Google may be using recursive self-improvement to speed up Gemini 4
The author says multiple Google employees describe RSI as key to recent progress, with new checkpoints showing stronger coding while Argon is still unrolled, suggesting Gemini 4 may be further along than expected.
I think google is making significant progress with recursive self-improvement. multiple google employees have been talking about RSI being a key part of their recent progress, with one researcher saying it's in full swing. and looking at what's happening with gemini 4, it makes sense. cause argon hasn't even been rolled out yet, and they're testing newer checkpoints that reportedly show significant improvements in coding capabilities. if google is already using its own models to accelerate the development of the next ones, that could explain why they're iterating so quickly. so they could be much further along than we think.
This post quotes another post: “Google reportedly tests Gemini 4 model codenamed Carbon”