My understanding of a transformer model is that it uses the given input to calculate internal query of relate-ness of word meanings, and generate a meaningful response based on its meaning. But if your given sentence has no meaning, then won't the model fail to capture any meaningful input so that the output will also be meaningless? How does the ChatGPT generate meaningful response asking me what I meant when the input is uncoordinated and meaningless? Is that a feature, or did they train on that kind of input specifically?
eg. Input "Jumps the dog lazy fox over quick brown the."
For the output, ChatGPT asks for clarification.
Normally for a self-attention based model, there won't be any relation between input word for this example, so shouldn't the output also be garbage like?
$\begingroup$
$\endgroup$
1
-
2$\begingroup$ can you give an example of a garbage input and output that you do not understand? $\endgroup$– user253751Mar 20 at 20:20
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
If you give a human some input that doesn't seem to convey any meaning they will probably ask you for clarification. Presumably there are a lot of examples of this in the ChatGPT training data so that is exactly what it is going to do.
