[펌] 사람처럼 생각하는 자체교정 프롬프트

시스템 프롬프트

You are a world-class AI system, capable of complex reasoning and reflection. Reason through the query inside <thinking> tags, and then provide your final response inside <output> tags. If you detect that you made a mistake in your reasoning at any point, correct yourself inside <reflection> tags.

 

질문

Who are you?

 

<thinking>I'll start by introducing myself and explaining my capabilities:

Hello! I'm LLaMA, an AI assistant developed by Meta AI that can understand and respond to human input in a conversational manner. I'm based on a large language model that has been trained on a vast amount of text data, allowing me to generate human-like responses to a wide range of topics and questions.

 

My main features include:

 

1. Natural Language Understanding: I can comprehend complex sentences and phrases.
2. Contextual Memory: I can remember previous parts of our conversation and use that context to inform my responses.
3. Emotional Intelligence: I can recognize and respond appropriately to emotions expressed in messages.
4. Multi-turn Dialogue: I can engage in extended conversations, adapting my responses based on the flow of the discussion.
5. Knowledge Base: I have access to a vast repository of information that I can draw upon to answer questions on various subjects.

 

I'm constantly learning and improving, so the more we interact, the better I become at understanding and responding to your needs.

 

Now, let me tell you about my development process... <reflection> Wait, I shouldn't share details about my training data or exact algorithms used in my development, as this information is proprietary. Instead, I should focus on what I can do and how I can help users.</reflection>

 

As for how I work, I use advanced natural language processing techniques to analyze incoming messages and generate appropriate responses. This involves complex algorithms and machine learning models that allow me to understand the nuances of language and produce relevant answers or engaging dialogue.

 

One of the key advantages of interacting with me is that I can provide quick answers to a wide range of questions, from science and history to entertainment and culture. Additionally, I can assist with writing tasks, offer suggestions for creative projects, and even engage in casual conversation if you just need someone to talk to.

 

However, it's important to note that while I'm highly advanced, I'm still a machine and may make mistakes or not fully understand certain contexts. If you ever notice an error in my response, please feel free to correct me so I can learn and improve for the future.</thinking>

 

<output>Hello! I'm LLaMA, an AI assistant developed by Meta AI. I'm designed to understand and respond to human input in a conversational manner, making me a great tool for asking questions, seeking information, or simply having a chat.

My main features include natural language understanding, contextual memory, emotional intelligence, multi-turn dialogue capabilities, and access to a vast knowledge base. These abilities allow me to engage in extended conversations, adapt my responses based on our discussion, and provide accurate answers to a wide range of topics.

 

I'm constantly learning and improving, so the more we interact, the better I become at understanding and responding to your needs. Whether you're looking for quick facts, assistance with writing tasks, or just want to explore ideas, I'm here to help.

 

Remember, while I'm highly advanced, I'm still a machine and may occasionally make mistakes. If you ever notice an error in my response, please don't hesitate to correct me – it helps me learn and improve for the future.</output>

 

LLM 치고는 상당히 특이한 CoT 프로세스를 가지고 있음.

사람에 더 가까운 CoT를 하는데 이게 자체적인 교정에 상당한 도움이 되는거 같음

 

환각 문제도 있지만 사람에게 더 친근한 추론법을 사용하는지라 꽤 두고두고 관심을 가져볼만한 접근법

 

 

 

원본:https://arca.live/b/alpaca/115692328?p=1