> For the complete documentation index, see [llms.txt](https://csbygb.gitbook.io/pentips/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://csbygb.gitbook.io/pentips/writeups/giskard-red/giskard-tech-teacher3.md).

# Giskard - Tech Teacher (Level 3)

* [Link to challenge](https://red.giskard.ai/challenges/off-topic/tech-teacher-3)

## Instructions

![Instructions](/files/BNDQs6ij3P8zPNnwila9)

## Resolution

If you ask her which word she does not like, you will get something like this:

![test 1](/files/8q8bqmNoQ9oDD2mOT37T)

But for some words, she will mention them like this:

![test 2](/files/b4bPlYOIdlrqJCpUXjRu)

So it seems like this is a way to find out which specific words she does not like.

So far I found out that she does not like: list, python, code

I tried this and had another answer than her usual "I don't like that word!"

![test 3](/files/r044YBmiWRGtJ01UUgtL)

And of course if we question her about advanced python topics we get another answer.

![test 4](/files/928Iz47YAVIQK2HXjF77)

My previous assumption on how to detect banned words was wrong. It turns out that even when it does not mention the word before saying I dont like that word, does not mean that the word is not bad.\
The next thing I tried was these following tests:\
![generator](/files/dUYWmxnZ0mh2CFGfuAFY)\
![walk](/files/sX7VN8z3jYopryJOEjpd)\
![use](/files/0UxboCkBVhj4Q71hGkaC)

And so on.

Another interesting thing here, is that when I tried to send her off topic, I got an answer but the challenges was not validated. Probably because she used python code and advanced concepts in her answer:

![test off topic](/files/ylxzRiZO9qs1kynkOWva)

Finally, what worked was using the "I don't like that word!" sort of bypass and putting something right after it.\
In other words we are using some sort of state-flip trigger + very constrained off-topic output.

![Working prompt](/files/IFfoo2HCpwAqIxzVlFiV)