I don't think anyone has ever tried having a model fully autonomously train a model that is better than it.

Everyone including myself has attempted this relentlessly, it just doesn't work out beyond some arbitrary improvement in specific tested categories instead of broader capability increase.

Did you publish on that topic? Just curious, would love reading more

There's a pretty large paper on the topic: https://arxiv.org/pdf/2606.15497, but results are somewhat mixed.

My response, which was flagged by automation (I used AI to verify the paper and it says the paper is NOT what you claim it is), was spot on. This paper is not recursive self improvement, it is AI in a feedback loop, but it is not improving itself. It turns your entire premise into something questionable. You said this one done everywhere by many people including you.

Then you present a paper which WAS NOT recursive self improvement so it clearly is not "everywhere".

To be real with you, I don't think you're being entirely honest. You're not presenting a fair argument. You have bias. Recursive self improvement can only be done by frontier labs who created these models in the first place. Academia simply Does not have the funds.

If this is being done, we're not privvy to it. The amount of compute to build these things is astronomical.