Is it possible make LLM attributable?

Reasoning is the disadvantage of ChatGPT, Are there some method to make ChatGPT and Auto-regress model attributable and interpretable?

Is "Objective-Driven AI architectures" just a vision? Is there some paper about that?