Why the training and error loss are similar?

Question

I am running a code to classify hand gestures. The code of the model is as follows:

model_d = tf.keras.models.Sequential([
    tf.keras.layers.Flatten(input_s