有人猜测 OpenAI 正走向递归自我改进
作者对比 Astra 发布时与当前表现，并提到 GPT-6.1 Sol 和 Luna，认为 Astra 在变弱而新模型反而变强，怀疑 OpenAI 已在悄悄尝试 RSI。
🚨 OpenAI RSI: Something Interesting Is Happening What if OpenAI is quietly moving toward recursive self-improvement? Recent model upgrades have me asking serious questions. Over the past week, I compared Astra at launch with its current performance, alongside GPT-6.1 Sol and Luna. The difference is WILD. We've watched Astra get weaker. We've seen Claude models take hits, too. But OpenAI's latest models seem to be moving in the opposite direction. Instead of declining, they appear to be getting SMARTER. And other people are starting to notice it as well. I'm not claiming OpenAI has achieved RSI, but the recent math results and consistent improvements are hard to ignore. Is OpenAI quietly taking its first steps toward recursive self-improvement?