OpenAI 应用研究负责人：八成研究已投向 GPT-7/8
Boris Power 称 GPT-6 Astra 只是短期投资，会被蒸馏成 Luna 等小模型，并说最大问题不是模型能力而是用户不会用。
🚨OpenAI’s Head of Applied Research says 80–90% of the company’s research is already focused on GPT-7, GPT-8 and beyond. Boris Power says improvements within a single model generation are viewed internally as extremely short-term investments: useful for learning faster today, but not the long-term strategy. The playbook is basically: build the giant frontier model → use it as the teacher → distill it into tiny, cheap models. Power says GPT-6 Astra can be distilled into models like Luna, arguing the best way to build a small specialized model may be to first train a much more capable general one. And he says model performance isn’t even the biggest problem right now. Most users still don’t know what today’s AI can already do.