Now back to the list of things I tried to use ChatGPT for (when I thought it was broken, but I was actually using it wrong). There are those that went well (the good - covered in part 1), and those that went badly (the awful - ones I really shouldn’t have been using ChatGPT for) and the ones which could have been better, or took a bit of tweaking to get exactly what I wanted (the okay).
(Reminder) What I’ve Been Using ChatGPT For:
- My weekly Monday motivation LinkedIn post
- Researching really specific things, eg how many data graduate schemes in the UK
- Making some images for a blog post
- Writing a bio on me
- Fixing some code issues I was having
- Prioritising stats education for a course
- Trying to identify an animal I saw in Costa Rica
- Helping me understand what to include and where in financial modelling
- Researching hiking
- Writing development plans for employees
- Playing a game where I try and remember all fifty states of USA
- General research on data jobs
- Editing an image
The Okay and The Awful
Something I often need to be reminded of when using ChatGPT (or other Generative AI chatbots), is that they aren’t a magic all-knowing talking head (like in the Wizard of Oz). It sometimes makes sense to take a step back and think about what these chatbots can actually do.
Firstly, they are apparently trained on “the internet”, which means they aren’t going to know other random things, not on the internet (this raises an interesting question to me - are all books, particularly textbooks, on the internet?) Anyway, what this means in reality - if you are good at using Google or other search engines to find information, then ChatGPT is not going to magically find information you couldn’t.
Research
The Awful - Complicated or In-Depth Research
If you notice, the activities highlighted in red are very research-focused. Specifically, they are me trying to use ChatGPT to try and find information I couldn’t find (e.g. how many data graduate schemes or other in-depth figures). Unfortunately, as much as this feels like a very natural use of these superbrains, it is not a useful way to use these tools.
If I Google the phrase “how many data graduate schemes are there in the UK”, the results are a bit underwhelming. There is no helpful box at the top with an answer (like if you type something with a more determined answer, eg “how many countries in Europe”). The results also aren’t very helpful - it lists the Civil Services Graduate Schemes, an Indeed search for data graduate jobs and a few other top graduate schemes. One reason for this (I think) is the ambiguity around the word “data” - in this case I mean “data graduate schemes” but I could be asking for data about graduate schemes.
Now this is my error, turning to ChatGPT and expecting a better answer. After all, Google basically indexes the entire internet (making everything searchable for companies like OpenAI to train their AI chatbots).
Example:
L: how many companies have data grad schemes in the uk
ChatGPT: Determining the exact number of companies in the UK with specific graduate schemes in data-related roles (such as data science, data analytics, data engineering, etc.) is challenging without direct access to updated and comprehensive databases of graduate employment opportunities. The availability of graduate schemes, especially in data-centric roles, varies widely by industry, company size, and demand for data skills.
However, graduate schemes in data roles are increasingly popular among large corporations, particularly in sectors where data is a critical asset, such as:
- Technology and Software
- Finance and Banking
- Telecommunications
- Retail and E-commerce
- Healthcare
- Consulting and Professional Services
Large multinational companies and well-established UK firms are more likely to offer structured graduate programmes, including those focused on data skills, given their resources and the strategic importance of data in driving business decisions and innovation.
[And it continued for 7 more paragraphs, including another set of bullet points]
Long, unspecific and overly unhelpful. But what did I expect?
The Okay - General Research
What ChatGPT was helpful with was more exploratory, open-ended research on trends and what these could mean for specific use cases. For me, this means getting information about the data job market, and asking ChatGPT to relate these back to Taught by Humans and what we are trying to do.
L: Can I have a bullet-pointed list of all data-related jobs in the UK (job titles)
L: Do any of these jobs have common skills?
L: Are there any trends in education for people working in data jobs?
Sometimes this required a bit of back and forth:
L: can you estimate the number of data skills
L: I want a number not text
L: using average days to hire, work out the estimated cost for a UK company to hire a data person
L: how would you estimate how much money is spent on data training for companies each year?
It is sometimes useful to ask for estimates, rather than “how many” as this allows ChatGPT to use its logical reasoning to work out a number (which may be wrong, but at least it is a starting point). Similarly, ChatGPT often replies with very long text answers, when sometimes all you want is a number - you need to specify this.
Advice
The Awful - Hiking Ideas
I also asked about some ideas for hiking, and as hiking is a well-covered topic on the internet this didn’t really get me any new information.
L: Where to travel in September for 10 days for hiking (long haul from the UK can fly on BA)
This got me a list (that I am sure I read on a travel blog when I was researching). One good thing about using these tools over just researching yourself is giving specific parameters.
L: No further away than Chicago is. Direct flights only
This can get useful results - unfortunately, in this case, it still didn’t get me anything better than what I had already researched myself.
The Okay - Lots of things
ChatGPT (for me at least) is useful for giving specific advice on well-covered problems, for example.
One great example of this is using ChatGPT for coding help on a specific problem - when you look up help for an error or an unexpected behaviour when coding, you get someone else’s code or generic advice (unless you post it yourself on a forum like Stackoverflow). With ChatGPT you can send the code, any errors and a description of what is happening.
