What I really want to know is this: if nature itself provides the feedback, and reinforcement learning is the paradigm, can a model's intelligence surpass human intelligence?
I am using one of humanity's finest achievements to date—LLMs—to explore their interaction, agency, and evolution within real textual environments. For a language model, the harness is the environment in which it acts, receives feedback, and changes—its natural world.
Reconstructing knowledge from classic books, not merely taking reading notes. It models concepts, relationships, and learning paths at a high level of abstraction, with the goal of turning classic works into teaching indexes and curricula for the AI era. The structure is not designed for linear reading; it is designed for agents to understand, organize, and teach knowledge.
My personal research workspace, where agents participate in literature discovery, domain learning, idea review, experimental design, and execution—an exploration of how frontier intelligence can become part of real scientific workflows.
Not ready to talk about this yet. A new self-evolving harness paradigm.
-
Backend Engineer · NeuroXess
A life-science company focused on the research and clinical application of implantable, flexible brain-computer interfaces.
I also enjoy contributing to open-source projects around AI agents—submitting PRs, fixing issues, and extending useful tools. If you are building the next generation of agents, I would love to collaborate.
Computer Science student at Shenyang Agricultural University, 2023 cohort.