Self-Evolving AI: LLM and Harness Together (RSI)

Recursive Self-Improvement (RSI) as the bleeding edge of Self-learning LLMs in parallel w/ optimizing Harness elements in one system-wide optimization. Small and Smart AI = The future for all non 8xB200 NVIDIA owner. Finally the "holy grail of AI": Reinforcement Learning of the LLM in parallel with the harness object optimization (prompt engineering, context engineering, loop engineering, system engineering). The first steps. With the complete pseudo-code and the maths explained in detail. Coding Agent Harness Self-improvement. Note: The latest text by Lilian Weng is the perfect introduction to this video. Read her first to understand the current system complexity (she summed up the main literature of the past 6 months). Then watch this video, because I'll go one step further than Lilian. The "RSI Roadmap": Lilian writes (see "Harness Engineering for Self-Improvement", July 4, 2026, https://lilianweng.github.io/posts/20... ) about the future of AI and states that the optimization targets of the AI industry have evolved to Recursive Self-Improvement (RSI): instruction prompts → structured context → workflow → harness code → optimizer code Lilian Weng outlined how AI is no longer just optimizing prompts, but rewriting its entire operating system (the harness). Now today, we - on this YouTube channel - are looking at a new paper (see below), that represents the absolute bleeding edge of this (also Lilian's) RSI roadmap. Self-improving AI systems (LLM and Harness), recursive, without humans in the loop. all rights w/ authors: "Harness-Aware Self-Evolving: Co-Evolving Model Weights, Harness, and Task Solutions" Haochen Luo1,3, Yi Huang2, Sichun Luo1,3, Fengyuan Liu1,3, Lei Li1,3, Zefa Hu2, Junlan Feng2, Qi Liu1,3 from 1 School of Computing and Data Science, The University of Hong Kong 2 Jiutian Research, China Mobile 3 Grace Investment Machine arXiv:2607.03935 #airesearch #aitechnology #aitech #aiexplained #science #scienceexperiment