Manual curation of evaluation datasets is struggling to keep up with the rapidly expanding capabilities and deployment scenarios of language models. Towards scalable model profiling, we introduce and validate a framework for evaluating LLMs, called Adaptive Evaluations. Adaptive evaluations use scaffolded language models (evaluator agents) to search through a target model's behavior on a domain dataset and create difficult questions (tasks) that can discover and probe the model's failure modes. We find that frontier models lack consistency when adaptively probed with our framework on a diverse suite of datasets and tasks, including but not limited to legal reasoning, forecasting, and online harassment. Generated questions pass human validity checks and often transfer to other models with different capability profiles, demonstrating that adaptive evaluations can also be used to create difficult domain-specific datasets.

Lisa Amini (IBM Research), Henr y F. Kor th (Lehigh University), Nita Patel (Otis), Evan Peck (University of Colorado Boulder), Ben Zorn (Microsoft) It is believed by some that we are entering a new age of technology, characterized by advanced, per vasive Ar tificial Intelligence (AI), during which the rate of workforce and economic disruption will be substantially greater than previous periods. Regardless of whether a new era has commenced, AI is increasing in capability, speeding integration into the workplace and our homes, and prevailing in both technical and non-technical contexts and occupations. New skills and professions -- many of which are not yet conceived -- will arise, as will widespread job displacement. Just as the Information Age required national imperatives for computing education, similar imperatives exist for the rise of AI. In a sur vey of 4702 CEOs, 70 percent say AI will significantly change the way their companies create, deliver, and capture value over the next three years, and 45 percent believe their companies will no longer be viable in ten years if they continue on their current path.

An energy-efficient learning solution for the Agile Earth Observation Satellite Scheduling Problem Antonio M. Mercado-Mart ınez, Beatriz Soret Senior Member, IEEE, Antonio Jurado-Navas Member, IEEE Abstract --The Agile Earth Observation Satellite Scheduling Problem (AEOSSP) entails finding the subset of observation targets to be scheduled along the satellite's orbit while meeting operational constraints of time, energy and memory. The problem of deciding what and when to observe is inherently complex, and becomes even more challenging when considering several issues that compromise the quality of the captured images, such as cloud occlusion, atmospheric turbulence, and image resolution. This paper presents a Deep Reinforcement Learning (DRL) approach for addressing the AEOSSP with time-dependent profits, integrating these three factors to optimize the use of energy and memory resources. The proposed method involves a dual decision-making process: selecting the sequence of targets and determining the optimal observation time for each. Our results demonstrate that the proposed algorithm reduces the capture of images that fail to meet quality requirements by > 60% and consequently decreases energy waste from attitude maneuvers by up to 78%, all while maintaining strong observation performance. I NTRODUCTION One of the most relevant advances in the realm of Earth Observation (EO) has been the introduction of Agile Earth Observation Satellites (AEOS) [1]. Unlike Conventional Earth Observation Satellites (CEOS), which can only adjust their attitude along the roll axis, AEOS have a strong attitude adjustment capability along three axes (roll, pitch, and yaw).

Speech to speech translation (S2ST) is a transformative technology that bridges global communication gaps, enabling real time multilingual interactions in diplomacy, tourism, and international trade. Our review examines the evolution of S2ST, comparing traditional cascade models which rely on automatic speech recognition (ASR), machine translation (MT), and text to speech (TTS) components with newer end to end and direct speech translation (DST) models that bypass intermediate text representations. While cascade models offer modularity and optimized components, they suffer from error propagation, increased latency, and loss of prosody. In contrast, direct S2ST models retain speaker identity, reduce latency, and improve translation naturalness by preserving vocal characteristics and prosody. However, they remain limited by data sparsity, high computational costs, and generalization challenges for low-resource languages. The current work critically evaluates these approaches, their tradeoffs, and future directions for improving real time multilingual communication.

Fine-tuning of Large Language Models (LLMs) for downstream tasks, performed on domain-specific data has shown significant promise. However, commercial use of such LLMs is limited by the high computational cost required for their deployment at scale. On the other hand, small Language Models (LMs) are much more cost effective but have subpar performance in a similar setup. This paper presents our approach to finetuning a small LM, that reaches high accuracy in multiple choice question answering task. We achieve this by improving data quality at each stage of the LM training pipeline. In particular, we start with data structuring resulting in extraction of compact, semantically meaningful text chunks used by a retriever. This allows more efficient knowledge digestion by the LM. Further, we improve the retrieved context by training a lightweight Chunk Re-Ranker (CRR) that generates more accurate relative relevance chunk scores. Finally, we improve the model generalization ability by merging the models fine-tuned with different parameters on different data subsets. We present detailed procedure descriptions, and corresponding experimental findings that show the improvements of each one of the proposed techniques.

Intelligence is a crucial trait for species to find solutions within a limited number of trial-and-error attempts. Building on this idea, we introduce Survival Game as a framework to evaluate intelligence based on the number of failed attempts in a trial-and-error process. Fewer failures indicate higher intelligence. When the expectation and variance of failure counts are both finite, it signals the ability to consistently find solutions to new challenges, which we define as the Autonomous Level of intelligence. Using Survival Game, we comprehensively evaluate existing AI systems. Our results show that while AI systems achieve the Autonomous Level in simple tasks, they are still far from it in more complex tasks, such as vision, search, recommendation, and language. While scaling current AI technologies might help, this would come at an astronomical cost. Projections suggest that achieving the Autonomous Level for general tasks would require $10^{26}$ parameters. To put this into perspective, loading such a massive model requires so many H100 GPUs that their total value is $10^{7}$ times that of Apple Inc.'s market value. Even with Moore's Law, supporting such a parameter scale would take $70$ years. This staggering cost highlights the complexity of human tasks and the inadequacies of current AI technologies. To further investigate this phenomenon, we conduct a theoretical analysis of Survival Game and its experimental results. Our findings suggest that human tasks possess a criticality property. As a result, Autonomous Level requires a deep understanding of the task's underlying mechanisms. Current AI systems, however, do not fully grasp these mechanisms and instead rely on superficial mimicry, making it difficult for them to reach an autonomous level. We believe Survival Game can not only guide the future development of AI but also offer profound insights into human intelligence.

