The "alignment tax" of post-training is typically framed as a drop in task accuracy. We show it also involves a severe loss of calibration, making models overconfident, less reliable, and model outputs less diverse. We show that this trade-off can be navigated effectively via a simple post-hoc intervention: interpolating between a model's weights before and after alignment. Crucially, this is not a strict trade-off. We find that the process consistently reveals Pareto-optimal interpolations - models that improve accuracy beyond both parents while substantially recovering the calibration lost during alignment. Our work demonstrates that simple model merging provides a computationally efficient method for mitigating the full scope of the alignment tax, yielding models that are more capable and more reliable.

A popular method to adapt large language models (LLMs) to new tasks is in-context learning (ICL), which is effective but incurs high inference costs as context length grows. In this paper we propose a method to perform instruction induction, where we take training examples and reduce them to a compact but descriptive prompt that can achieve performance comparable to ICL over the full training set. Specifically, we propose PROMPT-MII, a reinforcement learning (RL) based framework to meta-learn an instruction induction model that can generate compact instructions on the fly for an arbitrary new dataset. We train on over 3,000 diverse classification datasets from the HuggingFace hub, and evaluate on 90 unseen tasks. PROMPT-MII improves downstream model quality by 4-9 F1 points (10-20% relative), matching ICL performance while requiring 3-13x fewer tokens.

Time series AI is crucial for analyzing dynamic web content, driving a surge of pre-trained large models known for their strong knowledge encoding and transfer capabilities across diverse tasks. However, given their energy-intensive training, inference, and hardware demands, using large models as a one-fits-all solution raises serious concerns about carbon footprint and sustainability. For a specific task, a compact yet specialized, high-performing model may be more practical and affordable, especially for resource-constrained users such as small businesses. This motivates the question: Can we build cost-effective lightweight models with large-model-like performance on core tasks such as forecasting? This paper addresses this question by introducing SVTime, a novel Small model inspired by large Vision model (LVM) forecasters for long-term Time series forecasting (LTSF). Recently, LVMs have been shown as powerful tools for LTSF. We identify a set of key inductive biases of LVM forecasters -- analogous to the "physics" governing their behaviors in LTSF -- and design small models that encode these biases through meticulously crafted linear layers and constraint functions. Across 21 baselines spanning lightweight, complex, and pre-trained large models on 8 benchmark datasets, SVTime outperforms state-of-the-art (SOTA) lightweight models and rivals large models with 10^3 fewer parameters than LVMs, while enabling efficient training and inference in low-resource settings.

In the domain of software development, LLMs have been utilized to automate tasks such as code translation, where source code from one programming language is translated to another while preserving its functionality. However, LLMs often struggle with long source codes that don't fit into the context window, which produces inaccurate translations. To address this, we propose a novel zero-shot code translation method that incorporates identifier replacement. By substituting user-given long identifiers with generalized placeholders during translation, our method allows the LLM to focus on the logical structure of the code, by reducing token count and memory usage, which improves the efficiency and cost-effectiveness of long code translation. Our empirical results demonstrate that our approach preserves syntactical and hierarchical information and produces translation results with reduced tokens.

The rise of Large Language Models (LLMs) has redefined Machine Translation (MT), enabling context-aware and fluent translations across hundreds of languages and textual domains. Despite their remarkable capabilities, LLMs often exhibit uneven performance across language families and specialized domains. Moreover, recent evidence reveals that these models can encode and amplify different biases present in their training data, posing serious concerns for fairness, especially in low-resource languages. To address these gaps, we introduce Translation Tangles, a unified framework and dataset for evaluating the translation quality and fairness of open-source LLMs. Our approach benchmarks 24 bidirectional language pairs across multiple domains using different metrics. We further propose a hybrid bias detection pipeline that integrates rule-based heuristics, semantic similarity filtering, and LLM-based validation. We also introduce a high-quality, bias-annotated dataset based on human evaluations of 1,439 translation-reference pairs. The code and dataset are accessible on GitHub: https://github.com/faiyazabdullah/TranslationTangles

The proliferation of mobile devices generates a massive volume of time series across various domains, where effective time series forecasting enables a variety of real-world applications. This study focuses on a new problem of source-free domain adaptation for time series forecasting. It aims to adapt a pretrained model from sufficient source time series to the sparse target time series domain without access to the source data, embracing data protection regulations. To achieve this, we propose TimePD, the first source-free time series forecasting framework with proxy denoising, where large language models (LLMs) are employed to benefit from their generalization capabilities. Specifically, TimePD consists of three key components: (1) dual-branch invariant disentangled feature learning that enforces representation-and gradient-wise invariance by means of season-trend decomposition; (2) lightweight, parameter-free proxy denoising that dynamically calibrates systematic biases of LLMs; and (3) knowledge distillation that bidirectionally aligns the denoised prediction and the original target prediction. Extensive experiments on real-world datasets offer insight into the effectiveness of the proposed TimePD, outperforming SOT A baselines by 9.3% on average. The widespread deployment of Internet-of-Things (IoT) sensors has produced massive time series data across domains (Sun et al., 2025; Wang et al., 2024a), including traffic (Kieu et al., 2024; Cirstea et al., 2022), weather (Hettige et al., 2024), and energy (Wu et al., 2020). Accurate time series forecasting is crucial, enabling effective decision-making across diverse domains (Liu et al., 2025a; 2024a; Campos et al., 2023). We are seeing impressive advances in machine learning, especially in deep learning, that are successful in effective feature extraction and value creation (Hettige et al., 2024; Liu et al., 2025b).

