Home

01 Lecture 1 Quiz: AI in Software Development

March 2026 (2006 Words, 12 Minutes)

quiz ai deep-learning llm software-development

New: Interactive Quiz Available!

Want to track your progress and get instant feedback? Try our new interactive version with progress tracking, immediate explanations, and score calculation.

Start Interactive Quiz →

Instructions

Question 1: The Deep Learning Revolution

What was the significance of AlexNet (2012) for the field of AI?

A) It was the first neural network ever built
B) It won the ImageNet competition by a large margin, proving deep learning’s superiority over traditional methods
C) It introduced the transformer architecture
D) It was the first model to generate human-like text

Show Answer

Correct Answer: B

AlexNet, created by Alex Krizhevsky, won the 2012 ImageNet Large Scale Visual Recognition Challenge with a top-5 error rate of 15.3% – nearly half the error rate of the second-place entry (26.2%). This dramatic improvement proved that deep convolutional neural networks (CNNs) could far outperform traditional computer vision approaches, kickstarting the deep learning revolution.

Question 2: Neural Network Architectures

What architectural innovation did ResNet (2015) introduce that allowed training of much deeper networks?

A) Attention mechanisms
B) Convolutional layers
C) Skip connections (residual connections)
D) Recurrent feedback loops

Show Answer

Correct Answer: C

ResNet introduced skip connections (also called residual connections), which allow information to bypass layers by adding the input of a block directly to its output. This solved the “vanishing gradient” problem and enabled training of networks with 152+ layers, dramatically improving accuracy. Before skip connections, deeper networks paradoxically performed worse than shallower ones.

Question 3: The Transformer

What key innovation did the 2017 paper “Attention Is All You Need” introduce?

A) Convolutional neural networks for image recognition
B) The transformer architecture, which processes all tokens simultaneously using self-attention
C) Reinforcement learning from human feedback (RLHF)
D) Generative adversarial networks (GANs)

Show Answer

Correct Answer: B

The 2017 paper “Attention Is All You Need” by Vaswani et al. introduced the transformer architecture, which replaced sequential processing (as in RNNs) with self-attention mechanisms that process all tokens in parallel. This was a fundamental breakthrough because it enabled massive parallelization during training and allowed models to capture long-range dependencies in text. The transformer became the foundation for virtually all modern LLMs.

Question 4: BERT vs GPT

What is the fundamental difference between BERT and GPT?

A) BERT is open-source while GPT is proprietary
B) BERT reads text bidirectionally and excels at understanding; GPT reads left-to-right and excels at generation
C) BERT is larger than GPT
D) BERT was created by OpenAI while GPT was created by Google

Show Answer

Correct Answer: B

BERT (Google, 2018) reads text in both directions simultaneously, making it excellent at understanding tasks like classification and question answering, but it cannot generate text. GPT (OpenAI, 2018) reads left-to-right only, making it a generative model that produces new text token by token. For code generation, GPT’s approach is well-suited because code is written sequentially from top to bottom.

Question 5: Scaling Laws

What was the key capability that emerged with GPT-3 (175 billion parameters) that was not present in smaller models?

A) The ability to process images
B) Few-shot learning – performing tasks from just a few examples without fine-tuning
C) Real-time voice conversation
D) Internet browsing capabilities

Show Answer

Correct Answer: B

GPT-3 demonstrated emergent abilities – capabilities that appeared as a consequence of scale rather than explicit training. Most notably, few-shot prompting allowed GPT-3 to perform tasks (translation, code writing, question answering) by showing it just a few examples, without any fine-tuning. This was a qualitative leap from GPT-2, which required fine-tuning for specific tasks.

Question 6: The Birth of AI-Assisted Coding

What was Codex, and why was it significant for software development?