Learning from Demonstrations (LfD) allows robots to learn skills from human users, but its effectiveness can suffer due to sub-optimal teaching, especially from untrained demonstrators. Active LfD aims to improve this by letting robots actively request demonstrations to enhance learning. However, this may lead to frequent context switches between various task situations, increasing the human cognitive load and introducing errors to demonstrations. Moreover, few prior studies in active LfD have examined how these active query strategies may impact human teaching in aspects beyond user experience, which can be crucial for developing algorithms that benefit both robot learning and human teaching. To tackle these challenges, we propose an active LfD method that optimizes the query sequence of online human demonstrations via Curriculum Learning (CL), where demonstrators are guided to provide demonstrations in situations of gradually increasing difficulty. We evaluate our method across four simulated robotic tasks with sparse rewards and conduct a user study (N=26) to investigate the influence of active LfD methods on human teaching regarding teaching performance, post-guidance teaching adaptivity, and teaching transferability. Our results show that our method significantly improves learning performance compared to three other LfD baselines in terms of the final success rate of the converged policy and sample efficiency. Additionally, results from our user study indicate that our method significantly reduces the time required from human demonstrators and decreases failed demonstration attempts. It also enhances post-guidance human teaching in both seen and unseen scenarios compared to another active LfD baseline, indicating enhanced teaching performance, greater post-guidance teaching adaptivity, and better teaching transferability achieved by our method.

Text-to-SQL, the task of translating natural language questions into SQL queries, plays a crucial role in enabling non-experts to interact with databases. While recent advancements in large language models (LLMs) have significantly enhanced text-to-SQL performance, existing approaches face notable limitations in real-world text-to-SQL applications. Prompting-based methods often depend on closed-source LLMs, which are expensive, raise privacy concerns, and lack customization. Fine-tuning-based methods, on the other hand, suffer from poor generalizability due to the limited coverage of publicly available training data. To overcome these challenges, we propose a novel and scalable text-to-SQL data synthesis framework for automatically synthesizing large-scale, high-quality, and diverse datasets without extensive human intervention. Using this framework, we introduce SynSQL-2.5M, the first million-scale text-to-SQL dataset, containing 2.5 million samples spanning over 16,000 synthetic databases. Each sample includes a database, SQL query, natural language question, and chain-of-thought (CoT) solution. Leveraging SynSQL-2.5M, we develop OmniSQL, a powerful open-source text-to-SQL model available in three sizes: 7B, 14B, and 32B. Extensive evaluations across nine datasets demonstrate that OmniSQL achieves state-of-the-art performance, matching or surpassing leading closed-source and open-source LLMs, including GPT-4o and DeepSeek-V3, despite its smaller size. We release all code, datasets, and models to support further research.

Test-time adaptation aims to adapt a well-trained model to potential distribution shifts at test time using only unlabeled test data, without access to the original training data. While previous efforts mainly focus on a single modality, test-time distribution shift in the multi-modal setting is more complex and calls for new solutions. This paper tackles the problem of multi-modal test-time adaptation by proposing a novel method named Attention Bootstrapping with Principal Entropy Minimization (ABPEM). We observe that test-time distribution shift causes misalignment across modalities, leading to a large gap between intra-modality discrepancies (measured by self-attention) and inter-modality discrepancies (measured by cross-attention). We name this the attention gap. This attention gap widens with more severe distribution shifts, hindering effective modality fusion. To mitigate this attention gap and encourage better modality fusion, we propose attention bootstrapping that promotes cross-attention with the guidance of self-attention. Moreover, to reduce the gradient noise in the commonly-used entropy minimization, we adopt principal entropy minimization, a refinement of entropy minimization that reduces gradient noise by focusing on the principal parts of entropy, excluding less reliable gradient information. Extensive experiments on the benchmarks validate the effectiveness of the proposed ABPEM in comparison with competing baselines.

Vision-Language Models (VLMs) excel in integrating visual and textual information for vision-centric tasks, but their handling of inconsistencies between modalities is underexplored. We investigate VLMs' modality preferences when faced with visual data and varied textual inputs in vision-centered settings. By introducing textual variations to four vision-centric tasks and evaluating ten Vision-Language Models (VLMs), we discover a \emph{``blind faith in text''} phenomenon: VLMs disproportionately trust textual data over visual data when inconsistencies arise, leading to significant performance drops under corrupted text and raising safety concerns. We analyze factors influencing this text bias, including instruction prompts, language model size, text relevance, token order, and the interplay between visual and textual certainty. While certain factors, such as scaling up the language model size, slightly mitigate text bias, others like token order can exacerbate it due to positional biases inherited from language models. To address this issue, we explore supervised fine-tuning with text augmentation and demonstrate its effectiveness in reducing text bias. Additionally, we provide a theoretical analysis suggesting that the blind faith in text phenomenon may stem from an imbalance of pure text and multi-modal data during training. Our findings highlight the need for balanced training and careful consideration of modality interactions in VLMs to enhance their robustness and reliability in handling multi-modal data inconsistencies.