What a Trained Model Is: Weights, Not Rules
A trained model is a large set of numbers fitted to examples, not a list of rules somebody wrote out. That is why nobody can point at the line that produced a given answer, and why more examples change what it does without anyone editing logic.
What a learner can do afterwards
- Contrast a rule a person could write with a number fitted from data, on the same task
- Say what training changes inside a model and what stays fixed once it has finished
- Explain why a model cannot give the reason for an answer the way a program can be read
1 · Read
A written rule says what to do, like: if the temperature is high, say flu. A trained model holds no such sentence. It holds numbers fitted to examples, and nobody typed out any logic at all.
Picture a tiny flu model with three weights, 0.7, 0.2, and minus 0.3, plus a bias of minus 0.3. It multiplies each patient measure by its weight, adds everything with the bias, and says yes if the total is big enough. Its whole brain is those four numbers.
Training changes only the numbers. Show it a patient it gets wrong, and training nudges each weight and the bias a little toward the right answer. After thousands of patients the numbers settle, the recipe stays fixed, and once training ends even the numbers freeze.
That is why a model cannot explain itself like a program. A spam filter sorts your inbox, but there is no line to point at that produced one answer. More examples change what it does without anyone editing logic.
Rules are written, weights are fitted, and training moves numbers only.
2 · Watch
Take it off screen
Where it sits
Where this leads
Jobs that lean on this skill. Follow one to see everything it is built on.
8 questions wait behind this lesson, each with its answer explained. Every answer feeds the sky: stars light as they are learned, and dim when it is time to come back.