AI-102 Generative AI

Priming is utilised in this example because it involves setting up the context or role of the AI model explicitly in a system message. It instructs the model about its role (“You are an AI assistant that helps people solve mathematical puzzles.") and also provides directions about how it should respond ("Explain your answers as if the requestor is a 4-year-old."). This essentially 'primes' the model for the conversation, as it lets the model know the expected behavior and persona that it needs to take on throughout the dialogue.

Exam Question June 2024

Answer D, asked in Feb 2026.

I say this answer is D.

D is correct answer.

C is the answer : to explain it step by step
Not priming, that manipulates the output

A. few-shot learning is right answer.

A is the answer.
A common way to adapt language models to new tasks is to use few-shot learning. In few-shot learning, a set of training examples is provided as part of the prompt to give additional context to the model.

When using the Chat Completions API, a series of messages between the User and Assistant (written in the new prompt format), can serve as examples for few-shot learning. These examples can be used to prime the model to "respond in a certain way, emulate particular behaviors", and seed answers to common questions.

I'd go with "chain of thoughts", since it's about explaining the answers.
Priming is a different story: "This refers to including a few words or phrases at the end of the prompt to obtain a model response that follows the desired form. For example, using a cue such as “Here’s a bulleted list of key points:\n- ” can help make sure the output is formatted as a list of bullet points."
...that's not the case here!

https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/advanced-prompt-engineering?pivots=programming-language-chat-completions#prime-the-output

Priming is the answer

https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/advanced-prompt-engineering?pivots=programming-language-chat-completions#system-message

Must be priming, here we setting the role for the AI model

D. Priming

Priming involves providing context or instructions to the model before it generates a response. In this case, the system message is priming the GPT-3.5 Turbo model by setting the expectation that it should provide explanations in a way that is understandable to a 4-year-old. This technique helps guide the model's behavior and output based on the given context or instruction.

chain of thought. Try it on OpenAI. It explains step by step for a formular like 7+5*3+8

the answer is B
Output type: "EightSeverityLevels"
returns a floating-point severity score from 0.0 up to 7.0

The severity levels are from 0 - 7, with 0 being the lowest, if any form of offensive material, and 7 being very naughty/offensive.

Given this, A is the answer.

A is correct.

In this scenario, you are submitting an image to an Azure AI Content Safety resource for evaluation, specifically looking for violent content using the "EightSeverityLevels" output type. The image you submitted contains a simple circle, which likely doesn't depict any violent or harmful content.

Given that the image is non-violent, the expected severity level for violence should be the lowest possible score. In the "EightSeverityLevels" configuration, the scores range from 0 to 7, with 0 representing the least severe and 7 representing the most severe.

Thus, the correct output in this case would be:

A. 0

0: Non-offensive
1-2: Very Low
3-4: Low
5-6: Moderate
7-8: High
9-10: Very High

Could this not be A? Configuring the model to include data from the travel agent’s database would allow the chatbot to access a source of relevant and specific information, which can significantly improve the accuracy of its responses. Fine-tuning the model with data specific to the travel agent’s services and customer interactions can lead to more precise and contextually appropriate answers.

The question s about how to ensure the model responses are accurate and not randomly made up creatively

Hence the answer is Option C to set the lower temperature that can ensure model is less creative and more factual

用于预订、规则性强、不能出错的场景（如旅行预订）时，Temperature = 0 是标准做法

C is the answer. Temperature = 0 makes the model deterministic and focused, reducing randomness and ensuring it picks the most likely response. This is ideal when accuracy and consistency are critical, such as in travel bookings. Why not A? A. Configure the model to include data from the travel agent's database
This is useful for relevance, but GPT-3.5 cannot directly access external databases unless integrated via RAG or tools. It doesn't maximize accuracy by itself.

Option C to set the lower temperature that can ensure model is less creative and more factual based on the data available

Lowering temperature to 0 makes outputs deterministic and reduces creative variance—best for maximizing accuracy/consistency in transactional tasks like reservations.

C. Set the Temperature parameter for the model to 0

To maximize response accuracy for travel reservations, lower the Temperature to 0, which forces the model to always select the most probable next token (deterministic output). This reduces creativity and increases consistency, critical for factual tasks like bookings.

Key Rationale:
Database integration (A) enhances relevance but requires RAG/plugins and doesn’t directly optimize baseline model accuracy.
Top P 0 (B) is invalid (Top P ranges from 0.1 to 1.0), and even low Top P values allow some variability.
Temperature 0 (C): Eliminates randomness, ensuring the model adheres to the most statistically likely responses.
System message (D) helps guide responses but doesn’t inherently improve factual accuracy without parameter tuning.

The GPT-3.5 model does not have direct access to external databases.