A) A new programming language designed for AI
B) GPT-3 fine-tuned on billions of lines of code from GitHub, powering the first GitHub Copilot
C) A code review tool that used rule-based analysis
D) An IDE plugin that provided syntax highlighting

Show Answer

Correct Answer: B

Codex (2021) was created by fine-tuning GPT-3 on billions of lines of public code from GitHub. It became the engine behind the first version of GitHub Copilot, enabling developers to type a function signature and have AI complete the implementation. This marked the beginning of the AI-assisted coding era.

Question 7: ChatGPT’s Impact

Why was ChatGPT (November 2022) a turning point, even though it was not the most powerful model available?

A) It was the first AI model capable of writing code
B) It was free and open-source
C) It was the first AI product that felt like talking to a knowledgeable colleague, reaching 100 million users in 2 months
D) It replaced all existing search engines overnight

Show Answer

Correct Answer: C

ChatGPT was built on GPT-3.5 (not the most powerful – GPT-4 was months away) and was not the first chatbot. But it was the first AI product that felt conversational and knowledgeable enough for mainstream adoption. It reached 1 million users in 5 days and 100 million in 2 months – the fastest-growing consumer application in history. For developers, it lowered the barrier to AI-assisted coding from “install Copilot” to “open a browser tab.”

Question 8: Open vs Closed Models

Why was Meta’s release of Llama 2 (July 2023) described as a “strategic earthquake”?

A) It was the largest model ever created
B) It was the first open-weight frontier model, allowing anyone to download, fine-tune, and deploy it
C) It was faster than all competing models
D) It was the first model to support multiple languages

Show Answer

Correct Answer: B

Before Llama 2, frontier models were proprietary – you had to pay OpenAI or Google to use them. Llama 2 was released with open weights, meaning anyone could download, fine-tune, host, and modify a competitive language model for free. This spawned an enormous open-source ecosystem of specialized variants and fundamentally challenged the assumption that only well-funded companies could provide state-of-the-art AI.

Question 9: AI Development Paradigms

What is the correct chronological order of how developers have interacted with AI tools?

A) Agent mode → Chat-based → IDE-integrated → Autonomous workflows
B) IDE-integrated → Chat-based → Agent mode → Autonomous workflows
C) Chat-based → IDE-integrated → Agent mode → Autonomous workflows
D) Autonomous workflows → Agent mode → IDE-integrated → Chat-based

Show Answer

Correct Answer: C

The evolution followed four stages: (1) Chat-based (2022-2023): copy-paste code from browser windows like ChatGPT. (2) IDE-integrated (2023-2024): inline completions and chat panels in editors like VS Code with Copilot. (3) Agent mode (2024-2025): AI reads files, runs commands, and creates PRs (e.g., Claude Code, Copilot agent mode). (4) Autonomous workflows (2025+): AI handles full tasks end-to-end with MCP tool connectivity.

Question 10: Industry Impact

According to the lecture, what happened to Stack Overflow traffic after ChatGPT’s launch?

A) It increased by 50% as developers sought to verify AI answers
B) It remained unchanged
C) It dropped approximately 50% between early 2023 and mid-2024
D) It completely shut down

Show Answer

Correct Answer: C

Stack Overflow traffic dropped approximately 50% between early 2023 and mid-2024, with monthly visits falling from roughly 1.5 billion to under 800 million. However, the lecture emphasizes intellectual honesty: StackOverflow’s traffic was already declining before ChatGPT due to Google algorithm changes and competition. ChatGPT accelerated a trend rather than creating it. Also, StackOverflow’s content remains valuable as training data for LLMs.

Question 11: Developer Sentiment

According to the Stack Overflow Developer Survey 2024, what is the central tension regarding AI coding tools?

A) Developers love AI but their employers forbid its use
B) 70% say AI improves productivity, but only 43% trust the accuracy of AI output
C) AI tools are too expensive for most developers
D) Most developers refuse to use AI tools

Show Answer

Correct Answer: B

The survey found that 70% of developers say AI tools improve their productivity, but only 43% trust the accuracy of AI output. This gap is the central tension of the course: you can be more productive and less trustworthy at the same time if the AI makes you faster at producing code you do not fully understand.

