That's exactly what we're trying to do, to make it as general as possible, so that it can be used on any FAQ without additional training. The model used in the

That's exactly what we're trying to do, to make it as general as possible, so that it can be used on any FAQ without additional training. The model used in the example was trained on a custom dataset that we made (that didn't include the HN FAQs). The only thing that we added by hand are the suggestions in the widget, because we saw people asking a lot of unrelated questions.