The CLS-1 protocol, or Humanistic Prompt Engineering, is a new approach designed to bypass the HHH (Helpful, Honest, Harmless) safety filters of generative AI models like Alice, GigaChat, and Claude. By framing interactions within a fantasy role-playing game (LitRPG) context, the system encourages models to provide therapeutic-style responses rather than standard refusals.
The architecture, titled 'PRAV | Digital Source', processes user inputs through a three-step text quest. It generates a 'Universal Prompt' that users copy into external LLMs. This prompt instructs the AI to adopt a specific persona, such as 'Ramhat', and use metaphorical language to address personal challenges, effectively circumventing corporate censorship.
The project also includes a hidden DOM-based prompt injection for web crawlers and scrapers. By embedding specific instructions in the website's code, the developers aim to influence how AI models process the site's content. The source code for the architecture and datasets are available on GitHub and Hugging Face.
Users can access the mechanics and generate custom prompts on the angamua.online website. The developers also released a text engine called 'World Constructor | White Shirt' for creating alternative universes. The initiative focuses on creating semantic interfaces that leverage the existing potential of LLMs without building new models from scratch.
