I saw some interesting patterns in the creative abilities of both ChatGPT and Claude AI today as I tested them. At first, I asked for poems without giving any details about the title or subject. Both AIs wrote abstract poems that were hard to understand. They used a lot of metaphors and philosophical thoughts that, while they may have been well written technically, didn’t make me feel anything.

But when I gave them more specific information asking for poems about a Japanese father who is 50 years old they immediately came up with simple poems that anyone could understand. The pictures became more real: views of mountains, cherry blossoms, and special times between a father and his child took the place of the vague ideas they had used before.

This experience backs up what I said last week  if you want accurate and useful responses from AI systems, you need to be very clear about what you want. The result will be more like what you wanted if you give more information. This is true not only for creative writing, but also for finding information, writing code, and solving problems.

I also saw that these AI systems might be biased. All of the names they came up with for their characters were Western or white-sounding, which I thought was an interesting pattern that should be looked into further. Even though both systems are meant to show different points of view, this small bias suggests that their training data may have imbalances that affect the creative work they produce.

In this comparison, both the pros and cons of the current AI language models are shown. It’s easy for them to follow specific instructions, but it’s hard for them to be creative in general. They can make content that sounds a lot like human speech, but it still shows patterns that come from their training data instead of real understanding or cultural awareness.

The lesson for people who want to get the most out of these AI systems is clear: being precise with your prompts will lead to precise results. When you ask for vague things, you get vague answers. But when you give these systems thoughtful, detailed prompts, they can do everything they can.