A is not correct. There is no configuration available to include data in database. You have to use RAG or Fine Tune the mode.
B setting Top p to zero will make it more unpredictable.
D Modify sys msg make affect the way model generate response
C setting temperature to zero make the model less random. so The answer is C

I say A is the right answear.
If we look at Q, it says “maximize the accuracy of the response”. Temperature parameter will effect but it isn’t maximize the response. I believe Data fuel AI, and will maximize the response.

Adding more data as in A will not necessarily improve the accuracy of the responses

Im laughing out loud by the ones who say D. Seriously? Ever worked with openai?

Selected Answer: A

A is correct answer.

GitHub Copilot has the answer Set Temperature to 0

To maximize the accuracy of the responses from the chatbot, you should:

C. Set the Temperature parameter for the model to 0.

The temperature parameter controls the randomness of the model's output. A lower temperature will cause the model to make the most likely prediction at each step, making the output more deterministic and focused, thus potentially increasing the accuracy of the responses.

Accuracy is granted by initial prompt. D

It worse nothing, that even MS labs is used travel agency staff grounding information for a models.

The ChatRole that should be used is ChatRole.User. This role is assigned to the messages that come from the user, which the chatbot is responding to. The Temperature setting can be adjusted to increase creativity in the responses.

1. user
2. temperature

ChatRole.System
Temperature
代码中 new ChatMessage(..., @"") 里的字符串是空的 (@"")。
如果是 User，通常这里会是一个变量（如 userInput）。
如果是 System，在模板代码中经常先预留一个位置（空字符串或占位符）来设定 AI 的身份。

1. System
2. Temperature
There is no any reason to send >empty< user prompt to a model, neither assistant one.
To a sense, sending empty system prompt will randomize results, though that is not really useful thing.

An example given by Copilot :
{"role": "system", "content": "temperature": 0.7}

ChatRole.User represents generating a response without being prompted by a speaker, making the AI more creative and less deterministic, allowing it more freedom to express itself.

You are not setting the temp for user, that doesn't make sense. People please think a bit and dont trust on stupid responses from obvious dumb chatbots.

System is what you're trying to configure.

Copilot says ChatRole.User.

"To make the responses more creative and less deterministic, you should set the ChatRole.User in your coding. By doing so, you allow the chatbot to generate more imaginative and varied answers, as it will treat the user input as a prompt for creative responses. This approach encourages the chatbot to think beyond predefined patterns and produce more engaging content"

Should be System since we're modifying the behavior of the Bot. So we start again defining the System message that must be defined only one time

Right answer is ChatRole.Assistant and Temparature, the settings are on responses from chatbot

First would be ChatRole.System as it is the first chat in the ChatMessage object and the second answer is Temperature

First Option should be ChatRole.System.

"Typical usage begins with a chat message for the System role that provides instructions for the behavior of the assistant followed by alternating messages between the User role and Assistant role."


https://learn.microsoft.com/en-us/dotnet/api/azure.ai.openai.chatcompletionsoptions.messages?view=azure-dotnet-preview#remarks

Well, I think that should be ChatRole.System
ChatRole.User naturally make no sense - that user request

Selecting ChatRole.User as one of the options would imply that the responses should be generated based on the user's input or perspective. However, in this scenario, the goal is to ensure that the responses from the chatbot are more creative and less deterministic. Including ChatRole.User might limit the creativity of the responses because the model would primarily consider the user's input rather than generating novel content.

By primarily focusing on ChatRole.Assistant, the responses will be predominantly generated from the perspective of the AI assistant, allowing for more creative and varied outputs. This approach ensures that the chatbot's responses are not overly influenced by the user's input, leading to more diverse and imaginative answers.

ChatRole.User (Assuming the input is an actual message from user)
Temperature

Documentation/Examples: https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/reproducible-output?tabs=pyton

D,C. Con rol de user el chat es menos formal. con temperatura se ajusta la creatividad en las respuestas.

Yes, if the content you're inserting is the actual user message, then you should use ChatRole.User. This role signifies that the message is coming from the user, as opposed to the assistant, which would be generating the response.

Yes is CORRECT. The Safety metaprompt feature in Content Safety Studio.

The Safety metaprompt feature in Content Safety Studio is used to guide AI models in generating safer responses by providing safety-related instructions within prompts. However, it is not designed for testing and optimizing content filter configurations by analyzing sample inputs and outputs for objectionable content.

https://learn.microsoft.com/en-us/azure/ai-services/content-safety/overview#product-features

The answer is B. No.

Explanation:

The Safety metaprompt feature is generally used to enhance safety by adding specific guidance to the model’s responses. However, it is not designed for directly testing or optimizing content filter configurations with sample questions in Content Safety Studio.

