I found the model is good at capturing the statistics of the underlying text - it's quite good at producing stories that "sound like the original books". Howeve

I found the model is good at capturing the statistics of the underlying text - it's quite good at producing stories that "sound like the original books". However, coherent reasoning is still in such models. One example is "... She is a man....". These outputs clearly breaks proper reasoning constraints (and common sense)