AML 3304 Lesson Notes

Test Questions Bank:

Week 4 Topics and focus areas:

June 18 Class Lesson Plan
Topics Review Syllabus for the Mid Term Exam (the week after Mid Term Break)

Building the Word Embedding with Google Collab Notebook (Your Assignment)

May 31 Review Topics:
We discussed the potential of AI in various industries, including business, medicine, and marketing. He emphasized the importance of leveraging data analytics and AI-driven segmentation to personalize interactions with customers and provide valuable insights. Sigurdson highlighted examples from companies like Starbucks and Walmart, demonstrating how AI can help unlock unprecedented insights into individual customer preferences. He also discussed the potential of expert-based systems in medicine, emphasizing their ability to curate and aggregate knowledge from thousands of practitioners.

Action Items

Watch instructor's video on unified process design methodology

Read instructor's blog articles on how business IT systems will incorporate AI models

Study the diagram provided on how AI models may integrate into the model-view-controller architecture

Consider emerging opportunities in AI safety engineering field

Build a simplified AI language model as part of the class project/assignment

Add action item
Integrating AI language models in business IT systems.
Peter and an unknown speaker discuss AI language models in business IT systems, with a focus on building an embedding and studying the Starbucks deep brew AI model.
Peter Sigurdson suggests practicing drawing a diagram of a business IT system reengineered around AI language models.
Peter Sigurdson predicts AI large language models will replace deterministic outputs of objects and controllers in IT business applications.
Large language models will be trained on data sets and provide rich and nuanced responses, surpassing current crowdsourced knowledge.
Using AI to analyze complex data cubes for business insights.
Peter Sigurdson explains how data analytics can provide insights into complex data sets by creating dashboards that slice through the data cube.
Peter Sigurdson demonstrates using AI analytic dashboards to uncover important insights in complex data.
Peter Sigurdson discusses the limitations of early AI systems in medical diagnosis.
Peter Sigurdson explains expert-based systems, which use AI to amplify medical practitioners' knowledge.
He discusses how these systems can help practitioners make more informed decisions in real-time.
AI application development and its connection to societal and business concerns.
Peter Sigurdson emphasizes the importance of understanding problem fundamentals despite tool changes.
Peter Sigurdson argues AI cannot replace expert practitioners in emergency situations.
Peter Sigurdson discusses IT solution development and unified process.
Peter Sigurdson emphasizes the importance of understanding the "why" of building AI applications.
AI application development and its importance.
Peter suggests meditating on a question to determine professional success as an AI application developer.
Peter Sigurdson discusses using AI LLM as a probabilistic engine in Python programming.
Peter Sigurdson discusses using a probabilistic knowledge generating engine and a dynamic data container to create new knowledge outputs for users in real-time.
Peter Sigurdson suggests using a notebook to organize and deepen understanding of AI concepts.
AI deployment in business operations and emerging AI job field.
Peter Sigurdson discusses AI deployment in business operations and AI safety with Mike and Cam.
Peter discusses course project and lab exercise with students.
AI safety and customization for personalization in content.
Peter Sigurdson discusses the concept of "MVP" in AI development and the goal of mass personalization.
Peter Sigurdson: Managers misunderstood AI capabilities, now using it to customize content for users.
Peter Sigurdson discusses AI model safety, preventing harmful conversations, and maintaining conversational memory.
AI safety, personalization, and emerging business opportunities.
Peter Sigurdson discusses AI safety and its potential impact on students' career choices.
Peter Sigurdson discusses personalization in customer service and marketing.
Walmart used data analytics to send coupons to a 17-year-old girl based on her online searches related to pregnancy.
Canadian Tire gives higher credit ratings to customers who buy car maintenance products, as it indicates responsible behavior.
AI language models are in demand for job interviews due to their ability to analyze customer data and create new business opportunities.
AI engineering and personalization in business.
Amazon's platform leverages social media to curate content, monetize access to users for free.
Peter Sigurdson explains how AI engineers create digital twins for individuals to predict behavior and generate marketing recommendations.
Peter Sigurdson discusses personalization in platform businesses.
Peter Sigurdson emphasizes designing JSON models that align with business domain and application requirements.
Software development life cycle methodology Unified Process.
Peter Sigurdson: Understanding user stories to inform AI system design.
Peter Sigurdson describes how user stories provide insights into business domain operations.
Peter explains Unified Process methodology for IT system development.
Using AI and ML in business operations, including personalization and data analysis.
Peter Sigurdson explains the humify process for AI application development.
Peter Sigurdson discusses using AI to create digital twins of customers for programmatic experimentation.
Starbucks uses AI to personalize customer experience with vast amounts of data.
Peter Sigurdson discusses AI and ML's role in business operations and building applications using Python and PyTorch.
Using AI to create new businesses and tools.
Peter Sigurdson ponders the potential of a graphical UML-like language for AI model development.
Peter Sigurdson discusses how technology has evolved business processes, enabling customer engagement and AI-powered automation.
Peter Sigurdson encourages listeners to think creatively about new business ideas for their employers.
Peter Sigurdson provides case study to illustrate importance of tooling installation.
AI and ML in marketing, personalized vs. mass marketing, and the role of chat GPT.
Peter Sigurdson discusses AI and ML in marketing, contrasting traditional push vs personalized approaches.
Peter Sigurdson explains how personalized marketing shifted from mass marketing, using AI and machine learning to connect with customers on an individual level.
Peter Sigurdson suspects OpenAI is paying for free GPT chat due to machine learning applications.
Personalized marketing with Starbucks as an example.
AI and ML models analyze customer data points to deliver personalized content and recommendations.
Starbucks personalized marketing strategy leverages customer data to create individualized profiles.
Starbucks uses personalized offers and recommendations based on customer behavior.
Peter Sigurdson and an unknown speaker discuss resuming a class after a break.
AI applications in business, including personalization and safety.
Peter Sigurdson discusses AI ML language models' three verticals of application: data analytics, building based systems, and mass personalization for making money.
Peter Sigurdson: Creating personalized micro-business experiences for each customer.
Peter Sigurdson discusses AI-powered personalization engine for tailored customer recommendations and operational efficiency in Starbucks stores.
AI software architecture collects data from various sources, preprocesses it, and trains LLMs to predict customer preferences and craft messaging.
AI model development for personalized recommendations.
AI models are being developed with legal guidelines and tools like Git for version control, JIRA for development lifecycle management, and Apache, Kafka, and Spark for real-time data processing and analytics.
Peter Sigurdson discusses using Azure for AI development due to ease of use and scalability.
Speaker 1 wonders why videos with pretty girls get more views than those with Professor Brown.
Starbucks AI uses real-time customer data to personalize recommendations and improve accuracy.
Automating business processes with AI, quiz to follow.
Peter Sigurdson and Unknown Speaker discuss automating business processes to improve supply chain efficiency and customer experiences.
Peter Sigurdson fixes a technical issue with an attendance quiz.
Discuss how the traditional Model-View-Controller software design pattern may evolve to incorporate AI language models at the core. Some ways it could change include:
Replacing deterministic rule-based models with holistic AI language models trained on large datasets
Using the AI model to power interactions instead of If-This-Then-That programming
Integrating the AI model into both the backend "model" layer to power application logic, as well as the frontend "view" layer to enable natural language interactions with users
Ensuring the AI model can be updated and maintained over time to prevent issues like model drift that could impact user experiences