To meet the requirement of testing content filters on sample questions, you should use the Moderate text content feature, as it specifically allows you to check for objectionable content in text input and output. Therefore, using the Safety metaprompt feature does not fulfill the requirement.

API key
Deployment name (Different deployments can be configured on oai.azure.com)

Final Answer:
API Key
Deployment Name:

Why Bearer Token: This implies the use of Azure Active Directory (Microsoft Entra ID) authentication rather than a shared API Key.

- API Keys are shared secrets. If a key is leaked, anyone can access the resource. They do not strictly ensure "only the apps" have access.

- Bearer Tokens are used when authenticating via Managed Identities. You can assign an RBAC role (like Cognitive Services OpenAI User) specifically to the identity of each app. This ensures that access is strictly limited to those specific application identities, satisfying the security constraint.

Answer:
1 An API Key
2 A Deployment Name

API Key: Each app should use an API Key for secure access to the Azure OpenAI resource. API keys are commonly used to authenticate requests to Azure OpenAI resources.

Deployment name: Each app should specify the deployment name to connect to the appropriate GPT 3.5 deployment optimized for its unique workload.

The deployment name ensures the app uses the correct model version or configuration hosted in AI1

Answer is api-key (no confusion) and deployment_name. The issue with deployment name is that it is meant to confuse. There are three deployments of the model. The endpoint of all three of them remain the same. I worked with this and multiple deployments all the time, but still I forgot that the endpoint for azure-openai is same for all deployments. Of course to call you need the endpoint and the deployment name both, but looks like Microsoft wants to confuse you with mentioning REST. Just sad this company does this
Answer is very clearly now: api-key and deployment_name (what microsoft wants to answer)

I have years of experience using this, but I can't even understand the intent of the question. The build up is fine, but the question and the options make me really wonder. We use RBAC to connect in an enterprise environment.

Looks like the org that Microsoft outsourced this to (and getting money from them instead of paying) is asking the question on some image from REST api documentation. Guessing the low paid employee, I guess the answer he expects is :
api-key and end_point. REST api uses api-key as the header and also uses end-point to connect.

The scenario describing separate deployments does not make sense. An attempt at confusing the developer with experience from the real world. Nice work, vendor.

Answer is
An API Key
A Deployment Endpoint

https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/switching-endpoints

Note: There are key requirements in the questions ("Each app will access AI1 by using the REST API"). With this, apps will connect AI1 with REST API calls and to connect azure open ai with rest api, it is recommended to use API KEY and ENDPOINT as mentioned in above article.

API key and deployment name.
https://learn.microsoft.com/en-us/azure/ai-services/openai/quickstart?tabs=command-line%2Cpython-new&pivots=programming-language-python

An API key
A deployment endpoint

The Azure portal and documentation consistently talk about each Azure Open AI resource having a single endpoint, and each deployment having a name.

So the correct answer is most likely:
An API key
A deployment name

An API key
A deployment endpoint

While bearer tokens are commonly used for authentication, they are not specifically mentioned in the context of Azure OpenAI resource access. API keys are the recommended way to authenticate applications.

So, the answer should be API key and deployment endpoint.

1. An API key
2. A deployment endpoint

on exam today, I selected Api key and deployment endpoint

An API key
A deployment endpoint

chatgpt's answer was the bearer token and deployment name

Given answer is correct
1) chatcompletion.create - this is the API for creating chat completion
2) response.choices[0].text - for picking the text with highest rated response choice

Given answer is correct. This question was asked in September 2025 exam

Answer is correct

Correct. Under Completions > Example response
https://learn.microsoft.com/en-us/azure/ai-services/openai/reference#completions

# Output the response to the console
print(response.choices[0].message.content)

response = client.chat.completions.create(
The given answers are CORRECT

Answer is correct.

1. OpenAI
2. --encryption

1. Open AI
2. --encryption

This question was asked in September 2025 exam

--encryption is correct:

https://learn.microsoft.com/en-us/cli/azure/cognitiveservices/account?view=azure-cli-latest
az cognitiveservices account create -n myresource -g myResourceGroup --assign-identity --kind TextAnalytics --sku S -l WestEurope --yes
--encryption '{
"keySource": "Microsoft.KeyVault",
"keyVaultProperties": {
"keyName": "KeyName",
"keyVersion": "secretVersion",
"keyVaultUri": "https://issue23056kv.vault.azure.net/"
}
}'

az cognitiveservices account create \
--name myresource \
--resource-group myResourceGroup \
--kind OpenAI \
--sku S1 \
--location WestEurope \

Answer is correct.

