How to Build A GenAI Application and A Neopolitan Pizza
GenAI is like cooking, understanding the ingredients is half the battle.
Most of us will never make a true Neopolitan pizza with fresh mozzarella, San Marzano tomatoes, and a wood-fired oven, but that doesn’t stop us from marveling at the taste and wanting to know more!
That’s why today’s reading is important. Like learning about the ingredients of pizza, today’s read will help us understand more about GenAI and what goes on behind the scenes, even if we never build one ourselves.
The next time you use a GenAI app and have a good or bad experience, you’ll know which ingredients made it that way.
👉TAKEAWAYS
Components of a GenAI Application
Large Language Models (LLMs): LLMs are deep learning models trained on vast amounts of text data. They can generate coherent and contextually relevant text based on given prompts. Examples: GPT-4, Gemini, Llama 3, Claude 3, Poro and Gemma.
Knowledge Bases: A knowledge base is a centralized repository that stores information, enabling easy access and retrieval for various applications. It supports the integration and retrieval of data necessary for generative AI systems to function effectively. Examples: Pinecone, Weaviate, MySQL, MongoDB
Embedding Models: These models convert data into numerical vectors that capture semantic meaning, facilitating tasks like similarity search and clustering.
Backend Logic: The backend logic integrates various components and handles the business logic, API calls, data processing, and more. It ensures that the application functions smoothly and efficiently. Examples: Flask, Node.js.
Orchestration and Workflow Tools: These tools help in chaining together different AI models and services to create end-to-end applications. They manage the flow of data and tasks between components. Examples: Langchain, LlamaIndex.
User Interface: The front-end component that interacts with users, providing an accessible and user-friendly way to interact with the AI application. Examples: React, HTML/CSS, Javascript.
Seven different types of bias are recognized, showing how avoiding bias with AI is a very complicated business. In pizza terms, it's as foundational as getting the wood-fired oven to the right temperature. If you get it wrong, nothing else matters.
👊STRAIGHT TALK👊
I am hard-pressed to give technical commentary on this paper, but I did manage to highlight some sections of key importance to banks.
Obviously, banks are not going to ship all of their data offsite, and what I found fascinating is how elements of LLM app design have been adapted to hybrid cloud implementations where the LLM is on the cloud, but the data is on-premises.
While most of us assumed this was happening spelling out the details was interesting for me, and hopefully you too!
I recommend pages 17-21, covering hallucinations, data privacy, and bias, including their mitigation strategies, to all readers.
Sadly, hallucinations aren’t going away.
Let me know what you think about hallucinations and bias!
Join our community by subscribing. You’ll be joining an exciting journey down the rabbit hole to our shared digital future—you’ll be glad you did!
Subscribing is free, but I am honored by readers volunteering to opt for a paid subscription to recognize my high-quality writing. Thank you!
What’s in it for you if you subscribe?
Don’t get left behind: Get the expert insights on Central Bank Digital Currencies (CBDC), AI, Payments, and Financial Inclusion, that you need to stay ahead of the curve, all delivered directly to your inbox weekly;
See the future: Profit from a unique point of view directly from Asia that focuses on how the region is “leapfrogging” the West and showing us the future.
Be independent: My message doesn’t follow corporate diktats it’s a message that’s often controversial and does not conform with mainstream outlets;
Be prepared: CBDCs are no longer theoretical but coming soon, so keep up with the latest developments on the digital euro, yuan, sterling and dollar;
Understand AI: In-depth analysis of how our AI revolution impacts finance and will change how we interact with financial services.
Trust: My writing is trusted by over 50,000 executives, innovators, investors, policymakers, journalists, academics, and open-minded crypto holders daily
Sponsor Cashless and reach a targeted audience of over 50,000 fintech and CBDC aficionados who would love to know more about what you do!
