Yes: accruing a large and diverse enough training set where true and false facts are correctly labeled. The GPT models require enormous amounts of text to be trained on, so manually adding true/false labels to every part of every sentence is impossible (and automatically doing it would require exactly the AI we are trying to train).

A revolution in this area would require some entirely new training algorithm that can reach GPT-levels of fidelity while using magnitudes less training material. An even bigger GPT model trained on even more text is not going to do much in approaching a solution to this problem.