AI Degeneration – three thoughts on ChatGPT in marketing
Written by Paul Maher
The hype is high. It took a decade or so for the full effect of smartphones to be realised, and longer still for the web to recover from the dotcom crash of over twenty years ago.
Tech, from world processing to social media, always shows its societal benefits, and drawbacks, eventually. Despite the reality that it’s taken around 30 years of AI false- dawns to get here, the current buzz around Chat GPT is extraordinary.
Is it panicky when thousands of tech entrepreneurs, spooked by a chatbot, write open letters to the world asking for AI development to be frozen? Is it an over-reaction when entire countries ban weeks-old technology because of a data leak? Maybe.
But, for most of us, our interest is in the short to medium term societal effects of Large Language Models on how we consume and manipulate the digital data which so defines our lives. And, how it will define our job prospects.
While philosophy is not our smart suit, award-winning PR and content certainly is and we have seen a lot of chatter recently about the job losses which are `inevitable`. And, as an avid user of Generative AI, we are here to separate the BS from the LLM. Here’s how we see it today.
1. The robots are taking some people’s jobs – but not all of them
The first interaction with ChatGPT copy in a work environment came courtesy of some suspiciously competent and quickly-worked copy from a very junior but very tech-savvy colleague. In this case, it was for one of the more tedious, but client-scrutinised parts of our role, industry award entries.
Called out, the colleague held up their hands. Their resourcefulness was applauded and yet the copy delivered did not quite ring true and was obviously not created by them. No matter, because it arrived on time and needed way less editing to look and feel on-message for the client.
The issue is they had not learnt how to take it from good to great and potentially award-winning. That relied on older, more manual interventions. It is not an exaggeration to say that whether they learn the final part of the process will dictate their value to the profession and his future earning power.
2. We will all adapt – some faster than others
There was a super-interesting thread on one of our favourite podcasts, on the way in which detailed the multi-stage approach one of the most popular hosts wrote blogs. He researched multiple technical articles, with the AI summarising them, combining them into a single article, guided by his prompt which acted as his editorial angle. He then edited the results to put it in his voice in one final sweep to make sure the quality was right and published.
The results have changed forever how they will create content and as they themselves commented it has reduced their fear of the ‘blank sheet of paper’. Having a smart virtual researcher is a godsend for a writer, who can focus on themes and narration.
For better or worse, this blog was not written by a bot. Hopefully, you can tell this was hand-crafted the old fashioned caffeinated way. Which leads us onto the last point.
3. We had all better get used to the copy blandizer
But the reason the copy for the award was so glaringly ‘better’ is that it took a junior’s work up a level, by raising its ‘quality’ to a higher average than the writer’s at that stage in his career. How that happens matters.
Almost all Large Language models scrape the web for copy ideas. This means the results are, like most professional copy on the web, better-written than the majority of folks can, even those with college degrees. After all, most web copy has been professionally created and reviewed by several humans before being published.
This means what comes back is more than acceptable to most of us and even exceptionally good for those who struggle with writing. However, it is somehow blander, because it has been ‘averaged down’. This means while the average is pulled up, truly spectacular copy still ‘pops’.
Now, the latest versions of copy-generating AIs offer a way around this. They will add in ‘randomness’ and ‘quirkiness’, ironically by looking at long tail results in datasets and including more of that ‘flavour’. But let’s face it, most web copy, almost all SEO, and much of the promotional materials we consume online does not need to inspire us to ponder the meaning of life. Bland, but correct and well-written, is good enough for most advertising copy.
Want to buy a new widget on Amazon and basic, but well-written, details will suffice. Plus, if I am the product owner, I would rather have a hundred well-written Tweets than a single finely crafted one.
So, in summary, the new era of Generative AI will radically change the way Content and PR is delivered. Just as in all automation from the emergence of type to the spread of the World Wide Web this will require reskilling our teams and this will favour higher-order skills like editing rather than churning out copy. There will be a decrease in demand for low-level tasks like SEO and basic social media posting, and salaries will reflect this, but new higher value and higher paying opportunities will emerge. As with the web, smartphones and social media, those who adapt best to tech will thrive.
Banning, slowing down and regrets about tech progress are yesterday’s thinking.