olympiads (IMO) 2000-2024 geometry problems from 66% to 88%. The search process of AlphaGeometry2 has also been greatly improved through the use of Gemini architecture for better language modeling, and a novel knowledge-sharing mechanism that combines multiple search trees. Together with further enhancements to the symbolic engine and synthetic data generation, we have significantly boosted the overall solving rate of AlphaGeometry2 to 84% for $\textit{all}$ geometry problems over the last 25 years, compared to 54% previously. AlphaGeometry2 was also part of the system that achieved silver-medal standard at IMO 2024 https://dpmd.ai/imo-silver. Last but not least, we report progress towards using AlphaGeometry2 as a part of a fully automated system that reliably solves geometry problems directly from natural language input.

Pretraining large language models (LLMs) is resource-intensive, often requiring months of training time even with high-end GPU clusters. There are two approaches of mitigating such computational demands: reusing smaller models to train larger ones (upcycling), and training computationally efficient models like mixture-of-experts (MoE). In this paper, we study the upcycling of LLMs to MoE models, of which the scaling behavior remains underexplored. Through extensive experiments, we identify empirical scaling laws that describe how performance depends on dataset size and model configuration. Particularly, we show that, while scaling these factors improves performance, there is a novel interaction term between the dense and upcycled training dataset that limits the efficiency of upcycling at large computational budgets. Based on these findings, we provide guidance to scale upcycling, and establish conditions under which upcycling outperforms from-scratch trainings within budget constraints.

Citation Worthiness Detection (CWD) consists in determining which sentences, within an article or collection, should be backed up with a citation to validate the information it provides. This study, introduces ALPET, a framework combining Active Learning (AL) and Pattern-Exploiting Training (PET), to enhance CWD for languages with limited data resources. Applied to Catalan, Basque, and Albanian Wikipedia datasets, ALPET outperforms the existing CCW baseline while reducing the amount of labeled data in some cases above 80\%. ALPET's performance plateaus after 300 labeled samples, showing it suitability for low-resource scenarios where large, labeled datasets are not common. While specific active learning query strategies, like those employing K-Means clustering, can offer advantages, their effectiveness is not universal and often yields marginal gains over random sampling, particularly with smaller datasets. This suggests that random sampling, despite its simplicity, remains a strong baseline for CWD in constraint resource environments. Overall, ALPET's ability to achieve high performance with fewer labeled samples makes it a promising tool for enhancing the verifiability of online content in low-resource language settings.

Service robots operating in unstructured environments must effectively recognize and segment unknown objects to enhance their functionality. Traditional supervised learningbased segmentation techniques require extensive annotated datasets, which are impractical for the diversity of objects encountered in real-world scenarios. Unseen Object Instance Segmentation (UOIS) methods aim to address this by training models on synthetic data to generalize to novel objects, but they often suffer from the simulation-to-reality gap. This paper proposes a novel approach (ZISVFM) for solving UOIS by leveraging the powerful zero-shot capability of the segment anything model (SAM) and explicit visual representations from a selfsupervised vision transformer (ViT). The proposed framework operates in three stages: (1) generating object-agnostic mask proposals from colorized depth images using SAM, (2) refining these proposals using attention-based features from the selfsupervised ViT to filter non-object masks, and (3) applying K-Medoids clustering to generate point prompts that guide SAM towards precise object segmentation. Experimental validation on two benchmark datasets and a self-collected dataset demonstrates the superior performance of ZISVFM in complex environments, including hierarchical settings such as cabinets, drawers, and handheld objects. Our source code is available at https://github.com/Yinmlmaoliang/zisvfm.

Large Language Models (LLMs) have shown exceptional capabilities in Natural Language Processing (NLP) across diverse domains. However, their application in specialized tasks such as Legal Judgment Prediction (LJP) for low-resource languages like Arabic remains underexplored. In this work, we address this gap by developing an Arabic LJP dataset, collected and preprocessed from Saudi commercial court judgments. We benchmark state-of-the-art open-source LLMs, including LLaMA-3.2-3B and LLaMA-3.1-8B, under varying configurations such as zero-shot, one-shot, and fine-tuning using QLoRA. Additionally, we used a comprehensive evaluation framework combining quantitative metrics (BLEU and ROUGE) and qualitative assessments (Coherence, legal language, clarity). Our results demonstrate that fine-tuned smaller models achieve comparable performance to larger models in task-specific contexts while offering significant resource efficiency. Furthermore, we investigate the effects of prompt engineering and fine-tuning on model outputs, providing insights into performance variability and instruction sensitivity. By making the dataset, implementation code, and models publicly available, we establish a robust foundation for future research in Arabic legal NLP.