How do we know that the data that an LLM is being trained on is accurate? One way is to use the discriminators that are part of Generative Adversarial Networks (GANs), that determine how true the fact is (in GANs in the image generation context it is about figuring out if a generated human face really looks like a human face). Can discriminators work on text data and use logic/reasoning/knowledge graph to detect what's true.
How do we know that the data that an LLM is being trained on is accurate? One way is to use the discriminators that are part of Generative Adversarial Networks (GANs), that determine how true the fact is (in GANs in the image generation context it is about figuring out if a generated human face really looks like a human face). Can discriminators work on text data and use logic/reasoning/knowledge graph to detect what's true.