[sandos]

Prompt: The way forward to general AI is to quickly

The way forward to general AI is to quickly get as many neural net models into operation as possible, with each neural net performing a simple and specialized task, and then to quickly connect the neural nets together to work as a coordinated team. By putting more neural nets together we increase their collective understanding and ability to accomplish greater tasks than any one of the nets would be able to achieve alone. But we can't just stack neural nets on top of each other -- a neural net needs to see how its inputs are connected to its outputs. We don't want a single neural net to be dependent on other neural nets to determine how it behaves. Instead we need to connect the neural nets together in an organic way.

Let's say we have two neural nets, one that can recognize cats, and another that can recognize dogs. How do we connect these two neural nets together? One solution might be to make the two networks send their output to some central coordinator that could evaluate the input and come to a decision based on the output from both networks.

[22c]

> Let's say we have two neural nets, one that can recognize cats, and another that can recognize dogs. How do we connect these two neural nets together? One solution might be to make the two networks send their output to some central coordinator that could evaluate the input and come to a decision based on the output from both networks.

I guess the AI has been trained on naive opinions on AI from people such as myself:

https://news.ycombinator.com/item?id=30280785