2. 12. 2024 15:52 Quiz [IBM watsonx.
ai Generative AI Tools Level 2]: Attempt review
Quiz [IBM watsonx.ai Generative
AI Tools Level 2] Back Next
You must receive a score of 75% or higher on the quiz to complete
the course.
Started on Monday, December 2, 2024, 7:49 AM
State Finished
Completed on Monday, December 2, 2024, 9:51 AM
Time taken 2 hours 1 min
Feedback Congratulations, you passed the quiz!
Question 1
Correct
Points out of 1.00
IBM watsonx.ai provides many sample prompts for different
use cases for clients to explore. What are some use cases for
which examples are provided?
Translation, Summarization, Image generation, and
Classification
Code, Translation, Question Answering, and Graphing
Generation, Extraction, Question Answering, and Code
Classification, Extraction, Summarization, and Audio
extraction
https://learn.ibm.com/mod/quiz/review.php?attempt=3571710&cmid=309104 1/14
2. 12. 2024 15:52 Quiz [IBM watsonx.ai Generative AI Tools Level 2]: Attempt review
Question 2
Correct
Points out of 1.00
Back Next
Privacy is a huge concern for a client you are working with. They
are concerned that even providing a small set of masked data
will inadvertently disclose confidential data. How can you
assure them?
Since the data is synthetically generated - they are all fake
so there is no need to worry about disclosing anything.
Clients should choose the largest foundation models
available for synthetic data generation as that will
introduce more "creativity" into the data.
They should just never use real data as seed data. They
should just manually create a schema and let watsonx.ai
generate synthetic data.
No input data will be included in the output. Watsonx.ai
can apply differential privacy to further minimize any
privacy risks.
Question 3
Correct
Points out of 1.00
InstructLab is a new component in the IBM generative AI
toolbox. It trains models using taxonomy-guided synthetic data
generation. What is this technology based on?
The latest research on Linguistic Alignment and Boundary
conditions which is an iterative method of evaluating
model performance
A Red Hat methodology called Linked model Algorithm for
ChatBots (LAB)
A new methodology innovated by IBM Research called
Large scale Alignment for ChatBots (LAB).
The rich experience and lessons learned from IBM's own
implementation in its Advanced Incubation Lab.
https://learn.ibm.com/mod/quiz/review.php?attempt=3571710&cmid=309104 2/14
2. 12. 2024 15:52 Quiz [IBM watsonx.ai Generative AI Tools Level 2]: Attempt review
Question 4
Correct
Points out of 1.00
Back Next
A client wants to exploit generative AI for their new
applications. They feel the only thing they need to think about
when selecting their new AI platform is the foundation models
it supports. You suggest to them that a full generative AI stack
really consists of which of the following set of components?
Prompt engineering tools, shot prompt capabilities,
inference parameters selection, and model options.
Infrastructure, a user-friendly User Interface (UI), an
Integrated Development Environment (IDE) for fine-tuning,
and model documentation.
Infrastructure, foundation models, support for C++ and
Cobol connectors.
A generative AI platform, AI assistants, infrastructure,
and foundation models.
Question 5
Correct
Points out of 1.00
Generative AI can be very talkative, and at times it can repeat
the same output text over and over again. What should a client
do to mitigate this?
Clients can set a Max tokens value to something low
Clients can set a high value for the Repetition Penalty
parameter
Clients must use the Greedy mode where there will not be
any repetition
Clients can set a high value for the Temperature parameter
https://learn.ibm.com/mod/quiz/review.php?attempt=3571710&cmid=309104 3/14
2. 12. 2024 15:52 Quiz [IBM watsonx.ai Generative AI Tools Level 2]: Attempt review
Question 6
Correct
Points out of 1.00
Back Next
A client is evaluating generative AI and is not sure how much
they will need to use it initially. They want to know the
difference between the Standard watsonx.ai plan vs. the
Essential plan. What is one key difference between the two?
The Standard plan has all the watsonx.ai models available
but the Essential plan only has a subset of the models.
The Standard plan includes all the features and tooling,
whereas the Essential plan does not include watsonx.ai's
Synthetic Data Generator.
The two plans have the same capabilities but the
Essential plan is a pay-as-you-go plan.
The Essential plan is a free tier for clients to try out
watsonx.ai. It has a limit of 20 CUH/month whereas the
Standard is a pay-as-you-go plan.
https://learn.ibm.com/mod/quiz/review.php?attempt=3571710&cmid=309104 4/14
2. 12. 2024 15:52 Quiz [IBM watsonx.ai Generative AI Tools Level 2]: Attempt review
Question 7
Correct
Points out of 1.00
Back Next
A client has a use case that requires a lot of creativity. They
want to know what watsonx.ai has to offer. They do not mind
some hallucinations so long as the model isn't repeating words
(like wasting tokens or producing output with the same word 10
times in a row). What capabilities within watsonx.ai can help
them?
Clients can set the inference mode to Sampling instead of
Greedy. They can then set the Temperature value high but
set minimal values for Top P and Top K. Watsonx.ai has
optimized the models so they will never repeat words.