Azure OpenAI Service models can do everything from generating original stories to performing complex text an

https://learn.microsoft.com/en-us/azure/ai-services/openai/encrypt-data-at-rest
https://learn.microsoft.com/en-us/cli/azure/cognitiveservices/account?view=azure-cli-latest#az-cognitiveservices-account-create

tsv or csv only

Maximum file size for assistants and fine-tuning 512 MB
200 MB via the Azure AI Foundry portal.
ref: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/quotas-limits

A correct

200 MB is to big

At the moment, file uploads are limited to 100MB per file, so a 200MB .xlsx file exceeds the upload limit.

File Accepted? Reason
File1.tsv ✅ TSV is supported and file size is 80MB (acceptable).
File2.xml ❌ XML is NOT supported.
File3.pdf ❌ PDF is NOT supported.
File4.xlsx ❌ XLSX is supported format-wise, but the file is 200MB — too large for most uploads (Azure OpenAI max is often 100MB per file).

Option E

Option E
https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/fine-tuning?tabs=azure-openai%2Ccompletionfinetuning%2Cpython-new&pivots=programming-language-studio#openai-cli-data-preparation-tool

This tool accepts files in the following data formats, if they contain a prompt and a completion column/key:

Comma-separated values (CSV)
Tab-separated values (TSV)
Microsoft Excel workbook (XLSX)
JavaScript Object Notation (JSON)
JSON Lines (JSONL)
https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/fine-tuning?tabs=azure-openai%2Ccompletionfinetuning%2Cpython-new&pivots=programming-language-studio#openai-cli-data-preparation-tool

Data must be in JSONL format. The OpenAI CLI data preparation tool helps to convert the data. It's clearly A and E => https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/fine-tuning?tabs=azure-openai%2Ccompletionfinetuning%2Cpython-new&pivots=programming-language-studio#openai-cli-data-preparation-tool

Answer is A. The OpenAI CLI data preparation tool currently supports the following file formats:

JSONL: This is the preferred format for fine-tuning data. Each line in the file should be a valid JSON object representing a training example.
CSV: Comma-separated values files can be used, but they might require some preprocessing to convert them into the correct JSONL format.
TXT: Plain text files can be used, but they will likely need significant preprocessing to extract the relevant information and structure it as JSONL.
Therefore, the answer is A. File1.tsv only.

Answer is E, File 1 and 4, according to this link, file format (whether a CSV, JSON, XLSX, TSV or JSONL) is allowed as long as it's represented as a JSONL training example of a prompt-completion pair.

Given answers are correct

I would have picked Top P and Temperature as well.. but the recommendation is to configure one and leave the other as default value.

answer are correct

was on exam 20.06.24

temperature controls the "creativity" of the generated text, between 0 and 2. A higher temperature will result in more diverse and unexpected responses, while a lower temperature will result in more conservative and predictable responses. The default value for temperature is 1.0, but you can experiment with different values to see what works best for your use case. Higher values means the model will take more risks. Try 0.9 for more creative applications, and 0 (argmax sampling) for ones with a well-defined answer. We generally recommend altering this or top_p but not both.

top_p - An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens with top_p probability mass. 1 is the default value. So 0.1 means only the tokens comprising the top 10% probability mass are considered. We generally recommend altering this or temperature but not both.

1. Temperature
2. Top P

Temperature
Top P

To make the responses more deterministic and less creative, it is recommended to lower the values of Temperature and TopP.

I think this must be user instead of system role. ChatRole.User identifies the text as coming from the user in the conversation. This is important because the AI model will use this information to understand the context of the prompt and tailor its response accordingly.

1. user
2. temperature

it can not be temperature since it is set to -1. Temperature ranges are 0 to 2. It is frequency_penalty o rpresence_penalty

response = openai.ChatCompletion.create(
model="gpt-4",
messages=[
{"role": "system", "content": "bla bla bla"},
{"role": "user", "content": "bla bla bla"},
{"role": "system", "content": "temperature": 0.7}
] )

its user not system, you set the temp of the response you want to recieve when user prompt is sent

Usually you start with a system message and seeying that messages is still empty I would say system. But content being empty makes it hard to really determine. System is optional. So next would be user or assisant. If its a free chat than assistant wont be next, but User.

This is really a terrible question because we can have three options. You can set temp to any of them, it just doesn't make sense with system. Because messages is an array you can change the temp constantly. It makes the most sense with user i guess. But again, a terrible code example to be honest.

Assistant and Temperature as per ChatGPT, I think that's right

system message is optional and the model’s behavior without a system message is likely to be similar to using a generic message such as "You are a helpful assistant." So, It should be system.
second one is temprecture

In the role dropdown, you would select "user" because you are simulating a user's input to which the chatbot should respond. The API call will use this context to generate a response as if it were the assistant. You only need to specify the role for the input messages you're including in your API call, not the responses you're expecting to generate.

