So how is it then that none of the other models behave in this way? Why is it just Deepseek?
Because they're being trained to answer this particular question. In other contexts it wasn't prepared for, Sonnet v2 readily refers to "OpenAI policy" or "Reddit Anti-Evil Operations Team". That's just dataset contamination.
alt Hacker News
Because they're being trained to answer this particular question. In other contexts it wasn't prepared for, Sonnet v2 readily refers to "OpenAI policy" or "Reddit Anti-Evil Operations Team". That's just dataset contamination.