arXiv Giant Watch - Weekly - 2026-09-24

Welcome to this week’s arXiv Giant Watch. The strongest organization-backed signals are not centered on another larger base model. They are about the systems that make advanced models useful in the world. Chapters 00:00 Headline 01:03 DeepSeek Elastic Compute (DSec): A Sandbox Infrastructure for Effective Agentic Training at Scale 02:21 CORAL: An LLM-Native Harness for Production Recommender Systems 03:38 DeepSeek-V4-Flash on AMD gfx90a: Correctness Recovery and Inference Performance Engineering 04:52 Continuous Optimization for p-adic Models 06:12 Research Themes 07:29 Takeaway 08:36 Sources Papers 1. DeepSeek Elastic Compute (DSec): A Sandbox Infrastructure for Effective Agentic Training at Scale https://arxiv.org/abs/2609.22978 2. CORAL: An LLM-Native Harness for Production Recommender Systems https://arxiv.org/abs/2609.02730 3. DeepSeek-V4-Flash on AMD gfx90a: Correctness Recovery and Inference Performance Engineering https://arxiv.org/abs/2609.15627 4. Continuous Optimization for p-adic Models https://arxiv.org/abs/2609.25501 About Narrated from an automatically generated arXiv trending report. Rankings are inferred, not readership statistics; arXiv publishes no official trending chart. Claims described here are the papers' authors' own, with their stated caveats. #arXiv #arxiv-giant-watch #research #AI