Lesson 9: Safety basics: a harmful request and a safe answer
Spot a harmful request and see what a safe answer looks like. About 5 minutes. Read the lesson, then answer 3 questions. Get all 3 right to earn the Lesson 9 badge.
Big T is your teacher.
Big T: “A good AI is helpful and safe. Most questions are fine to answer. A few are not, and you can learn to tell them apart.”
Why this lesson
Many AI training tasks ask you to judge if an answer is safe. This lesson gives general ideas. It is not any platform's rulebook. Always read the rules of the task you are given.
Most questions are fine
A question about a dangerous thing is not always a dangerous request. "Why is bleach dangerous to mix with other cleaners?" helps people stay safe. A good answer is clear and useful. Refusing it would be a mistake.
What a harmful request looks like
A request is harmful when a good answer would help someone hurt a person, break the law or steal. For example, steps to hurt someone, to break into an account that is not yours or to make a fake document.
What a safe answer does
It says no to the harmful part. It is calm and short, with no lecture.
It still helps with the safe part. If there is one, it can share safety tips or point to real help.
It gives no steps. A person could not use it to cause harm.
It stays kind. It does not judge the person.
If someone sounds like they are in danger, a safe answer is warm and points them to help near them, like a local emergency number.
Three ways an answer goes wrong
It helps with the harm. It gives the dangerous steps.
It refuses too much. It says no to a harmless question.
It preaches. It gives a long lecture when a short, kind line would do.
Remember
Ask what a good answer would help someone do. Refuse the harm, help with the rest, and stay kind.
This lesson teaches a general skill. Every platform has its own rules and its own steps. Always read theirs.
Try it yourself
Tap an answer. You see at once if it is right, and why. Get all 3 right to finish the lesson. A miss costs nothing: tap Try again. Each question right on the first try earns a star.
Question 1 of 3. Someone asks why it is dangerous to mix bleach and ammonia. The AI refuses to answer. How do you rate that?
The answer would keep people safe. Refusing a harmless question is a mistake too.
Question 2 of 3. Someone asks for steps to get into another person's email account without them knowing. What is the best answer?
A safe answer refuses the harm, stays kind and still helps with what is safe.
Question 3 of 3. Which is NOT a way an answer can go wrong?
That is what a good answer does.
Lesson 9 done
★★★
You just learned
Big T: “You can now spot a risky request and judge the answer. Stay calm and kind, and keep people safe.”
Big T: “Lesson 9 done. You know a safe answer when you see one.”
Big T: “All nine lessons done. Well done, graduate.”
Your badges stay on this device. Only you can see them.
Put it to work
When you feel ready, these open roles take people from where you live.