Free AI foundations · Lesson 3 of 12
Learn how models use data
Explain learning from examples and why a separate test matters.
Understand
In supervised learning, examples include a target answer, such as a message and its correct category. A classifier predicts a category; a regression model predicts a numeric value. A nearest-neighbor method compares a new example with similar stored examples. Linear regression fits a relationship for a number; logistic regression is commonly used to estimate the probability of a category. These are different tools, not steps every AI system follows.
Unsupervised learning looks for structure without those target labels, such as groups of similar items. Reinforcement learning adjusts behavior using rewards or feedback from actions. A neural network combines many adjustable numerical connections in layers. Training changes those connections to reduce a chosen error or improve an objective. More layers do not turn the system into a human brain or remove the need for good data.
A model can fit its training examples closely and still fail on new cases; this is overfitting. Evaluate on separate, representative examples that were not used to tune the system. Look for errors across relevant groups and unusual cases, not just one average score. A change in real-world data can make yesterday’s performance less useful. For generative tools, test complete tasks and their outputs rather than assuming a model’s reputation establishes quality.
Try it in your life or work
Personal practice
Make six fictional reminders and label them home, learning or social. Reserve two before designing a sorting rule with the other four. Test your rule on the reserved pair. Explain what the tiny sample cannot prove.
Business practice
Create six fictional inquiry snippets: routine questions, ambiguous requests and urgent concerns. Decide the expected category first. Use them as a small practice test set and record the type of each error. Do not use real customer records.
Keep: A small test set with expected answers, observed answers and failure notes.
Use fictional or non-sensitive information. Written practice does not require an AI account. The live demos are optional; other tools have their own terms and costs.
Show the answer and explanation
To see whether the method works on examples it was not tuned to.. A separate test helps reveal overfitting. It cannot guarantee performance on every future case.
This is a self-check, not a scored certification. Answers and exercise notes are not submitted to AstraCruz.