Comment by LudwigNagasena

Why is it a bombshell? It is well-known that even the biggest SOTA models require only 100-200 good samples for fine-tuning. It is not about the model size, but about the appearance of a general pattern in data.