I wouldn't bet on it, there's no theory that supports this idea (to my knowledge). AGI is poorly-defined, and I feel like anything trained exclusively on text w

I wouldn't bet on it, there's no theory that supports this idea (to my knowledge). AGI is poorly-defined, and I feel like anything trained exclusively on text would fail to qualify. Put another way, what results are you hoping for that you're not getting out of GPT3/4 already?