Things you may not know about ChatGPT

Chat GPT is brilliant!

Not what you were expecting to hear from a copywriter?

I can’t deny it though, the technology behind ‘write-source’ AI is phenomenal. Mind blowing even!

Just a few simple prompts and you're presented with an in-depth article ready to publish. Maybe you want to get your copy checked for readability? Just paste it into ChatGPT and let it evaluate it for you, amazing right?

‘But surely you’re worried you’ll be out of a job soon then Suzanne?’ I hear you holler from the back of the bus.

Far from it, because you see the more I learn about ChatGPT, Claude or Bard then the more sure I am that AI won’t be replacing me anytime soon. Because as great as all these language model based- chatbots are, they still have limitations and pitfalls that leave them a way off taking my job just yet!

 What I’m saying is, you should absolutely use programmes such as ChatGPT to help you formulate your written content but just take the time to learn a little more about it.

Being as informed as we can be has got to be an advantage, and there are some things I have learned about them recently that I really want to share with you!

Lost Property

Firstly, any text within ChatGPT becomes the property of ChatGPT.
So, not just the text that ChatGPT serves up as a result of user prompts, but anything input by the user can be used again and served up to other users in the future.

AI is after all a learning model, whatever you put into it, it stores and learns for the next time a similar query is made. Worth bearing in mind when asking the system to check your copy.

Be careful what information you are entering into the prompt, you should certainly avoid any company data and figures that you want to be kept private, and definitely never enter any customer data as that would be a breach of data protection regulations.

What would Louis say?

As I mentioned, ChatGPT is, by its very nature, a learning tool for AI. But it’s a savvy piece of tech, because it’s able to prevent itself from regurgitating its own answers as fact at a future time.
Every piece of text produced by the programme will include code that enables its own 'crawler bots' to identify and therefore ignore it in the future, so as not to reel out the same text over and over.

Here's the kicker though, this code also signals to the SEO crawler bots that the text it's attached to should be ignored, meaning that no matter how great the blog you just generated and copied into your site is, it will actually act negatively for the SEO of your site!
So, as Louie Walsh used to say 'Make it your own', edit the text generated by ChatGPT to really reflect your own brand identity and tone of voice, and always ensure it is crafted to speak to your ideal audience too.

Consider the words and phrases used and make sure you’re certain these are things you would say yourself. Because it’s getting easier to spot AI generated text, and if a reader does that it will turn them off almost immediately. Your readers value honesty and transparency, if they perceive you don’t even have the time to write a blog yourself then why should they invest in the time to read it?

Can you teach a computer programme to roll over?

I had the pleasure of listening to Joyann Boyce speak recently and what I learned from her really made an impression.

You know by now that AI is always learning, but have you considered who it is learning from?

Joyann used the analogy that AI is like a puppy, and a puppy will only learn the information it is given by those who train it. So, by the same token, AI is trained by those who use it and enter information into it.

This then creates a situation that I, in my white, able-bodied, cis-female privilege, had failed to even comprehend may be an issue.

AI is built on bias because humans are biased. We must consider when using it that bias is always at play.

How do we counter this? When entering your prompts you need to be specific about who you want to reach and how the language should be presented. For example, when asking for a social media post to advertise running trainers, you could specifically tell the programme to add inclusivity to the copy, because trainers can be worn by anyone not just able bodied athletes.

Maybe, if we all do this consistently then eventually the language models will learn to do it automatically but it will take a very long time so we must always bear in mind that the language being produced will contain some level of bias.

Pants on fire

This is a shocking one . . . AI LIES!!!!

That’s right, you can’t even rely on a tech bot for the truth these days! It’s a weird one, and I learned all about this from the gorgeous girls Louise and Naomi at Lucky Night Studio who really know their stuff!

Think of ChatGPT as the school jock, they’re confident, cocky and loud, and they DO NOT want you to find fault in anything they do. So they naturally present themselves as being far better than they actually are.

For example, if you were to paste a URL into ChatGPT and ask it to follow that link to check something for you, it would most likely come back and say yes, I did that. But, if you were to dig a little deeper, and ask in the same chat for it to return to the website and tell you what the second paragraph on the page said, then ChatGPT would have to ‘fess up and admit that it cannot in fact got to websites. It is a limitation within the programming and it simply cannot do it.

But the point, in case you didn’t get it, is that it had lied to you when it said it been there in the first place!

Equally, these programmes only know what they have been taught so if the information they have been fed is incorrect then the answers they provide in the future will also not be accurate.

Louise of Lucky Night gave the very strong (and valid) advice that you should never ask ChatGPT to write you an article on something you know nothing about, if you don’t have the necessary knowledge to check the facts and be certain of the accuracy then how can you be sure that what you have been provided with is truthful?

Part of your toolkit

All this being said, language model based programmes are fantastic for helping you to generate ideas for your next blog, or to outline the bullet points of an article you need to write.

But please don’t use it verbatim as a factual resource.

It’s a tool to help with your copywriting, not an ‘out-and-out’ replacement for you doing the work.

Phew . . that was a lot right? And I’m pretty sure I’m only scraping the surface. If you want to know more about the potential pitfalls of AI, then why not enter a prompt into ChatGPT and ask it? I did . . and it made for some VERY interesting reading!

Let me know what you thought of this blog in the comments, and please, please share it with someone who you think may benefit, I would really appreciate that!!

Thank you!