A user on Less Wrong has developed a method for controlling large language model output by embedding specific promptware rules within the context window. These rules, such as 'CoT-forcing' and 'tree_rule', guide the model's generation process to produce more detailed and structured responses. The user details a system directive that includes various rules for termination, length control, jargon usage, and response pruning, aiming to eliminate distracting elements and ensure precise output, particularly for the Gemini model. AI
IMPACT Provides advanced techniques for controlling LLM output and eliciting more structured responses.
RANK_REASON User-generated guide on prompt engineering techniques for LLMs.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