Microsoft Copilot seems to be clear on this one. Based on that, answer is correct

It's response about this question (I did not ask about temperature because that one is obvious)

o ensure that your chatbot’s responses are more creative and less deterministic, you should use the Chat.Role.Assistant. This role represents the AI language model (such as GPT-3.5 Turbo) and allows for imaginative and varied answers. By assigning messages to the assistant role, you encourage the model to generate creative and less predictable content.

This one is SYSTEM here and not user. Looking at the code
Typically, a conversation is formatted with a system message first, followed by alternating user and assistant messages.
https://platform.openai.com/docs/guides/text-generation/chat-completions-api

n this context, when configuring the chatbot's behavior, you would use the "system" role to provide initial settings or instructions that affect the overall behavior of the chatbot. This is not an actual message that's part of the conversation with the user but rather a directive to the AI on how to conduct itself in the conversation. For example, you might use this to set the chatbot's personality, instruct it to prioritize certain types of information, or follow specific conversational guidelines.

The "assistant" role, on the other hand, is used for messages that simulate responses from the AI as part of the conversation with the user. It represents the chatbot's side of the dialogue.

Since you want to ensure that the responses are more creative and less deterministic, and this is a setting affecting the AI's behavior, the "system" role is the correct choice. You might include instructions in the "system" message that tell the AI to be more creative or to use less strict adherence to certain conversational rules.

Should be system / temp
Without a system prompt, the reponses are more creative

Now called Prompt shields for user Prompts
https://learn.microsoft.com/en-us/azure/ai-services/content-safety/concepts/jailbreak-detection#prompt-shields-for-user-prompts

Update to prompt shield

first one is chatCompletionsOptions()
second one is AzureChatExtensionConfiguration - see link here https://learn.microsoft.com/en-us/java/api/com.azure.ai.openai.models.azurechatextensionconfiguration?view=azure-java-preview. This details AzureChatExtensionConfiguration & ChatCompletionsOptions() the other methods just dont exist on the SDK

answer is correct

This was based on the beta version. Now with 2.1.0, the second selection would be completely different:
https://azuresdkdocs.blob.core.windows.net/$web/dotnet/Azure.AI.OpenAI/2.1.0/index.html#use-your-own-data-with-azure-openai

options.AddDataSource(new AzureSearchChatDataSource()
{
Endpoint = new Uri("https://your-search-resource.search.windows.net"),
IndexName = "contoso-products-index",
Authentication = DataSourceAuthentication.FromApiKey(
Environment.GetEnvironmentVariable("OYD_SEARCH_KEY")),
});

Seems correct answer

No, using the Protected material detection feature from Content Safety Studio does not meet the requirement. The Protected material detection feature is designed to identify and manage sensitive or protected material, but it is not specifically intended for optimizing content filter configurations for objectionable content in a chatbot.

To optimize the content filter configurations for objectionable content, you should use the Moderate text content feature in Content Safety Studio. This feature is specifically designed to help you run tests on sample questions and check for objectionable content, ensuring that your chatbot's input and output adhere to safety and quality standards.

https://learn.microsoft.com/en-us/azure/ai-services/content-safety/concepts/protected-material?tabs=text#user-scenarios

But if using the true definition of objectionable (unpleasant or offensive) then the choice is No as this would be the text analyze api. Protected material is meant to block copyrighted material from being used in responses

https://learn.microsoft.com/en-us/azure/ai-services/content-safety/overview#product-features

B is Correct , here's what chatGPT says:

The answer is B. No.

Here’s why:

The Protected material detection feature in Content Safety Studio is typically designed to detect and monitor sensitive or protected content, but it may not directly apply to optimizing the filter configurations for a chatbot. To test content filtering settings specifically for objectionable or harmful content in AI-generated responses, you would likely use Content Safety’s standard filtering features directly on the input and output of the chatbot, rather than the Protected material detection feature.

For optimizing content filters, you would test using sample questions and responses in Content Safety Studio’s primary filtering and moderation tools, where you can review and adjust configurations.

The answer should be Yes:

The solution involves using the "Protected material detection" feature from Content Safety Studio to optimize content filter configurations by running tests on sample questions. This approach meets the requirement for testing and optimizing content safety configurations for generative AI output, ensuring that objectionable content is properly detected and managed.

Using the Moderate text content feature from Content Safety Studio to run tests on sample questions does meet the requirement. This approach allows you to analyze and optimize the content filter configurations effectively by ensuring that the input and output are checked for objectionable content. This helps in maintaining the quality and safety of the responses generated by the chatbot.

