CAIC 7 Flashcards
What should you always keep in mind when using ChatGPT?
The limitations mentioned in previous chapters, as ChatGPT may provide partial or incorrect information.
It is a good practice to double-check the information provided.
What type of questions should be avoided when using ChatGPT?
Vague, open-ended questions.
Examples include ‘What can you tell me about the world?’ or ‘Can you help me with my exam?’
What is a best practice when expecting a specific output structure from ChatGPT?
Specify that structure in your prompt.
What is the knowledge base limit of ChatGPT?
Limited to 2021.
What is the purpose of the Moderator API in ChatGPT?
To prevent engagement in unsafe conversations.
What are the classes used by the Moderator API to classify content?
- Violence
- Self-harm
- Hate
- Harassment
- Sex
What is the hidden bias in GPT-3’s training data attributed to?
Mainly written by white males from Western countries.
What did the study by OpenAI researchers reveal about racial bias in GPT-3?
Sentiment associated with racial categories varied across different models.
What does the concept of responsible AI encompass?
Bias and ethics within AI models.
What is the historical evolution of machine learning (ML)?
From checker game-playing programs in the 1950s to advanced AI like ChatGPT.
What significant change has occurred in the technology infrastructure for ML?
Evolved from single machine/server to complex end-to-end ML platforms.
What new professional roles have emerged due to hyper-growth in AI/ML?
- ML Engineers
- Data Scientists
- AI Ethics Researchers
- Data Analysts
- AI Product Managers
What is the role of an ML solutions architect?
To support end-to-end ML initiatives.
What is the first stage in the ML lifecycle?
Business understanding.
What must be defined to measure the success of an ML project?
Business goals and business metrics.
What is a common business goal for ML projects?
Cost reduction for operational processes.
What does the saying ‘data is the new oil’ imply in the context of ML?
The necessity of having the required data to move forward with an ML project.
What is involved in the data acquisition and understanding stage of the ML lifecycle?
Gathering and comprehending available data.
What is feature engineering?
The process of using domain knowledge to extract useful features from raw data.
What must be validated before deploying a model into production?
Model quality using relevant technical metrics.
What is a validation dataset also known as?
Test dataset.
Why is model accuracy not always a suitable validation metric?
It may not reflect performance well in cases like fraud detection where the number of frauds is small.
What type of project structure is typical in an ML project?
- Business understanding
- Data acquisition and understanding
- Data preparation
- Model building
- Model evaluation
- Model deployment
What was the author’s previous experience before working in AI/ML?
Building computer software platforms for large financial services institutions.
What does the iterative process in ML involve?
Numerous runs of data processing and model development to find optimal performance.
What challenges did the author face in deploying the model?
Integrating it into the existing business workflow and system architecture.
What is essential to ensure before proceeding with an ML project?
Sufficient justification and measurable outcomes.
What is the purpose of model validation in machine learning?
To gauge how the model performs on unseen data
Model validation ensures that the model generalizes well beyond the training dataset.
What factors determine the appropriate metrics for model validation?
ML problems and the dataset used
Different problems and datasets require different evaluation metrics.
Why would model accuracy not be a good metric for evaluating fraud detection models?
The number of frauds is small, resulting in potentially high accuracy despite poor performance
A model predicting not-fraud all the time could still achieve high accuracy.
What are the two main deployment concepts in machine learning?
Deployment of the model for client applications and integration into business workflow applications
These concepts ensure that the model’s predictions are utilized effectively.
How can a credit fraud model be deployed?
Hosted behind an API for real-time prediction or as a package for batch predictions
This allows flexibility in how predictions are generated and used.
What is a key post-deployment step in the ML lifecycle?
Model monitoring
Monitoring is crucial for detecting performance degradation and changes in data distribution.
What is model drift?
Model performance degradation due to changes in production data characteristics
This phenomenon can significantly impact the effectiveness of deployed models.
What should be tracked to measure the actual business impact of a deployed model?
Business metrics before and after model deployment
This helps in assessing the model’s effectiveness and overall impact.
What is A/B testing in the context of model evaluation?
Comparing business metrics between workflows with and without the ML model
A/B testing helps determine the model’s contribution to business outcomes.
What should be done if a deployed model does not deliver expected benefits?
Re-evaluate the model for improvement opportunities or consider framing the problem differently
This may involve exploring alternative ML approaches to address the business problem.
Fill in the blank: The ML lifecycle does not end with _______.
[model deployment]
True or False: Software behavior is highly deterministic while ML models can behave differently in production.
True
This difference arises because ML models learn from data rather than being explicitly coded.
