The speaker believes that while there is still a lot to learn, we have the potential to make significant improvements in ensuring AI models match public expectations by insisting companies solve problems before releasing models.