AI Interview Questions for Beginners: 40 Basics With Short Answers

Entry-level interviews for AI and machine learning roles ask forty or so basic questions again and again: what supervised learning is, why models overfit, which metric to use when, what an embedding is, how a transformer differs from what came before, what prompting and retrieval do, and how you would talk about a project when you are new. Below are those questions with one-sentence answers you can expand on aloud. For deeper role-specific banks, see machine learning engineer interview questions, LLM interview questions and generative AI interview questions.

Machine learning fundamentals

1. What is machine learning? Building models that learn patterns from data to make predictions or decisions without being explicitly programmed for each case.
2. Supervised vs unsupervised learning? Supervised learns from labelled examples; unsupervised finds structure, such as clusters, in unlabelled data.
3. What is reinforcement learning? An agent learns a policy by acting in an environment and receiving rewards.
4. Classification vs regression? Classification predicts a category; regression predicts a continuous value.
5. What is a feature? An input variable the model uses; feature engineering is creating useful ones from raw data.
6. What is a loss function? A measure of how wrong the model's predictions are, which training minimises.
7. What is gradient descent? An optimisation method that adjusts parameters step by step in the direction that reduces the loss.
8. What is a learning rate? The step size in gradient descent; too high diverges, too low is slow.
9. What is regularisation? Techniques such as L1, L2 or dropout that penalise complexity to reduce overfitting.
10. What are hyperparameters? Settings chosen before training, such as learning rate or tree depth, usually tuned by validation.

Data, evaluation and overfitting

1. What is overfitting? The model learns the training data's noise and performs worse on new data.
2. What is underfitting? The model is too simple to capture the pattern and performs poorly even on training data.
3. Bias-variance trade-off? Simpler models have high bias and low variance; complex ones the reverse; you balance them.
4. Train, validation and test sets? Train to fit, validation to tune, test to estimate real performance once.
5. What is cross-validation? Rotating which fold is held out so every example is used for validation once.
6. Precision vs recall? Precision is the share of positive predictions that were right; recall is the share of actual positives found.
7. When is accuracy misleading? With imbalanced classes; predicting the majority class scores high while missing everything that matters.
8. What is a confusion matrix? A table of true and false positives and negatives from which most classification metrics are computed.

Deep learning

1. What is a neural network? Layers of weighted connections with non-linear activations that learn representations from data.
2. What is backpropagation? Computing gradients of the loss with respect to each weight by applying the chain rule backwards through the network.
3. What is an activation function? A non-linearity such as ReLU that lets networks model non-linear relationships.
4. What is a CNN good for? Grid-like data such as images, using convolutions to detect local patterns.
5. What is a transformer? An architecture built on attention that processes sequences in parallel and underlies modern language models.
6. What is attention? A mechanism that weights how much each part of the input influences each output position.

Language models and prompting

1. What is a large language model? A transformer trained on very large text corpora to predict tokens, which lets it generate and follow instructions.
2. What is a token? The unit of text a model reads and writes, often a word piece.
3. What is an embedding? A vector representation of text where similar meanings are close together.
4. What is fine-tuning? Further training a pre-trained model on task-specific data.
5. What is prompting? Shaping a model's output through instructions, examples and context in the input.
6. What is retrieval-augmented generation? Fetching relevant documents and putting them in the prompt so the model answers from them rather than from memory.
7. What is a hallucination? Confident output that is false or unsupported; mitigated by retrieval, grounding and verification.
8. What is temperature? A sampling setting; higher gives more varied output, lower more deterministic.
9. What is a context window? The maximum number of tokens the model can consider at once.
10. How do you evaluate an LLM application? With task-specific test sets, human review, and automated checks for accuracy, safety and format.

Talking about projects and practice

1. Describe a project. Problem, data, approach, result with a metric, what you would do next.
2. How did you handle messy data? Name the specific issues and the specific fixes.
3. Why did you choose that model? Baseline first, then complexity justified by validation results.
4. What went wrong? One honest failure and what it taught you.
5. How do you keep learning? Named sources and a recent thing you built.
6. How do you use AI tools in your own work? Specific tasks, how you check the output, what you do not use them for.

A structure for the project question is in AI engineer interview questions. Say every answer above aloud; definitions you can recite in writing fall apart when spoken for the first time in an interview.

Where CoPilot Interview fits

CoPilot Interview is an AI interview assistant whose free AI mock interview can be pointed at an entry-level AI role's posting and your résumé; it asks questions like these aloud, scores each spoken answer out of ten and shows a model answer, with Coding mode for the technical rounds. The free plan gives 15 minutes a day of live help in sessions of up to 15 minutes, with AI mock interviews included. Rehearse the basics until they are automatic, and in any live technical round follow your employer's rules and the platform's terms.

Disclosure: CoPilot Interview is our product, so weigh what we say about it accordingly.

Frequently asked questions

What AI interview questions are asked at entry level?

Fundamentals such as supervised versus unsupervised learning, overfitting, evaluation metrics, neural networks and transformers, embeddings, prompting and retrieval, plus questions about a project you have done.

How should a beginner answer 'describe a project'?

Problem, data, approach, result with a metric, and what you would do next, with one honest thing that went wrong.

Do I need to know transformers for a beginner AI interview?

At the level of what attention does and why transformers replaced earlier sequence models, yes; derivations are rarely asked at entry level.

How do I practise these questions?

Aloud, with a timer, ideally with a mock interviewer that scores the answer and follows up; written recall does not transfer to speaking.