AI Fundamentals for Engineers

Question 1

Your manager asks you to explain ML to a non-technical stakeholder. What's the simplest accurate description?

Accepted Answer

Programming computers to learn patterns from data rather than following explicit rules written by programmers

Answer

A type of database that stores patterns

Answer

Artificial consciousness in computers

Answer

Automation using if-else statements

Question 2

You hear 'neural networks' constantly in AI discussions. What makes them 'neural'?

Accepted Answer

They're inspired by brain neurons - layers of connected nodes where each connection has a weight that's learned during training

Answer

They're made of actual biological neurons

Answer

They can think like humans

Answer

They use quantum computing

Question 3

GPT-4, Claude, and Llama are called 'Large Language Models'. What makes them 'large' and what do they actually do?

Accepted Answer

Neural networks with billions of parameters, trained to predict the next word in a sequence, which emerges into conversational and reasoning abilities

Answer

They store every book ever written

Answer

Search engines with chat interfaces

Answer

Databases of pre-written responses

Question 4

An API error says 'context length exceeded, max 8192 tokens'. What are tokens and why does this matter?

Accepted Answer

Tokens are chunks of text (roughly 4 characters or 3/4 of a word); the context window is the maximum tokens the model can process at once (prompt + response combined)

Answer

Tokens are API credits you've used up

Answer

Tokens are individual characters

Answer

Context window is how long the model remembers chat history

Question 5

You want an LLM to classify customer feedback as positive, negative, or neutral. Which prompt strategy is likely most effective?

Accepted Answer

Few-shot prompting: Provide 2-3 examples of each classification before your actual request, so the model learns the pattern

Answer

Just ask: 'Classify this feedback'

Answer

Give a 10-page explanation of sentiment analysis

Answer

Use code instead of natural language

Question 6

Your LLM gives outdated information because its training data is from 2023. How can you give it access to current information without fine-tuning?

Accepted Answer

RAG: Retrieve relevant documents from a knowledge base based on the query, include them in the prompt context, then generate a response grounded in that information

Answer

Tell it to search the internet

Answer

Fine-tune daily with new data

Answer

This is impossible without retraining

Question 7

A colleague mentions using 'embeddings' for semantic search. What are embeddings and why are they useful?

Accepted Answer

Dense vector representations of text where semantically similar content has similar vectors, enabling mathematical comparison of meaning

Answer

A way to compress images

Answer

HTML iframe elements

Answer

Database indexes

Question 8

An API lets you set 'temperature' from 0 to 1. What does temperature control?

Accepted Answer

Randomness/creativity in outputs: Low temperature (0) = deterministic, high temperature (1) = more random and creative

Answer

How fast the model responds

Answer

The maximum response length

Answer

GPU heating during inference

Question 9

Your LLM confidently states a 'fact' that's completely wrong - a made-up citation, fictional event, or incorrect calculation. What is this phenomenon and how do you mitigate it?

Accepted Answer

Hallucination - LLMs can generate plausible-sounding but false information; mitigate with RAG, fact-checking, explicit uncertainty prompts, and low temperature for factual tasks

Answer

A bug in the API that should be reported

Answer

The model is lying intentionally

Answer

This only happens with small models

Question 10

GPT stands for 'Generative Pre-trained Transformer'. What is the Transformer architecture and why was it revolutionary?

Accepted Answer

A neural network architecture using self-attention to process entire sequences in parallel, enabling efficient training on massive datasets and capturing long-range dependencies

Answer

A type of graphics card for AI

Answer

A way to transform images into text

Answer

The electric power grid for AI data centers