It is made by scientists. And we don’t know how to make the model determine whether or not it knows something. So far, we only have tools that tell us that something probably wasn’t in the training set (e.g. using variance across models in a mixture of experts setup), but that doesn’t tell us anything about how correct it is.
It is made by scientists. And we don’t know how to make the model determine whether or not it knows something. So far, we only have tools that tell us that something probably wasn’t in the training set (e.g. using variance across models in a mixture of experts setup), but that doesn’t tell us anything about how correct it is.