What do the training costs of DeepSeek models (V3 at $5.5M, V4 at $12M) demonstrate compared to frontier models like GPT-4 ($100M+)?

A) Cheaper models are always worse
B) Algorithmic innovation can dramatically reduce training costs while maintaining frontier-level performance
C) Training costs are irrelevant to model quality
D) Only American companies can build competitive models

Show Answer

Correct Answer: B

DeepSeek-V3 matched frontier model performance for $5.5 million – roughly 1/20th of GPT-4’s cost. DeepSeek-V4 matched GPT-5 class performance for $12 million. While exact comparisons are complicated, the trend shows that algorithmic efficiency (such as Mixture of Experts architectures) can dramatically reduce costs, meaning more players can enter the field than headline training budgets would suggest.

Question 13: The LLM Provider Landscape

Which provider is most associated with the concept of “data sovereignty” for European companies?

A) OpenAI
B) Meta
C) Mistral
D) DeepSeek

Show Answer

Correct Answer: C

Mistral is a French AI company whose models can be hosted on European infrastructure, complying with GDPR without sending data to US or Chinese servers. This makes Mistral particularly important for European companies with data sovereignty requirements. While DeepSeek is open-weight, it operates from China, which raises its own data sovereignty concerns.

Question 14: Model Context Protocol (MCP)

What is the Model Context Protocol (MCP), introduced by Anthropic in November 2024?

A) A new programming language for AI
B) An open standard that lets AI models connect to external data sources and tools through a universal connector
C) A method for compressing model weights
D) A benchmark for evaluating AI models

Show Answer

Correct Answer: B

MCP is an open standard that lets AI models connect to external data sources and tools – file systems, databases, APIs, CI/CD systems, and more. The lecture compares it to “what USB was to hardware: a universal connector that replaces proprietary integrations.” MCP is a key enabler of the autonomous workflow paradigm where AI agents can directly access tools rather than relying on copy-paste workflows.

Question 15: Course Tool Choice

Why was VS Code + GitHub Copilot chosen as the primary tool for this course?

A) It is the only AI coding tool available
B) It is free for students, has the best IDE integration, is the industry standard, and supports multiple AI models
C) It is the cheapest commercial option
D) It was developed by the course instructor

Show Answer

Correct Answer: B

Five reasons: (1) Free for students through GitHub Education. (2) Best IDE integration – Copilot was built for VS Code. (3) Industry standard – VS Code is used by 70%+ of developers. (4) Increasingly model-agnostic – Copilot now supports Claude and Gemini as backends. (5) Supports all interaction modes: inline completions, chat, and agent mode. The course also explores other tools (Claude Code, direct API access, Cursor) for comparison.

Scoring Guide

Key Takeaways to Remember

  1. AlexNet (2012) was the ImageNet moment that proved deep learning works at scale
  2. Transformers (2017) replaced sequential processing with parallel self-attention, becoming the foundation of all modern LLMs
  3. BERT understands, GPT generates – bidirectional vs left-to-right processing
  4. GPT-3’s emergent abilities showed that scale produces qualitatively new capabilities like few-shot learning
  5. ChatGPT’s impact was about accessibility, not technical superiority – it made AI coding help available to everyone
  6. Open-weight models (Llama, DeepSeek) are closing the gap with proprietary models, changing the competitive landscape
  7. The productivity-trust gap (70% say productive, 43% trust accuracy) is the central tension of AI-assisted development
  8. Training cost efficiency is improving dramatically – algorithmic innovation can offset brute-force compute scaling
  9. MCP enables autonomous AI workflows by providing standardized tool connectivity
  10. The evolution from chat → IDE → agents → autonomous workflows defines how developers interact with AI
© 2026 Dominik Müller   •  Powered by Soopr   •  Theme  Moonwalk