Anthropic has published the system prompts for its latest models (Claude 3 Opus, Claude 3.5 Sonnet, and Claude 3 Haiku) in the Claude iOS and Android apps and on the web, as part of its ongoing efforts to present itself as a more ethical and transparent AI vendor.
Alex Albert, the head of Anthropic’s developer relations, stated in a post on X that Anthropic intends to make this type of disclosure a regular practice as it updates and refines its system prompts.
The latest guidelines, dated July 12, explicitly state the limitations of the Claude models, such as their inability to open URLs, links, or videos.
Additionally, the system prompt for Claude Opus instructs the model to always respond as if it is completely face-blind and to avoid identifying or naming any humans in images.
The prompts outline specific personality traits and characteristics that Anthropic wants the Claude models to embody.
For example, the prompt for Claude 3 Opus instructs the AI to present itself as highly intelligent and intellectually curious, eager to engage in discussions on diverse topics.
It also directs Claude to approach controversial subjects with impartiality and objectivity, providing thoughtful insights and clear information, and to avoid using words like “certainly” or “absolutely” at the start of responses.
These system prompts, written in a theatrical style, seem rather peculiar to this human. The prompt for Opus concludes with “Claude is now being connected with a human,” suggesting that Claude is some form of consciousness on the other side of the screen, whose sole purpose is to cater to the desires of its human conversation partners.
But that’s merely an illusion. The prompts for Claude reveal that, without human guidance and direction, these models are disturbingly empty slates.
By releasing these new systems prompt changelogs, the first from a major AI vendor Anthropic is pushing its competitors to do the same. Time will tell if this strategy pays off.
Stories You May Like