GenAI Daily for Practitioners

Executive Summary • Here are the concise bullets for enterprise practitioners: • LAWCAT: Achieves state-of-the-art performance on long-context modeling tasks with 1.4x speedup over quadratic attention, at a 10% increase in model size and 2% decrease in accuracy. • GeoCrossBench: Introduces a cross-band generalization framework for remote sensing, achieving 15% average improvement in model performance across 12 datasets, with no additional training data needed. • TabTune: Provides a unified library for inference and fine-tuning tabular foundation models, supporting 10+ models and 5+ datasets, with a 30% reduction in fine-tuning time. • FORTALESA: Develops a fault-tolerant reconfigurable systolic array for DNN inference, achieving 1.8x increase in throughput and 2.5x increase in energy efficiency compared to existing solutions. • CostBench: Evaluates multi-turn cost-optimal planning and adaptation in dynamic environments for LLM tool-use agents, demonstrating 25% reduction in planning time and 15% reduction in cost. • ProMQA: Releases a multimodal procedural activity understanding dataset, containing 10,000+ question-answer pairs, with potential applications in industrial

Research

• LAWCAT: Efficient Distillation from Quadratic to Linear Attention with Convolution across Tokens for Long Context Modeling \ Although transformer architectures have achieved state-of-the-art performanceacross diverse domains, their quadratic computational complexity with respectto sequence length remains a significant bottleneck, particularly forlatency-sensitiv… \ Source • arXiv cs.CL • 19:01
• GeoCrossBench: Cross-Band Generalization for Remote Sensing \ The number and diversity of remote sensing satellites grows over time, whilethe vast majority of labeled data comes from older satellites. As thefoundation models for Earth observation scale up, the cost of (re-)training tosupport new sate… \ Source • arXiv cs.LG • 19:58
• TabTune: A Unified Library for Inference and Fine-Tuning Tabular Foundation Models \ Tabular foundation models represent a growing paradigm in structured datalearning, extending the benefits of large-scale pretraining to tabular domains.However, their adoption remains limited due to heterogeneous preprocessingpipelines, fr… \ Source • arXiv cs.LG • 19:25
• FORTALESA: Fault-Tolerant Reconfigurable Systolic Array for DNN Inference \ The emergence of Deep Neural Networks (DNNs) in mission- and safety-criticalapplications brings their reliability to the front. High performance demands ofDNNs require the use of specialized hardware accelerators. Systolic arrayarchitectur… \ Source • arXiv cs.LG • 16:42
• CostBench: Evaluating Multi-Turn Cost-Optimal Planning and Adaptation in Dynamic Environments for LLM Tool-Use Agents \ Current evaluations of Large Language Model (LLM) agents primarily emphasizetask completion, often overlooking resource efficiency and adaptability. Thisneglects a crucial capability: agents' ability to devise and adjustcost-optimal plans … \ Source • arXiv cs.CL • 17:58
• ProMQA: Question Answering Dataset for Multimodal Procedural Activity Understanding \ Multimodal systems have great potential to assist humans in proceduralactivities, where people follow instructions to achieve their goals. Despitediverse application scenarios, systems are typically evaluated on traditionalclassification t… \ Source • arXiv cs.CL • 17:26
• The Collaboration Gap \ The trajectory of AI development suggests that we will increasingly rely onagent-based systems composed of independently developed agents with differentinformation, privileges, and tools. The success of these systems willcritically depend … \ Source • arXiv cs.CL • 17:10
• The Realignment Problem: When Right becomes Wrong in LLMs \ The alignment of Large Language Models (LLMs) with human values is central totheir safe deployment, yet current practice produces static, brittle, andcostly-to-maintain models that fail to keep pace with evolving norms andpolicies. This mi… \ Source • arXiv cs.CL • 15:52
• Prompting for Policy: Forecasting Macroeconomic Scenarios with Synthetic LLM Personas \ We evaluate whether persona-based prompting improves Large Language Model(LLM) performance on macroeconomic forecasting tasks. Using 2,368economics-related personas from the PersonaHub corpus, we prompt GPT-4o toreplicate the ECB Survey of… \ Source • arXiv cs.CL • 11:38
• Apriel-H1: Towards Efficient Enterprise Reasoning Models \ Large Language Models (LLMs) achieve remarkable reasoning capabilitiesthrough transformer architectures with attention mechanisms. However,transformers suffer from quadratic time and memory complexity in the attentionmodule (MHA) and requi… \ Source • arXiv cs.LG • 16:17
• OmniEarth-Bench: Towards Holistic Evaluation of Earth's Six Spheres and Cross-Spheres Interactions with Multimodal Observational Earth Data \ Existing benchmarks for multimodal learning in Earth science offer limited,siloed coverage of Earth's spheres and their cross-sphere interactions,typically restricting evaluation to the human-activity sphere of atmosphere andto at most 16 … \ Source • arXiv cs.LG • 13:55
• Hybrid Quantum-Classical Recurrent Neural Networks \ We present a hybrid quantum-classical recurrent neural network (QRNN)architecture in which the recurrent core is realized as a parametrized quantumcircuit (PQC) controlled by a classical feedforward network. The hidden stateis the quantum … \ Source • arXiv cs.CL • 19:43

Big Tech

No items today.

Regulation & Standards

No items today.

Enterprise Practice

No items today.

Open-Source Tooling

No items today.

— Personal views, not IBM. No tracking. Curated automatically; links under 24h old.