According to CoinMeta, Musk shared a comparison chart of “AI then / AI now” and only commented on one term: “accurate.” The chart shows that in the past, machine learning research required designing core methods, proving the effectiveness of models, and building probabilistic models. However, in the era of large models, there are only three steps left: scaling up the transformer, feeding in more data, and adding more GPU; thereby, the model can continue to improve. This reflects the real dilemma of the large model industry: in the past, principles were studied first, followed by model design, but now, the focus is often on scaling up first, and then investigating why certain capabilities emerge.