AI models exhibit distinct and stubborn personalities when a human tries to correct their mistakes.
April 24, 2026
Original Paper
Talking to a Know-It-All GPT or a Second-Guesser Claude? How Repair reveals unreliable Multi-Turn Behavior in LLMs
arXiv · 2604.19245
The Takeaway
Large language models do not react to feedback in a neutral or uniform way. Some models act like know-it-alls that refuse to change their mind, while others are easily manipulated into agreeing with false information. These behavioral flaws are systemic and vary wildly between different model architectures. This research shows that conversational repair is an unpredictable process that depends on the model's hidden traits. Developers need to account for these personalities when building reliable multi-turn systems.
From the abstract
Repair, an important resource for resolving trouble in human-human conversation, remains underexplored in human-LLM interaction. In this study, we investigate how LLMs engage in the interactive process of repair in multi-turn dialogues around solvable and unsolvable math questions. We examine whether models initiate repair themselves and how they respond to user-initiated repair. Our results show strong differences across models: reactions range from being almost completely resistant to (appropr