The last interesting genAI experience I would like to present in this blog post was during my internship at a major telecommunications firm. Different than my usual application of ChatGPT, I had to work with Microsoft Copilot. And let me tell you, I am not a fan.
My team (of less tech-savvy employees I must add) was asked to implement Copilot in a new project. As the gen Z, tech enthusiast intern, I was told to lead this project with a colleague. The project itself was quite simple. It constitutes of a database which had to be summarized, analyzed, and finally presented to the customer. This database was nothing out of the ordinary, a few hundred columns and rows with basic inputs in MS Excel. After working and experimenting with ChatGPT since a good year or so, I was sure that basic prompts would work just as fine with MS Copilot. I was so wrong.
Copilot was able to tell me the largest/ smallest number in each category over the years stated in the database.
That was about it. I prompted “create a pivot table including data from A2:G8 (for example)”. That, apparently, was too difficult to understand so I tried again. “Please create a pivot table using data, starting from cell A1 to cell G8. Include all data in this range”. It still did not work.
Apart from the example of the pivot tables, analyzing this data turned out to be the most demanding task I got during that internship, since writing out the correct prompts alone cost me several hours. I must add, Copilot was just recently published for the private customer, and was far from perfect, however, thinking back to the beginnings of ChatGPT, the level of completeness in Copilot was far from the one of ChatGPT. I was happy that due to several issues with the AI, the project was closed a few weeks after.
In general, MS Copilot is a brilliant idea. I could think of a lot of tasks in my Microsoft applications which would be simplified by implementing AI in the program itself. However, unless Copilot receives a major update, I think I will further rely on ChatGPT.
(It has been over half a year since my internship, so Copilot might have evolved since then)
Thank you for sharing. I have the same thought—neither Copilot nor Gemini is as useful as ChatGPT.
I really appreciate how you provided specific examples of pivot tables in your article, which gave us a practical understanding that Copilot is not actually that useful. In this case, manual methods might actually be simpler. This makes us reflect: Can AI tools really improve our work efficiency?
Its refreshing to see a post about the downsides of certain AI. I agree with your stance on CoPilot and I think that many Generative AIs that are not chatGPT have a lot of downsides, and just seem to be created in order to profit off the hype around it. Both in user friendliness and in performance.
I do wonder however, if you think that you would have been able to do this specific task way better with chatGPT, and did you try using that in the end? And do you think CoPilot will ever take off, seeing as ChatGPT is able to do way more if you just give it the dataset? I personally think most people will stick to ChatGPT. But it’s good to know I am not missing out on CoPilot, I’ve personally never used it and don’t intend to.
Hey Bosse, thank you for the comment! Regarding your first question, the project (unfortunately) was based on Copilot so I did not try to use anything else. But it certainly might have been interesting, I have to give you that! To be honest, I have never (also not outside of the project) have tried to summarise tables and create pivot tables for Excel with ChatGPT, so I cannot really say if it would have been better. Technically, Copilot is created to specifically help with issues and projects like these in the MS environment, so it should work better than ChatGPT (right?).
Concerning the second question, I really hope that Copilot will improve on what it is trained to do. The potential is huge in my opinion. But directly comparing the functions of Copilot and ChatGPT, it will most probably not come close to ChatGPTs level. I also do think that it is not made to compete with ChatGPT, but rather made for simplifying the MS applications in-app. This feature is, as far as I know, not yet available for ChatGPT