GenAI Daily for Practitioners

Executive Summary • Here are the concise, non-sensationalist bullets for enterprise practitioners: • LiveTradeBench: Achieved 12.1% average return on investment with large language models in real-world trading scenarios. Estimated costs: $10,000 - $50,000 per year. (Source: arxiv.org/abs/2511.03628v1) • VoiceAgentBench: Tested voice assistants on 14 agentic tasks, achieving 64.2% average accuracy. Results suggest significant improvement potential with task-specific training. (Source: arxiv.org/abs/2510.07978v2) • PDE-SHARP: Developed a hybrid PDE solver with 2.5x speedup and 1.2x accuracy improvement compared to state-of-the-art methods. Potential applications in physics, engineering, and finance. (Source: arxiv.org/abs/2511.00183v2) • TabTune: Introduced a unified library for inference and fine-tuning tabular foundation models, achieving 1.4x speedup and 0.8x memory reduction. (Source: arxiv.org/abs/2511.02802v2) • SHIELD: Developed an anomaly detection system for Healthcare IoT

Research

• LiveTradeBench: Seeking Real-World Alpha with Large Language Models \ Large language models (LLMs) achieve strong performance acrossbenchmarks--from knowledge quizzes and math reasoning to web-agent tasks--butthese tests occur in static settings, lacking real dynamics and uncertainty.Consequently, they evalu… \ Source • arXiv cs.CL • 17:47
• VoiceAgentBench: Are Voice Assistants ready for agentic tasks? \ Large-scale Speech Language Models (SpeechLMs) have enabled voice assistantscapable of understanding natural spoken queries and performing complex tasks.However, existing speech benchmarks primarily focus on isolated capabilitiessuch as tr… \ Source • arXiv cs.CL • 08:44
• PDE-SHARP: PDE Solver Hybrids through Analysis and Refinement Passes \ Current LLM-driven approaches using test-time computing to generate PDEsolvers execute a large number of solver samples to identify high-accuracysolvers. These paradigms are especially costly for complex PDEs requiringsubstantial computati… \ Source • arXiv cs.LG • 18:58
• 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 • 18:36
• SHIELD: Securing Healthcare IoT with Efficient Machine Learning Techniques for Anomaly Detection \ The integration of IoT devices in healthcare introduces significant securityand reliability challenges, increasing susceptibility to cyber threats andoperational anomalies. This study proposes a machine learning-driven frameworkfor (1) det… \ Source • arXiv cs.LG • 18:20
• RAGBoost: Efficient Retrieval-Augmented Generation with Accuracy-Preserving Context Reuse \ Retrieval-augmented generation (RAG) enhances large language models (LLMs)with retrieved context but often suffers from downgraded prefill performance asmodern applications demand longer and more complex inputs. Existing cachingtechniques … \ Source • arXiv cs.LG • 14:59
• Watermarking Large Language Models in Europe: Interpreting the AI Act in Light of Technology \ To foster trustworthy Artificial Intelligence (AI) within the European Union,the AI Act requires providers to mark and detect the outputs of theirgeneral-purpose models. The Article 50 and Recital 133 call for marking methodsthat are ''suf… \ Source • arXiv cs.CL • 18:00
• R2R: Efficiently Navigating Divergent Reasoning Paths with Small-Large Model Token Routing \ Large Language Models (LLMs) achieve impressive reasoning capabilities at thecost of substantial inference overhead, posing substantial deploymentchallenges. Although distilled Small Language Models (SLMs) significantlyenhance efficiency, … \ Source • arXiv cs.CL • 17:39
• HaluMem: Evaluating Hallucinations in Memory Systems of Agents \ Memory systems are key components that enable AI systems such as LLMs and AIagents to achieve long-term learning and sustained interaction. However, duringmemory storage and retrieval, these systems frequently exhibit memoryhallucinations,… \ Source • arXiv cs.CL • 15:37
• Silenced Biases: The Dark Side LLMs Learned to Refuse \ Safety-aligned large language models (LLMs) are becoming increasinglywidespread, especially in sensitive applications where fairness is essentialand biased outputs can cause significant harm. However, evaluating the fairnessof models is a … \ Source • arXiv cs.CL • 12:24
• LexTime: A Benchmark for Temporal Ordering of Legal Events \ Understanding temporal relationships and accurately reconstructing the eventtimeline is important for case law analysis, compliance monitoring, and legalsummarization. However, existing benchmarks lack specialized languageevaluation, leavi… \ Source • arXiv cs.CL • 09:52
• PhysicsEval: Inference-Time Techniques to Improve the Reasoning Proficiency of Large Language Models on Physics Problems \ The discipline of physics stands as a cornerstone of human intellect, drivingthe evolution of technology and deepening our understanding of the fundamentalprinciples of the cosmos. Contemporary literature includes some works centeredon the… \ Source • arXiv cs.CL • 08:50

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.