Clients can set the inference mode to Greedy instead of
Sampling. Greedy means the model will try to provide as
many variations as possible.
Clients can set the inference mode to Sampling instead
of Greedy. They can then set the Temperature, Top P,
and Top K to high values and also set Repetition
penalty to 2.
Clients will need to choose the larger models. Any models
with less than 20b parameters do not hallucinate as they
are domain-specific.
https://learn.ibm.com/mod/quiz/review.php?attempt=3571710&cmid=309104 5/14
2. 12. 2024 15:52 Quiz [IBM watsonx.ai Generative AI Tools Level 2]: Attempt review
Question 8
Correct
Points out of 1.00
Back Next
Testing the model with a large body of data is a big concern for
your client. However, they are not allowed to use their own data
and there is no public data that resembles what they do. What
would be your recommendation?
Instead of testing with a large amount of data, the client
should just use a bigger model.
There is no better option than using real data so the client
needs to work with their security team, asking the security
team to mask any sensitive data before using it for testing.
Testing is in-house so security should not really be a
concern.
The client should go to the Resource hub in watsonx.ai and
look through the set of sample data to find something that
closely resembles their own data and use that for testing.
Clients can use the Synthetic Data Generator. They can
manually provide a schema or a small set of sample
data to generate a large volume of synthetic data.
https://learn.ibm.com/mod/quiz/review.php?attempt=3571710&cmid=309104 6/14
2. 12. 2024 15:52 Quiz [IBM watsonx.ai Generative AI Tools Level 2]: Attempt review
Question 9
Incorrect
Points out of 1.00
Back Next
In a discussion about handling multiple downstream tasks, the
client raises a concern that there is no time to learn a different
model's characteristics and then tune it for every downstream
task. You want to assure them that it may not be as
complicated. What would you highlight regarding watsonx.ai
that can mitigate this concern?
Clients can provide different sets of labeled data to train
the same foundation model (versus using a different model
as a starting point).
In prompt tuning, a soft prompt is created. All the client
has to do is attach the soft prompt to different models.
There is no need to perform different prompt tuning.
IBM watsonx.ai comes with a variety of models and clients
can simply choose a large model which in general provides
better performances for more tasks than smaller models.
If they are concerned, clients should use different shot
prompts to deal with different downstream tasks.
https://learn.ibm.com/mod/quiz/review.php?attempt=3571710&cmid=309104 7/14
2. 12. 2024 15:52 Quiz [IBM watsonx.ai Generative AI Tools Level 2]: Attempt review
Question 10
Incorrect
Points out of 1.00
Back Next
When you described the Chat interface to a client, they
responded with: "Oh, that's just IBM's version of ChatGPT". You
assure them that IBM provides more. What else is true about
the watsonx.ai Chat mode?
The Chat interface in watsonx.ai has been optimized for all
use cases including code generation, extraction,
classification, multi-shot prompting, as well as Synthetic
Data Generation.
The Chat capability can work with any model including
all models that clients can access from Hugging Face.
ChatGPT cannot do that.
The IBM watsonx.ai Chat mode utilizes advanced models
and will never hallucinate, which is an advantage over
ChatGPT.
Clients can examine the underlying code for the chat in
different programming languages such as Node.js and
Python.
https://learn.ibm.com/mod/quiz/review.php?attempt=3571710&cmid=309104 8/14
2. 12. 2024 15:52 Quiz [IBM watsonx.ai Generative AI Tools Level 2]: Attempt review
Question 11
Incorrect
Points out of 1.00
Back Next
A client has been fine-tuning models for quite some time. While
it does perform well, they feel that the effort of creating the
huge set of labeled data is not practical or scalable. What would
you say to them regarding how InstructLab can help them?
InstructLab has been optimized so it can perform fine-
tuning based on tens of input samples versus the typically
thousands of examples required in fine-tuning.
InstructLab uses a multi-epochs approach where a smaller
set of input data is processed 10 times and provides the
effect of a larger set of data.
InstructLab leverages synthetic data generation to create
many more training data points based on a small amount of
input, so they don't need to generate all the data required
to train a model.
InstructLab provides a Structured interface to guide the
user through entering and storing the data necessary
for fine-tuning.
https://learn.ibm.com/mod/quiz/review.php?attempt=3571710&cmid=309104 9/14
2. 12. 2024 15:52 Quiz [IBM watsonx.ai Generative AI Tools Level 2]: Attempt review
Question 12
Correct
Points out of 1.00
Back Next
You described shot-prompting to the client and they were
impressed with this watsonx.ai capability. They want to know
more. Which of the following is a true statement regarding shot-
prompting in the watsonx.ai Structure mode?
From the watsonx.ai Structured mode, the client can
perform zero-shot, one-shot, two-shot, up to a maximum
of five-shot prompting.
While shot-prompting is powerful, clients should not need
more than 2 shots. They should consider fine-tuning if
more than 2 shots are required.
The Structured mode provides a guided way for clients
to perform shot-prompting using as many shots as they
like.
Shot-prompting is only available in the watsonx.ai Chat
mode.
