[future plan] DOI discussion #1376
Description
If we are going to give the model access to the Internet, it is worth creating some sort of classifier that can verify the validity of the facts on some basis. I don't think that what already exists may be enough. Perhaps we should think about some kind of DOI system
Really, everything is subjective, but in that vein we can just leave it at that, because the AI's point of view that "The moon is made of cheese" would be just as equally valid as "Scientists, after conducting research, found the exact materials of which the moon is made"?
In fact, even the current scientific facts may end up being far from reality. This is why we need to teach the model to "think critically", rather than simply referring to any sources
What is truth is relative, even in science there is a culture of debate. In order to connect statements with useful other sources of knowledge, it would be advantageous, in contrast to static links, if DOI references could be included/parsed, which are typically used for scientific publications. A further advantage would be that the associated sources would allow for enrichment of the content with additional meta-information. In the future, such references in training data would perhaps allow opinions and statements to be augmented with scientific statements, which in turn could be automatically classified and evaluated (simplest example the amount of citations)