Misleading Questions / Answers

[Karan]

Some questions are misleading and topcis seem to be outdated. The first question marks the correct answer incorrect (see first two attachedscreenshots) and the second question is very misleading by stating that retraining is followed after labeling images so readers assume that in combination labelling and retraining the model is done in combination …”You upload the images and use the Smart Labeler tool to automatically tag and label the images before retraining the model. Does this process effectively update the classifier with the new species images?” Correct answer –> “No” Explaination: A standout feature of Custom Vision is its Smart Labeler tool, which assists in speeding up the data preparation process. The Smart Labeler uses active learning techniques to automatically suggest labels for new images based on patterns recognized in previously labeled data. However, while the Smart Labeler can help automate and accelerate the labeling process, the labeled data does not automatically update the model. To effectively incorporate new images and labels into the model’s performance, a new training run must be initiated.

[Nikee-TutorialsDojo]

Hello Karan,

Thank you for taking the time to share your thoughtful feedback. We sincerely appreciate your attention to detail and understand how some questions may appear unclear or potentially misleading.

Regarding your comments:

We’ve reviewed and updated the first question you referenced to align with the correct answer. If you’re still encountering issues, please feel free to share the latest screenshots or details, and we’d be happy to re-investigate.

On your statement, “You upload the images and use the Smart Labeler tool to automatically tag and label the images before retraining the model. Does this process effectively update the classifier with the new species images?” Correct Answer: No.

We understand how this question can seem misleading at first glance, especially since it mentions both labeling and retraining. However, the key detail lies in how the process is structured and what is explicitly done in the scenario.

The Smart Labeler tool helps automate tagging and labeling based on previously trained data, significantly streamlining the data preparation phase. However, simply labeling the images—even with retraining mentioned afterward—does not guarantee that the classifier has been updated unless the model retraining is actually performed using the newly labeled data.

The question’s phrasing focuses on whether this entire process effectively updates the classifier. Since labeling and retraining are not necessarily executed together automatically, the correct answer is “No”—manual intervention to initiate retraining is still required after labeling.

We hope this clarification helps! Please continue to share any areas where things seem unclear—we’re always working to improve the learning experience.

Regards,

Nikee @ Tutorials Dojo

[Karan]

Thanks for clarifiying!

I have another issue considering the following question:

Category: AI-102 – Plan and manage an Azure AI solution

You are working with an Azure subscription that includes the following resources: OpenAIResource1, an Azure OpenAI resource used to generate generative answers for user queries in a chatbot, ContentFilter1, an Azure AI Content Safety resource designed to filter out objectionable content from both the chatbot’s input and output, and SpeechToTextResource1, an Azure Cognitive Services Speech-to-Text resource for converting spoken language into text.

Your goal is to improve the accuracy of the content filter configurations by conducting tests with sample questions to ensure the content filter is appropriately fine-tuned.

Solution: You plan to use the Moderate text content feature from Content Safety Studio to perform the tests.

Which solution will meet this requirement?
Correct answer: “Yes”

What I don’t understand here:
We are asking to “fine-tune” a model which requires retraining in Machine-Learning-Lingo. Simply applying the “moderate text” feature improves our solution but does not incorporate retraining a model, hence fine-tuning is not applied. Therefore, compared to our previous case where we should have considered the complete process including retraining, not just labelling, we have exactly the same case as before – as retraining is not part of the solution and no fine-tuning will be applied. The answer should therefore be “No”. Your correct answer is “Yes”. This again is misleading and contradicting your previous answer.

[Karan]

Also please take a look at the first to screenshots – I think there is a mistake in the solution Azure AI language is marked wrong but mentiones as correct in the review. Sorry for mixing up some of my questions and thank you so much.

[Nikee-TutorialsDojo]

Hello Karan,

Thanks for posting.

This has been fixed. Kindly check if the changes have been updated.

Cheers,

Nikee @ Tutorials Dojo