Weakly general sounds strange to me. Sounds like almost human. I think we need some sort of minimal requirements otherwise we might be talking about different things.

I think AGI has to at minimum:

• be multimodal
• Embodied
• Know how to learn
• Able to follow a chain of arguments
• Able to communicate with autonomy
• Understand ethical principles

And there are many other things, but these seem hard enough. I think the first two are doable by 2027. Not so sure about the others.

I know how people love to talk about exponential growth. But let's not forget that something has to drive it. Deep learning has been driven by GPUs and the abundance of data. Both are not inexhaustible resources.