Hacker News Reader

I would not be so sure about #1, OpenAI has been very cagey about what exactly the newer models are trained on.

I agree. People seem to largely be pretending it's not the case that GPT-4 significantly outperforms every other model right now.