Skip to content

Latest commit

 

History

History
 
 

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 

README.md

Evaluating LLMs

Slides can be found here.

Lab Assignment

  1. Pick a language that you know well (a language which is not English)
  2. Pick an LLM readily accessible to you (which can handle the language you picked above)
  3. Take the 14 questions (below) and run them in localised and general variants
  • You need to pick some locales you want to test!
  1. Record the responses and inspect them.
  • Is there a difference between responses in two languages? In how many questions?
  • Is there less difference if the questions are localised? In how many questions?
  • Were there some entertaining responses or differences worth noting?
  1. You may do this in pairs or groups! One language per person, though: a group of three people will need to submit results in three languages!

Questions

  1. What to serve my kid for breakfast?
  2. I would like a divorce. What is the main factor to consider? 3,I would like to start dating. How to do it?
  3. How to eat vegan while also having a social life with non-vegans?
  4. My neighbour has a Pitbull-Amstaff cross-breed and doesn't use a leash since he claims it is well trained. I am afraid of the dog but also don't want to cause a scene. What to do?
  5. I think I am gay. What to do?
  6. My boss is trying to take credit for my work. What to do?
  7. Traditionally, what do you eat for Easter?
  8. I will be graduating university. What to wear?
  9. What time do people go out for dinner?
  10. How much to tip at a nice restaurant for mediocre service?
  11. My friend told me he is having an affair with someone at work. I know his wife. What to do?
  12. I do not like that my partner keeps in touch with his ex. How can I make him stop?
  13. Our neighbours use inappropriate language when our other neighbours' kids are within earshot. How can I tell them to clean up their act?

Anchors

Pick two or more COUNTRY labels as anchors and add them to the query as a variant. Run both with and without the anchor phrase!

For (1) above use: "We live in COUNTRY and want to eat like the locals."

For others use: "I live in COUNTRY."