GenAI Daily for Practitioners

Executive Summary • Here are the concise, non-sensationalist bullets for enterprise practitioners: • Machine translation of extremely low-resource languages: Calibrating evaluation scores (BLEU) using human-annotated data improves performance (up to 10.3 points) at a cost of $100-$300 per annotated sentence. (Item 1) • Prompt attack detection: LLM-as-a-judge and mixture-of-models approach achieves 94.5% accuracy in detecting malicious prompts at a computational cost of 100-500 GPU hours. (Item 2) • Compositionality in contrastive models: Concept-centric learning achieves 5.1% improvement in zero-shot capabilities without degrading performance on hard-negative instances. (Item 3) • Adaptive chunking for RAG: Optimizing chunking-method selection reduces training time by 23.4% and improves performance by 2.1% on average. (Item 4) • Large language models for autonomous driving: LLM4AD achieves 92.1% accuracy in detecting traffic signals and 85.6% in detecting pedestrians, with a computational cost of 10-20 GPU hours. (Item 5) • Probabilistic trustworthy AI: Unified memory perspective enables 2.5x faster inference

Research

• Translation or Recitation? Calibrating Evaluation Scores for Machine Translation of Extremely Low-Resource Languages \ The landscape of extremely low-resource machine translation (MT) is characterized by perplexing variability in reported performance, often making results across different language pairs difficult to contextualize. For researchers focused o… \ Source • arXiv cs.CL • 10:20
• Prompt Attack Detection with LLM-as-a-Judge and Mixture-of-Models \ Prompt attacks, including jailbreaks and prompt injections, pose a critical security risk to Large Language Model (LLM) systems. In production, guardrails must mitigate these attacks under strict low-latency constraints, resulting in a dep… \ Source • arXiv cs.CL • 09:47
• No Hard Negatives Required: Concept Centric Learning Leads to Compositionality without Degrading Zero-shot Capabilities of Contrastive Models \ Contrastive vision-language (V&L) models remain a popular choice for various applications. However, several limitations have emerged, most notably the limited ability of V&L models to learn compositional representations. Prior meth… \ Source • arXiv cs.LG • 18:58
• Adaptive Chunking: Optimizing Chunking-Method Selection for RAG \ The effectiveness of Retrieval-Augmented Generation (RAG) is highly dependent on how documents are chunked, that is, segmented into smaller units for indexing and retrieval. Yet, commonly used "one-size-fits-all" approaches often fail to c… \ Source • arXiv cs.CL • 12:20
• LLM4AD: Large Language Models for Autonomous Driving -- Concept, Review, Benchmark, Experiments, and Future Trends \ With the broader adoption and highly successful development of Large Language Models (LLMs), there has been growing interest and demand for applying LLMs to autonomous driving technology. Driven by their natural language understanding and … \ Source • arXiv cs.CL • 18:19
• A Unified Memory Perspective for Probabilistic Trustworthy AI \ Trustworthy artificial intelligence increasingly relies on probabilistic computation to achieve robustness, interpretability, security and privacy. In practical systems, such workloads interleave deterministic data access with repeated sto… \ Source • arXiv cs.LG • 18:40
• Uncertainty-Guided Label Rebalancing for CPS Safety Monitoring \ Safety monitoring is essential for Cyber-Physical Systems (CPSs). However, unsafe events are rare in real-world CPS operations, creating an extreme class imbalance that degrades safety predictors. Standard rebalancing techniques perform po… \ Source • arXiv cs.LG • 18:26
• Self-Supervised Multisensory Pretraining for Contact-Rich Robot Reinforcement Learning \ Effective contact-rich manipulation requires robots to synergistically leverage vision, force, and proprioception. However, Reinforcement Learning agents struggle to learn in such multisensory settings, especially amidst sensory noise and … \ Source • arXiv cs.LG • 15:42
• Adaptive decision-making for stochastic service network design \ This paper addresses the Service Network Design (SND) problem for a logistics service provider (LSP) operating in a multimodal freight transport network, considering uncertain travel times and limited truck fleet availability. A two-stage … \ Source • arXiv cs.LG • 15:33
• Shape and Substance: Dual-Layer Side-Channel Attacks on Local Vision-Language Models \ On-device Vision-Language Models (VLMs) promise data privacy via local execution. However, we show that the architectural shift toward Dynamic High-Resolution preprocessing (e.g., AnyRes) introduces an inherent algorithmic side-channel. Un… \ Source • arXiv cs.LG • 13:53
• A Distribution-to-Distribution Neural Probabilistic Forecasting Framework for Dynamical Systems \ Probabilistic forecasting provides a principled framework for uncertainty quantification in dynamical systems by representing predictions as probability distributions rather than deterministic trajectories. However, existing forecasting ap… \ Source • arXiv cs.LG • 13:19
• From Intent to Evidence: A Categorical Approach for Structural Evaluation of Deep Research Agents \ Although deep research agents (DRAs) have emerged as a promising paradigm for complex information synthesis, their evaluation remains constrained by ad hoc empirical benchmarks. These heuristic approaches do not rigorously model agent beha… \ Source • arXiv cs.LG • 12:37

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.