Ask HN: What did our AI efforts of the past get wrong about AI models?