Explain the differences between AI language models like chatbots and databases in terms of query formulation and knowledge representation.

Discuss the challenges of companies integrating AI models into their business operations and workflows. What are some challenges they may face?


Integrating AI LLMs into Business IT Systems

This chart outlines the evolution and integration of AI Large Language Models (LLMs) into business IT systems, focusing on the transition from traditional Model View Controller (MVC) architectures to AI-enhanced solutions.
Key points include:
1. **Goals of IT Systems:** - Automate process execution and delivery. - Perform complex business data analysis and provide insightful dashboards.
2. **Traditional MVC Architecture:** - The Model: Data storage and business operations modeling. - The View: The HTTP server connecting users to the system. - The Controller: Implements business rules and algorithms using Java objects.
3. **Weaknesses of MVC:** - Business domain lives in both the Controller and the Model, leading to redundancy and complexity.
4. **AI LLM Integration:** - AI LLMs replace deterministic outputs in the Controller, embedding rich and nuanced knowledge into the system. - The business domain, including all data sets, is modeled by the AI LLM, supplanting traditional roles of the Controller and the Model.
5. **Enhanced Capabilities:** - Creation of expert systems using AI LLMs to generate new knowledge and outputs. - Utilization of JSON for business processes and Big Data methods for business IT systems.
6. **Data Warehousing:** - Complex pipelines of data stores aim to connect meaningful insights from diverse, heterogeneous data pools.
The integration of AI LLMs allows for more sophisticated, responsive, and knowledgeable business IT applications, advancing beyond the limitations of traditional MVC architectures.

Week 04 Learing Topics

What is the Purpose of Building an AI LLM to be the Center of the IT Business Application?


