Trust but verify. We’ll get processes built around this for accuracy sensitive applications. I imagine it will look something like a GAN configuration with two or more LLMs trained to adversarially critique and fact check the outputs of the other(s). Might even mimic the relationship between the hemispheres of our brains.