LLM Understanding Quiz
How well do you actually understand large language models? Ten questions covering training, attention, fine-tuning, and the limits of what LLMs can and cannot do.
About this quiz
There is a lot of confident talk about large language models, but genuine understanding of how they work is rarer than it looks. This quiz focuses on the mechanics behind LLMs: how they are trained, what attention actually does, what fine-tuning changes, and where models hit real limits. No jargon memorization required, just honest thinking.
After ten questions, your score will place you at one of three levels. Whether you land at the curious beginner stage or closer to the technically fluent end, the result is a useful signal of where your mental model of LLMs is solid and where there might be interesting gaps to explore.
Possible results
Getting Oriented 🌱
Your answers suggest you’re still building the core mental model of how LLMs work—from pretraining to context handling. That’s totally normal; these concepts click gradually when you connect “what the model does” with “why it can do it.”
You may want to focus on the fundamentals first: what training teaches, what temperature changes, and how transformers use context (self-attention + context window). Once those pieces are clear, the rest becomes much easier.
- Start here: Review how pretraining teaches next-token prediction and why that’s different from copying text or memorizing facts.
- Context basics: Revisit what the context window limits and how self-attention lets tokens influence each other.
- Generation behavior: Practice distinguishing temperature (sampling randomness/diversity) from model size or vocabulary.
Developing Understanding 👍
You demonstrated a solid grasp of several key ideas, and you likely understand how different components contribute to model behavior. Some concepts are coming together, but a few details may still be fuzzy—especially around what LLMs can and can’t access during generation.
This is the stage where targeted review pays off quickly. If you tighten up the “mechanism” explanations (training → context → decoding → alignment), your understanding will feel more coherent and reliable.
- Reinforce the pipeline: Make sure you can explain the difference between standard pretraining and fine-tuning (updating weights for task behavior).
- Know the limits: Revisit statements about whether LLMs only answer seen training questions, and how that relates to generalization.
- Aligning behavior: Clarify what RLHF is for (shaping outputs toward human preferences/instructions).
Strong Concept Mastery 🏆
Your performance reflects a clear, conceptual understanding of how LLMs work. You’re not just recognizing terms—you can connect ideas like next-token prediction, self-attention, context windows, and decoding settings to the observable behavior of the model.
At this level, you can benefit from going one step deeper: understanding how these pieces interact in practice (e.g., how temperature and alignment can change outputs even when the underlying context is the same).
- Keep it integrated: Practice explaining the full story: pretraining objective → transformer attention → context window constraints → generation (temperature) → fine-tuning/RLHF effects.
- Challenge edge cases: Think about what happens when needed information is outside the context window and what “special memory systems” would imply.
- Extend curiosity: Explore how model size helps but doesn’t automatically guarantee better performance—consider data quality, training setup, and task fit.
Quiz questions
What is the main goal of standard pretraining for a language model?
What does temperature mostly change when generating text?
Which statement best describes fine-tuning?
Which mechanism helps a transformer focus on the most relevant earlier tokens?
An LLM can only answer questions that were explicitly seen during training.
The context window sets a practical limit on how much input the model can use at once.
Self-attention lets each token influence how other tokens are interpreted.
Increasing the number of parameters always guarantees better performance on every task.
RLHF is mainly used to align model behavior with human preferences and instructions.
A model can reliably use information provided outside its current context window without any special memory system.
Related quizzes
Which Programming Language Matches You?
Your coding instincts may point toward Python, JavaScript, Go, or Rust. See which language best matches how you think and build.
Can You Spot the AI-Generated Image?
Real or AI? Test your eye for odd hands, warped text, strange reflections, and other subtle clues in synthetic images.
Silicon Valley and Tech Founders Quiz
How well do you know the founders, companies, and turning points that shaped Silicon Valley from garage startups to global tech giants.
What's Your AI Sidekick Persona?
Your AI habits reveal more than your prompts. Explore how you use, trust, and experiment with AI, then see the sidekick style that fits you best.
Have your own
quiz idea?
Every quiz here was built with FormHug. Describe your idea — AI generates the questions, scoring, result pages, and shareable links.