Reinforcement learning with verifiable rewards (RLVR) has recently enhanced the reasoning capabilities of large language models (LLMs), particularly for mathematical problem solving. However, a fundamental limitation remains: as the sampling budget increases, the advantage of RLVR-trained models over their pretrained bases often diminishes or even vanishes, revealing a strong dependence on the base model's restricted search space. We attribute this phenomenon to the widespread use of the reverse Kullback-Leibler (KL) divergence regularizer, whose mode-seeking behavior keeps the policy trapped inside the base model's support region and hampers wider exploration. To address this issue, we propose RAPO (Rewards-Aware Policy Optimization), an algorithm to promote broader yet focused exploration. Our method (i) utilizes the forward KL penalty to replace the reverse KL penalty for out-of-distribution exploration, and (ii) reweights the reference policy to facilitate adaptive in-distribution exploration. We train Qwen2.5-3B and 7B models with RAPO on the 8K SimpleRL-Zero dataset, without supervised fine-tuning, and evaluate them on AIME2024 and AIME2025. Results show that RAPO consistently improves problem-solving performance. Notably, RAPO enables models to surpass the base model's performance ceiling and solves previously intractable problems, advancing the frontier of RLVR for challenging reasoning tasks.

Benchmarks for competition-style reasoning have advanced evaluation in mathematics and programming, yet physics remains comparatively explored. Most existing physics benchmarks evaluate only final answers, which fail to capture reasoning processes, while recent stepwise methods rely on heuristic LLM-as-judge scoring or restrictive linear assumptions, limiting reliability and diagnostic validity. We introduce PRISM-Physics, a process-level evaluation framework and benchmark for complex physics reasoning problems. Solutions are represented as directed acyclic graphs (DAGs) of formulas, explicitly encoding causal dependencies among intermediate steps to enable fine-grained, interpretable, and theoretically grounded scoring. We prove the optimality of the DAG representation and the corresponding scoring policy. Combining with a fully rule-based method for symbolic formula equivalence matching that we developed, we ensure consistent validation across diverse formulations without heuristic judgments. Results show that our evaluation framework is more aligned with human experts' scoring. Experiments on state-of-the-art LLMs reveal persistent reasoning failures in physics, while step-level scoring offers both diagnostic insight and rich signals for later training. By combining structural rigor, theoretical guarantees, and symbolic validation, PRISM-Physics provides a principled foundation for advancing process-level evaluation and guiding the development of models with deeper scientific reasoning capabilities.

Model Organisms (MOs) research involves intentionally training models to exhibit specific failure modes, to serve as a testbed for study and development of mitigations (Hubinger et al., 2024; Denison et al., 2024; Marks et al., 2025). Prior work has introduced several types of MOs, including models that conceal capabilities unless a specific trigger is present in the input (Greenblatt et al., 2024b; van der Weij et al., 2025), fake alignment to evade safety measures (Greenblatt et al., 2024a), and display broad misalignment after being fine-tuned on a narrow distribution of harmful data (Bet-ley et al., 2025). The secret-keeping models trained in this work represent a novel class of MOs that refrain from revealing that they have certain factual knowledge. Auditing Language Models Our work contributes to the growing field of alignment auditing, which aims to systematically investigate whether a model pursues undesired or hidden objectives, rather than merely evaluating its surface-level behavior (Casper et al., 2024). A central methodology for validating such audits is to construct a testbed with a known ground truth, a principle applied in prior work (Schwettmann et al., 2023; Rager et al., 2025).

High-dimensional decision-making tasks, such as business partner selection, involve evaluating large candidate pools with heterogeneous numerical, categorical, and textual features. MAS, a hierarchical multi-agent framework that decomposes evaluation into three layers: a Planner Agent that designs strategies, Specialized Agents that perform role-specific assessments, and a Supervisor Agent that integrates their outputs. To support systematic evaluation, we also introduce a curated benchmark dataset of venture capital co-investments, featuring diverse firm attributes and ground-truth syndicates. MAS consistently outperforms single-agent and debate-based multi-agent baselines, achieving up to 10-15% higher match rates. Analysis of agent reasoning shows that planners are most responsive to domain-informed prompts, specialists produce complementary feature coverage, and supervisors play an important role in aggregation. Our implementation is available at this anonymous link. In real-world decision-making, practitioners often navigate high-dimensional data including extensive option sets and numerous evaluative features (Sandanayake et al., 2018; Sigle et al., 2023). Business partner selection which includes partner shortlisting and strategic alliance formation exemplifies this challenge (Mindruta et al., 2016): firms often face a vast pool of potential candidates, each described by diverse attributes ranging from quantitative indicators (e.g., financial metrics, geographic presence) to text-rich information (e.g., strategic fit, investment preferences) (Shah & Swaminathan, 2008). The scale and complexity of such data can easily overwhelm human decision-makers, incurring significant costs (Li et al., 2008). This underscores the need for intelligent systems capable of analyzing large candidate sets and diverse features. Large language models (LLMs) have emerged as promising tools for addressing reasoning tasks in data-rich domains (Lee et al., 2025; Mischler et al., 2024). With appropriate prompting (e.g., few-shot learning) or information retrieval techniques (e.g., RAG), these models can identify salient features using only feature and task descriptions, achieving performance comparable to established methods (Li et al., 2025a; Jeong et al., 2024).