GenAI Exam Prep
Home Mock Exam
✦ EXAM TOOLKIT

Exam Tips & Common Traps

Knowing the content is half the battle — the other half is exam technique. Here is the playbook for the MCQ + coding paper, plus the misconceptions that quietly cost marks.

MCQ + coding strategy 18 examiner traps ⏱ ~12 min read

A.1 Before the exam

A.2 MCQ technique

🔑 Read every option before you commit Examiners deliberately place a "nearly right" distractor as option A and the fully-correct answer later. Reading only until the first plausible option is the #1 cause of lost MCQ marks.

A.3 Coding-question technique

🔑 Partial code earns partial marks Never leave a coding question blank. Even the correct import lines and the right model class name score marks. Write the skeleton first, then fill the logic.

A.4 Time & nerves management

B Common Traps & Misconceptions

These are the misunderstandings that examiners love to test. For each: the Trap students fall for, and the Truth.

Logistic Regression · L4
The Trap

"Logistic Regression is a regression algorithm — it has 'regression' in the name."

The Truth

It is a classification algorithm. It predicts the probability of a class and applies a threshold to decide the label.

Generative AI · L11
The Trap

"Inference happens during training."

The Truth

Inference happens after training — it is the phase where the finished model generates output from a prompt.

Tokens · L10–L11
The Trap

"One token always equals one word."

The Truth

A token is a sub-word chunk. A single word like "microtransactional" can be split into several tokens.

Model Evaluation · L5
The Trap

"A model with 99% accuracy is excellent."

The Truth

On imbalanced data a lazy model that always predicts the majority class scores 99% yet is useless. Check precision & recall.

Fine-Tuning · L14
The Trap

"Fine-tuning fixes hallucinations."

The Truth

It does not reliably — and can worsen them if the data has errors. RAG is better for factuality because it grounds answers in retrieved text.

NLP · L9
The Trap

"Always remove stop words to clean text."

The Truth

Removing "not" flips meaning ("not good" → "good"). Simple models remove stop words; RNNs/Transformers keep them.

Model Evaluation · L5
The Trap

"A higher R² always means a better model."

The Truth

Plain R² always rises when you add features — even useless random ones. Use Adjusted R² to compare models fairly.

Data Preprocessing · L2
The Trap

"Fill missing numeric values with the mean."

The Truth

The mean is dragged by outliers. For skewed data or data with outliers, the median is the safe choice.

Data Preprocessing · L2
The Trap

"Encode any text column as 0, 1, 2, 3…"

The Truth

For nominal data (Red/Green/Blue) that invents a fake ranking. Use one-hot encoding; label-encode only ordinal data.

Responsible AI · L10
The Trap

"A biased AI is biased because it is malicious or evil."

The Truth

Bias ≠ malice. The model simply repeats patterns in its historical training data — bias is a data problem, not intent.

Agentic AI · L18
The Trap

"ollama pull downloads a model and starts the chat."

The Truth

ollama pull only downloads. ollama run downloads if needed and starts the chat session.

Neural Networks · L7
The Trap

"A single perceptron can learn any pattern."

The Truth

A single perceptron is a linear classifier — it cannot solve XOR (not linearly separable). Hidden layers are required.

Managing State · L15
The Trap

"The LLM remembers the whole conversation by itself."

The Truth

LLMs are stateless. The application re-sends the full conversation history with every request to simulate memory.

Model Evaluation · L5
The Trap

"100% training accuracy means the model is excellent."

The Truth

If test accuracy is much lower, that gap is overfitting (high variance) — the model memorised noise instead of learning the pattern.

RAG vs Fine-Tuning · L14 / L16
The Trap

"To give a model your latest company data, fine-tune it."

The Truth

Use RAG — it updates instantly by editing the database and can cite sources. Fine-tuning needs slow, costly retraining.

Neural Networks · L8
The Trap

"Use sigmoid activation for the hidden layers."

The Truth

Use ReLU for hidden layers. Sigmoid in deep hidden layers causes the vanishing gradient problem. Sigmoid belongs on a binary output.

Decision Trees · L6
The Trap

"A Gini index of 1 means a perfectly pure node."

The Truth

Gini = 0 is perfectly pure. For a binary node the maximum impurity is 0.5 (a 50/50 mix) — Gini never reaches 1.

Neural Networks · L7
The Trap

"Backpropagation updates the network's weights."

The Truth

Backpropagation computes the gradients (using the chain rule). Gradient descent then uses those gradients to update the weights.

🎯 The one-line summary Read every MCQ option, compute numerical questions, never leave a coding answer blank, and watch the 18 traps above — most of them are a single confused word away from a lost mark.