Day 19 - Foundation Models → LLMs → Deployed AI Models: How They Connect

Context:

In my previous post, I explored the difference between Foundation Models and LLMs. Today, I’m adding the third piece of the puzzle: Deployed AI Models — the ones we actually interact with in real-world applications.

What I Learned:

• Foundation Models:
  • The “brain base” of AI, trained on massive, diverse datasets (text, images, audio, etc.).
  • Examples: GPT-4, LLaMA, Gemini, Claude.
• Large Language Models (LLMs):
  • A specialized type of foundation model focused on language understanding and generation.
  • Examples: GPT-4, Claude 3, Gemini Pro, Mistral Medium.
• Deployed AI Models:
  • Real-world, user-facing versions of models, integrated into apps, APIs, and tools with performance and safety layers.
  • Examples:
    • GPT-4 → Foundation Model (LLM)
    • ChatGPT → Deployed AI Model powered by GPT-4
    • GitHub Copilot → Deployed AI Model powered by Codex

Why It Matters for QA / AI Testing:

• Testers need to understand this hierarchy to design effective test strategies.
• Foundation Models define capabilities, LLMs define language-specific behavior, and deployed models define user experience.
• Knowing the layers helps identify where issues may arise — model logic vs application integration.

My Takeaway:

Foundation Model = Broad intelligence base
LLM = Language expert built on that base
Deployed AI Model = The product you actually use in apps & tools