Another helpful way to use ChatGPT for advice is how to apply frameworks to your company or team. For example, at Taught by Humans I had two interns and needed to do some development planning during their internships. I was unsure what would be the best framework to use for this - we are a small business, with limited resources; these were short fixed-term part-time internships; we have a particular company culture which I wanted to be reflected in our people management. ChatGPT allows me to mention all of this, and can recommend different frameworks with the pros and cons for our business specifically.
Random and Obscure Without Detail or Needed Context (the Awful)
Similar to the issues with research, I have been guilty of expecting ChatGPT to work magic. While on holiday in Costa Rica earlier this year (apologies if that makes anyone else sad about the British weather…), me and my partner saw this skittish little animal darting around. It was quite small, but kind of looked like a weasel or a ferret. Didn't think much of it, but asked the staff at the bar what it would be. Spent about half an hour with them suggesting things, us googling them and everyone being stumped.
I decided to try ChatGPT who suggested (quite impressively) all the same animals as the locals had. But eventually, it got fed up (or so it seemed) with me, and started suggesting birds (even though I said it was like a ferret) and the same animals again.
Annoyingly I reckon if I'd had a picture it would have been able to (but then probably so would the locals).
L: Hello I am trying to find an animal I saw. In Costa Rica in Monteverde. It was small and thin. It has a long fluffy tail. Mostly black with a white marking on the head. Quite like a ferret
ChatGPT: Hello! Based on your description, the animal you saw in Monteverde, Costa Rica, sounds like it could be a Tayra. Tayras are weasel-like mammals that are indeed similar to ferrets. They are known for their slender bodies, long, bushy tails, and are generally dark in colour, often with distinctive markings on their heads, which could include white. They are native to Central and South America, including Costa Rica, making them a likely candidate for the animal you encountered.
L: Much smaller and fluffier
ChatGPT: Considering the additional details of the animal being smaller and fluffier, it might be a Central American squirrel. These squirrels are native to the region, including Costa Rica. They are known for their smaller size, long fluffy tails, and agility. Their colouring can vary, but they often have dark fur with some lighter markings, which could include a white patch on the head, similar to what you described. Their appearance and behaviour can indeed resemble that of a ferret to some extent.
L: More black and much thinner
ChatGPT: Given your description of a much thinner, mostly black animal with a fluffy tail and a white marking on the head, you might have seen a Black-tailed Trogon. These birds are found in Costa Rica and are known for their slender bodies, long tails, and distinctive colouring. Although they are primarily birds and not mammals like ferrets or squirrels, their long, thin appearance and tail characteristics could be mistaken for a small mammal at a distance or in brief sightings. Their colouring typically includes darker shades with specific lighter markings, which aligns with your observation.
Recapping on these has also made me realise how much of this was actually on my terrible prompts.
I continued (causing the awfulness):
L: Differently not a bird - it had four legs and was very fast
L: Not an iguana. It was a mammal like a weasel or stoat
L: Smaller blacker
L: Does not look like a picture of those
L: You already said that - it was much smaller
L: Just over a foot long. No more than 2 inches wide
L: That looks more like it. But it did not have a white bib
I think it was a weasel with strange markings, for those interested. But something that is said a lot in the data world - garbage in, garbage out (that is the polite version) really applies here. ChatGPT is a tool, and tools need to be used correctly to get the required results.
Image Editing (the Awful)
I tried getting ChatGPT to update the text in an image (as I wanted a way to do this quicker than updating them all in Canva by hand).
L: can you add this to this image in the place of the words:
Title should be Google Sheets Functions - SUM
Content type should be text - info - 0-2 mins
The output was not what I asked for:
I respect the effort - but not what I needed.
L: I wanted the same image I sent you
ChatGPT: I apologise for the oversight. To rectify the text on the original image you provided, I'll need to manually edit the image. I'll update the text as per your request and maintain the design and style of your original image. Please give me a moment to complete this for you.
The image has been updated with the specified text in the same style as the original. You can view and download the edited image below:
And some code to create the images myself - which I will be honest I haven’t tried as I was in a rush. But it is interesting that ChatGPT cannot directly recognise text in an image and update it (something which can be done in both Adobe and Canva).
Upgrading the Okay and the Awful to the Good
Using ChatGPT (or any other AI tool) it is often about how you use it, and making sure you use it for the right use cases. These are some tips worth remembering:
1. Remember what they actually know - ChatGPT is trained on the internet, if you can’t find the answers on Google search, it is unlikely it has the information either.
2. If you want an estimate, say so - When asking for estimates of the things which aren’t easy to find on Google, it can be helpful to specifically ask for a number and say you want an estimate “can you estimate how many data skills there are? This should be output as a number”.
3. ChatGPT is useful for adding context and specifics to well-documented problems - this helps you work out how to actually solve business problems or which frameworks to use in your work, eg fixing code, setting up spreadsheets for business development.
4. Images still don’t work as I magically expect - the chatbots nearly always produce new images, even when asking for small tweaks on an image it created.