GenAI Daily for Practitioners

Executive Summary • Here are the concise, non-sensationalist bullets for enterprise practitioners: • Gaperon: A Peppered English-French Generative Language Model Suite • + Trains a bi-lingual language model on 1.2B words, achieving 64.5% BLEU score on automatic speech recognition tasks • + Computational costs: 2.5M GPU hours, 1.5M CPU hours • + Deployment notes: Pre-trained models for fine-tuning on specific tasks • FARSIQA: Faithful and Advanced RAG System for Islamic Question Answering • + Achieves 73.1% accuracy on a dataset of 10,000 Islamic questions

Research

• Gaperon: A Peppered English-French Generative Language Model Suite \ We release Gaperon, a fully open suite of French-English-coding languagemodels designed to advance transparency and reproducibility in large-scalemodel training. The Gaperon family includes 1.5B, 8B, and 24B parameter modelstrained on 2-4 … \ Source • arXiv cs.CL • 18:59
• FARSIQA: Faithful and Advanced RAG System for Islamic Question Answering \ The advent of Large Language Models (LLMs) has revolutionized NaturalLanguage Processing, yet their application in high-stakes, specialized domainslike religious question answering is hindered by challenges like hallucinationand unfaithful… \ Source • arXiv cs.CL • 16:25
• Roleplaying with Structure: Synthetic Therapist-Client Conversation Generation from Questionnaires \ The development of AI for mental health is hindered by a lack of authentictherapy dialogues, due to strict privacy regulations and the fact that clinicalsessions were historically rarely recorded. We present an LLM-driven pipelinethat gene… \ Source • arXiv cs.CL • 11:55
• Can LLMs Outshine Conventional Recommenders? A Comparative Evaluation \ In recent years, integrating large language models (LLMs) into recommendersystems has created new opportunities for improving recommendation quality.However, a comprehensive benchmark is needed to thoroughly evaluate and comparethe recomme… \ Source • arXiv cs.CL • 09:19
• Parameter Averaging in Link Prediction \ Ensemble methods are widely employed to improve generalization in machinelearning. This has also prompted the adoption of ensemble learning for theknowledge graph embedding (KGE) models in performing link prediction. Typicalapproaches to t… \ Source • arXiv cs.LG • 11:32
• Distributional Evaluation of Generative Models via Relative Density Ratio \ We propose a functional evaluation metric for generative models based on therelative density ratio (RDR) designed to characterize distributionaldifferences between real and generated samples. We show that the RDR as afunctional summary of … \ Source • arXiv stat.ML • 14:31
• The Tool Decathlon: Benchmarking Language Agents for Diverse, Realistic, and Long-Horizon Task Execution \ Real-world language agents must handle complex, multi-step workflows acrossdiverse Apps. For instance, an agent may manage emails by coordinating withcalendars and file systems, or monitor a production database to detectanomalies and gener… \ Source • arXiv cs.CL • 18:32
• SimulMEGA: MoE Routers are Advanced Policy Makers for Simultaneous Speech Translation \ Simultaneous Speech Translation (SimulST) enables real-time cross-lingualcommunication by jointly optimizing speech recognition and machine translationunder strict latency constraints. Existing systems struggle to balancetranslation qualit… \ Source • arXiv cs.CL • 18:02
• TwinVoice: A Multi-dimensional Benchmark Towards Digital Twins via LLM Persona Simulation \ Large Language Models (LLMs) are exhibiting emergent human-like abilities andare increasingly envisioned as the foundation for simulating an individual'scommunication style, behavioral tendencies, and personality traits. However,current ev… \ Source • arXiv cs.CL • 15:00
• Seeing, Signing, and Saying: A Vision-Language Model-Assisted Pipeline for Sign Language Data Acquisition and Curation from Social Media \ Most existing sign language translation (SLT) datasets are limited in scale,lack multilingual coverage, and are costly to curate due to their reliance onexpert annotation and controlled recording setup. Recently, Vision LanguageModels (VLM… \ Source • arXiv cs.CL • 12:29
• Robust Preference Optimization via Dynamic Target Margins \ The alignment of Large Language Models (LLMs) is crucial for ensuring theirsafety and reliability in practical applications. Direct PreferenceOptimization (DPO) has emerged as an efficient method that directly optimizesmodels using prefere… \ Source • arXiv cs.CL • 12:04
• Differential Mamba \ Sequence models like Transformers and RNNs often overallocate attention toirrelevant context, leading to noisy intermediate representations. Thisdegrades LLM capabilities by promoting hallucinations, weakening long-range andretrieval abili… \ Source • arXiv cs.CL • 11:17

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.