Recently I was invited to attend an Oxford/Wales biotech collaboration event hosted by Sue Rees MediWales and Mischon de Reya. There was an incredible line-up of speakers. And I had a lightbulb moment regarding ChatGPT and content.
The presentation on AI and ChatGPT was given by Professor Nigel Crook.
Professor Crook has one of the longest lists of credentials I’ve seen. As a Professor of AI and Robotics he is ‘Director of the Institute for Ethical AI and Associate Dean of Research and Knowledge Exchange’ at Oxford Brookes University. He also has that rare gift of being able to explain the highly technical in a way that anyone can understand.
In his presentation, he considered the history of the chatbot, sharing his not-insignificant role in that history, and the current capabilities of the chatbot everyone knows; ChatGPT by OpenAI.
From his talk, I was able to extrapolate how ChatGPT directly impacts marketing content – and it explains why it can be so infuriating.
One of my talents is taking bad photos, so you will have to take my word for it that the person to the right is Professor Crook!
Summary of how ChatGPT works
The summary of how written content generating AI, or more specifically ChatGPT, works at the moment as explained by Professor Crook went something like this:
- Start on ChatGPT by typing in a few words such as ‘Once upon a…’
- ChatGPT trawls the whole internet within seconds to come up with a number of potential probable next words in the sequence e.g. ‘time’ ‘journey’ ‘story’.
- ChatGPT doesn’t know what the correct word is, so it effectively ‘rolls an electronic dice’ to choose between ‘time’ ‘journey’ ‘story’.
- You can type the same sequence of words again and it may give you a different next word depending on how the dice rolls.
- To ChatGPT there is little difference between ‘Once upon a time’ ‘Once upon a journey’ ‘Once upon a story’
This explains why the same query can generate different answers each time. Why ChatGPT is ‘unpredictable’, the word I’ve circled on the slide.
This also explains why, when using ChatGPT to help write copy, it can be a useful tool to get you started when faced with a blank page, but it can come up with some barmy phrases.
It also explains why the text usually lacks an argument flow, because ChatGPT isn’t logically building an argument flow in the way that a human does, it’s merely reacting to the last word typed.
To me ChatGPT in terms of marketing written copy generation therefore feels more like an intelligent aggregator of content which then requires the human touch to make sense to an audience.
I’m sure this will evolve.
P.S. This article was written solely by a human.
Interested in learning the basics about content writing and SEO?