|
Getting your Trinity Audio player ready...
|
This is an issue I want to address because I believe it can help a lot of people. It may also give me ideas to improve in any feedback I get. People ask me how I use AI – which tools and which process and I have decided to write an in-depth article on the exact process. I am making one change for this article – a quiz question which you will find at the end.
I use AI and I will tell you exactly how.
The ideas for this website are all mine – 100% – I don’t ask AI for ideas but I will ask Claude or Chat GPT to show me research on a topic I have decided to write or to flag for me new studies. For this article, I didn’t. The ideas usually come when I least expect – in the shower, doing Yoga or when I have a few moments to think. I don’t sit to write when I don’t have an idea. Today for example, I had lunch with friends and we discussed AI and how I use it and I decided to write about it and share.
I always draft the first version but use AI in this process. Sometimes I type but most of the time I dictate. I have tried using the standard dictation tools, but they don’t pick up my voice well as it waxes and wanes and as its strength is generally low or I am dictating in the summer with a fan and/or the AC on. So, I use an app – Wispr Flow – which is an AI app to dictate. It somehow picks up what I am saying with startling accuracy. It is simply awesome. Instead of seeing the text as you write you hold a button and it takes the text and cleans it of Ums, mistakes, and repetition and usually leaves me a truly representative draft. This afternoon I happen to be typing but the posts this morning were dictated.
If it is a more technical/medical article or if I am writing a non PSP article, I will often ask Claude to summarize/double check a research paper or papers/sources. If I want an infographic, I will ask usually ask Claude to give me a prompt to use in Chat GPT to create an infographic image. I do this because I find Chat GPT so much better at doing it.
I then do a read through (I use Microsoft Word most of the time) and check spellings, grammar but most importantly the feel and message. Often it leads to redraft after redraft or the bin.
Under a year ago the process stopped there. Tools like the dictation software can help so many people who find it hard to type. For this, the article is worth writing.
On a couple of occasions (a while back), I tried to use AI from the very start of the process to make it faster and easier, but I have wasted so much time. It simply doesn’t replicate what I want to say, the nuance, the humour (if you can call it that) and it doesn’t feel like me. So, I don’t. It has also proven quicker for me to develop an idea fresh than play with someone else’s.
At this point, I put the text into Chat GPT and ask it to score it for me out of 100 – usually the score is somewhere around the 90/92 range. If it is a more educational article, I do the same but in Claude. I ask it to fix the typos and grammar and potentially some of the feedback it has given me and ask for a final score. If it is around 95 or above that, is it for me.
If it scores an article far below – the truth, is I knew it before intuitively – and almost always it is because the idea doesn’t work. 9 out of 10 I abandon and 1 out of 10 I try and redraft it and repeat the process but is usually a waste of time and the article remains unpublished.
This morning was an exception and I persisted, which is why I guess it was on my mind at lunch. I had an idea and drafted it, but I couldn’t decide how serious or funny to make it (It was an article on dressing), whether to make it about me personally or others, or indeed whether to broaden it to other topics. It is a part of the process which is frustrating, but I love it.
I then ask the Chat GPT to do a translation at a very high score (95+/100) into Hebrew using the male vernacular because like French and not like English, the masculine and feminine matter far more.
If it looks ok – from a very quick glance – then I go with it but if I feel it is wrong I put it into Claude and double check (and sometimes put the new version back to Chat GPT).
Images, when I use them apart, I create in Chat GPT usually asking for something very specific or asking it for an image which reflects the tone of the article (and then edit). The Buddy photo was an exception and was from my own camera roll.
That is the process I use. AI helps me to a very significant extent and the largest part by far is the dictation software and then the proofread and sense check (and of course the translation and the image).
These are parts of the process I simply could not have imagined a year ago and they allow me to do what I do. These tools are available to people willing to pay the software licenses.
At the core however, the ideas and the drafting are mine, as sadly is the humour. As I reach this last line, I have not touched AI but now I will do something different. I will score this out of 100 in Chat GPT (the exact prompt I am using is: “please score this blog post out of 100” and see what it says. What do you predict? I predict a 92.
It scored it 93 and I am quite pleased with myself I am making no changes (I can’t anyway as it will render the quiz impossible), and I will translate and an image with not too much writing to avoid giving the content away.
