23/09/2026
Understanding What an AI Model Actually Learns
=================================================
After many years of working with traditional software development, I have started exploring AI model development more deeply.
One fundamental difference immediately stands out.
In traditional software development, **we define the rules**.
For example:
`Salary = Experience × Rate + Base Amount`
The program executes the logic that we have explicitly written.
With Machine Learning, the approach is different.
We provide historical examples to a model, and during **training**, the model adjusts its internal parameters to learn relationships in the data.
A simplified flow looks like this:
*********************************************
**Historical Data → Training Dataset → Model Training → Learned Parameters → Trained Model → New Input → Prediction**
Two concepts I found particularly important are **training** and **inference**.
**Training** is the process in which model parameters are adjusted based on data and a training objective.
**Inference** happens after training, when we give new input to the trained model and ask it to produce a prediction or output.
As a software engineer, I find this distinction interesting because the development mindset changes from:
**“Write all the rules”**
to
**“Define the model/training process and allow useful patterns to be learned from data.”**
I am continuing to explore this from an engineering perspective, gradually moving from ML fundamentals toward neural networks, Transformers, LLMs, RAG, and eventually AI applications around Java, Spring Boot, and Microservices.
=================================================
