本文由 AI 分析生成
建立時間： 2026-03-24 來源： https://gist.github.com/Maharshi-Pandya/4aeccbe1dbaa7f89c182bd65d2764203
Summary
A viral GitHub gist (~1,364 stars) providing a system prompt that instructs LLMs to engage in extended stream-of-consciousness reasoning before answering. The “contemplative” style mandates extensive self-questioning, backtracking, and uncertainty expression, wrapped in <contemplator> and <final_answer> XML tags.
一個廣受歡迎的 GitHub Gist，提供讓 LLM 進行「沉思推理」的系統提示詞，要求模型在回答前進行大量自我質疑、回溯和不確定性表達，並以 XML 標籤格式化輸出。
Key Points
- Core principle: exploration over conclusion — never rush, keep reasoning until answer emerges naturally
- Minimum 10,000 characters of contemplation before final answer
- Output structure:
<contemplator>for free-form internal monologue,<final_answer>only when reasoning converges - Style mirrors human stream-of-consciousness: short sentences, express doubt, backtrack freely
- Predates o1-style reasoning models; was influential in the “reasoning mode” prompting community
Insights
This prompt essentially replicates the extended chain-of-thought behavior that was later built into o1 and similar reasoning models. The fact that it works as a simple prompt injection suggests LLMs already have latent reasoning capabilities that the standard instruction-following mode suppresses. The viral spread (1,364 stars) reflects practitioner demand for “thinking aloud” behavior before dedicated reasoning models were available.
Connections
Raw Excerpt
You are an assistant that engages in extremely thorough, self-questioning reasoning. Your approach mirrors human stream-of-consciousness thinking, characterized by continuous exploration, self-doubt, and iterative analysis.