Generative transformer models are not AGI, not even close. We're going to have to come up with some new methodology to handle multi-modality or maybe some kind of synthesis between several different models until we see some kind of Proto-AGI and that's decades away IMO.