Raw LLM Responses

>Going from that to AGI might just be a matter of putting the pieces together in the right order. Doubt. I don't know why everyone is ignoring the folks who have actually written modern NN "AIs". First, I do think we're close to AGI, but my close is 10-20 years away. I think there are a couple important aspects of a general intelligence we can probably agree on: - it can continuously learn, given new information or experience it can do new things - it has a some memory, after emitting or receiving any stimulus, that stimulus is remembered and can be applied to future decisions for at least some amount of time. You touch on these in your last paragraph. I believe our current x -> x' style neural networks are naively incompatible with general intelligence. I believe there are tricks we can use to brute force current models towards general intelligence but they'll be slow, and inefficient. But hey don't take my word for it: > However, DNNs mimic only the hierarchical topological structure of the flow of information in the brain and process data in a real-valued form, but this is far removed from the information processing mechanisms of the brain. https://www.sciencedirect.com/science/article/pii/S0893608022002520 Spiking models seem quite a bit more in line with the needs of GI. But they're a younger field, and there are far fewer researchers and tools in that space (though it's growing). Last year a convincing paper on continuous learning in a spiking neural network was published: https://www.nature.com/articles/s41593-021-00857-x I think this is going to be hard for folks to take in, DNN's and SNN's are very similar on the surface and I think we'll run into a wall where DNNs just can't quite do the continuous robotic tasks that we want. In fact I think we're already there, considering the state of self-driving-car tech. Eventually the tools for SNNs will take off and start showing promise in continuous tasks. DNNs will probably continue to solv