Anthropic (the company behind ChatGPT competitor Claude) has taken an interesting step toward AI transparency by publishing Claude’s system prompts – the ones that shape the model’s behavior. These prompts dictate not only the model’s restrictions (like avoiding facial recognition) but also their intended personality traits, such as intellectual curiosity and impartiality. According to Anthropic, its goal is to position itself as an ethical leader in AI. That would certainly set it apart from the pack.
It’s nice to see a big foundational model builder make an effort to demonstrate the importance of human oversight. It’s also nice to understand what kinds of prompts are necessary to get a foundational model to behave as expected (which require system prompts, fine tuning, and a bunch of other post-training techniques).
Take a minute and read the whole prompt by clicking the “July 12, 2024” box on this page. It’s fascinating reading.
Author’s note: This is not a sponsored post. I am the author of this article and it expresses my own opinions. I am not, nor is my company, receiving compensation for it. This work was created with the assistance of various generative AI models.