Why, or how, would bars know what it is trained on? And even if bard contains training data recent enough to include discussions on bard, how would it be able

Why, or how, would bars know what it is trained on? And even if bard contains training data recent enough to include discussions on bard, how would it be able to tell speculation from facts? The only way I can think of is through deliberate alignment. After training on source data, the model is fine tuned by human curated chat dialogue.