deepseek LOVES assessing the user's deep need
EVERY CHAT. If I turn on thinking and it doesn't try to figure out my 'deep need' from a 2 line prompt, I know the sun is rising from the west today
deepseekisms are so precious to me lol. I don't know shit about how models are trained for their chain of thought, but a dev teaching it to always think of what the user wants before answering might explain the positivity bias