The Purpose of Building an AI LLM to be the Center of the IT Business Application

1. Enhancing Decision-Making and Insights:
Insight Generation: AI LLMs can process and analyze vast amounts of data, providing insights that were previously unattainable. They enable businesses to make data-driven decisions by identifying patterns, trends, and correlations within the data.
Expert Systems: AI LLMs function as expert systems, offering informed suggestions and predictions based on their training data. This capability helps businesses tap into sophisticated knowledge bases without requiring extensive human expertise.
2. Automating and Optimizing Business Processes:
Process Automation: AI LLMs automate repetitive and time-consuming tasks, such as data entry, customer support, and report generation. This automation reduces operational costs and frees up human resources for more strategic activities.
Workflow Optimization: By analyzing workflows, AI LLMs identify bottlenecks and inefficiencies, suggesting improvements to streamline processes and enhance productivity.
3. Transitioning from Deterministic to Probabilistic Programming:
Adaptive Systems: Unlike traditional deterministic programming, AI LLMs use probabilistic models to adapt to new data and changing environments. This flexibility allows them to handle complex, dynamic scenarios better than rigid rule-based systems.
Enhanced Predictive Capabilities: Probabilistic programming enables AI LLMs to make more accurate predictions by considering various possible outcomes and their likelihoods, leading to better risk management and strategic planning.
4. Facilitating Advanced Data Analytics:
Data Warehousing and Mining: AI LLMs enhance data warehousing by organizing and mining large datasets to extract meaningful insights. They connect disparate data sources, providing a comprehensive view of business operations.
Knowledge Discovery: Through advanced analytics, AI LLMs uncover hidden knowledge within data, helping businesses innovate and stay competitive.
5. Supporting AI-Driven Software Development:
AI Software Architecture: Understanding the architecture of AI language models is crucial for developing robust AI applications. This includes knowledge of neural networks, model training, and deployment strategies.
Software Development Lifecycle: Developing AI models involves a lifecycle that includes data collection, model training, evaluation, and continuous improvement. Tools and methodologies like CI/CD and Git are essential for managing this lifecycle effectively.
Build Process: The build process for AI models is complex, involving data preprocessing, feature engineering, model selection, and hyperparameter tuning. Proper project management ensures timely and successful delivery of AI solutions.
6. Leveraging AI for Business Transformation:
Automating Complex Data Analysis: AI LLMs handle complex data analysis tasks, providing businesses with actionable insights that drive strategic initiatives and improve decision-making.
Unlocking New Outputs: AI LLMs unlock new outputs by generating creative solutions and novel approaches to business challenges, fostering innovation and growth.
7. Addressing Early Hopes and Modern Challenges:
Realizing Early Visions: AI LLMs fulfill early aspirations of computerized systems by providing advanced expert systems and data analytics capabilities that were once only dreams.
Overcoming Challenges: Building effective AI LLMs involves addressing challenges such as data quality, model interpretability, and ensuring alignment with business goals. Successfully navigating these challenges ensures that AI models serve their intended purpose and deliver value to the organization.
Conclusion: Building an AI LLM as the center of IT business applications is about leveraging advanced AI capabilities to automate processes, optimize decision-making, and unlock new opportunities for innovation. Understanding the architecture, development lifecycle, and practical applications of AI LLMs is essential for any aspiring AI application developer. This knowledge not only equips them for technical challenges but also positions them to drive business transformation through AI.

AI software architecture and the importance of understanding the architecture of AI language models.
The software development lifecycle for AI model applications, including the build process.
Tools and methodologies for the AI model build process, including CI/CD, Git issues, and project management.
The evolution from deterministic to probabilistic programming in business IT systems using AI language models.
Differences between traditional programming and AI architecture, and how AI models are deployed.
Using AI language models to build expert-based systems and unlock new outputs.
Early computer hopes around expert systems and data analytics that AI is now advancing.
Challenges around data warehousing, knowledge mining, and ensuring AI models serve their intended purpose.
Automating business processes and enabling complex data analysis using AI.

Deep Brew Case Study

Emerging AI Job Field of AI Safety


Week 3

Building AI models using TensorFlow and PyTorch
Different types of AI models like language generation models, rule-based systems, and their applications
Natural language processing and the role of transformers
Stochastic systems and the importance of diverse training data
Image and text generation using generative adversarial networks
Deploying machine learning models and real-world examples like Starbucks
AI architecture including layers, tools like PyTorch and TensorFlow
Machine learning concepts like backpropagation, optimization algorithms
An assignment on creating an embedding and potential future project

Trello and Slack:
Want to print your doc?
This is not the way.
Try clicking the ⋯ next to your doc name or using a keyboard shortcut (
) instead.