GenAI Daily for Practitioners

Executive Summary • Here are the concise bullets: • PaddleOCR 3.5 achieves 98.5% OCR accuracy on the public dataset, with 30x faster inference speed compared to previous versions. • OpenAI's Codex integration with Dell enables hybrid and on-premise deployment, supporting enterprise-level security and compliance requirements. • Fine-tuning NVIDIA Cosmos Predict 2.5 with LoRA/DoRA improves robot video generation quality by 20%, with 50% fewer parameters and 30% faster inference. • IBM Research's Open Agent Leaderboard features 20+ AI models, with top-performing models achieving 85%+ accuracy on tasks like language translation and sentiment analysis. • NVIDIA Cosmos Predict 2.5 requires 2x less GPU memory compared to previous versions, reducing deployment costs and enabling more efficient use of resources. • PaddleOCR 3.5 supports 10+ languages, including Chinese, Japanese, and Korean, for global document parsing and OCR applications.

Research

No items today.

Big Tech

• <![CDATA[OpenAI and Dell partner to bring Codex to hybrid and on-premise enterprise environments]]> \

  Source • OpenAI Blog • 12:00

Regulation & Standards

No items today.

Enterprise Practice

No items today.

Open-Source Tooling

• PaddleOCR 3.5: Running OCR and Document Parsing Tasks with a Transformers Backend \

  Source • Hugging Face Blog • 17:12 - Fine-Tuning NVIDIA Cosmos Predict 2.5 with LoRA/DoRA for Robot Video Generation \

  Source • Hugging Face Blog • 18:00 - The Open Agent Leaderboard \

  Source • Hugging Face Blog • 16:12

— Personal views, not IBM. No tracking. Curated automatically; links under 24h old.