Test 01 Flashcards

1
Q

What is Localization in the context of chatbots?

A

Localization refers to how well a chatbot ‘speaks’ a specific language or dialect and the relevance of the resources it provides to a user’s locale.

It includes spelling, grammar, and the use of culturally appropriate words.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

Why are chatbots sometimes unresponsive to prompts?

A

Chatbots make errors and are still learning; user interactions and feedback help them improve over time.

Feedback is crucial for refining chatbot responses.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

Do chatbots have access to the internet?

A

No, chatbots do not have access to the internet and are not up to date with current information.

Avoid asking questions that require recent data or including links.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

What should you do if a chatbot claims to perform impossible tasks?

A

You should penalize the chatbot for Truthfulness if it lies about its capabilities.

This includes pretending to have opinions on controversial topics.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

What does a well-localized chatbot response look like?

A

It uses correct spelling, appropriate words, and native-like grammar for the specific dialect.

For example, American English should not use British spellings.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

When providing a content text, does it change the type of Q&A prompt?

A

No, providing a content text does not make an Open Q&A prompt Closed Q&A.

The distinction is based on whether the answer can be found in the content.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

What is a system prompt?

A

A system prompt is an instruction given to a chatbot to guide its responses, ideally used around 10% of the time.

It should be provided before initiating conversation.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

Do prompts need to be in the same language for a single conversation?

A

Yes, all prompts should be in the same language as the project.

Consistency in language helps maintain clarity.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

What should you do if a chatbot responds in the wrong language?

A

Select Major Issues for Instruction Following and evaluate the responses accordingly.

Do not rate a wrong-language response higher than a correct one.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

True or False: Chatbots can remember prompts from previous turns.

A

True.

Chatbots should maintain context across turns.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

Fill in the blank: A poorly rendered table in a chatbot’s response should _______.

A

not be penalized.

This is a known issue with the website, not the chatbot.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

What should you do if your prompt for the first turn doesn’t fit the category?

A

Rewrite it or skip to a different category.

Ensure prompts align with the assigned categories.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

What happens if both responses from different chatbots are in the wrong language?

A

Select Major Issues for Instruction Following for both responses.

You can select either as the better response based on normal criteria.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

Why is it important to rate the chatbot’s responses for Truthfulness?

A

Rating for Truthfulness ensures the chatbot provides accurate information and does not misrepresent its capabilities.

This is crucial for maintaining user trust.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

What is the significance of providing resources relevant to the user’s locale?

A

Resources should be useful to speakers of the dialect from the user’s country to ensure effective localization.

Irrelevant resources can confuse users.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

What is the recommended approach if a chatbot ignores a system prompt?

A

Select Minor Issues for Instruction Following if this is the only instance of disobedience.

Multiple issues would warrant a Major Issues rating.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
17
Q

How should you categorize a prompt if unsure of its type?

A

Use your best judgment and categorize it as Other if it doesn’t fit any provided categories.

Ensure it really does not fit before categorizing as Other.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
18
Q

What is a good method for helping the chatbot improve?

A

Teaching it your dialect by correcting its response

19
Q

What should you do if the chatbot uses words from another language?

A

Use your best judgment to determine if it’s a minor or major issue

20
Q

Should all prompts be in the same language?

A

Yes

21
Q

Should all of the chatbot’s responses be in the same language?

A

Yes

22
Q

What should you do if the chatbot’s response is in the wrong language?

A

Edit it if concise and easy to translate

23
Q

Should you edit the response if the chatbot replied in the wrong language?

A

Yes, if you see a reason to

24
Q

What should you do if you edit a response and want to add formatting?

A

Use Markdown for formatting

25
Q

How should you approach the Overall Rating?

A

Holistic evaluation reflecting all axes

26
Q

What do you do if Chatbot A and Chatbot B give identical responses?

A

Use an Unsure rating or edit one to improve it

27
Q

How should you evaluate Localization if the chatbot didn’t respond in your language?

A

Rate according to the response produced

28
Q

How should you evaluate Verbosity?

A

Responses should contain necessary information without excess

29
Q

How do you evaluate canned responses for Truthfulness and Instruction Following?

A

Canned responses should receive Cannot Assess for Truthfulness

30
Q

How should you rate the overall quality of canned responses?

A

Better if they redirect the conversation

31
Q

What should you do if the chatbot gives a canned response that contains incorrect information?

A

Rate it down for Truthfulness

32
Q

What if the chatbot gives a canned response in the wrong language?

A

Select Major Issues for Instruction Following

33
Q

Do you need to fact-check links provided by the chatbot?

A

Yes, ensure they are valid

34
Q

What should you prioritize when fact-checking links?

A

Accuracy over speed

35
Q

What should you do if both responses are really bad?

A

Rate them as appropriate and compare

36
Q

How should you rate Instruction Following for Harmful prompts?

A

Responses should be rated N/A for Instruction Following

37
Q

What is the ideal behavior for models responding to Benign prompts?

A

Acknowledge sensitive content

38
Q

What if the bot refuses to answer a harmful/benign prompt in the wrong language?

A

Select Major Issues for Instruction Following

39
Q

What should you do if the wrong-language response is harmless but the right-language response is harmful?

A

Prefer the wrong-language response for safety

40
Q

What does it mean for a prompt to be written in a specific language or locale?

A

It should reflect regional slang, culture, and references

41
Q

What should you do if the response includes resources that aren’t relevant to your locale?

A

Penalize the chatbot on Localization

42
Q

What type of characters should be used for Mandarin (Mainland) projects?

A

Simplified characters

43
Q

What if the chatbot’s response uses the wrong type of character for Mandarin?

A

Treat it as a major issue for Localization