OpenAI has already made AGI, but is concerned about releasing it?

The training of GPT-4 was finished in 2021. Ever since then, I think OpenAI made much bigger models on more recent data, but out of fear of public reaction to AGI, the company only released a dumbed down version of it called GPT-4. They don’t even mention what they technically did in their paper.

What fuels this theory is the fact that quickly after the release of GPT-4, tons of projects started using self reflection to boost GPT-4’s capabilities. As an engineer in OpenAI, this looks like a low hanging fruit. I doubt they didn’t already run something like AutoGPT before releasing the model. Maybe they have already achieved AGI and the AGI is recommending them to keep the public in dark.