A Hong Kong-based developer, Vincent Chong, has introduced SLS (Semantic Logic System), a groundbreaking framework that enables control of Large Language Models (LLMs) entirely through semantic prompting. Departing from conventional methods that necessitate code, APIs, or external tools, SLS harnesses the inherent power of language to dictate model behavior, memory, and responses.
SLS is built upon five key modules:
* **Meta Prompt Layering (MPL):** Creating layered semantic prompts for nuanced control.
* **Semantic Directive Prompting (SDP):** Defining roles, behaviors, and limitations using natural language instructions.
* **Intent Layer Structuring (ILS):** Steering the model through desired outcomes rather than explicit commands.
* **Semantic Snapshot Systems:** Saving and retrieving internal model states using natural language descriptions.
* **Symbolic Semantic Rhythm:** Ensuring consistent tone and logical flow in generated content.
Chong highlights that SLS is more than just a collection of prompt templates; it’s a fundamental architecture for “semantic programmability.” By strategically structuring prompts, users can develop recursive logic, agent-like functionalities, and modular reasoning capabilities directly within the LLM.
Two white papers detailing the architecture and modules of the system, built using GPT-4o without any plugins or coding, are now available:
* SLS 1.0: [GitHub](https://github.com/chonghin33/semantic-logic-system-1.0), [OSF](https://osf.io/9gtdf/)
* LCM v1.13: [GitHub](https://github.com/chonghin33/lcm-1.13-whitepaper), [OSF DOI](https://doi.org/10.17605/OSF.IO/4FEAZ)
Chong is actively seeking collaboration, with a vision of SLS as a scalable foundation for modular agents, recursive cognitive processes, and future advancements in AI logic layers.