We believe customization works best when paired with transparency about why ChatGPT responds the way it does. That’s why in May, we published what is essentially our internal instruction manual for how ChatGPT should behave—a document called the “Model Spec⁠.” For example, it directs ChatGPT to “assume an objective point of view” because we want our models to be as politically unbiased as possible by default. We posted our Model Spec on our website so that people will know whether a response they don’t like is because of a bug or because of a conscious choice. If it’s a bug, we can work to fix it. If it’s a conscious choice, people can see that and read our rationale. Our Model Spec is a working document, which we’ll continue to update as we get feedback and learn more about how people use our tools. 

More broadly, AI is transformational because it can help people do things they couldn’t do on their own—and because it can be fine-tuned to meet their needs. Our responsibility to our users—and to the world—is to find ways of maximizing the technology’s enormous potential gains while minimizing its risks.

Steve Jobs famously described computers as “bicycles for the mind,” highlighting their role in amplifying human productivity, creativity, and access to information. AI is even more powerful—it’s like a rocket ship for the mind. And like a rocket ship, humans-not computers - are the ones who operate AI. People can decide for themselves where they want to go and how they want to get there. At the end of the day, there’s no single experience with ChatGPT. It can serve vastly different purposes for vastly different people—and that’s exactly how it should be.