A mechanism-first reading of SchedCP, an agentic control plane that turns Linux scheduler tuning from kernel craft into workload-aware infrastructure optimisation.
A mechanism-first guide to why Agentic RL reframes LLMs as trainable policies operating across tools, memory, environments, and long-horizon business workflows.
L-MARS shows that retrieval helps legal AI when the answer depends on current authority, but adds little when the task is mainly legal reasoning.
A practical reading of GenDataCarto: how per-sample training dynamics can expose memorization hotspots before they become privacy, contamination, or governance failures.
A mechanism-first reading of how LLMs can make supply chain optimizers usable without pretending to replace the optimizer.
A mechanism-first reading of how sparse autoencoders can expose and steer the concepts an LLM uses when processing financial news.
A practical reading of why simple observation masking can beat expensive LLM summaries for software-engineering agents.
A mechanism-first look at why LLM scoring systems need both reasoning and calibration when text must become a precise number.
A mechanism-first reading of EconAgentic, a DePIN market simulation showing how token incentives, node-provider patience, and LLM-style decision rules may affect network inclusion, stability, and market value.
A practical reading of AICL as a testability layer for LLM applications, where QA shifts from exact assertions to observable, replayable behaviour.