| ▲ | OutOfHere 6 hours ago | |
I am not convinced that memoryless large models are sufficient for AGI. I think some intrinsic neural memory allowing effective lifelong learning is required. This requires a lot more hardware and energy than for throwaway predictions. | ||