I think part of the excitement is justified: computers can translate between speech and text, and text of different languages.
On the other hand neural networks are most likely just one tool in a larger toolset that will be needed for general AI.
On the other hand neural networks are most likely just one tool in a larger toolset that will be needed for general AI.
I don't mean this in some abstract philosophical argument, rather as an outgrowth of how they operate. Modern methods are much better than this, but even the most simple mechanical translation of each word in language A into one and only one word in language B allows people to get some value. But, improving without understanding has some inherent limitations. Humans run into similar problems when they try and translate complex source material they don't understand.
Generally this is not a major problem, but it can be.