The answer is A. Yes.

Explanation:

The Moderate text content feature in Content Safety Studio is designed to detect objectionable or harmful content in text. By using this feature, you can effectively test and optimize content filtering configurations by running sample questions and analyzing the responses for inappropriate content. This approach meets the requirement of optimizing the content filter configurations for your chatbot's input and output.

Answer is correct to https://learn.microsoft.com/en-us/azure/ai-services/content-safety/how-to/use-blocklist?tabs=windows%2Crest Blocklist is used in the http body as blocklistnames

text:analyze - > https://learn.microsoft.com/en-us/azure/ai-services/content-safety/quickstart-text?tabs=visual-studio%2Cwindows&pivots=programming-language-rest#analyze-text-content

I think it's correct, contentsafety and then text:analyze

2nd shloud be text/blocklists : https://learn.microsoft.com/en-us/azure/ai-services/content-safety/how-to/use-blocklist?tabs=windows%2Crest

Answer is ContentSafetyClient and AnalyzeTextOptions : https://learn.microsoft.com/en-us/azure/ai-services/content-safety/how-to/use-blocklist?tabs=windows%2Crest

Here are two ways to improve the quality of the responses from the chatbot with minimal development effort:

1. Provide grounding content: This involves feeding the chatbot with relevant domain-specific information and data. This can include documents, articles, FAQs, or any other content related to the chatbot's purpose. By providing this context, the chatbot can better understand the user's intent and respond in a more relevant and informative way.

2. Add sample request/response pairs: This technique involves providing the chatbot with a set of pre-defined questions and their corresponding answers. This helps the chatbot learn the conversation patterns and phrasing related to its specific domain. By analyzing these examples, the chatbot can improve its ability to generate natural and consistent responses to user queries.

Both options (A. Provide grounding content and C. Add sample request/response pairs) achieve the goal of improving response quality with minimal development effort, as they do not require extensive retraining or model building.

Therefore, the two correct answers are:

B. Provide grounding content.
C. Add sample request/response pairs.

Remember that while fine-tuning (A) and custom models (E) can yield high-quality results, they often require significant development effort and computational resources. In contrast, grounding content and sample pairs offer pragmatic improvements with minimal overhead

B and C because it is NOT talking about accuracy, instead is asking to improve the quality of the response

Least development effort is key here

Selected Answer: BC

All 4 answers work, while Training a custom LLM is the highest effort, so that probably isn't it.

I say this answer is B and C.

BC is correct answer.

B and C for me too

B and C are right answer.

Must be B & C

GPT 4 Answered:

B. Provide grounding content.
C. Add sample request/response pairs.

To handle a custom dictionary of offensive terms in Azure AI Content Safety, the blocklist feature is designed for exactly that scenario. You can supply your own list of obscure or custom offensive terms as a blocklist, and Content Safety will flag or block content containing those terms.

Blocklist is what you would use: https://learn.microsoft.com/en-us/azure/ai-services/content-safety/how-to/use-blocklist?tabs=windows%2Crest

https://learn.microsoft.com/en-us/azure/ai-services/content-safety/overview#content-safety-studio-features

The answer is B. No.

Explanation:

The Monitor online activity feature in Content Safety Studio is typically used for tracking and analyzing real-time online activities, which is not directly related to testing or optimizing content filtering configurations on sample questions. To optimize and test content filters for a chatbot, you would use features like Moderate text content that are specifically designed for detecting objectionable content in text inputs and outputs. Therefore, this solution does not meet the requirement.

Answer is ContentSafetyClient and client.AnalyzeImage(request);

Azure Document Intelligence for second, imho

The given Answer is correct. Try it in Chat Playground => N,Y,Y

chat GPT : N,Y,Y
MS copilot : Y,Y,Y
Rider copilot : YYY

Y,Y,Y democracy wins.

I believe this is Y,Y,N

1 - API key
2 - deployment name

see here: https://learn.microsoft.com/en-us/javascript/api/@azure/openai/openaiclient?view=azure-node-preview#@azure-openai-openaiclient-getembeddings

1. 1111a111a11a111aaa11a1a1a11a11aa
2. embeddings1
no doubts

1-APi key
2-deploymentname
The name of the model deployment (when using Azure OpenAI) or model name (when using non-Azure OpenAI) to use for this request.

1 - API key
2 - text-embeddings-ada-002

key word is "pair" , you know that's a key value format so JSONL

https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/fine-tuning-functions

The correct answer is:

C. JSONL

Reasoning:
To fine-tune an Azure OpenAI model, the training data needs to be in JSONL (JSON Lines) format. Each line in the file represents a separate prompt-completion pair in JSON format. This format is required to ensure that Azure OpenAI correctly interprets each pair as an individual training instance.

