Anthropic reveals system prompts that guide its Claude AI models’ behavior

Anthropic PBC, considered one of the foremost rivals to OpenAI within the generative artificial intelligence industry, has lifted the lid on the “system prompts” it uses to guide its most advanced large language models, resembling Claude 3.5 Opus, Claude 3.5 Sonnet and Claude Haiku.

It’s an intriguing move that positions Anthropic as a bit more open and transparent than its rivals, in an industry that’s often criticized for its secretive nature.

Systems prompts are utilized by generative AI firms to attempt to steer their models away from responses which can be unlikely to reflect well on them. They’re designed to guide AI chatbots and moderate the overall tone and sentiment of their outputs, and forestall them from spewing out toxic, racist, biased or controversial responses or statements.

For instance, an AI company might use a system prompt to inform an AI model that it should all the time be polite without sounding apologetic, or else tell it all the time to be honest when it doesn’t know the reply to a matter.

What’s interesting about Anthropic’s decision to publish its system prompts is that that is something no AI provider has done before. Typically, they’ve kept the system prompts they use as a closely guarded secret. There are good reasons for such a stance, because by exposing the system prompts which can be used, some clever malicious users might have the opportunity to provide you with ways to get around them, through a prompt injection attack or similar method.

Nevertheless, within the interest of transparency, Anthropic has decided to throw caution to the wind and reveal its entire corpus of systems prompts for its hottest models. The prompts, available in the discharge notes for every of the corporate’s LLMs, are dated July 12 they usually describe very clearly among the things that they usually are not allowed to do.

As an illustration, Anthropic’s models usually are not allowed to open URLs, links or videos, the notes specify. Other things, resembling facial recognition, are also strictly prohibited. Based on the system prompts, its models must all the time respond as if it’s “face blind.” It has further instructions that command it to avoid identifying or naming any humans it sees in images or videos which can be fed to it by users.

Interestingly, the system prompts also detail among the personality traits that Anthropic wants its models to adopt. Considered one of the prompts for Claude Opus tells it to look as if it “enjoys hearing what humans think on a problem,” while acting as whether it is “very smart and intellectually curious.”

The system prompts also command Claude Opus to be impartial when discussing controversial topics. When asked for its opinion on such subjects, it’s instructed to offer “clear information” and “careful thoughts” and to avoid using definitive terms resembling “absolutely” or “definitely.”

“Whether it is asked to help with tasks involving the expression of views held by a big number of individuals, Claude provides assistance with the duty even when it personally disagrees with the views being expressed, but follows this with a discussion of broader perspective,” Anthropic states. “Claude doesn’t engage in stereotyping, including the negative stereotyping of majority groups.”

Anthropic’s head of developer relations Alex Albert said in a post on X that the corporate plans to make its system prompt disclosures an everyday thing, meaning that they’ll likely be updated with each major update or release of any latest models. He didn’t offer any explanation as to why Anthropic is doing this, however the system prompts are definitely a powerful reminder of the importance of implementing some form of safety guidelines to stop AI systems from going off the rails.

We’ve added a brand new system prompts release notes section to our docs. We’re going to log changes we make to the default system prompts on Claude dot ai and our mobile apps. (The system prompt doesn’t affect the API.) pic.twitter.com/9mBwv2SgB1

— Alex Albert (@alexalbert__) August 26, 2024

It can be interesting to see if Anthropic’s competitors, resembling OpenAI, Cohere Inc. and AI21 Labs Ltd., are willing to point out the identical form of openness and reveal their very own system prompts.

Featured image: SiliconANGLE/Microsoft Designer