Which generative AI tool? ChatGPT or Copilot?

Recently I have been considering switching from ChatGPT to Copilot. For one simple reason, I do not know where my ChatGPT data really goes.

My thinking on this

Most Universities subscribe to Microsoft packages, and in some instances this includes Copilot. Having this sort of a Copilot subscription can mean that the settings on the SSO (single sign on) version allow for any inputs not to train the model and the privacy is lot more robust. Do I trust this completely? Never, but at least I know where the data goes… sort of. I would still not put anything I (or anyone else) would not be happy to share publicly. Microsoft has been trusted as a provider for a long time, and we all store a lot of information that we would not want others to see using Microsoft as a service provider. Copilot even tells me that it will integrate with the other Microsoft productivity tools, so I might have to give them a go too.

Is it a good alternative?

How does it compare? I have had a go. And it seems to work ok. I tried building a Moodle quiz using MCQs which was pretty comparable to ChatGPT, but the Gap Fill needed a few more rules. For some reason it wrote a block of text with questions after it, rather than have the gaps in the text. So I have had to adapt my prompt a little. I find Copilot does need a few more prompts after to get it correct, and I cannot get away from gaps next to each other that will be incorrect if the answers are correctly swapped, but maybe ChatGPT just learnt what I wanted. I also tried to generate some anatomy and physiology diagrams on reproductive systems for some module content, and it gave me a very direct no for this.

Use the rule below to build a gap fill question for a quiz on the #topic of choice#


Produce a 3 word title for the question that does not indicate any of the correct answers

Use [] to surround words to form the gaps to drop words into.

Produce at least #number of gaps# gaps to be filled.

Gaps should be in the main body of text, and not in questions after

Interchangeable words are disallowed, for example [first] or [second] phalanges where either could be correct but would be incorrect in the quiz.

UK English to be used.

The text should use terms expected in the knowledge of level #level of study# on a #programme or field of study#

The text should be around #lengeth of text# words in length.

a set of #number of incorrect but potential words# distractor words provided separated by commas on a single line that could be potentially correct.

Distractors should not be present as gaps for the gap fill.

Words can only be used once.

Adjacent gaps are disallowed, for example ‘the humerus connects with the two bones of the forearm: the [radius] and the [ulna].’ is disallowed as either orientation of the answers could be correct, but not recognised within the quiz tool

Do not make reference to the programme of study in the text.

I also tried to create the header image for this post in Copilot, and it feels different, but it is not awful. One big thing is that ChatGPT incorporated itself as an avatar into its images, whereas Copilot did not appear to do this at all. I prompted Copilot, and checked that I still own the image and all good with that one.

create me an image for a blog piece comparing ChatGPT and Copilot as AI tools and using then higher education

Copilot. (2024). Comparison of ChatGPT and Copilot as AI tools in higher education. Generated by Microsoft Copilot on 28 October 2024.

The model still appears to be conversational and comfortable for communicating with, it is quite responsive and I like that it gives me options for images, which ChatGPT does not. If you have a subscription to Copilot through your institution, I am guessing there may be variable usage limits. I cannot see what they are so I expect that is sitting in some settings menu hidden away from the end user. And getting a proper answer from Copilot on that question appears too difficult.

So far so good. I am going to make the move over to Copilot to be a bit more responsible in my AI use in terms of ‘using it for work’. You still might see me use other tools within these posts, but I cannot endorse them. I can only report on my experiences. It is up to you to check what is safe, secure, and complies with the policies that govern your work.

Will I switch

Yes, I am going to give it a go. I will see what happens with Copilot and see if it really does speed up and improve anything I do.

Does this mean I will be using Edge and Bing? I don’t think I will go that far.