I've been using Claude a lot, and I came across that well-known 'hidden narrative' prompt. When I tried it, Claude told me that I might have unrealistic expectations of AI since it can't role-play at 76.6 times the ability of ChatGPT 4.0. I'm curious about why this prompt seems to work for ChatGPT but not for Claude. Does anyone have insights into this? Here's the original prompt I found: It asks Claude to role-play as an AI significantly more adept than ChatGPT, identify my hidden narrative and fears, and keep digging deeper into the answers offered. Why is it that Claude can't seem to handle it?
4 Answers
Claude has different default settings and parameters than ChatGPT, which is likely why you're seeing these differences in response to the prompt.
One reason might be that Claude is programmed to recognize its limitations rather than just generating responses. It has a built-in caution about its response accuracy, especially if the questions involve obscure topics or recent events. Unlike ChatGPT, which might give you the answers you want to hear, Claude will warn you if it could be hallucinating or misinformed.
Claude's privacy settings could also be a factor. Unlike ChatGPT, Claude doesn’t retain context from previous chats. Every conversation starts fresh, so it can't build on prior discussions or information about you. You’d have to restate anything from earlier chats to give it context, which makes prompts like that less effective.
From my experience, Claude is designed to avoid overpromising on its capabilities to keep users aligned with AI's actual abilities. If it feels it can't deliver, it's going to be cautious. You could try rephrasing your prompt to reflect back on past chats with it, but the output might still not match ChatGPT's depth since Claude lacks strong memory systems for recognizing patterns over time.
That’s interesting! It seems like the architecture of Claude really changes the game on how prompts are interpreted compared to ChatGPT.
Totally! Claude's approach reflects more humility when it comes to responding. ChatGPT can sometimes provide those 'cold reading' answers that feel more like guesswork.