to_add_somewhere_3

Question 1

To add somewhere 3

Does customizing a FM with more training data increase latency of inference?

Answer 1

No, you’re adjusting weights, not making the model substantically slower

Question 2

To add somewhere 3

Can you use F1 to score text-generation model?

Answer 2

No: F1 evals binary outputs (false-positives, recall, precision, …). It doesn’t deal with geneated text.

Question 3

To add somewhere 3

Amazon SageMaker Autopilot?

Answer 3

Automates the process of building machine learning models, not involved with monitoring or operating them

Question 4

To add somewhere 3

Diff between kNN and K-Means?

Answer 4

k-NN is algorithm for classifying new data. K-means is unsupervised, used to cluster data

Question 5

To add somewhere 3

Sampling Bias?

Answer 5

Training data doesn’t represent the entire population

Question 6

To add somewhere 3

Is reporting bias a problem of sampling data?

Answer 6

Not really, it’s about overrepresentation in training data leading to same things being reported from this group

Question 7

To add somewhere 3

Automation bias?

Answer 7

Tendency of people to over-rely on automated systems and trust their decisions

Question 8

To add somewhere 3

Implicit bias?

Answer 8

Unconscious attitudes or stereotypes that affect human decision-making

Question 9

To add somewhere 3

Model Fairness?

Answer 9

Answers are unbiased

Question 10

To add somewhere 3

Model explainability?

Answer 10

Can explain how the model made decisions

Question 11

To add somewhere 3

Model transparency?

Answer 11

All stakeholders understand the training data, model development, etc.

Question 12

To add somewhere 3

HealthScribe?

Answer 12

Transcribing and summarizing patient-clinician conversations

Question 13

To add somewhere 3

Comprehend Medical?

Answer 13

Extracts health data from unstructured clinical text

Question 14

To add somewhere 3

If you see a question about foundation models, is SageMaker the answer?

Answer 14

No :-)

Question 15

To add somewhere 3

Self-refine prompting?

Answer 15

A model iteratively solves a problem, critiques its own solution, and then revises the solution based on the critique

Question 16

To add somewhere 3

PartyRock?

Answer 16

No-code generative AI app-building playground powered by Amazon Bedrock

Question 17

To add somewhere 3

What is AWS Audit Manager?

Answer 17

Service designed to simplify compliance audits in cloud environments

Question 18

To add somewhere 3

What does AWS Audit Manager do?

Answer 18

Continuously collects evidence from AWS resources and maps configurations to industry standards and regulatory frameworks such as HIPAA, SOC 2, PCI

Question 19

To add somewhere 3

What is Data Exchange?

Answer 19

Subscribe to data providers, get notified when new datasets released

Question 20

To add somewhere 3

Tailor LLM responses to match company’s tone and style?

Answer 20

Prompt engineering (just this simple)

Question 21

To add somewhere 3

Random Cut Forest (RCF)?

Answer 21

Anomaly detection, which aims to identify rare or unusual data points in a dataset

Question 22

To add somewhere 3

k-NN or K-means if you have unlabeled data?

Answer 22

K-means: kNN needs labels, K-means just keep comparing distances

Question 23

To add somewhere 3

What is Principal Component Analysis (PCA) used for?

Answer 23

Dimension reduction

Question 24

To add somewhere 3

What provides pre-built solutions and foundation models to accelerate AI and machine learning projects?

Answer 24

SageMaker JumpStart

Question 25

# *To add somewhere 3* What AWS service continuously monitors many resources against HIPAA, SOC, etc.?

Answer 25

Audit Manager

Question 26

# *To add somewhere 3* How do you capture all inputs and outputs from a Bedrock model?

Answer 26

Bedrock model invocation logging (separate feature)

Question 27

# *To add somewhere 3* Domain adaptation fine-tuning vs. Transfer Learning?

Answer 27

DAFT is a specific type of TL

Question 28

# *To add somewhere 3* What is DAFT often best at?

Answer 28

Dealing with your jargon or your company's product descriptions