Thursday, August 6, 2026

1. DASyR-LLM: Domain-Aware Symbolic Regression with LLMs for Kinetic Model Discovery
   T2arXiv AI/CL/LG（论文）

   Kinetic model discovery in chemical engineering depends on accurate rate expressions, yet traditional symbolic regression often lacks domain knowledge and tends to explore physicochemically implausible models. This study introduces DASyR-LLM, which embeds a large language model into an iterative symbolic regression algorithm. The LLM plays dual roles: providing qualitative physicochemical critiques of candidate models and proposing new rate expressions guided by chemical knowledge. Across four in silico case studies, the framework reduces iterations needed to identify the ground-truth model by 41.7%-79.3%, with the LLM directly proposing correct model structures in over half of guided runs, while maintaining equivalent predictive performance (R²>0.98) on independent validation sets. This substantially reduces wet-lab experimental effort and paves the way for fully automated, domain-aware kinetic modeling pipelines.

—

—

1. Hardware Design and Security in the Era of Chiplets and LLMs: A Unified Analysis
   T2arXiv AI/CL/LG（论文）

   This paper presents a unified analysis of security challenges arising from heterogeneous 2.5D chiplet systems and LLM-driven EDA flows. It covers attacks across architectural, logical, and physical levels—including hardware stacks for LLM acceleration—and reviews defenses such as 2.5D split manufacturing and active interposers for isolated Root of Trust, as well as protections for LLM-based EDA pipelines. It also discusses how LLM systems can advance hardware security for chiplet-era systems. The authors argue that these two revolutions compound security risks, requiring co-design of trust architecture and AI-assisted toolchains rather than isolated fixes.

2. Mirendil inks $100M+ Google Cloud deal to scale self-improving AI
   T2TechCrunch：AI（RSS）

   AI startup Mirendil has signed a multi-year partnership with Google Cloud valued at over $100 million to significantly expand its compute infrastructure for self-improving AI research. The collaboration will leverage Google Cloud's AI-optimized platforms to scale systems designed to accelerate scientific discovery and speed up AI development. The deal underscores the heavy reliance of frontier AI research on massive computational resources and highlights how cloud providers are becoming central players in the AI infrastructure race. It marks one of the largest publicly reported commitments tied specifically to self-improving AI, signaling growing commercial interest in this emerging field.

—