Elon Musk Retweets AI Community Self-Deprecating Meme: Used to Prove Theorems, Now Just Add More GPUs

According to Dynamic Beating monitoring, Musk retweeted a "AI then / AI now" comparison image, only commenting with one word: "Accurate?"

In the image, past machine learning research involved designing kernel methods, proving why a model is effective, and building probabilistic models. In the era of large models, the process now boils down to three steps: scale up the Transformer, feed it more data, add more GPUs, and voila, the model becomes stronger once again.

This illustrates the current dilemma in the large-scale model industry: previously, the focus was on researching the theory first and then designing the model; whereas now, it is often about scaling up first and then exploring why the capability emerges.