This is the Schematic Structure of an autoencoder

You should be able to enlarge this schematic by clicking on it if it is useful to you to do that. The HTML input is to the left which then goes down to the code in the middle. Then it is decoded into output.

However, this situation only works completely in ideal circumstances because you have different programmers with different goals working where you might be quoting from and then where you want to put it might have different rules than those programmers put within their code. So, in some ways it is like translating from one language to another one in actuality. For example, with Google Translate you still likely cannot count on more than a 70% to 90% correct translations between languages and even then different languages emphasize different things all over the world because cultures are different. So, we have the same problem when programmers have different sets of goals on different levels and then code hits new parameters and then you wind up with strange outputs. I run into this problem almost every day here at my site.

But, it is still much much less time consuming for me to run quotes through an autoencoder than to spend 10 to 100 times more trying to do it all in HTML. It I had to program everything in HTML directly myself at this point I wouldn't blog at all. It was okay to do this from 1999 to 2007 when I was almost 20 years younger though.