There are some pretty egregious errors here. Decreasing K increases the variance and is more prone to overfitting. Increasing K increases the bias and is not necessarily more accurate. It all depends on the underlying distribution of the data.
What is the time complexity of knn? Why should I use it? When do you not want to use it?
I do agree that increasing K does not mean it will increase the accuracy, distribution and nature of the data will be key to understand the effect of the K in the accuracy.
I think the author has mixed up underfitting and overfitting.
thanks for your interest, can you please elaborate.