Recently, artificial intelligence company Anthropic has set up a "System Prompt" page on its release announcement website to record and publicly disclose updates to the default prompts of its AI chatbot Claude. According to Alex Albert from Anthropic, this move aims to increase transparency and allow users to understand the changes in Claude's underlying workings.

The information released this time shows that the latest version of Claude, Claude 3.5 Sonnet, has made adjustments in system design, particularly in handling requests that cannot or do not want to be executed. This change reflects Anthropic's continuous efforts in optimizing user experience and enhancing AI conversation logic.

Specifically, Claude's web interface (Claude.ai) and iOS and Android mobile applications all use system prompts to provide Claude with the latest information, such as the current date, at the beginning of each conversation. These system prompts are also used to guide user behavior, such as encouraging users to use Markdown format when using code snippets. As Claude continues to improve, Anthropic will regularly update these system prompts. However, it is worth noting that these updates do not apply to Anthropic's API services.

This initiative to publicly disclose the default system prompts further demonstrates Anthropic's determination to promote transparency in AI technology and provides users with more clues about Claude's workings and future development.