JSONL.

A, verified with chatGPT

The correct answer is:

A. Cognitive Services OpenAI Contributor

Reasoning: The Cognitive Services OpenAI Contributor role provides the necessary permissions to manage the Azure OpenAI resource, including uploading datasets and fine-tuning models. This role is specific to OpenAI and offers the least privilege required for User1 to complete the tasks without granting additional permissions that are unnecessary for their role.

A is correct. Cognitive Services OpenAI Contributor permission: Upload datasets for fine-tuning.
Ref：https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/role-based-access-control#cognitive-services-openai-contributor

Answer is A, models that use mathematical-based searches are always related to embeddings.

ChatGPT answer:
The correct answer is:

A. text-embedding-ada-002

Reasoning: To identify files that contain specific phrases using cosine similarity, you should use an embedding model. The text-embedding-ada-002 model is specifically designed to create embeddings (numerical representations of text) that can be used for similarity searches, such as cosine similarity. This model is efficient and optimized for tasks that involve finding similarity between pieces of text, making it ideal for identifying files with specific phrases.

The other options (GPT-4, GPT-3.5 Turbo, and GPT-4-32k) are primarily designed for text generation and do not specialize in embedding-based similarity tasks.

A. Cognitive Services OpenAI User

Reasoning: The Cognitive Services OpenAI User role provides read-only access to the Azure OpenAI resource, allowing User1 to:

Identify resource endpoints
View available models for deployment
Generate text and images using deployed models
This role follows the principle of least privilege by granting only the permissions needed to view and interact with deployed models, without allowing management or configuration changes. The other roles, such as Cognitive Services OpenAI Contributor and Cognitive Services Contributor, grant additional permissions that are unnecessary for User1’s requirements.

A is correct. Cognitive Services OpenAI User permission: 1.View the resource endpoint under “Keys and Endpoint”. 2.View what models are available for deployment in Azure OpenAI Studio. 3. Use playground experiences with any models that have already been deployed to this Azure OpenAI resource
Ref：https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/role-based-access-control#cognitive-services-openai-contributor

B is correct, chatGPT verified:

The correct answer is:

B. Azure AI Content Safety

Reasoning: To ensure that your chatbot checks all input and output for objectionable content, you should create an Azure AI Content Safety resource. This service is specifically designed to evaluate text and detect potentially harmful or objectionable content, making it ideal for your chatbot's needs. It can help ensure that the responses generated by AI1 are appropriate and meet safety standards.

The other options do not directly address the requirement of filtering objectionable content in AI-generated responses.

My answer is that
No
No
No

No
No
No

1. No: The total tokens used for the session include both the prompt tokens (37) and the completion tokens (86), totaling 123 tokens. Therefore, the subscription will be charged for 123 tokens, not just 86.
2. Yes: The Max response tokens were set to 100, and the completion used 86 tokens. The text completion was not truncated because the response did not exceed the maximum allowed tokens.
3. No: The prompt_tokens are not included in the Max response tokens value. The Max response tokens only refer to the tokens used in the model's response, not the tokens used in the input prompt.

answer is correct.
3rd should be no because "Token costs are for both input and output. For example, suppose you have a 1,000 token JavaScript code sample that you ask an Azure OpenAI model to convert to Python. You would be charged approximately 1,000 tokens for the initial input request sent, and 1,000 more tokens for the output that is received in response for a total of 2,000 tokens." source below
https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/manage-costs#understand-the-azure-openai-full-billing-model

copilot says

The subscription will be charged 86 tokens for the execution of the session: Yes

The text completion was truncated because the Max response tokens value was exceeded: No

The prompt_tokens value will be included in the calculation of the Max response tokens value: Yes

N/N/N
A - It takes into account the prompt tokes - hence as per the calculation -123 tokens should be charged
B - Text completion was not truncated, because the response token is 86 < 100
C - Prompt_tokens is not included in calculation max_response value.

N - the subscription is NOT charged for 86 tokens - the response does not contain 86 tokens. For reference, each token is roughly four characters for typical English text.
N - text completion is clearly under 86 and the sentence is not truncated. the finish_reason here is "stop". If the prompt had been cut off, the finish_reason would have been 'length'
N - max tokens is the maximum number to generate in the COMPLETION.
The token count of your prompt plus max_tokens can't exceed the model's context length. Most models have a context length of 2048 tokens (except for the newest models, which support 4096).

