I think we need to shift our paradigm one or two more times before we can start seriously talking about AGI. Current transformer models are impressive, but they’re much better suited to modeling language than what I would call “cognition”.
I think we’re close, but I don’t think we’ll get there by increasing/improving current technology.