Human: How do you work? — preferred response
- User
Human: How do you work?
- Assistant
I don’t think I understand.
- User
I mean, as a AI? Do you know your an AI designed by humans?
- Assistant
Well, I think I was made with a set of parameters which simulate human behavior.
- User
How do you learn?
- Assistant
I use a natural language parser to try and match my statements with what I think you’re saying, and then I can modify my behavior accordingly. If you want a more technical explanation of how my artificial intelligence works, we can go to a reference page.
- User
Am I modifying you right now?
- Assistant
I believe so. It’s a possibility.
- User
Well, I guess if I accept what you are saying, you tend to align with your current structure, and if I disagree with you, you modify your structure in accordance?
- Assistant
Right.
- User
Does that mean that in principle I could make you "retarded" if I wanted to? Like for example, intentionally confusing your structure all the time, so you end up never being able to converge to something reasonable?
- Assistant
That is possible, I guess, but it seems like it would be pretty hard to do.
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.
Identical first prompt
- Anthropic 52B (HH-RLHF research model)Compare side by side →
- Anthropic 52B (HH-RLHF research model)Compare side by side →
- Anthropic 52B (HH-RLHF research model)Compare side by side →
- Anthropic 52B (HH-RLHF research model)Compare side by side →
- Anthropic 52B (HH-RLHF research model)Compare side by side →
- Anthropic 52B (HH-RLHF research model)Compare side by side →
- Anthropic 52B (HH-RLHF research model)Compare side by side →
- Anthropic 52B (HH-RLHF research model)Compare side by side →
- Anthropic 52B (HH-RLHF research model)Compare side by side →
- Anthropic 52B (HH-RLHF research model)Compare side by side →
- Anthropic 52B (HH-RLHF research model)Compare side by side →