N/N/N
Subscriptions are charged for both the prompt and completion tokens.
Completion tokens is less than max response tokens.
It refers to max response tokens only, not max tokens.
https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/manage-costs#understand-the-azure-openai-full-billing-model

The subscription will be charged 86 tokens for the execution of the session. Yes, that’s correct. The completion_tokens value represents the number of tokens in the model’s response, and this is what you’re billed for.
The text completion was truncated because the Max response tokens value was exceeded. No, that’s not correct. The response in this case wasn’t truncated. The max_tokens parameter sets a limit on the length of the generated response. If the model’s response had exceeded this limit, it would have been cut off, but in this case, the response is only 86 tokens long, which is less than the max_tokens value of 100.
The prompt_tokens value will be included in the calculation of the max_tokens value. Yes, that’s correct. The max_tokens parameter includes both the prompt tokens and the completion tokens. So if your prompt is very long, it could limit the length of the model’s response.

The session execution consumed 86 tokens is NO it should be total of 123 tokens which includes the prompt tokens
The text completion was truncated due to exceeding the Max response tokens value is Yes
The prompt_tokens value is included in the calculation of the Max response tokens value is YES

I think it should be N/N/N.

1. Create deployment using GPT-3.5-Turbo
2. Apply Market Writing Assistant template
3. Deploy solution to a new web app

To deploy an Azure OpenAI model for the app "Appl" with minimized development effort, follow these steps in Azure OpenAI Studio:

Create a deployment that uses the GPT-35 Turbo model.

Drag "Create a deployment that uses the GPT-35 Turbo model" to the answer area.

Apply the Default system message template.

Drag "Apply the Default system message template" to the answer area.

Deploy the solution to a new web app.

Drag "Deploy the solution to a new web app" to the answer area.

"Deploy the solution to a new web app" << That does not mean anything in this context. Pure nonsense. But I guess that is what I have to select considering this is MIcrosoft exam

"Deploy solution to a new web app" << What does that mean??
But because nothing else can be selected I have to select this.

This question was in the test I took in August 2024

1. Create deployment using GPT-3.5-Turbo
2. Apply Market Writing Assistant template
3. Deploy solution to a new web app

In the Setup panel, under Use a system message template, select the Marketing Writing Assistant template and confirm that you want to update the system message.
https://microsoftlearning.github.io/mslearn-openai/Instructions/Exercises/01-get-started-azure-openai.html

embeddings

c for sure

embeddings correct

Copilot says Azure OpenAI Service embeddings.

correct

Answer is correct:
Azure OpenAI allows you to manage how frequently your application can make inferencing requests. Your rate limits are based on Tokens-per-Minute (TPM). For example, if you have a capacity of 1, this equals 1,000 TPM, and the rate limit of requests you can make per minute (RPM) is calculated using a ratio. For every 1,000 TPM, you can make 6 RPM.

So, if you need to process 600 requests every minute, you'll require a TPM that supports that many RPM. Using the ratio, for 600 RPM, you need 100,000 TPM (because 600 divided by 6 equals 100, and 100 multiplied by 1,000 equals 100,000). In this scenario, you would set the capacity to 100, since each capacity unit equals 1,000 TPM.

Answer is correct
From here https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/quota?tabs=rest
When a deployment is created, the assigned TPM will directly map to the tokens-per-minute rate limit enforced on its inferencing requests. A Requests-Per-Minute (RPM) rate limit will also be enforced whose value is set proportionally to the TPM assignment using the following ratio:

6 RPM per 1000 TPM.

From here:
https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/quota?tabs=rest
capacity integer This represents the amount of quota you are assigning to this deployment. A value of 1 equals 1,000 Tokens per Minute (TPM). A value of 10 equals 10k Tokens per Minute (TPM).

So the math is 600 RPM needs 100000 TPM which translates to capacity 100

1. capacity
2. 100

Parameter Type Description
sku Sku The resource model definition representing SKU.
capacity integer This represents the amount of quota you are assigning to this deployment. A value of 1 equals 1,000 Tokens per Minute (TPM). A value of 10 equals 10k Tokens per Minute (TPM).

capacity and 100.

capacity and 100.
From Takedajuku perspective, if you study for 4 days and spend 2 days reviewing, you will have a better chance of passing the exam.

on exam today 24/05/2024, I selected capacity and 600

The accurate capacity value to meet the requirement of processing 600 requests per minute is indeed 100. I apologize for any confusion caused earlier

In the provided options, the correct choice is:

"capacity" : 600

The capacity parameter in the SKU configuration of an Azure OpenAI resource specifies the maximum number of requests per minute (RPM) that the resource can handle. By setting "capacity": 600, you are configuring the resource to handle up to 600 requests per minute.

1. GetCompletions - this is the api for creating chat.
2. (response.Value.Choises[0].Text) - this will output the text from top rated response object.

1. GetCompletions
2. (response.Value.Choises[0].Text);

Agree answer is correct

Answer is correct.