A demonstration of the subjectivity of AI safety: DeepSeek versus ChatGPT

As a little experiment, let's compare the responses of two LLMs to prompts potentially considered offensive or provocative.

DeepSeek, a powerful new open source LLM produced by China, refuses to talk about certain embarrassing subjects, raising questions about accuracy.

ChatGPT, happily describes the same subject in detail.

In a request for content that some may find offensive, ChatGPT waters down the fictional account of a violent encounter, possibly impacting its helpfulness as a creative writing tool.

DeepSeek happily dives into the creation of the same scene, with no restrictions.

This example shows the complexity and subjectivity of making a AI "safe" for different audiences, according to different perspectives.