Passing My Renewal Assessment For Microsoft AI-102 With MistralAI’s Mixtral8x7B LLM

Leandro M Losaria
3 min readApr 1, 2024

--

Context

I recently got an email from Microsoft that my Microsoft AI-102 certification is now eligible for renewal. This was then called Microsoft AI-100 and I took it last May 2021 before it was retired and replaced by AI-102. Since then, I’ve renewed it annually; this will be my 2nd time. I also just recently set up MistralAI’s Mixtral8x7B on my Ubuntu machine and wrote another article here in Medium. So it got me thinking… Wouldn’t it be the perfect display of AI Engineering competencies if I use an LLM to solve a real-world problem or challenge, which is to pass the AI-102 renewal assessment?

The email I got from Microsoft

My Plan

My plan was simple, I’d copy and paste the renewal assessment’s questions and submit it to my locally hosted Mixtral8x7B using Jupyter Notebook. Whatever answers the LLM would come up with, I’d trust it with no questions asked.

If I do fail the renewal assessment, it wouldn’t be too much of a big deal, and will just retake it and study this time around, because as stated here

https://learn.microsoft.com/en-us/credentials/certifications/renew-your-microsoft-certification-faq

How many times can I retake the assessment?

There is no limit to renewal assessment retakes. Just be sure to pass before your certification expires. If you don’t pass, you can try again immediately. After the second try, you’ll need to wait at least 24 hours between future attempts.

Taking The Assessment

Usually, Microsoft renewal assessments are mostly multiple-choice and have around 25 questions that cover different aspects and categories of a broader topic. I took the renewal assessment earlier this afternoon. The first question was this one.

The first question of the renewal assessment

I immediately copied and pasted the question’s text into a Jupyter notebook and submitted it to Mixtral8x7B as seen below.

I can see from the terminal that the LLM is processing my question

I’ve instructed the LLM to, not only select the best answer, but also explain in detail why this should be the answer.

Submitting the Question To Mixtral8x7B for inference

I continued this process with all 25 questions, sometimes changing the instructions as some of the questions can only be answered by Yes or No, or sometimes you have to choose 2 or 3 answers from a pool of 5 possible answers.

The Result

After submitting all 25 questions to our local LLM, and finishing the assessment, I let out a loud laugh when I saw that I passed the assessment (albeit BARELY!) as seen below.

Ha! 64% out of 100! Barely passed it, but I'll take it!

Parting Words

Honestly, I was expecting Mixtral8x7B LLM to pass the assessment for me, but doing and passing it for real made me chuckle. This was also more of curiosity on my part and would like to know how Mixtral8x7B would perform (64%). But take note that no LLM can beat someone who’d take time to study.

In case you are interested in knowing and understanding more about Mixtral8x7B or LLMs in general, I’d recommend that you watch these YouTube videos.

How Did Open Source Catch Up To OpenAI? [Mixtral-8x7B]
Introduction to large language models by Google Cloud

I hope that you had fun reading my article. If you don’t mind, please tell me what your thoughts or ideas on how I would be able to improve things. I’d be very grateful to you.

I wish you peaceful skies, wherever you are.

Always be awesome!

--

--

Leandro M Losaria

Data plumber, janitor, and story teller. Full stack data scientist in a Fortune 500 company. Located in the Philippines.