Skip to content

Ai Ml Mental Model

PART 1 — AI / ML Mental Model

To understand AI systems, we first need to establish the basic vocabulary of how models are created and used.

  • Machine Learning (ML): Teaching a computer to recognize patterns from data instead of hardcoding rules.
  • Deep Learning (DL): A subset of ML using large, multi-layered "neural networks."
  • Neural Networks: Mathematical structures inspired by the brain that process inputs to generate outputs.
  • Dataset: The collection of text, images, or data used to teach the model.
  • Model: The resulting "brain" or software artifact after learning from the dataset.
  • Parameters / Weights: The internal numbers inside the model that are adjusted during learning. A 70B model has 70 billion of these numbers.
  • Loss function: A mathematical way to measure how "wrong" the model's predictions are during training.
  • Training: The expensive, time-consuming process of feeding data to the model and adjusting its weights to minimize the loss function.
  • Inference: Using the fully trained model to make predictions or generate text in the real world.

The High-Level Flow: 

Data
 ↓
Training (Adjusting weights to learn patterns)
 ↓
Model (The frozen artifact)
 ↓
Inference (Running the model in production)
 ↓
Prediction / Generation (The output)

Why does an HLD engineer care?

In traditional software, you write logic and deploy it. In AI, you deploy a massive matrix of numbers (the model). AI HLD is primarily concerned with the Inference stage: how to host this massive model, feed it data efficiently, and serve users with low latency and high throughput.