GenAI Daily for Practitioners

Executive Summary • Here are the bullets summarizing the news items for enterprise practitioners: • Forging Time Series with Language: A Large Language Model Approach to Synthetic Data Generation • + Generates synthetic time series data using language models, achieving competitive performance with human-generated data. • + Potential applications in data augmentation and anomaly detection. • + Research paper available at http://arxiv.org/abs/2505.17103v2. • Retrieval-Augmented Defense: Adaptive and Controllable Jailbreak Prevention for Large Language Models • + Proposes a retrieval-augmented defense mechanism for large language models, achieving up to 94% jailbreak prevention.

Research

• Forging Time Series with Language: A Large Language Model Approach to Synthetic Data Generation \ SDForger is a flexible and efficient framework for generating high-qualitymultivariate time series using LLMs. Leveraging a compact data representation,SDForger provides synthetic time series generation from a few samples andlow-computatio… \ Source • arXiv cs.CL • 17:31
• Retrieval-Augmented Defense: Adaptive and Controllable Jailbreak Prevention for Large Language Models \ Large Language Models (LLMs) remain vulnerable to jailbreak attacks, whichattempt to elicit harmful responses from LLMs. The evolving nature anddiversity of these attacks pose many challenges for defense systems, including(1) adaptation to… \ Source • arXiv cs.CL • 16:40
• Deep Video Discovery: Agentic Search with Tool Use for Long-form Video Understanding \ Long-form video understanding presents significant challenges due toextensive temporal-spatial complexity and the difficulty of question answeringunder such extended contexts. While Large Language Models (LLMs) havedemonstrated considerabl… \ Source • arXiv cs.CL • 09:39
• MedREK: Retrieval-Based Editing for Medical LLMs with Key-Aware Prompts \ LLMs hold great promise for healthcare applications, but the rapid evolutionof medical knowledge and errors in training data often cause them to generateoutdated or inaccurate information, limiting their applicability in high-stakesclinica… \ Source • arXiv cs.CL • 09:12
• RL-100: Performant Robotic Manipulation with Real-World Reinforcement Learning \ Real-world robotic manipulation in homes and factories demands reliability,efficiency, and robustness that approach or surpass skilled human operators. Wepresent RL-100, a real-world reinforcement learning training framework built ondiffus… \ Source • arXiv cs.LG • 15:09
• Contextual Tokenization for Graph Inverted Indices \ Retrieving graphs from a large corpus, that contain a subgraph isomorphic toa given query graph, is a core operation in many real-world applications. Whilerecent multi-vector graph representations and scores based on set alignment andconta… \ Source • arXiv cs.LG • 11:11
• Beyond Empathy: Integrating Diagnostic and Therapeutic Reasoning with Large Language Models for Mental Health Counseling \ Large language models (LLMs) hold significant potential for mental healthsupport, capable of generating empathetic responses and simulating therapeuticconversations. However, existing LLM-based approaches often lack the clinicalgrounding n… \ Source • arXiv cs.CL • 18:03
• XIFBench: Evaluating Large Language Models on Multilingual Instruction Following \ Large Language Models (LLMs) have demonstrated remarkableinstruction-following capabilities across various applications. However, theirperformance in multilingual settings lacks systematic investigation, withexisting evaluations lacking fi… \ Source • arXiv cs.CL • 10:40
• GTAlign: Game-Theoretic Alignment of LLM Assistants for Social Welfare \ Large Language Models (LLMs) have achieved remarkable progress in reasoning,yet sometimes produce responses that are suboptimal for users in tasks such aswriting, information seeking, or providing practical guidance. Conventionalalignment … \ Source • arXiv cs.LG • 19:54
• Scalable Multi-Task Learning for Particle Collision Event Reconstruction with Heterogeneous Graph Neural Networks \ The growing luminosity frontier at the Large Hadron Collider is challengingthe reconstruction and analysis of particle collision events. Increasedparticle multiplicities are straining latency and storage requirements at thedata acquisition… \ Source • arXiv cs.LG • 14:06
• Diversity-Aware Policy Optimization for Large Language Model Reasoning \ The reasoning capabilities of large language models (LLMs) have advancedrapidly, particularly following the release of DeepSeek R1, which has inspireda surge of research into data quality and reinforcement learning (RL)algorithms. Despite … \ Source • arXiv cs.LG • 13:40
• Image Hashing via Cross-View Code Alignment in the Age of Foundation Models \ Efficient large-scale retrieval requires representations that are bothcompact and discriminative. Foundation models provide powerful visual andmultimodal embeddings, but nearest neighbor search in these high-dimensionalspaces is computatio… \ Source • arXiv cs.LG • 11:21

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.