ChatGPT will never go through the Looking Glass

By: Rob Corbidge, 15 December 2022

Alice through the looking glass by Stable Diffusion

Impressive as its public debut is, AI writing tool ChatGPT isn't transforming the written word world any time soon. It can create only what already is, not what can be.

As a veteran of AI-enabled content creation of some four months standing, an amount of time that obviously yields complete mastery of any discipline as long as no-one is scrutinising, much has already been learned about the abilities and limitations of such technology.

The first and biggest lesson is that it won't do your job for you. Crap in, crap out, as the old phrase perfectly describes it. To get the fullest use of an application such as Stable Diffusion, Midjourney or the latest star, AI article-writer ChatGPT, the user does need to know how to get the system to do want they want.

"The good news about computers is that they do what you tell them to do. The bad news is that they do what you tell them to do!", as the esteemed Ted Nelson once observed. Nothing has changed in that regard.

That said, on the first few passes ChatGPT does produce coherent results - so coherent in fact that this chap  produced a children's book on it, with the results fed into Midjourney. It was pretty awful by all accounts, grotesque fingers for one thing being a regular feature of AI-generated art. Drawing hands is hard. Lewis Carroll didn't have to worry about that when putting Alice in Wonderland.

In common with everyone in tech and their mothers, we pointed ChatGPT at our own sector of the publishing industry, the glamorous world of the Content Management System. What could we ask it to see what it do

"Write a whitepaper on why developing your own CMS is a bad idea" was the prompt, knowing full well that we already know why it's bad. Surely we'd have ChatGPT on the ropes, and it would produce nothing but vaguely relevant words and phrases!

Anyway, off it went and assembled the following:

Building your own Content Management System is a very attractive idea, especially for those who have ambitious web development goals. It assumes that with the right skills and knowledge, you’ll be able to create a tailor-made CMS tailored to your exact needs. However, as with any ambitious project, there are risks and challenges involved, and developing your own CMS can often be a bad idea.

First, developing your own CMS can be an expensive endeavor. Not only will you need to purchase the necessary software and hardware to develop the CMS, but you’ll also need to pay for the time and resources required to design and develop the system. This can quickly add up, especially if you’re not an experienced web developer.  

Developing a CMS is a long and complex process. It takes considerable time to create a CMS that’s both functional and secure. This means you’ll need to dedicate a significant amount of time and resources to complete the project. This can be difficult to do if you have a limited time frame or if you’re already busy with other projects.

One of the most important aspects of a CMS is security. A poorly-secured CMS can leave your website vulnerable to hackers and other malicious actors, which can lead to data breaches, website downtime, and other serious issues. Developing a secure CMS requires specialized knowledge and expertise, which may be difficult to find if you’re not an experienced web developer.

Once your CMS is developed, you’ll need to maintain it on an ongoing basis. This can be time-consuming and costly, as you’ll need to keep up with the latest software updates and security patches. This can be difficult to do on your own if you don’t have the necessary technical expertise.

Developing your own CMS is a risky endeavor. It can be costly, time-consuming, and difficult to ensure the necessary security and maintenance. For these reasons, it’s often a bad idea to develop your own CMS, and it’s usually better to use an existing CMS that is already secure and maintained by professionals.

Much to our surprise, there is no dispute from our perspective with what it 'wrote', our perspective born of many decades of cumulative experience in the publishing industry, and a love for that industry.

For this straightforward cribbed-from-simple example, ChatGPT does indeed produce a serviceable result. It certainly has practical utility in producing a first organisational draft for technical writing, for example. 

So, should writers be afraid? Are they all to be redundant, the predicted fate for musicians when electronica announced itself on the scene? Well no, of course not. The musicians are still here, and so will the writers be. Form can change without substance being lost.

ChatGPT is derivative and sits on what is already understood - and then only a portion of that. As a wiser head than me has observed, if we are worried about kids using it to cheat in school, then make them produce their essays using it, then mark them on how they critique the essay. Problem solved.

We're also facing the prospect of layers upon layers of automation. AI-generated writing curated by algorithms. What dataset is all this being trained on? Dinner will eat itself. Certainly Google can detect such content very easily and decide what to do with it.

And if you're interested in how you can manipulate the publicly available version of ChatGPT to reveal more about the way the way it works, then this is the thread for you.

The fact is that the vital spark, that of the creative process, can't be replicated. 

Can it ever? Possibly, if humans gain greater understanding of themselves and can transfer that knowledge to machines. If anyone thinks we already have that complete understanding of ourselves then I have a four month course in AI to sell you.