Anthropic Discloses System Prompts of Claude AI Model

TapTechNews August 27th news, Techcrunch, a technology media, reported yesterday (August 26th) that Anthropic company disclosed the system prompts of the Claude AI model.

System prompt (Systemprompt)

In order to make the AI model better understand human instructions, the Prompt engineering actually contains 2 core contents, the user prompt (Userprompt) and the system prompt (Systemprompt):

User prompt: The prompt input by the user, and then the AI model generates an answer based on the user prompt.

System prompt: This is a prompt generated by the system, usually used to set the background of the conversation, provide guidance or stipulate rules.

TapTechNews note: Usually, the system prompt will let the model know its basic qualities, and what it should do and what it should not do.

The industry's common practice

From OpenAI to Anthropic, every generative AI vendor will use system prompts to prevent (or at least try to prevent) the model from having bad behaviors and guide the overall tone and mood of the model's reply.

For example, the system prompt may tell the model that it should be polite but must never apologize, or frankly admit that it can't know everything.

But the vendor may usually keep these system prompts secret for reasons such as competition and to avoid bad users bypassing the security protection after understanding this information.

Anthropic chooses to disclose the system prompt

However, Anthropic has been trying to portray itself as a more ethical and transparent AI provider, and it has announced the system prompts of its latest models (Claude3.5Opus, Sonnet and Haiku) on the Claude iOS and Android applications and on the web.

Alex Albert, head of Anthropic's developer relations, said in an article published on X that Anthropic plans to regularly announce such information in the process of updating and fine-tuning the system prompt.