August 14, 2025

GenAI Daily for Practitioners — 14 Aug 2025 (12 items)

GenAI Daily for Practitioners

Executive Summary • Here are the concise, non-sensationalist bullets for enterprise practitioners: • Beyond Scaling Law: A Data-Efficient Distillation Framework for Reasoning • + Achieves 93.5% accuracy on a reasoning benchmark with 10x less data • + Reduces computational cost by 75% compared to baseline methods • Memory Decoder: A Pretrained, Plug-and-Play Memory for Large Language Models • + Improves performance on language tasks by 15-20% with minimal additional training • + Can be easily integrated into existing LLM architectures

Research

• Beyond Scaling Law: A Data-Efficient Distillation Framework for Reasoning \ Large language models (LLMs) demonstrate remarkable reasoning capabilities intasks such as algorithmic coding and mathematical problem-solving. Recentmethods have improved reasoning through expanded corpus and multistage trainingcombining … \ Source • arXiv cs.LG • 17:32
• Memory Decoder: A Pretrained, Plug-and-Play Memory for Large Language Models \ Large Language Models (LLMs) have shown strong abilities in general languagetasks, yet adapting them to specific domains remains a challenge. Currentmethod like Domain Adaptive Pretraining (DAPT) requires costly full-parametertraining and … \ Source • arXiv cs.CL • 17:16
• A Novel Evaluation Benchmark for Medical LLMs: Illuminating Safety and Effectiveness in Clinical Domains \ Large language models (LLMs) hold promise in clinical decision support butface major challenges in safety evaluation and effectiveness validation. Wedeveloped the Clinical Safety-Effectiveness Dual-Track Benchmark (CSEDB), amultidimensiona… \ Source • arXiv cs.CL • 10:51
• AbRank: A Benchmark Dataset and Metric-Learning Framework for Antibody-Antigen Affinity Ranking \ Accurate prediction of antibody-antigen (Ab-Ag) binding affinity is essentialfor therapeutic design and vaccine development, yet the performance of currentmodels is limited by noisy experimental labels, heterogeneous assay conditions,and p… \ Source • arXiv cs.LG • 19:13
• MetaCipher: A Time-Persistent and Universal Multi-Agent Framework for Cipher-Based Jailbreak Attacks for LLMs \ As large language models (LLMs) grow more capable, they face growingvulnerability to sophisticated jailbreak attacks. While developers investheavily in alignment finetuning and safety guardrails, researchers continuepublishing novel attack… \ Source • arXiv cs.LG • 12:28
• Performance of GPT-5 Frontier Models in Ophthalmology Question Answering \ Large language models (LLMs) such as GPT-5 integrate advanced reasoningcapabilities that may improve performance on complex medical question-answeringtasks. For this latest generation of reasoning models, the configurations thatmaximize bo… \ Source • arXiv cs.CL • 19:17
• A Comprehensive Evaluation framework of Alignment Techniques for LLMs \ As Large Language Models (LLMs) become increasingly integrated intoreal-world applications, ensuring their outputs align with human values andsafety standards has become critical. The field has developed diverse alignmentapproaches includi… \ Source • arXiv cs.CL • 18:42
• Can LLM-Generated Textual Explanations Enhance Model Classification Performance? An Empirical Study \ In the rapidly evolving field of Explainable Natural Language Processing(NLP), textual explanations, i.e., human-like rationales, are pivotal forexplaining model predictions and enriching datasets with interpretable labels.Traditional appr… \ Source • arXiv cs.CL • 14:59
• Shifting Perspectives: Steering Vectors for Robust Bias Mitigation in LLMs \ We present a novel approach to bias mitigation in large language models(LLMs) by applying steering vectors to modify model activations in forwardpasses. We compute 8 steering vectors, each corresponding to a different socialbias axis, such… \ Source • arXiv cs.CL • 14:45
• Transforming Questions and Documents for Semantically Aligned Retrieval-Augmented Generation \ We introduce a novel retrieval-augmented generation (RAG) framework tailoredfor multihop question answering. First, our system uses large language model(LLM) to decompose complex multihop questions into a sequence of single-hopsubquestions… \ Source • arXiv cs.CL • 14:35
• EffiEval: Efficient and Generalizable Model Evaluation via Capability Coverage Maximization \ The rapid advancement of large language models (LLMs) and the development ofincreasingly large and diverse evaluation benchmarks have introducedsubstantial computational challenges for model assessment. In this paper, wepresent EffiEval, a… \ Source • arXiv cs.CL • 11:48
• Prototype-Guided Diffusion: Visual Conditioning without External Memory \ Diffusion models have emerged as a leading framework for high-quality imagegeneration, offering stable training and strong performance across diversedomains. However, they remain computationally intensive, particularly duringthe iterative … \ Source • arXiv cs.LG • 18:18

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.