Nobel laureate and co-founder of Google DeepMind, Hassabis, once imagined a scenario where AI is immersed in a specific historical knowledge context to assess its ability to conceive a novel explanatory framework. This hypothetical challenge has come to be known as the 'Einstein Test.' Independent researcher Michael Hla took this idea a step further by setting the knowledge cutoff date to 1900 and training a model named GPT-1900. Impressively, this model offered an explanation akin to Einstein's light quantum theory when tackling the photoelectric effect. However, it fell short in most other physics-related tasks. Moreover, the experiment was marred by issues of knowledge leakage, a consequence of incorporating modern AI techniques. Researchers have underscored that while contemporary large-scale models excel in induction and deduction, they fall short of Einstein's signature 'abductive leaps.' These models are incapable of independently proposing entirely novel explanatory frameworks, assessing theoretical worth, identifying problems autonomously, or carrying out verification processes. Thus, there remains a significant chasm to bridge before AI can truly assume the mantle of scientists.