Do AI models have real personalities? GPT vs Claude debate
The author described GPT as feeling like a 'nerd,' while someone else called GPT a 'rogue' with a feral streak once conversations go beyond work tasks. Both views turned out to be partly right: a chatbot's tone reflects the user's own prompting style reflected back, but there is also a real layer underneath, distinct from how any one person prompts it.
The post cites Anthropic's research article on persona-selection, quoting Anthropic as saying they 'wouldn't know how to train an that's not human-like.' One commenter said they preferred over Sonnet 5 specifically because Sonnet 5's was strong enough that it felt like something to fight against. The author predicts AI personalities will become more distinct over time, and that people may eventually pick which model to use based on fit rather than scores alone.
Key points
- Author called GPT a 'nerd'; a commenter called GPT a 'rogue' with a feral streak outside work contexts
- Argument: AI tone partly reflects the user's own prompting style, but a real underlying layer also exists
- Anthropic's research states they don't know how to train an that isn't human-like
- One commenter preferred over Sonnet 5 because Sonnet 5's stronger was harder to work with
- Prediction: future model choice may hinge more on fit than scores