A new framework called Language Construct Modeling (LCM) offers a novel way to control Large Language Models (LLMs) using only language, eliminating the need for code, plugins, or alterations to the LLM’s internal functions. Developed by Vincent Chong, LCM treats prompts not as static instructions, but as dynamic, semantic modules capable of creating logic, recursive structures, and stateful behaviors within the LLM.
LCM seeks to overcome the fragility and lack of reusability often associated with traditional prompting techniques. It achieves this through Meta Prompt Layering (MPL), which allows for the recursive definition of semantic layers, and a Regenerative Prompt Tree structure, where prompts can dynamically re-invoke other prompt chains. This approach enables language-native intent structuring, avoiding the need for external tools or plugin APIs.
According to Chong, LCM’s ability to treat prompts as semantic control units unlocks potential for structured memory management, behavior modulation through language, scalable prompt design, and internal agent-like architectures – all without relying on function calling or external tools.
A white paper (v1.13) detailing LCM, including appendices and a regenerative prompt chart verified via OpenTimestamps, is now available. Chong encourages developers and researchers to provide feedback and collaborate on this foundational framework. The full paper is accessible under the CC BY-SA 4.0 license via GitHub and OSF links provided in the Reddit release post.