Question 13
Correct
Points out of 1.00
IBM watsonx.ai provides sample code for any Chat session so
clients can examine and leverage it. In what formats does
watsonx.ai generate the sample code?
Python, R, and Neo.4j graph
Python, Java Script, and cURL
cURL, Python, and C++
Python, Node.js, and cURL.
https://learn.ibm.com/mod/quiz/review.php?attempt=3571710&cmid=309104 10/14
2. 12. 2024 15:52 Quiz [IBM watsonx.ai Generative AI Tools Level 2]: Attempt review
Question 14
Correct
Points out of 1.00
Back Next
InstructLab can be used to provide training for a Large
Language Model. What are the four stages of the InstructLab
workflow?
Taxonomy-based skill and knowledge representation,
Synthetic Data generation, Skill and Knowledge training,
Visual model generation.
Taxonomy-based skill and knowledge representation,
Manual Data generation, Manual Data validation, Skill and
Knowledge training.
Taxonomy-based skill and knowledge representation,
Synthetic Data generation, Synthetic Data validation,
Skill and Knowledge training.
Taxonomy-based visual representation, Synthetic Data
generation, Synthetic Data validation, Skill and Knowledge
training.
https://learn.ibm.com/mod/quiz/review.php?attempt=3571710&cmid=309104 11/14
2. 12. 2024 15:52 Quiz [IBM watsonx.ai Generative AI Tools Level 2]: Attempt review
Question 15
Correct
Points out of 1.00
Back Next
A client heard the terms prompt engineering and prompt tuning
and asked what the differences are. What should you tell them?
There is no difference except that clients access prompt
engineering via the Promt Lab and prompt tuning via the
Tuning Studio.
Prompt engineering involves changing the input, or
providing multiple shots (sample input/output) as part of
the prompt input. Prompt tuning includes everything
prompt engineering does but also includes changing
inference parameters.
Prompt engineering is meant to engineer a new model,
whereas prompt tuning is to tweak the inference
parameters to get optimal performance.
Prompt engineering involves modifying the input
prompt and inference parameters to get optimal
results. Prompt tuning requires training the model with
a set of labeled data to create a tunable soft prompt.
Question 16
Correct
Points out of 1.00
Generative AI is projected to have a large economic impact. A
2023 Goldman Sachs study predicted that it will have a sizable
and noticeable impact on the Global Domestic Product (GDP).
What is Goldman Sachs prediction for the potential impact of
Generative AI on GDP?
A 6% potential drop in GDP within 10 years
A 15% potential rise in GDP within 10 years.
A 15% potential drop in GDP within 10 years
A 7% potential rise in GDP within 10 years
https://learn.ibm.com/mod/quiz/review.php?attempt=3571710&cmid=309104 12/14
2. 12. 2024 15:52 Quiz [IBM watsonx.ai Generative AI Tools Level 2]: Attempt review
Question 17
Incorrect
Points out of 1.00
Back Next
Shot prompting is an important way to help improve prompt
performance. How does watsonx.ai help guide users to perform
shot prompting?
When a user puts in a prompt, watsonx.ai can
automatically generate additional sample prompts and
outputs for shot prompting
IBM watsonx.ai provides a Structured mode where users
are guided through input/output fields so clients can
provide sample prompts and expected completions.
Shot prompting is a simple case of prompt tuning (with
a smaller number of sample input/output pairs).
IBM watsonx.ai does not work with shot prompting yet, but
this capability is available when the InstructLab feature is
available.
Question 18
Correct
Points out of 1.00
A client has partnerships with many other enterprises and they
are interested in jointly developing new open source generative
AI applications for some common use cases. What would you
suggest is a good way for data scientists from different
companies to collaborate on generative AI model
development?
This cannot be done as there are AI regulations and privacy
issues preventing companies from contributing, even for an
open source project.
The client can prompt tune a model, then pass that onto
each of the participating companies for them to
successively prompt tune it further.
The client should collaborate with their partners to create a
set of labeled data and let IBM tune the model for them.
The IBM InstructLab provides an infrastructure for
different Subject Matter Experts (SMEs) to collaborate
and contribute skills and models to tune a model.
https://learn.ibm.com/mod/quiz/review.php?attempt=3571710&cmid=309104 13/14
2. 12. 2024 15:52 Quiz [IBM watsonx.ai Generative AI Tools Level 2]: Attempt review
Question 19
Correct
Points out of 1.00
Back Next
Enterprises have been rapidly adopting generative AI.
According to a Gartner survey in 2023, what is the adoption
rate for organizations in the U.S., Germany, and U.K.?
65%
34%
7%
29%
Question 20
Correct
Points out of 1.00
The InstructLab offers a slightly different way of tuning by using
a taxonomy to learn new things. What kind of learning can
InstructLab do?
InstructLab can learn complex art forms such as abstract
painting and even music composition
InstructLab can learn from other foundation models by
linking to different model files
InstructLab can learn knowledge (information) and
skills (interpretation)
InstructLab can learn new image and audio creation
https://learn.ibm.com/mod/quiz/review.php?attempt=3571710&cmid=309104 14/14
