Copilot Exam Assistant... Epic Fail!

So ChatGPT, large language models, generative AI etc is so awesome, right 

Well, let me tell you a little story. Yesterday I had to renew two Microsoft certifications. Renewal exams are open book so you can search the internet to find the correct answer. I'll interpret that as, "you may use ChatGPT / Copilot / whatever to help with exam questions". They were the two "solution architect expert" certifications too that I really don't want to let expire. I won't say I was lazy, but I was feeling particularly unenthusiastic about having to use my brain on that day when I just wanted to get out of the house ASAP and go spend my time on something useful, such as... building a boat, for example. A brilliant idea popped into my head - hey, why don't I copy paste the exam questions into Copilot as fast as I can and smash through this renewal exam in like 15 minutes. I did just that. Now I won't say I'm proud of doing that cause it felt like the ultimate form of cheating, but fuck it, it's open book, and I know this shit, but let's say I was a bit pressed for time and really didn't want to waste it on some silly exam questions.

Aaaanyway, to cut a long story short: I FAILED THE EXAM! I was flabbergasted. How?! This is Microsoft Copilot, and it gave me the WRONG ANSWERS to some "simple" Microsoft renewal exam questions. How could that be when generative AI is so awesome and, you know, has all this intelligence, distilled knowledge and reasoning ability. I was mad because then I had to do a second attempt at passing the exam and I actually had to READ the questions and THINK about the correct answer. I passed both exams an hour later.

Going back to doing a bit of root cause analysis on the epic Copilot exam assistant failure, it boiled down to two things:

  1. If it gave me the wrong answer and I then prompted it with something like "why isn't option y the correct answer" the AI quickly changed its tune and responded to something like "you're right, I apologise for the confusion. Option y is the correct answer because such and such..." It was basically like arguing with a compulsive liar or a particularly spineless person who would just agree with you on whatever the moment you challenged them.
  2. The references it generated to support its answers were quite often just made up of hyperlinks that led to web pages that didn't exist.

So there you go. Next time you're feeling depressed and Google AI thingy tells you to consider jumping off a bridge, I recommend applying some critical thinking. Likewise, adding non-toxic glue to pasta sauce to make it stick to pasta and cheese is probably not the right answer.

Comments

Popular posts from this blog

That pesky hard-coding

Poor Man's T-SQL Formatter

Wingfoiling app for Garmin Watch