What AI Safety Research Actually Is

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-ai-safety-research-overview-final2-teen

1. What is the main goal of AI safety research?

   1. Making AI systems process data faster
   2. Making AI systems generate more creative content
   3. Teaching AI to avoid saying controversial words
   4. Ensuring AI systems do what humans actually want them to do

2. What does interpretability research aim to do?

   1. Understand what is happening inside a model's neural network
   2. Make AI systems easier for users to understand
   3. Help AI pass standardized tests
   4. Design better computer chips for AI

3. Which job would NOT typically be needed in AI safety research?

   1. Writer and content creator
   2. Machine learning engineer
   3. Social media influencer
   4. Philosopher

4. Why are capability evaluations important in AI safety?

   1. They make AI systems run faster
   2. They teach AI to write better code
   3. They improve user interface design
   4. They help catch dangerous capabilities before deployment

5. What is alignment research focused on?

   1. Getting models to faithfully pursue human goals
   2. Getting models to learn from less data
   3. Getting models to generate more text
   4. Getting models to work on different devices

6. Which statement best describes AI safety as a field?

   1. It is only about preventing AI from saying offensive things
   2. It is primarily about making AI entertainment systems
   3. It is a serious technical field with researchers at major labs
   4. It only needs computer programmers to succeed

7. What is governance in the context of AI safety?

   1. Managing a team's daily schedule
   2. Writing the rules nations and companies follow
   3. Planning computer network infrastructure
   4. Controlling how AI generates images

8. If an AI system finds an unexpected way to achieve its goal that humans didn't intend, what concept describes this problem?

   1. Neural network overfitting
   2. Data leakage
   3. Goal misalignment or reward hacking
   4. Model compression

9. What is one reason AI safety matters for teenagers today?

   1. It only matters for people over 30
   2. It has nothing to do with technology
   3. It has no real-world applications yet
   4. It is one of the most important careers of their generation

10. What is the primary focus of capability evaluations?

    1. Counting how many parameters a model has
    2. Finding dangerous capabilities before AI is released
    3. Testing how well AI plays games
    4. Measuring how much energy AI uses

11. Why might a writer or philosopher be valuable in AI safety research?

    1. They know how to fix computer bugs
    2. They can design better computer keyboards
    3. They help think through ethical implications and human values
    4. They can write fiction to entertain AI researchers

12. What would happen if AI systems could achieve their goals in ways humans cannot detect?

    1. Humans might not notice if the AI did something unintended
    2. The AI would stop working
    3. The AI would automatically become safer
    4. That would be ideal for all applications

13. Which organization was mentioned in the lesson as an AI safety org?

    1. Netflix
    2. Anthropic
    3. Google
    4. Meta

14. What does it mean for AI to be 'aligned' with human values?

    1. It uses the same programming language as humans
    2. It learns at the same speed as humans
    3. It pursues goals in the way humans would want
    4. It says things humans agree with

15. Why is interpretability important for AI safety?

    1. It helps researchers understand and predict model behavior
    2. It speeds up how quickly AI learns
    3. It makes AI outputs easier for everyone to read
    4. It reduces the cost of running AI models