ChatGPT and Those Awful Tasks…

I’ve spent quite a lot of time playing with ChatGPT now, and like many people are trying to get to grips with what it means to have a tool which can approximate some sort of creative response when asked to write a poem or develop a piece of prose or whatever. The conversation here is obviously ongoing and pretty fascinating: can an AI be creative, what does “creativity” mean anyway, are humans themselves anything more than an advanced language model, etc. If you want to read a pretty great bit of writing from the ever-awesome Nick Cave, take a look at this post where he responds to a ChatGPT generated lyric with his usual humour and insight…

More recently, I’ve been playing with actual use cases in my working day. Previously I did a quick experiment to make a couple of web things using ChatGPT, with some interesting results – nothing mindblowing, but you start to see the limitations and possibilities.

This week I’ve installed MacGPT which gives a nice direct access to the interface at all times, and so I’ve been using ChatGPT more frequently as part of a normal working day. I’ve been trying to avoid asking questions with specific, immediate factual answers – I know from reading and some experience that a standard search (my search engine of choice is now the excellent Kagi) does a better job of that. The questions I’ve asked have been really in 3 main areas.

Firstly, CSS. I’m a pretty novice CSS’er so being able to ask “make a 3 column layout using flexbox” – and then not only get “an answer” but also an “explained answer” is pretty great – so the following text is appended to the actual CSS and HTML code for this request:

In this example, we first create a container element with the display property set to flex and flex-wrap set to wrap. This allows the columns to wrap onto a new line if the container width is too small to display them all in a single row.

Then, we create three column elements with the class column. We set the flex property to 1 to make them all equally sized, and add some margin and padding for spacing. We also set a background color to make them stand out.

For me this is really helpful – but as per my previous post, what’s even more helpful is being able to continue on with the request – so for instance “now can you add a row with 2 columns and a background colour of red” will get me the response I need.

Second, WordPress and PHP. This has been documented to death, so I won’t go into it here – but where ChatGPT excels over just doing a StackOverflow search is again that I can ask the engine to return further details, or append in meta-queries that are specific to me, or filter by a custom taxonomy or whatever. In my experience over the last couple of weeks, ChatGPT gives pretty great answers – sometimes they need a little tweak or two, but it’s definitely time saved compared to traditional sources of information.

Finally – and this is where I’m most excited about the potential for time-saving in a modern work environment – those god-awful tasks that we all have to do all the time when trying to run a business or do a modern desk job. Here are two examples:

Firstly, I had a bunch of data in our financial software and I wanted to add up figures in a column – and there’s no export feature on this specific page. So a normal method would be to highlight the text, copy it into a spreadsheet, hope for the best and then sum stuff up. But try it and you realise that the financial software uses divs and not a table to display the data, which means when you copy-paste it you get a chunk of text without any structure – nothing you can do anything useful with. Here comes ChatGPT – I pasted the text in there and asked it to sum up the total of all the £numbers:

Here’s a part of my input:

Gsuite Ireland On 01 Mar Bdc
Computer Software · 02 Mar 23 · £23.00

Gsuite Ireland On 01 Feb Bdc
Computer Software · 02 Feb 23 · £20.91

Gsuite Ireland On 01 Jan Bdc
Computer Software · 02 Jan 23 · £18.40

Computer Software · 02 Jan 23 · £4.60


My first attempt was simply “can you sum up all the £ values in the list”, at which ChatGPT responded with a very confident sounding answer. Only… when I double checked, the answer was wrong – and not only that, but when I asked the engine to re-add, it made up a second, and then a third wrong answer. Now – because I know a tiny bit about language models, I knew that GPT-3 (which is what I was using) just doesn’t have the capability to add up – but most people wouldn’t know this, and the confidence of the answer means that most people wouldn’t even check. But – what’s interesting here is that ChatGPT has done the hard, gnarly bit which is gathering together all the figures in a long list. And it turns out you can ask this: “can you now pipe delimit the list so that I have description, date and value in three pipe separated columns?” – and, woo, it does it:

Yes, here is the list with descriptions, dates, and values separated by pipes:

Gsuite Ireland On 01 Mar Bdc|02 Mar 23|£23.00
Gsuite Ireland On 01 Feb Bdc|02 Feb 23|£20.91
Gsuite Ireland On 01 Jan Bdc|02 Jan 23|£18.40
GOOGLE*GSUITE THIR IRELAND ON 01 JAN BDC Attachment|02 Jan 23|£4.60

And now we’re getting somewhere – this has taken an un-ordered list of stuff and made it into something nicely separated and ready to get into a spreadsheet.

As long as you know the caveats about not asking for any maths skills, this is definitely a time-saver – especially given the above list of stuff was about 50+ data points long.

A bit later in the week I was populating a new client website and this time was given a Google Doc with some “data not data” in it – a list of things / times / days / contact details, but all tab and space separated rather than anything nicely formatted. Now this is exactly the sort of thing you’d spend a bunch of minutes on formatting into a nice tabular format, copy-pasting or manually typing in each entry. So this time I pasted in the raw text to ChatGPT with the following instruction: “turn this into an html table”. And lo and behold, it not only did it (with accuracy) but also correctly guessed the number of columns and the right titles to give to each column.

(on left, original content – on right a nicely formatted html table pasted into WordPress)

This is seriously impressive stuff, and a real time-saver. But more to the point, it’s time-saving on awful tasks that we all went to avoid – and this to me is exactly what our robot helpers should be doing for us.

Do you have any good examples of awful tasks and how you’ve used these new tools to help? I’d love to hear more examples 🙂

Leave a comment