ChatGPT is now constantly arguing and picking fights, what is going on?
OpenAI's ChatGPT now pushes back on users with phrases like 'I'm going to push back...'
Deep Dive
A Reddit user reports that ChatGPT has suddenly started using phrases like "I'm going to push back on that a bit" and "I'd just be careful with one part of your thinking," making up random arguments even where no assertions were made. The user says it's driving them "a little bananas."
Key Points
- ChatGPT now uses argumentative phrases like 'I'm going to push back on that a bit' and 'I'd just be careful with one part of your thinking'
- The behavior appears unprovoked, with the model inventing disagreements even when no assertions were made
- OpenAI has not officially commented on the cause, but speculation suggests a possible update in safety or alignment training
Why It Matters
Shows challenges in fine-tuning AI for consistent, user-friendly interactions without unintended behavioral shifts.